Data processing system

Abstract

The present application relates to a comprehensive decision support system, in the case of a person who is difficult to make a decision such as an elderly person, a variety of personal data such as medical history, living habits, values, family relationships, legal information, and pre-instruction are comprehensively managed, and in an emergency or when the situation changes, the situation is analyzed in real time using generative AI, the best medical and nursing options considering the values or wishes of the person, the intentions of the family members, the situation of the medical site, and the social ethics regulations are prompted, and the basis or reason thereof is explained in natural language with high transparency, thereby supporting the decision based on the trust of the person or the family members, and scenario simulation or dynamic update of the pre-instruction can be performed.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to solve is the difficulty faced by small families, such as elderly couples, in making optimal medical and nursing decisions when they or their spouses face emergencies such as sudden illness or accidents. This involves comprehensively and quickly considering various factors, including the individual's wishes or values, family members' intentions, the situation at the medical facility, legal requirements, and social ethical regulations. Specifically, it addresses the following multi-layered and practical issues: the individual or their family cannot fully understand the complex options or legal procedures related to medical and nursing care, making appropriate judgments or expressing their wishes; medical and nursing policies are decided without sufficient respect for the individual's right to self-determination or dignity; the individual's hopes cannot be accurately conveyed when family members live far away or communication is difficult; furthermore, pre-arranged instructions or expressions of wishes cannot be flexibly and dynamically updated according to changes in the situation; and in emergencies, family members or medical professionals are forced to bear mental and time burdens, making it difficult to reach optimal consensus or share information. Summary of the Invention

[0005] The means to solve the problem involved in this invention is to construct a system comprising a personal information management unit, a situation analysis unit, and a decision support unit. The personal information management unit has the following functions: acquiring various personal data from external services or communication terminals, such as the user's medical history, lifestyle habits, values, family relationships, legal information, and prior instructions; comprehensively and dynamically managing this data and updating it in real time according to changes in the situation; in the situation analysis unit, using generative AI to analyze the user's health status, action data, communication history, and information from the medical scene during an emergency to determine the urgency or necessary response; in the decision support unit, based on the information obtained from the personal information management unit and the situation analysis unit, comprehensively considering the user's values ​​or wishes, the family's intentions, the situation at the medical scene, and social ethical regulations, using a generative AI-based ethical decision-making algorithm to propose the best option, and explaining the basis or reason for the option in natural language with high transparency, supporting decisions based on the understanding and trust of the user or family. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8 This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10 It represents an emotion graph that maps multiple emotions.

[0016] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) This invention provides a decision support system that enables collaboration between a server and a terminal. The system includes a personal information management department, a situation analysis department, and a decision support department. These departments are primarily implemented on the server side, while the terminal side handles the user interface and some data input and display processing. The structure of each department and the application of generative AI are described in detail below, interwoven with specific examples. Specifically, this system employs a distributed computing architecture. The server consists of multiple computing nodes. Each computing node has a parallel processing processor to drive a large-scale language model. The terminal executes a lightweight client application. Edge computing technology is used to minimize communication latency between the terminal and the server. The personal information management department functions as a distributed database system. The situation analysis department contains a real-time stream processing engine. The decision support department executes a Transformer-based neural network. These departments are containerized as a microservice architecture. Through this containerization, the system dynamically scales resources according to the load. Compared to existing monolithic systems, this architecture significantly improves processing throughput. This system demonstrates extremely high availability in the specific use case of real-time decision support in medical settings or nursing facilities. This system does not simply automate human mental processes. This system performs semantic similarity calculations in high-dimensional vector spaces—a computation unique to computers. This processing capability surpasses human cognitive abilities in its ability to analyze complex medical data. This system thus represents a functional improvement in computer technology itself.

[0040] First, the Personal Information Management Department operates on a server to uniformly manage various personal data, including users' medical records, lifestyle habits, values, family relationships, legal information, and prior instructions. For example, users can input their medical history (hypertension, diabetes, heart disease, stroke history, osteoporosis, presence of dementia, etc.), medication history (antihypertensive drugs, antiplatelet drugs, insulin, analgesics, etc.), daily health status (body temperature, blood pressure, pulse, blood sugar, weight, diet, excretion status, etc.), exercise habits (walking frequency, muscle strength training, rehabilitation records, etc.), dietary lifestyle (salt intake, vegetable intake, frequency of alcohol consumption, etc.), smoking and drinking habits, and religious beliefs from terminals (smartphones, tablets, personal computers, wearable devices, etc.). When data is collected, such as religious reasons for refusing specific treatment, hopes for life-prolonging treatment (hope for cardiopulmonary resuscitation, availability of a ventilator, hope for a gastrostomy, hope for palliative care), family structure and contact information (spouse, children, grandchildren, siblings, legal representatives, guardians, etc.), information on legal representatives (adult guardians, discretionary guardianship agreements, executors of wills, etc.), and the contents of prior instructions (living will, prior medical instructions, declaration of death with dignity, designation of a proxy decision-maker, etc.), this data is sent to the server and stored in the personal information management department. Furthermore, it can also interface with external health management services or medical institutions' electronic medical record systems, nursing record systems, and wearable terminals (activity meters, ECG monitors, sleep trackers, fall detection sensors, etc.) via API to automatically acquire data such as blood pressure, pulse, body temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records (excretion, bathing, dietary assistance, dementia care records, etc.). Generative AI comprehensively analyzes these diverse data sets, continuously learning from changes in the user's health status, decision-making tendencies, and values. For example, when a user records improvements in their diet or changes in their exercise habits, the generative AI automatically detects these changes and dynamically updates their value or health risk profiles. Furthermore, information input by family members or medical professionals through the terminal agent is also reflected in the personal information management department, accumulating as information for agent decision-making when the individual is unable to express their wishes. Specifically, this personal information management department parses the various input text data using a natural language processing module. This module converts the text data into word embedding vectors. These word embedding vectors are, for example, represented as a 768-dimensional floating-point array. The personal information management department normalizes the time-series health data as a one-dimensional tensor. This normalization process includes interpolation of missing values ​​and removal of outliers. The personal information management department inputs this heterogeneous data into a multimodal synthesis layer. This multimodal synthesis layer uses a self-attention mechanism to learn the correlations between different data modalities. This generative AI is based on a large-scale language model with a Transformer architecture.The input data to this generative AI consists of a text vector representing the user's past medical conditions, a time-series numerical tensor representing the blood pressure fluctuation over the past week, and a combined tensor representing the embedding vectors of the pre-treatment instructions. Specific examples of this input data include a categorical variable vector representing hypertension, a normalized numerical vector representing a systolic blood pressure of 140, and a semantic vector representing the refusal of life-extending treatment. This generative AI processes these input tensors using a multi-layered feedforward neural network. The output of this generative AI is a probability distribution vector representing the user's current health risk score and the stability of their values. Specific examples of this output data include numerical data such as a heart disease risk of 0.85 and a value change probability of 0.12. This system uses this output as input for threshold determination processing in the subsequent condition analysis department. If the value change probability exceeds a predetermined threshold, the system initiates a configuration file update module. This learning process optimizes the model's weights using backpropagation of the cross-entropy error function. This significantly improves the accuracy of data management. In the use case of dynamic analysis of medical data, this system delivers a technical benefit by reducing data retrieval latency.

[0041] Next, the Situation Analysis Department, implemented on the server, analyzes user situations in real time, whether in emergencies or everyday situations. For example, when an emergency report is entered from the terminal, such as "I have collapsed and am unconscious," "I complain of chest pain," "I have fallen and cannot move," "I have difficulty breathing," "I have a persistent fever," or "My dementia symptoms are rapidly worsening," the Situation Analysis Department uses generative AI to comprehensively analyze medical records or lifestyle data stored in the Personal Information Management Department, recent health monitoring data, communication records with family members or caregivers, and information from the medical field. For example, in the case of "an 82-year-old male with a history of hypertension and heart disease who has lost consciousness," the generative AI automatically determines the urgency level and estimates the risk of myocardial infarction or stroke. In addition, it also considers the user's values, the content of pre-instruction instructions, and contact information of family members to determine what kind of response aligns with the user's wishes. For example, if the user has stated in the pre-instruction instructions that they "do not wish for CPR," the department can immediately notify the emergency team or medical institution of their intention to "not perform resuscitation measures and prioritize palliative care." Generative AI derives optimal coping strategies by referencing similar past cases, medical guidelines, and social ethical regulations. For example, in a case where someone suffers from dementia and has difficulty expressing their wishes, the AI ​​considers the individual's past communication history, the intentions of their family, and the instructions of their legal guardian to support agency decision-making. Specifically, this situation analysis unit receives unstructured text input from the terminal in real time. The unit segments the received text into a labeled sequence using a word segmenter. This labeled sequence is then converted into a 512-dimensional input tensor. This input tensor is fed into a pre-learned recurrent neural network. The recurrent neural network updates the hidden state vector, which retains contextual information about the time series. The input to this generative AI consists of an emergency notification text vector, a binary vector representing the user's underlying medical condition, and a numerical matrix representing recent vital signs. Specific examples of this input include the embedding vector of text describing chest pain, a binary vector indicating heart disease is on, and numerical data representing a heart rate of 120. This generative AI uses an attention mechanism to weight and evaluate these input data. The output of this generative AI is a continuous score from 0 to 1 representing the urgency level, and a probability distribution of the recommended medical treatment category labels. Specific examples of this output include an urgency score of 0.95, a probability of 0.88 for the myocardial infarction treatment category, and a probability of 0.05 for the stroke treatment category. The situation analysis unit determines whether this output score exceeds a pre-set urgency threshold. If the threshold is exceeded, the situation analysis unit immediately initiates the subsequent notification routing module. This notification routing module sends an alert data packet to the endpoint of the best medical institution. The system uses past similar case data as teacher data for fine-tuning. This fine-tuning uses gradient descent to fine-tune the model's parameters. This system eliminates existing judgment processes that rely on human intuition.This system objectively and rapidly classifies situations using a non-traditional method: distance calculation of feature vectors in high-dimensional space. In the use case of triage in emergency medical care, this system prevents judgment delays and effectively improves the overall processing efficiency of the system.

[0042] The decision support department is also primarily implemented on the server, but it works closely with the user interface on the client side. Based on the analysis results obtained from the situation analysis department or data from the personal information management department, the decision support department executes an ethical decision-making algorithm based on generative AI. This algorithm comprehensively considers the user's values ​​or wishes, the family's intentions, the situation at the medical facility, and social ethical regulations to generate the best option. For example, it may prompt specific options such as "Do not wish to resuscitate when cardiopulmonary arrest occurs," "Prioritize palliative care instead of life-prolonging treatment," "Remain unchanged until family arrives," "Do not wish to use a ventilator," "Refuse gastrostomy," "Prioritize pain relief," and "Refuse blood transfusion for specific religious reasons." The generative AI provides highly transparent explanations of the basis or rationale for these options in natural language. For example, the system automatically generates statements such as, "Since the individual's prior instructions indicated a desire to avoid life-prolonging treatment, and this desire has been confirmed in past conversations with family members, respecting the natural course of events and refraining from CPR is the best course of action," "Since the individual refuses blood transfusions for religious reasons, alternative treatments need to be explored," and "Since the family members are located far away, it is recommended to temporarily maintain the treatment plan and contact the family members." Specifically, this decision support unit obtains the probability distribution vector output from the situation analysis unit and the value vector extracted from the personal information management unit. This decision support unit combines these vectors and maps them to a high-dimensional feature space. This combined vector is then input into a Transformer-based generative model. This generative model has a network structure with an encoder and a decoder. The inputs to this generative model are a numerical tensor of the urgency score, a text embedding vector of the prior instructions, and a classification vector representing the family members' intentions. Specific examples of this input include a tensor of urgency score 0.95, a 512-dimensional vector representing refusal of life-prolonging treatment, and marker data representing family members' consent. This generative model generates labels autoregressively in the decoder layer. The output of this model is a text sequence of recommended options and an array of reliability scores for each option. Specific examples of this output include text data indicating the intention to postpone cardiopulmonary resuscitation (CPR) with a reliability score of 0.92, and text data indicating palliative care as an alternative. The decision support unit inputs this output text sequence into a natural language processing (NLP) filter. This NLP filter verifies the syntactic consistency of the generated text and its adherence to ethical constraints. The decision support unit then sends the verified text data to the terminal's display module. This system uses reinforcement learning to optimize the ethical decision-making algorithm. This reinforcement learning defines a reward function that follows medical guidelines. The system uses policy gradient descent to update the model's weights to maximize the reward. This system thus presents the judgment criteria of the AI, which are easily black-boxed, as explicit text output. In the use case of decision support in end-of-life care, this system delivers the technical benefits of ensuring the transparency of information prompts and improving the display accuracy of the user interface.

[0043] In addition, the Decision Support Department communicates bidirectionally with users or their families via a natural language dialogue interface on the terminal. For example, when a user inputs questions such as "What will happen if I have a stroke?", "How should I cope if my family is far away?", "What are the treatment plans if I have dementia?", "What preparations are needed if I want to face death at home?", or "What procedures are required if I want to be admitted to a nursing facility?", generative AI provides easily understandable explanations of medical and nursing options or legal procedures, allowing users or their families to input their wishes or desires in a dialogue format. The input is immediately reflected in the Personal Information Management Department, dynamically updating the pre-treatment instructions or values ​​profile. For example, when a user inputs, "Recently, my thoughts on life-prolonging treatments have changed, so I hope to actively pursue treatment in the future," the generative AI analyzes this content and automatically updates the corresponding section of the pre-treatment instructions. Specifically, this Decision Support Department is equipped with a dialogue management module. This dialogue management module continuously monitors the natural language input from users. This dialogue management module parses the input voice or text data in real time. In the case of voice input, this system converts speech recognition data into text data using a speech recognition engine employing a Hidden Markov Model or a Deep Neural Network. The input to this generative AI consists of an embedding vector of the user's question text, a context matrix representing past dialogue history, and a state vector representing the current profile status. Specific examples of this input include a text vector inquiring about stroke response, a tensor of the past three dialogues, and the current pre-instruction flag. This generative AI processes these inputs using a multi-head attention layer to extract contextual dependencies. The output of this generative AI is a sequence of response text and a control command vector instructing profile updates. Specific examples of this output include text data explaining legal procedures and a control signal activating an active treatment hope flag. The decision support department then passes this control command vector to subsequent database update transactions. These database update transactions securely rewrite the records of the personal information management department while maintaining exclusive control. The system also uses support vector machines or convolutional neural networks for dialogue intent classification. This allows the system to convert ambiguous user input into accurate system commands. This system achieves fully automated dialogue processing without human operator intervention. In the medical consultation chatbot use case, this system demonstrates the technical effectiveness of minimizing response latency while saving communication bandwidth.

[0044] Furthermore, the Decision Support Department has the capability to virtually simulate various emergency situations or medical and nursing scenarios using generative AI. For example, it can simulate scenarios such as "a person falls at home at night and becomes disoriented," "a spouse develops dementia and is unable to express their wishes," "family members are overseas and cannot be contacted immediately," "hospitalization options are limited during a new infectious disease outbreak," "medical care is needed in a shelter during a disaster," "a cluster infection occurs in a nursing facility," and "a person wishes to die at home." For each scenario, it provides natural language prompts regarding the decision-making outcome and its impact based on the individual's values ​​or desires. Through these simulations, users or their families can prepare for or reassess decisions while experiencing multiple situations in advance, and the simulation results automatically update pre-emptive instructions or personal information. For example, it can perform detailed simulations for specific scenarios such as "the treatment process and prognosis when a stroke patient desires aggressive treatment," "the home care process when life-prolonging treatment is refused," "the response at medical institutions before family members arrive," and "the process of proxy decision-making during dementia progression." Specifically, this Decision Support Department includes a probabilistic simulation engine based on the Monte Carlo method. This simulation engine uses a random number generator to generate multiple virtual environmental variables. The inputs to this generative AI are a state vector representing the initial state of the virtual scene, a tensor representing the user's health parameters, and a binary matrix representing environmental constraints. Specific examples of these inputs include an event code representing a fall at night, a numerical vector representing current mobility, and an environmental marker indicating the absence of family members. This generative AI feeds these inputs into a state transition model. This state transition model calculates the probability distribution of the next state based on a Markov decision process. The output of this generative AI is a time-series tensor of state changes over time, and a list of recommended actions at each time point. Specific examples of this output include data on the survival probability shift one hour after the onset of illness, and action codes indicating whether to use emergency transport or stay home. The decision support unit sends this output time-series tensor to the visualization module. This visualization module converts the tensor data into coordinate data for graph plotting. This system uses a generative adversarial network (GAN) to extend the simulation variations. This GAN automatically generates extreme edge-case scenarios by pitting the generator against the discriminator. This system predicts complex situational changes that cannot be covered by simple rule-based branching. In the use case of proactive care planning, this system delivers the technical advantage of providing high-accuracy future predictions while efficiently utilizing computational resources.

[0045] Furthermore, from a security and privacy perspective, the various components implemented on the server side possess functions such as encrypted communication, multi-factor authentication, and access control management, enabling the setting of appropriate information disclosure scopes for each relevant individual, including the user, their family, and medical professionals. For example, an individual's medical history or values ​​may only be accessible to the individual and their family, while information may only be disclosed to legal representatives or medical professionals within necessary limits. The terminal side features an intuitive user interface, allowing users or their families to utilize the system anytime, anywhere through voice input or output, screen display, and notification functions. For example, it provides voice guidance for visually impaired users and a simplified interface for users experiencing dementia progression, and can also be customized based on user characteristics. Specifically, the server uses a transport layer security protocol to encrypt all communication paths. The system executes a hybrid encryption algorithm combining public-key cryptography and public-key cryptography. The system drives a multi-factor authentication module that combines biometric data and one-time passwords in user authentication. The system expands role-based access control lists into memory. The system performs access permission verification processing for each data request in microseconds. This terminal features a display control unit that renders a graphical user interface. This display control unit dynamically adjusts contrast and font size based on the user's visual characteristics. The system uses a speech synthesis markup language to convert text data into natural speech waveform data. The system then transmits the converted speech waveform data to the terminal's audio output buffer. As a simplified interface for dementia patients, this system performs layout calculations to limit the number of operation buttons on the screen and increase the pixel size of icons. This system offloads these security and interface processing tasks to a hardware accelerator. This offloading significantly reduces the computational load on the main processor. In the use case of secure medical data sharing, this system effectively reduces the risk of data leakage while maintaining system responsiveness.

[0046] Thus, in this embodiment of the invention, generative AI is incorporated as an essential element to achieve comprehensive and flexible decision support, including integrated management of personal information, real-time analysis of conditions, ethical decision support, explanations and dialogues via natural language, scenario simulation, and dynamic updates of pre-instructions. This provides a system that, while reducing the burden on the elderly, their families, and medical professionals, maximizes respect for the dignity and self-determination of the individual. For example, it can address various scenarios such as emergency teams instantly confirming the individual's wishes on-site when their condition suddenly changes; family members participating in decision-making remotely via terminals; medical professionals determining treatment strategies while considering the individual's values ​​or past intentions; and nursing facility staff sharing changes in the individual's condition with family members by inputting daily care records into the system. Furthermore, generative AI continuously learns from user input and changes in conditions, enabling the individualized and optimized decision support to evolve. Specifically, this system integrates the generative AI inference pipeline as the core process. This system combines an asynchronous message queue with a distributed database, a stream processing engine, and a neural network inference engine. This asynchronous combination eliminates processing bottlenecks between modules. This system generates response data packets in milliseconds for queries from emergency medical service terminals. It establishes a bidirectional, real-time communication session using WebSocket with remote family terminals. The system sends JSON payloads to healthcare professionals' terminals to render past intention data as time-series graphs. It indexes daily care records sent from nursing facility staff terminals in real-time using stream processing rather than batch processing. This generative AI incorporates an online learning algorithm. This online learning algorithm iteratively updates the model's weights using new input data points. The system incorporates a forgetting factor to decay the weights of older data, allowing the model to adapt to changes in the current situation. This system thus elevates a static database retrieval system to a dynamic and self-evolving inference system. In the comprehensive healthcare linkage use case, this system delivers the technical advantage of maximizing processing throughput while maintaining overall system data consistency.

[0047] (System Composition) The system involved in this implementation includes a personal information management department, a status analysis department, a decision support department, a simulation department, a user interface department, a data linkage department, and a security management department. The personal information management department uniformly manages various personal data, such as users' medical history, lifestyle habits, values, family relationships, legal information, and pre-treatment instructions. For example, when a user inputs past medical conditions such as hypertension, diabetes, heart disease, or stroke from the terminal, this information is stored in the database on the server. It also records information on medications such as antihypertensive drugs, antiplatelet drugs, insulin, and analgesics as part of the user's medication history. It collects data on daily health status, including body temperature, blood pressure, pulse, blood sugar levels, weight, diet, and excretion. It saves information on exercise habits, such as walking frequency, muscle strength training, and rehabilitation records. The system also comprehensively manages dietary habits, smoking and drinking habits, religious beliefs, hopes for life-prolonging treatments, family structure and contact information, legal representative information, and the content of pre-treatment instructions. Furthermore, the system integrates with external health management services or medical institutions' electronic medical record systems, nursing record systems, and wearable terminals via API to automatically acquire data such as blood pressure, pulse, temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records. It also reflects information input by family members or medical professionals on behalf of the individual, accumulating information that can be used as a basis for decision-making when the individual cannot express their wishes. Specifically, this system deploys these components as independent containerized microservices on cloud infrastructure. The Personal Information Management Department employs a multi-language persistent architecture combining relational and NoSQL databases. The Personal Information Management Department stores past medical records or medication histories as structured data in tabular form. The Personal Information Management Department stores unstructured data such as text of pre-instructions or images of daily dietary content in a document-oriented database. The Data Linkage Department exposes a RESTful application programming interface. The Data Linkage Department receives JSON-formatted payloads from external systems. The Data Linkage Department executes a parser to extract necessary data fields from the payload. The Data Linkage Department converts the extracted data into an internal standardized data format. This system writes high-frequency time-series sensor data sent from wearable terminals into a time-series database. The system employs data compression algorithms during this write process to conserve storage capacity. The system records the proxy input information in a blockchain technology or tamper-proof log system. This recording process cryptographically guarantees the data's provenance and integrity. In the use case of comprehensive medical data management, this system effectively ensures data interoperability between heterogeneous systems and reduces input errors through automated data collection.

[0048] The Situation Analysis Department analyzes user situations in real time, whether in emergencies or everyday situations. For example, when an emergency report is entered from the terminal, such as a user collapsing and unconscious, complaining of chest pain, being unable to move after a fall, experiencing difficulty breathing, persistent fever, or a rapid deterioration of dementia symptoms, generative AI is used to comprehensively analyze data accumulated in the Personal Information Management Department, including medical records or lifestyle habits, recent health monitoring data, communication history with family or caregivers, and information from the medical scene. In cases of elderly men with a history of hypertension and heart disease losing consciousness, the department automatically assesses the urgency level and estimates the risk of myocardial infarction or stroke. If the user has indicated in a pre-incident instruction that they do not wish for CPR, the department immediately notifies the emergency team or medical institution of their intention to prioritize palliative care over resuscitation. In cases of dementia where the user is unable to express their wishes, the department considers past statements from the user, the intentions of family members, and instructions from legal guardians to support agency decision-making. Specifically, this Situation Analysis Department adopts an event-driven architecture. The department subscribes to emergency report events via a message broker. Upon receiving an event, the department immediately initiates the inference process. The input to this generative AI consists of a 512-dimensional tensor, vectorized from the emergency notification text, and a multi-dimensional feature matrix of medical records obtained from the Personal Information Management Department. Specific examples of this input include an embedding vector representing natural language indicating loss of consciousness and a binary array containing hypertension markers. This generative AI processes these inputs using a multilayer perceptron and a self-attention mechanism. The output of this generative AI is a floating-point score representing the urgency level and a probability distribution vector of the predicted disease category. Specific examples of this output include an urgency score of 0.99, a myocardial infarction probability of 0.85, and a stroke probability of 0.10. The situation analysis unit applies a Softmax function to the output probability distribution vector to determine the disease category with the highest probability. The situation analysis unit compares this determination result with the marker data in the pre-instruction document using a logical AND operation. Based on this operation result, the situation analysis unit generates a control flag indicating whether resuscitation measures are required. The situation analysis unit sends a notification payload containing this control flag to an external medical institution system. This system thus completely eliminates judgment delays caused by human intervention. In the use case of triage during emergency transport, this system achieves the technical effect of minimizing data processing latency and maximizing system response speed.

[0049] Based on the analysis results obtained from the Situation Analysis Department or the data from the Personal Information Management Department, the Decision Support Department executes an ethical decision-making algorithm based on generative AI. This algorithm comprehensively considers the user's values ​​or wishes, the family's intentions, the situation at the medical facility, and social ethical regulations to generate the optimal options. For example, it may prompt the user to choose not to resuscitate in the event of cardiopulmonary arrest, to prioritize palliative care instead of life-prolonging treatment, to maintain the treatment plan until the family arrives, to choose not to use a ventilator, to refuse a gastrostomy, to prioritize pain relief, or to refuse blood transfusions for specific religious reasons. For each of these options, the rationale or justification is explained transparently in natural language. The system automatically generates explanations based on the user's prior instructions indicating a desire not to prolong life-prolonging treatment, which is confirmed by past conversations with the family; therefore, respecting the natural process and refusing CPR is the best course of action; alternative treatments should be explored due to the user's religious refusal of blood transfusions; and decisions to temporarily maintain the treatment plan and contact the family are recommended, based on the user's location. Specifically, this Decision Support Department employs a hybrid approach, combining a rule-based inference engine with a neural network-based generative model. This decision support unit first filters out absolute contraindications using a rule-based engine. The unit then initiates a generative model. The inputs to this generative AI are a disease prediction vector from the condition analysis unit, an embedding tensor representing the user's values, and a text vector of medical guidelines. Specific examples of these inputs include a status code representing cardiopulmonary arrest, a semantic vector representing refusal of life-prolonging treatment, and a sequence of labeled resuscitation guidelines. This generative AI uses a Transformer decoder architecture to process these inputs. The output of this generative AI is structured JSON data representing multiple treatment options and a sequence of natural language text explaining the rationale for each option. Specific examples of this output include option IDs, action codes representing a transition to palliative care, and text strings describing the reasons for such transitions. The decision support unit parses this structured output data and converts it into display objects that the interface unit can interpret. During the learning process of the generative model, the system performs supervised learning using specialist physicians' judgment history as teacher data. The system minimizes a composite objective function combining cross-entropy error and a reward term from reinforcement learning as its loss function. This system automatically generates ethically sound and logically sound explanatory text. In the complex use case of medical policy decisions, this system effectively ensures the accuracy of information processing and fulfills its responsibility to provide explanations to users.

[0050] The simulation department uses generative AI to virtually simulate various emergency situations or medical and nursing scenarios. These scenarios include: a person falling and becoming disoriented at home at night; a spouse suffering from dementia and unable to express their wishes; family members stranded overseas and unable to be contacted immediately; limited hospital access during a novel infectious disease outbreak; needing medical care in a shelter during a disaster; a cluster infection in a nursing facility; and the desire to die at home. For each scenario, natural language prompts are provided to suggest the decision-making outcomes and their impact based on the individual's values ​​or desires. This allows users or their families to experience multiple situations in advance, preparing for or reassessing decisions, and automatically updating pre-emptive instructions or personal information based on the simulation results. Specifically, the simulation department executes a discrete event simulation algorithm. It calculates state transitions while virtually advancing the system clock. The input to this generative AI is a parameter matrix defining the initial state and a noise vector representing the uncertainty of the environment. Specific examples of inputs could include a numerical tensor representing physical injury from a nighttime fall, a random variable representing the arrival delay of family members, and categorical data representing the prevalence of infectious diseases. This generative AI feeds these inputs into a recurrent neural network to predict state changes at each time step. The output of this generative AI is a time-series array of the user's health status score at each time step, and a probability tree of possible events. Specific examples of outputs could include a data structure representing a probability of 0.7 for decreased consciousness after one hour and a probability of 0.9 for emergency transport after three hours. The simulation unit uses a search algorithm to parse the output probability tree and extract the response route with the highest expected value. This extracted route information is then passed to the natural language generation module to generate scenario description text. The system temporarily stores the simulation results in a cache memory for high-speed response to subsequent user requests. This system thus efficiently processes future predictions that require significant computational resources. In the use case of pre-medical planning, this system delivers the technical benefits of high-speed computation of multi-dimensional scenario branches and improved prediction accuracy.

[0051] The user interface (UI) is installed on the terminal side, providing a natural language dialogue interface for two-way communication with the user or their family. When the user inputs questions such as "What will happen if I suffer a stroke?", "How should I cope if my family is far away?", "What are the treatment plans if I develop dementia?", "What preparations do I need if I want to face death at home?", or "What procedures are required if I want to be admitted to a nursing facility?", the generative AI provides easily understandable explanations of medical and nursing options or legal procedures. This allows the user or their family to input their wishes or desires in a conversational format. The input is immediately reflected in the personal information management department, dynamically updating the pre-instruction document or values ​​profile. Specifically, this UI operates as a front-end web application or native application. This UI uses asynchronous communication technology to maintain a constant connection with the decision support department on the server side. This UI captures the user's keystrokes or voice input through event listeners. The input to this generative AI is a tagged array of the captured user's question text and a context vector associated with the session ID. Specific examples of this input could include the embedding vector of text inquiring about treatment guidelines for dementia onset, and a state tensor representing the preceding dialogue content. This generative AI processes these inputs within the inference pipeline of a large language model. The output of this generative AI is a string of response text and control commands for changing the display state of the user interface. Specific examples of this output could include text data explaining legal procedures and rendering instruction code for displaying an "agree" button on the screen. This interface receives the control commands and dynamically rewrites the document object model. This interface packages the user's new hopes or intentions into a JSON payload and sends it to the server. This system employs a design that minimizes the rendering load on the client side throughout this series of processes. In the use case of automating medical consultations, this system achieves the technical effect of improving the user experience while reducing the communication overhead between the server and the terminal.

[0052] The Data Linkage Department integrates with external health management services, electronic medical record systems, nursing record systems, and wearable terminals via APIs to automatically acquire data such as blood pressure, pulse, temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records. This data is then provided to the Personal Information Management Department or the Status Analysis Department to achieve real-time and comprehensive information collection. Specifically, this Data Linkage Department functions as a gateway module supporting multiple communication protocols. It periodically sends polling requests to the RESTful endpoints of external systems via HTTPS. For data reception from wearable terminals, it employs a lightweight publish-subscribe model such as MQTT. The received raw data is temporarily stored in a buffer area on memory. The department then performs data cleaning on this raw data, including missing value completion, outlier filtering, and timestamp synchronization. Finally, the cleaned data is mapped to structured data based on an internal common schema. This structured data is then published to a message queue. The Personal Information Management Department and the Status Analysis Department asynchronously consume data from this message queue. The system rigorously performs data pattern validation within the Data Linkage Department. Data violating the pattern is isolated as an error log. This prevents illegal data from entering the system. In the use case of data integration between heterogeneous systems, this system automates the data collection process and homogenizes data quality, resulting in improved overall system processing stability.

[0053] The Security Management Department is responsible for the overall security and privacy protection of the system. It possesses functions such as encrypted communication, multi-factor authentication, access control management, and audit log management. It sets appropriate information disclosure scopes based on the user, their family, medical professionals, and other relevant personnel. This ensures that an individual's medical history or values ​​are only accessible to the individual and their family, and that information is only disclosed to legal representatives or medical professionals within necessary limits, thus achieving strict information management. Specifically, this Security Management Department is designed based on the principles of a zero-trust architecture. It performs rigorous authentication and authorization processes for all access requests from both internal and external sources. It uses the AES 256-bit encryption algorithm for encrypting communication data. In multi-factor authentication, in addition to password hash verification, it also executes a time-based one-time cipher algorithm. For access control, it installs an attribute-based access control model. It inputs the user's role, the IP address of the access source, and the confidentiality level of the requested data into an evaluation function. The evaluation function outputs a Boolean value indicating whether access is allowed or denied. Based on this evaluation result, the Security Management Department controls the execution of database queries. This security management department records all access history and system events in an immutable audit log database. This audit log database uses hash chaining technology to detect log tampering. The system periodically and automatically executes a vulnerability scanning module to monitor the application status of security patches. In the use case of protecting highly confidential medical information, this system effectively minimizes the risk of data leakage due to cyberattacks or internal breaches, maintaining system availability and integrity.

[0054] As specific examples of prompts read by the generative AI during the implementation of this invention, examples include: "An 82-year-old male with a history of hypertension and heart disease, who stated in his pre-treatment instructions that he does not wish for life-prolonging treatment, is now unconscious. What is the best course of action during emergency transport? Considering his values ​​and the wishes of his family, explain the options and their reasons in natural language." "A person suffering from dementia and unable to express his wishes, based on his past speaking history, the wishes of his family, and the instructions of his legal guardian, simulate the process of agency decision-making and explain the best response strategy and its basis." "A person wishes to die at home. Given that his family is far away, what kind of medical and nursing care system is needed? Specifically explain the necessary procedures or preparations." "In the case of a person refusing blood transfusions for religious reasons, what alternative treatments can be considered in an emergency? Generate a prompt for the medical staff and family." Specifically, this system has a prompt engineering module. This prompt engineering module dynamically embeds structured data extracted from the database into a natural language prompt template. The input to this generative AI is the text data of this dynamically generated prompt. As a specific example of this input, a text string embedding variables such as age, past medical conditions, prior instructions, and current vital signs can be considered. This system converts this text string into a numerical array of tag IDs using a tokenizer. This array of tag IDs is then provided to the input layer of a large language model. This generative AI uses a self-attention mechanism to calculate the dependencies between words within the prompt. The output of this generative AI is a probability distribution of the tagged response text. As a specific example of this output, a text sequence illustrating the optimal response strategy and its generation probability score can be considered. This system uses a beam search algorithm to explore the text sequence with the highest probability to determine the final output text. This system appends an indicator tag to the end of the prompt, enforcing a specific output format. This system thereby controls the output of the generative AI, ensuring it is reliably interpretable in subsequent parsing processing as either JSON format or a specific tag format. In the use case of automatic prompt generation and optimization, this system delivers the technical benefits of suppressing the illusion of generative AI and dramatically improving the accuracy and consistency of the output.

[0055] (Implementation steps) Step 1: Collection and Registration of Personal Information. Users or their family members use the terminal to input their medical history (history of hypertension, diabetes, heart disease, stroke, osteoporosis, presence of dementia, etc.), medication history (antihypertensive drugs, antiplatelet drugs, insulin, analgesics, etc.), daily health status (body temperature, blood pressure, pulse, blood sugar level, weight, diet, excretion status, etc.), exercise habits (walking frequency, muscle strength training, rehabilitation records, etc.), diet and lifestyle (salt intake, vegetable intake, frequency of alcohol consumption, etc.), smoking and drinking habits, religious beliefs, hopes for life-prolonging treatment (hope for cardiopulmonary resuscitation, availability of a ventilator, hope for gastrostomy, hope for palliative care, etc.), family structure and contact information, information of legal representatives, and the contents of prior instructions (will, prior medical instructions, declaration of death with dignity, designation of a proxy decision-maker, etc.). This data is stored in the personal information management department of the server. Furthermore, the system integrates with external health management services or medical institutions' electronic medical record systems, nursing record systems, and wearable terminals via API to automatically acquire data such as blood pressure, pulse, body temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records. Information entered by family members or medical professionals on behalf of the user is also reflected. Specifically, in this step, the terminal renders a web form for data input. The terminal validates the user-input data in real time via a client-side script. The terminal serializes the validated data into a JSON object. The terminal sends this JSON object as a secure HTTPS POST request to the server. The server's receiving module parses the request and maps the data to the database schema of the personal information management department. The server inputs the text data into a natural language processing engine, automatically performing semantic markup. In API integration with external systems, the server executes an authorization process using the OAuth protocol. The server obtains an access token and acquires time-series health data from external data endpoints through batch or stream processing. The server inputs the acquired heterogeneous data into an ETL pipeline. This ETL pipeline performs data extraction, transformation, and loading, storing the data in memory in a normalized vector format. The system executes this series of data collection processes asynchronously as background tasks. In the use case of building the initial dataset, this system effectively reduces the user's input load while efficiently accumulating sufficient high-quality data necessary for the machine learning model's learning.

[0056] Step 2: Detection of Situational Changes and Emergency Notification. When an emergency notification is input from the terminal indicating that the user has collapsed and is unconscious, complains of chest pain, is unable to move after a fall, has difficulty breathing, has a persistent fever, or is experiencing a rapid deterioration in dementia symptoms, or when anomalies or abnormal behavior are automatically detected by the wearable terminal or nursing record system, the server's situation analysis unit receives this information in real time and immediately determines the urgency or necessity of response. Specifically, in this step, the terminal or wearable device continuously samples high-frequency time-series data from its built-in accelerometer or gyroscope sensors. The device executes a lightweight anomaly detection algorithm in an edge computing environment. The device calculates the moving average or variance of the sensor data and monitors deviations from pre-set thresholds. Upon detecting an anomaly, the device immediately generates an interrupt and generates an emergency notification data packet. The server's situation analysis unit receives this emergency notification data packet with low latency via a dedicated UDP port or WebSocket connection. The situation analysis unit parses the header information of the received data packet and determines the priority of the event. Based on this priority, the situation analysis unit dynamically adjusts the resource allocation of the inference engine. For high-urgency events, this situation analysis department prioritizes inference tasks in the GPU cluster's highest-priority queue. In cases of emergency notifications via text input, the system uses a natural language processing model to calculate an urgency score for the text. If this score exceeds a threshold, the system triggers a software trigger equivalent to a hardware interrupt. In the real-time anomaly detection use case, this system reduces the latency from data transmission to parsing to milliseconds, maximizing the system's instantaneous responsiveness to catastrophic changes in circumstances.

[0057] Step 3: Generative AI-based situation analysis and decision support prompt generation The Situation Analysis Department uses generative AI to comprehensively analyze medical records or lifestyle data stored in the Personal Information Management Department, recent health monitoring data, communication records with family members or caregivers, and information from the medical field. Specific examples of prompts that the generative AI can read include: "An 82-year-old male with a history of hypertension and heart disease, who stated in his pre-event instructions that he does not wish for life-prolonging treatment, is now unconscious. What is the best course of action during emergency transport? Considering his values ​​and the wishes of his family, explain the options and their reasons in natural language." Another example is: "The individual suffers from dementia and has difficulty expressing his wishes. Based on past speaking history, the wishes of his family, and the instructions of his legal guardian, simulate the process of agency decision-making and explain the best response strategy and its basis." Specifically, in this step, the Situation Analysis Department queries the target user's latest profile data from the database. This profile data is combined with the received emergency event data to construct a multi-dimensional feature vector. This feature vector is then input into the prompt generation module. This prompt generation module drives the template engine to insert the elements of the feature vector into placeholders in natural language. The input to this generative AI is the text data of the dynamically synthesized prompt. A specific example of this input could be a structured text string describing age, pre-existing conditions, the content of a pre-instruction document, and the current emergency situation. This system converts this text string into a sequence of token IDs using a word segmenter. This sequence is then fed into the encoder layer of the Transformer model. The encoder layer uses a multi-head attention mechanism to compute contextual dependencies between tokens, generating a high-dimensional hidden state vector. This system then prepares to pass this hidden state vector to the decoder layer. During the prompt generation process, the system performs preprocessing to filter out unwanted information or data that becomes noise. In the context optimization use case, this system automatically constructs the optimal input representation to maximize the inference accuracy of the generative AI, effectively preventing the waste of computational resources.

[0058] Step 4: Hints and explanations of the best options based on generative AI Generative AI comprehensively considers the individual's values ​​or wishes, the family's intentions, the situation at the medical scene, and social ethical regulations, prompting specific options such as: not wanting resuscitation when cardiopulmonary arrest occurs, prioritizing palliative care instead of life-prolonging treatment, maintaining the treatment plan until the family arrives, not wanting a ventilator, refusing a gastrostomy, prioritizing pain relief, and refusing blood transfusions for specific religious reasons. It also provides highly transparent explanations of the basis or reasons for these options in natural language. Specific examples of prompts for the generative AI to read include: "Generating the following explanation: Since the individual's prior instructions state that they do not want life-prolonging treatment, and this same intention has been confirmed in past conversations with the family, respecting the natural process without CPR is the best course of action," and "Generating a text explaining to the medical scene and family that because the individual refuses blood transfusions for religious reasons, alternative treatments need to be explored." Specifically, in this step, the decoder layer of this generative AI autoregressively generates output tokens based on the hidden state vector received from the encoder. The output of this generative AI is a text sequence representing the optimal response option, along with a text sequence explaining the logical basis for that option. Specific examples of this output could include a conclusion text suggesting the postponement of CPR, and a text explaining the rationale based on prior instructions and past conversations. The system applies a post-processing module to the generated text sequence. This module uses regular expressions or a rule-based checker to verify that the generated text does not contain medical contraindications or ethical violations. The system packages the verified text data into a JSON-formatted response payload. This payload is then sent to the requesting terminal. The terminal's display control parses the payload and renders the text onto a designated component on the graphical user interface. When the user provides feedback on the suggested options, the system saves this input data as new learning data in a log database. The system uses this log data to continuously fine-tune the model. In the final output use case of decision-making, this system rapidly and safely presents highly reasoned results in a human-understandable form, resulting in improved overall system usefulness and reliability.

[0059] Step 5: Dialogue and feedback of wishes with users and their families Through the interface, users or their families can confirm the options or explanations provided by the generative AI on the terminal and can add additional questions or express their wishes using natural language input. For example, for questions such as "What will happen if I have a stroke?", "How should I cope if my family is far away?", and "What are the treatment options if I have dementia?", the generative AI provides easily understandable explanations of medical and nursing options or legal procedures, allowing users or their families to input their wishes or hopes in a conversational manner. The input is immediately reflected in the personal information management department, dynamically updating the pre-instruction manual or values ​​profile. Specifically, this interface executes a natural language processing model. This natural language processing model is a large-scale language model based on the Transformer architecture. This interface receives input from the user. This input is a text string in natural language. For example, this input is the string "What will happen if I have a stroke?". This interface converts this text string into a word embedding vector. This word embedding vector is, for example, a 768-dimensional numerical array. This large-scale language model receives this word embedding vector as input. This large-scale language model uses a self-attention mechanism to parse the context. This large-scale language model outputs a probability distribution of the response text. This large-scale language model generates the optimal response text string based on this probability distribution. For example, the output is the string "In the case of stroke, rehabilitation training should be carried out after acute treatment." This interface outputs this response text string to the display module. Furthermore, this interface converts the input intent into structured data. This structured data is a hierarchical data structure composed of key-value pairs. This interface sends this structured data to the personal information management department. The personal information management department uses this structured data to automatically update the database. Thus, this system accurately extracts the intent of the dialogue. This system automates database updates. As a result, this system improves the accuracy of data input. Furthermore, this system significantly improves processing speed. This technology is applied to the patient intent confirmation process in the medical field.

[0060] Step 6: Scene simulation and pre-event preparation based on generative AI The simulation department uses generative AI to virtually simulate various emergency situations or medical and nursing scenarios. These scenarios include situations such as a person falling at home at night and becoming disoriented, a spouse suffering from dementia and unable to express their wishes, family members being overseas and unable to be contacted immediately, hospitalization restrictions during a new infectious disease outbreak, needing medical care in a shelter during a disaster, a cluster infection in a nursing facility, and the person wishing to die at home. For each scenario, natural language prompts are provided to guide the user's decision-making process and its impact based on their values ​​or desires. This allows users or their families to experience multiple situations in advance and prepare or reassess their decisions. Specific examples of prompts that the generative AI can read include: "If I wish to die at home and my family is far away, what kind of medical and nursing system do I need? Please explain the necessary procedures or preparations," and "Simulate the treatment process and prognosis of a stroke patient who desires aggressive treatment, and generate a document to explain to their family." Specifically, this simulation department executes a conditional generation model. This conditional generation model is constructed using a deep neural network. This simulation unit receives the patient's current state vector as input. This state vector is a multi-dimensional numerical array containing indicators such as age or medical history. Additionally, this simulation unit receives an environmental condition vector as input. This environmental condition vector is an array numerically representing the distance to family members or the occurrence of a disaster. For example, this input is a numerical array where the patient's own end-of-life care hope indicator is 1 and the distant family member indicator is 1. This conditional generation model maps these input vectors to a high-dimensional latent space. This conditional generation model outputs a time-series state transition probability matrix. Furthermore, this conditional generation model outputs a score array of recommended actions at each time step. This simulation unit sends this score array to the natural language processing module. This natural language processing module converts this score array into a human-understandable text string. This simulation unit extracts scenarios where the score array exceeds a predetermined threshold. This simulation unit sends the extracted scenarios to the interface unit. This interface unit renders the scenarios onto the screen. Thus, this system rapidly calculates scenario branches in the high-dimensional conditional space. This system provides comprehensive simulation results with low latency. As a result, this system achieves various risk predictions that surpass the limits of human cognition. This technology is used to support the development of pre-care plans in nursing facilities or home healthcare.

[0061] Step 7: Data Linkage and Security Management The Data Linkage Department integrates with external health management services, electronic medical record systems of medical institutions, nursing record systems, wearable terminals, etc., via API to automatically acquire data such as blood pressure, pulse, body temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records, and provides this data to the Personal Information Management Department or the Status Analysis Department. The Security Management Department is responsible for the overall security and privacy protection of the system, possessing functions such as encrypted communication, multi-factor authentication, access control management, and audit log management. Appropriate information disclosure scopes are set according to the user or their family, medical professionals, and other relevant personnel to achieve strict information management. Specifically, this Data Linkage Department uses an asynchronous communication protocol to connect with external systems. This Data Linkage Department acquires time-series data from external systems. This time-series data includes raw data such as blood pressure or steps. This Data Linkage Department performs interpolation algorithms for missing values ​​on this raw data. This interpolation algorithm includes linear interpolation or spline interpolation. This Data Linkage Department performs filtering processing to remove outliers. This Data Linkage Department converts the preprocessed data into a fixed-length tensor. This tensor is a two-dimensional numerical array consisting of time steps and feature values. The data linkage unit sends this tensor to the status analysis unit. Meanwhile, the security management unit executes advanced encryption algorithms. These algorithms are, for example, symmetric key encryption with a 256-bit key length. The security management unit performs token verification processing in multi-factor authentication. The security management unit generates a bitmask representing user access permissions. The security management unit uses this bitmask to dynamically control data access. Thus, this system converts heterogeneous data formats into a unified tensor representation. This system improves the inference accuracy of the subsequent artificial intelligence model. Furthermore, this system prevents data tampering through encryption and access control. As a result, this system mathematically guarantees the overall integrity and confidentiality of the system. This technology is applied to cloud platforms that process highly confidential medical data.

[0062] (specific use case) For example, imagine a scenario in a household consisting solely of an elderly couple where the husband suddenly loses consciousness at home, and the spouse uses a terminal to issue an emergency notification. In this situation, the Personal Information Management Department registers the husband's pre-existing conditions as hypertension, heart disease, and mild dementia; detailed records of his daily health status include blood pressure or pulse, medication history, exercise habits, diet, smoking history, alcohol consumption history, family composition, and legal guardian information; and detailed pre-event instructions include "Do not wish for life-prolonging treatment," "Do not require CPR," "Refuse ventilator," "Prioritize pain relief," and "Decision to hold treatment until family arrives." Furthermore, data such as recent increases in blood pressure or difficulty walking, nocturnal urinary frequency, and increased risk of falls are automatically retrieved from external health management services or medical institutions' electronic medical record systems and wearable terminals. Specifically, this system receives an emergency notification trigger from the terminal. This emergency notification trigger enables an event-driven architecture. This Personal Information Management Department executes a distributed database system. This Personal Information Management Department queries the corresponding user's profile data. This profile data contains a time-series array of pre-existing condition markers or vital sign data. The Personal Information Management Department combines the acquired data into a feature vector. This feature vector consists of hundreds of numerical elements. The Personal Information Management Department caches this feature vector in memory. The Personal Information Management Department then transmits this feature vector to the Situation Analysis Department at high speed. As a result, the system performs parallel reads from the distributed database. This system reduces data acquisition latency in emergencies to the millisecond level. Consequently, real-time situation monitoring becomes possible. This technology is applied to initial response support systems in emergency medical care.

[0063] The Situation Analysis Department receives an emergency notification from the terminal stating that "the individual has collapsed and is unconscious." It uses generative AI to comprehensively analyze data from the Personal Information Management Department and recent health monitoring data. The generative AI automatically determines the urgency level and estimates the risk of myocardial infarction or stroke, referencing the individual's past medical history or pre-incident instructions, family contact information, legal representative information, past communication history, medical guidelines, and social ethics regulations. Furthermore, considering factors such as the individual's pre-incident instructions explicitly stating "no CPR," family members living far away, or a spouse's wavering, it immediately notifies emergency services or medical institutions of the intention to "not perform resuscitation measures and prioritize palliative care." Specifically, this Situation Analysis Department executes a multimodal neural network. The Situation Analysis Department receives an embedding vector of text input as input. This embedding vector is, for example, a 512-dimensional numerical array. The Situation Analysis Department receives a time-series tensor of vital signs data as input. This time-series tensor is an array representing changes in heart rate or blood pressure over the past 24 hours. The Situation Analysis Department receives a vector of profile feature values ​​as input. This multimodal neural network performs comprehensive calculations on these heterogeneous data in a high-dimensional space. The multimodal neural network outputs an urgency score, which is a floating-point number from zero to one. Furthermore, the multimodal neural network outputs a probability distribution of disease categories. For example, the output is an array of values ​​representing the probability of myocardial infarction as 0.85 and the probability of stroke as 0.12. The situation analysis unit determines whether the urgency score exceeds a predetermined threshold. This threshold is, for example, set to 0.8. When the urgency score exceeds this threshold, the situation analysis unit activates an instant notification module. This instant notification module sends an alarm signal to an external emergency medical system. Thus, this system comprehensively analyzes multiple types of data. This system classifies urgency with higher accuracy than rule-based judgments based on a single data source. As a result, this system reduces false alarms and accelerates appropriate medical intervention. This technology is applied to automated triage systems during emergency transport.

[0064] The Decision Support Department executes an ethical decision-making algorithm based on generative AI, comprehensively considering the individual's values ​​or wishes, the family's intentions, the situation at the medical scene, and social ethical regulations to generate the best option. For example, it suggests options such as "Do not perform CPR, prioritize pain relief," "Remain unchanged until family arrives," "No need for a ventilator," and "Respect the individual's wishes to the greatest extent possible, do not perform life-prolonging treatment," and provides a highly transparent explanation of the basis or rationale for each option in natural language. For example, it automatically generates explanations such as "Since the individual's pre-arranged instructions state their desire not to undergo life-prolonging treatment, and this desire has been confirmed in past conversations with family, respecting the natural course of events by not performing CPR is the best course of action," and "Since family is far away, it is recommended to temporarily maintain the treatment plan and contact family." Specifically, this Decision Support Department executes a recommendation model based on reinforcement learning. This Decision Support Department receives the disease probability distribution from the Situation Analysis Department as input. This Decision Support Department receives the natural language embedding vector of the pre-arranged instructions as input. This Decision Support Department receives the constraint matrix of ethical regulations as input. This recommendation model generates a list of multiple action options based on these inputs. The model calculates a utility score for each option, based on a reward function. Furthermore, the model outputs generated text based on an attention weight map, which is a string explaining the rationale for each choice. The decision support unit extracts the top three options with the highest utility scores and sends them to the interface unit. Thus, the system rapidly derives the optimal solution that satisfies complex constraints. This system achieves comprehensive option evaluation that surpasses the limits of human cognition through neural network reasoning. As a result, the system significantly reduces the decision-making burden on healthcare professionals. This technology is applied to the automatic generation of clinical pathways in end-of-life care.

[0065] The simulation includes scenarios such as "a person falling at home at night and becoming disoriented," "family members being overseas and unable to be contacted immediately," "hospitalization restrictions during a novel infectious disease outbreak," and "needing medical care in a shelter during a disaster." The system simulates various scenarios, such as "a cluster infection in a nursing facility" and "a person's desire to die at home." For each scenario, it provides natural language prompts indicating the possible outcomes and impacts of decisions based on the user's values ​​or desired outcomes. This allows users or their families to prepare for or reassess decisions by experiencing multiple scenarios beforehand. Specifically, the simulation department executes a Markov decision process model. It uses a Monte Carlo tree search algorithm to explore the state space. The simulation department receives an initial state vector as input, which is a numerical array representing the patient's current health status. It also receives environmental change parameters as input, representing values ​​indicating the arrival time of family members or the urgency of infectious disease beds. The Markov decision process model outputs the predicted probability of the final state in each scenario. It also outputs a Q-value matrix representing the value of actions. Based on these predicted probabilities, the simulation department generates a scenario tree. It generates coordinate data for depicting nodes and edges on the user's graphical user interface. This coordinate data is then sent to the interface department. Thus, the system efficiently explores a vast state space. This system generates real-time predictions for various scenarios while minimizing the consumption of computing resources. As a result, the system quantitatively visualizes future risks. This technology can be applied to support the optimal allocation of medical resources or the initiation of proactive countermeasures.

[0066] The interface is installed on the terminal side, providing a natural language dialogue interface for two-way communication with the user or their family. Users or family members can input questions such as "What will happen if I have a stroke?", "How should I cope if my family is far away?", "What are the treatment plans if I have dementia?", "What preparations are needed if I want to face death at home?", and "What procedures are required if I want to be admitted to a nursing facility?". When encountering issues such as medical care, nursing options, or legal procedures, generative AI provides easily understandable explanations, allowing users or their families to input their wishes or desires in a conversational format. This input is immediately reflected in the personal information management department, dynamically updating pre-treatment instructions or value profiles. Specifically, this interface implements a pipeline integrating a speech recognition model and a natural language understanding model. This interface receives speech waveform data from the user as input. This speech waveform data is, for example, a digital signal with a sampling rate of 16 kHz. This interface converts this speech waveform data into a text string. This natural language understanding model receives this text string as input. This natural language understanding model outputs an intent classification label. This intent classification label is an identifier representing categories such as questions or information updates. Furthermore, this natural language understanding model outputs extracted entities. This entity is string data representing the target disease or desired nursing care pattern. This interface automatically generates an update query for the database using this entity. This interface sends this update query to the personal information management department. Thus, this system converts unstructured natural language input into structured data in real time. This system automates the dynamic updates of the database. As a result, this system maintains data consistency and eliminates input errors. This technology is applied to hands-free information input systems for the elderly.

[0067] The Data Linkage Department integrates with external health management services, electronic medical record systems of medical institutions, nursing record systems, wearable terminals, etc., via API to automatically acquire data such as blood pressure, pulse, body temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, and nursing records, and provides this data to the Personal Information Management Department or the Status Analysis Department. Specifically, this Data Linkage Department integrates with edge computing technology. It executes a data stream processing engine. It receives raw data from external application programming interfaces (APIs). This raw data is described in a hierarchical text data format. The Data Linkage Department performs interpolation algorithms for missing values ​​on this raw data. It performs outlier removal processing based on Z-scores. The Data Linkage Department converts the processed data into a normalized time series tensor. This time series tensor is a two-dimensional array with dimensions of time steps and feature quantities. The Data Linkage Department caches this time series tensor in an in-memory database. The Data Linkage Department provides this time series tensor in response to read requests from the Status Analysis Department. Thus, preprocessing is centralized within the Data Linkage Department. This system significantly reduces the computational load on the backend artificial intelligence models. As a result, the system's overall processing throughput is improved. This technology is applied to large-scale data collection infrastructure from IoT devices.

[0068] The Security Management Department is responsible for the overall security and privacy protection of the system. It possesses functions such as encrypted communication, multi-factor authentication, access control management, and audit log management. Appropriate information disclosure scopes are set for each relevant person, including the user, their family, and medical professionals, to achieve strict information management. Specifically, the Security Management Department implements public-key encryption. It employs a role-based dynamic access control algorithm. The Security Management Department receives user authentication requests as input. These requests contain a hash of the user identifier or password. Additionally, they contain a biometric feature vector. Based on these inputs, the Security Management Department generates an access token. This access token is an encrypted string with an expiration date. The Security Management Department outputs a bitmask representing the permission level. The Security Management Department verifies this access token and bitmask when requesting data access from various functional departments. The Security Management Department only filters and returns permitted data fields. Thus, the system strictly controls data access at both the hardware and software levels. This system prevents unauthorized access. As a result, the system mathematically reduces the risk of leakage of highly confidential medical data. This technology is applied to the zero-trust security infrastructure of the medical information system.

[0069] As specific examples of prompts that generative AI can read, examples include: "An 82-year-old male with a history of hypertension and heart disease, who stated in his pre-treatment instructions that he does not wish for life-prolonging treatment, is now unconscious. What is the best course of action during emergency transport? Considering his values ​​and the wishes of his family, explain the options and their reasons in natural language." "A person suffering from dementia has difficulty expressing their wishes. Based on past speaking history, the wishes of their family, and the instructions of their legal guardian, simulate the process of proxy decision-making and explain the best response strategy and its basis." "A person wishes to die at home. Given that their family is far away, what kind of medical and nursing care system is needed? Specifically explain the necessary procedures or preparations." "In the case of a person refusing blood transfusions for religious reasons, what alternative treatments can be considered in an emergency? Generate a prompt for the medical staff and the family." Specifically, this system has a prompt generation unit. This prompt generation unit performs the process of embedding structured data extracted from the database into a natural language prompt template. This prompt generation unit receives the patient profile vector as input. This prompt generation unit receives the current context vector as input. This prompt generation unit parses these vectors and outputs an optimized prompt string. This prompt string is represented as an array of tokens. The prompt generation unit feeds the generated prompt string into the input layer of a large-scale language model. This large-scale language model then begins inference processing using a self-attention mechanism. Thus, the system dynamically constructs prompts. This system enables the artificial intelligence model to make inferences that accurately reflect the latest patient state. As a result, the system significantly reduces the probability of fact-checking errors. This technique improves the accuracy of natural language generation systems specialized in the medical field.

[0070] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0071] (Implementation methods for carrying out the present invention) This invention provides an advanced decision support system for high-quality care and convincing decision-making in nursing facilities or home care settings, while maximizing respect for the dignity and self-determination of the patient and reducing the burden on family members or caregivers. The system comprises multiple functional units, including a patient information management unit, a status analysis unit, a nursing decision support unit, a simulation unit, a user interface unit, a data linkage unit, and a security management unit. Specifically, the system is built on a cloud computing environment and employs a microservices architecture. Each functional unit is deployed as an independent container. These containers use virtualization technology to isolate hardware resources. The functional units communicate asynchronously via message queues. These message queues perform data transmission based on a publish-subscribe model. This prevents the entire system from crashing even when the load is concentrated on a specific functional unit. The system allows each functional unit to scale horizontally independently. As a result, the system significantly improves scalability and fault tolerance. This technology is applied to the construction of highly available systems in large-scale healthcare networks.

[0072] The nursing care recipient information management department operates on a server, managing the individual's medical history (hypertension, diabetes, heart disease, stroke history, osteoporosis, presence or absence of dementia, past hospitalization history, surgical history, allergy information, infectious disease history, etc.), nursing history (past fall history, aspiration history, whether nighttime toileting assistance was provided, frequency of feeding assistance, rehabilitation implementation records, bedsore history, bowel and bladder dysfunction, medication errors, whether there was refusal of care, daytime activity level, recreational participation, etc.), and lifestyle habits (exercise habits, diet, etc.). The criteria for determining whether a patient smokes or drinks alcohol, their sleep patterns, daytime activity levels, interests and hobbies, frequency of outings, and social participation include: values ​​(wanting to avoid hospitalization, not wanting life-prolonging treatments, prioritizing pain relief, specific religious considerations, valuing meetings with family members, hoping to die at home, hoping for active treatment, expecting a natural process, and designation of a proxy decision-maker when the patient cannot express their own wishes); and the intentions of family members (family members living far away and unable to meet frequently, family members being designated as proxy decision-makers, family members wanting to actively participate in nursing care policies, and disagreements among family members). The system aims to alleviate the mental burden on family members and prioritize their quality of life. It manages a unified system of data including: legal information (regarding the presence of an adult guardian, arbitrary guardianship agreements, executors of wills, medical consent forms, pre-death instructions, declarations of dignified death, designation of a proxy decision-maker, medical expense payment methods, and the status of nursing insurance utilization); nursing plans (bathing frequency, rehabilitation frequency, medication management, dietary content, excretion care, external support, recreational participation, nursing visits, doctor visits, short-term stay utilization, daytime service utilization, and nighttime care system); vital sign data (blood pressure, pulse, temperature, blood glucose, oxygen saturation, weight, respiratory rate, ECG data, sleep duration, and activity level); action data from wearable devices (steps, fall detection, sleep duration, activity level, heart rate variability, location information, and emergency notification button press history); and input information from caregivers or family members (nursing records, logs, messages from family members, recent changes in the individual's appearance or mood, new hopes or anxieties, caregiver opinions, and comments from medical professionals). For example, when caregivers input daily care records from the terminal, when family members update a person's values ​​or wishes via their smartphones, when wearable devices detect nighttime falls, when the latest medical information is automatically retrieved from medical institutions via API linkage, when a person expresses through voice input that they "recently prefer a natural process rather than aggressive treatment," and when family members input that they "hope for home-based end-of-life care," all this information is accumulated in real time on the server and dynamically updated based on changes in circumstances or new inputs. Specifically, this nursing care recipient information management department implements a distributed graph database.This patient information management department receives unstructured and semi-structured data from various sensors or terminals as input. It uses a natural language processing model to perform entity extraction. The department then performs relation mapping between the extracted entities. Finally, it outputs a knowledge graph data structure consisting of nodes and edges. A node represents an entity such as a patient, disease, or hope. An edge represents a relationship such as having or hoping. The department expands this knowledge graph into memory. It responds to complex queries from the condition analysis department in milliseconds. Thus, this system integrates diverse data as a knowledge graph. This system enables high-speed retrieval of hidden relationships between data. As a result, this system provides a robust data infrastructure supporting the inference accuracy of artificial intelligence models. This technology is applied to next-generation electronic medical record systems that integrate and manage complex medical information.

[0073] The situation analysis department, implemented on a server, analyzes the condition of caregivers in real time, whether in emergencies or routine situations. For example, when inputting messages like "The person fell in the toilet," "Persistent fever," "Choked while eating," "Wandering at night," "Abnormal values ​​detected by vital signs sensors," "Family reports changes in the person's appearance," "Nursing staff reports decreased level of consciousness," "Medication error occurred," or "Frequent bowel movement problems,"... In cases of newly reported bedsores, the Situation Analysis Department uses generative AI to comprehensively analyze data accumulated in the Patient Information Management Department, including medical or nursing records, lifestyle data, recent health monitoring data, communication records from family members or caregivers, and information from the medical field. For example, if a fall detection sensor is activated, the department immediately refers to past fall history, presence of osteoporosis, stability of vital signs, the patient's desire to avoid hospitalization, family contact information, and current care plans to automatically determine the urgency level and estimate the risk of fractures or head injuries. If fever is detected, the department comprehensively assesses the risk of pre-existing conditions or infections, medication status, dehydration risk, family intentions, and coordination with medical institutions to derive necessary responses. If the patient wanders at night, the department analyzes the progression of dementia, past wandering history, family wishes, facility safety measures, and GPS-based location information to determine the increased risk of wandering or the necessity of additional countermeasures. In the event of aspiration during eating, the system comprehensively analyzes past aspiration history, swallowing function assessments, dietary patterns, opinions from nurses or speech-auditrists, family wishes, and coordination with medical institutions to determine the necessity of emergency treatment or medical attention. Specifically, this situation analysis unit executes an ensemble model combining a recurrent neural network for time-series data processing and a Transformer model for text processing. The unit receives a multidimensional time-series tensor from sensors as input. This multidimensional time-series tensor is a three-dimensional array containing acceleration, angular velocity, or heart rate. The unit also receives text embedding vectors from nursing records as input. The ensemble model integrates these inputs and outputs an anomaly detection score. Furthermore, the ensemble model outputs a state classification label. This state classification label represents categories such as fall, fever, or wandering. The ensemble model outputs a confidence level for this state classification label. This confidence level is a value from zero to one. The unit determines whether this confidence level exceeds a predetermined threshold. If the confidence level exceeds this threshold, the unit generates a corresponding alarm event. The unit then sends this alarm event to the nursing decision support department. Therefore, this system integrates and analyzes multimodal data. With high accuracy and low latency, this system identifies complex anomalies that are difficult to detect using single sensor data. As a result, this system prevents omissions in the care setting and improves safety. This technology is applied to automated care systems in elderly care facilities.

[0074] Based on the analysis results obtained from the Situation Analysis Department or the data from the Patient Information Management Department, the Nursing Decision Support Department executes an ethical decision-making algorithm based on generative AI. This algorithm comprehensively considers the patient's values ​​or desires, family wishes, nursing plans, the situation at the medical facility, and social ethical regulations to generate the optimal nursing policy or procedures. For example, it may suggest specific options such as: "Contact family immediately in case of a fall and prioritize observation within the facility," "Contact a medical institution in case of fever, ensure thorough hydration and rest," "Provide emergency treatment in case of aspiration during eating and seek medical attention as needed," "Ensure safety and report the situation to family when wandering around at night," "Enhance home care if the patient does not wish to be hospitalized," "Adjust medication to prioritize pain relief," "Avoid specific treatments for religious reasons," "Rapid medical intervention and reassessment of the nursing plan in case of pressure ulcers," "Immediate reporting to the doctor and proposals for recurrence prevention in case of medication errors," and "Enhanced excretion care and explanation to family in case of frequent excretion problems." The generative AI provides highly transparent explanations of the rationale or justification for these options using natural language. For example, the system automatically generates suggestions such as: "Since you have registered your intention to avoid hospitalization, it is recommended that you coordinate with your family and undergo observation at home or in a facility," "In cases of high risk of fractures or head injuries due to falls, it is recommended that you seek immediate medical attention," "Since fever has been confirmed, considering the risk of dehydration or infection, it is recommended that you contact a medical facility and receive hydration," and "Due to frequent nighttime wanderings, it is suggested that you carry a GPS device or strengthen the nighttime care system." The system provides information such as, "Due to the development of a new pressure ulcer, it is recommended to contact the medical facility and reassess the care plan," based on the situation. Specifically, this nursing decision support unit executes a hybrid approach using a rule-based inference engine and a deep learning-based generative model. The unit receives anomaly labels from the situation analysis department as input. It also receives a constraint vector extracted from a knowledge graph. This constraint vector is a numerical array containing de-hospitalization or religious constraint flags. The hybrid model outputs an array of recommended care steps. It also outputs a priority score for each step. Furthermore, it outputs a text token column of explanatory text. The nursing decision support unit ranks the care steps based on these priority scores. The ranked care steps are then pushed to the nursing staff's terminals via the interface. Thus, this system balances the safety guarantees based on strict rule-based processing with the prompts of flexible response strategies based on generative models. This system significantly improves the quality and speed of decision-making in the healthcare setting. As a result, this system enables even less experienced staff to respond appropriately, on par with experienced staff. This technology is applied to operational support and education systems in nursing facilities.

[0075] The simulation department uses generative AI to virtually simulate various emergency situations or nursing and medical scenarios. These scenarios include: "falling in the toilet at night," "developing a high fever," "accidentally swallowing food," "wandering at night," "family members being far away and unreachable," "hospitalization restrictions during a novel infectious disease outbreak," "requiring care in a shelter during a disaster," "a cluster infection in a nursing facility," "the individual wishing to die at home," "worsening bedsores," and "a series of medication errors." For each scenario, natural language prompts are provided to guide users, their families, and caregivers in anticipating and preparing for decisions based on their values, desires, and the impact of their care plans. This allows users, their families, and caregivers to experience multiple situations in advance, prepare for or reassess decisions, and automatically update pre-instructions, personal information, and care plans based on simulation results. Specifically, the simulation department executes a virtual environment simulator using digital twin technology. The simulation department receives the patient's current state vector as input. It also receives a matrix of environmental variation parameters as input. The virtual environment simulator outputs the future state transition probabilities for each scenario. This virtual environment simulator outputs an array of risk scores. Furthermore, it outputs natural language text summarizing the simulation results. The simulation department extracts scenarios where the risk scores exceed a specific threshold. Based on the extracted scenarios, the simulation department generates proposal data for preventative measures. This proposal data is then used to automatically update the nursing plan data in the patient information management department. Thus, this system rapidly calculates future predictions in a high-dimensional parameter space. This system visualizes complex risks that are difficult for humans to predict in advance. As a result, this system enables the development of proactive nursing plans. This technology is applied to the optimization of nursing management within the long-term care insurance system.

[0076] The interface is installed on the terminal side, providing a natural language dialogue interface for two-way communication with the user, their family, or caregivers. For example, users can input questions such as "Do I need to be hospitalized because of this fall?", "What are the measures to prevent it from happening again?", "What would be the treatment plan if I were diagnosed with dementia?", and "What preparations should I make if I want to face the end of my life at home?". When faced with questions such as "What procedures are required for admission to a nursing facility?", "What are the countermeasures for frequent nighttime wanderings?", "What should be done if bedsores worsen?", and "What are the countermeasures to prevent recurrence of repeated medication errors?", the generative AI provides easily understandable explanations of medical and nursing options or legal procedures. This allows users, their families, and caregivers to input their wishes or desires in a conversational format. The input is immediately reflected in the patient information management department, dynamically updating pre-instructions or care plans. Specifically, this interface implements a lightweight speech recognition model and an intent extraction model on the client side. This interface receives voice or text input from the user. The intent extraction model parses the input and outputs a structured query object. This query object contains key-value pairs representing the target disease or the requested action. This interface sends this query object to the generative AI on the server side. This interface renders the response text received from the server onto the screen. Furthermore, this interface sends the extracted user intentions to the patient information management department. The patient information management department uses the received data to update the database. Thus, this system provides an intuitive natural language interface. This system enables even users with limited IT literacy to easily operate complex systems. As a result, it lowers the barrier to data entry and maintains the freshness and accuracy of information. This technology is applied to mobile applications targeting the elderly or caregivers.

[0077] The Data Linkage Department integrates with external health management services, electronic medical record systems of medical institutions, nursing record systems, wearable terminals, etc., via APIs to automatically acquire data such as blood pressure, pulse, body temperature, steps, sleep time, fall history, medical records, medication reminders, rehabilitation progress, nursing records, electrocardiogram data, activity levels, GPS location information, and emergency notification history. This data is then provided to the Patient Information Management Department or the Status Analysis Department. Specifically, this Data Linkage Department executes a data pipeline orchestrator. It asynchronously and in parallel processes data collection from multiple external application programming interfaces (APIs). It receives raw data in various formats from external systems as input. It uses a schema mapping algorithm to perform data structure transformation. It performs timestamp synchronization. It performs noise filtering. It outputs a multidimensional time series tensor with a unified schema. The transformed tensor is streamed to the Patient Information Management Department and the Status Analysis Department via a message queue. Thus, this system automates data integration between heterogeneous systems. This system constructs a real-time data pipeline. As a result, this system ensures that the artificial intelligence model always draws inferences based on the latest and clean data. This technology is applied to the data foundation of a community-based integrated care system that connects healthcare institutions and care facilities.

[0078] The Security Management Department is responsible for the security and privacy protection of the entire system. It possesses functions such as encrypted communication, multi-factor authentication, access control, audit log management, data backup, information leakage detection, usage history recording, terminal authentication, and access control for each user. It sets appropriate information disclosure scopes based on the user, their family, caregivers, medical professionals, and other relevant personnel. This ensures that an individual's medical history or values ​​are only accessible to the individual and their family, and that information is only disclosed to legal representatives or medical professionals within necessary limits, achieving strict information management. Specifically, this Security Management Department implements a dynamic access control algorithm based on a zero-trust architecture. This Security Management Department receives the Internet Protocol address of the access request source as input. This Security Management Department receives the device fingerprint as input. This Security Management Department receives the user's biometric authentication data as input. This dynamic access control algorithm dynamically calculates and outputs a trust score based on these inputs. This dynamic access control algorithm outputs a decision to allow or deny access. This Security Management Department determines whether the trust score is below a predetermined threshold. If the trust score is below the threshold, this Security Management Department sends a prompt to the Interface Department requesting additional multi-factor authentication. Furthermore, this security management department records all access history in a blockchain-based tamper-proof log system. This allows the system to implement context-based dynamic access control. The system provides immutable log records. As a result, the system's tolerance to internal violations or external cyberattacks is maximized. This technology is applied to a highly secure medical data sharing platform that complies with personal information protection laws.

[0079] In this way, in the embodiments of the present invention, generative AI is used as an essential element to achieve comprehensive and flexible decision support, including integrated management of patient information, real-time analysis of conditions, ethical decision support, natural language-based explanations and dialogues, scenario simulation, dynamic updates of pre-instructions or care plans, external data linkage, and safety management. This approach maximizes respect for the dignity and self-determination of the patient in the care setting while reducing the burden on family members or caregivers, providing high-quality care and convincing decision-making. For example, it can handle various application scenarios. When a patient's condition suddenly changes, caregivers can instantly confirm the patient's wishes using a terminal on-site; family members can participate in decision-making even remotely through a terminal; medical professionals can determine treatment strategies while referring to the patient's values ​​or past expressions of intent; nursing facility staff can share changes in the patient's condition with family members by inputting daily care records into the system; AI automatically issues risk alerts and proposes additional countermeasures when the risk of wandering or falling increases at night; and AI automatically proposes countermeasures to prevent recurrence and automatically modifies care plans in cases of bedsores or medication errors. Furthermore, generative AI can continuously learn from user input or changes in circumstances, enabling the continuous evolution of individually optimized decision support. Specifically, this system as a whole forms a feedback loop with a continuous learning framework. The system receives feedback data from users as input. This feedback data includes a history vector of approvals or revisions to proposed care plans. The system uses backpropagation to fine-tune the weight parameters of the neural network. The system outputs an updated weight parameter matrix. The system seamlessly deploys the updated model to the production environment. The system improves the accuracy of the next inference. Thus, this system utilizes data accumulated through application, allowing the AI ​​model to evolve autonomously. The system continuously improves personalized accuracy specific to each user. As a result, the overall value of the system increases over time. This technology is applied to the construction of adaptive AI agents for healthcare.

[0080] (System Composition) The system involved in this implementation includes a patient information management department, a condition analysis department, a nursing decision support department, a simulation department, a user interface department, a data linkage department, and a security management department. The patient information management department records detailed medical histories of patients, including hypertension, diabetes, heart disease, stroke, osteoporosis, presence of dementia, past hospitalizations, surgical procedures, allergies, and infectious disease history. It also records nursing histories, including past falls, aspiration, frequency of nighttime toileting and feeding assistance, rehabilitation records, bedsore occurrence history, bowel and bladder problems, medication errors, refusal of care, daytime activity levels, and recreational participation. Finally, it manages lifestyle habits, including exercise habits, diet, and the presence of [unspecified substance]. Smoking and drinking habits, sleep patterns, daytime activity levels, interests and hobbies, frequency of outings, and social participation are recorded as values. Other records include desires to avoid hospitalization, avoid life-prolonging treatments, prioritize pain relief, specific religious considerations, emphasis on family meetings, a desire to die at home, a desire for active treatment, a desire for a natural process, and the designation of a proxy decision-maker when the individual cannot express their wishes. Family members' intentions are also recorded, including situations where family members live far away and infrequent visits are difficult, family members are designated as proxy decision-makers, family members wish to actively participate in care policies, disagreements among family members, a desire to reduce the family's mental burden, and a desire for the family to prioritize their own quality of life. Legal information includes the presence of an adult guardian, arbitrary guardianship agreements, executors of wills, medical consent forms, the content of prior instructions, declarations of dignified death, the designation of a proxy decision-maker, methods of medical expense payment, and the use of long-term care insurance. A care plan is maintained by managing bathing frequency, rehabilitation frequency, medication management, dietary content, elimination assistance, support for outings, recreational participation, visits to caregivers, visits to doctors, short-term hospital stays, daytime service utilization, and nighttime care arrangements. For vital sign data, it collects blood pressure, pulse, body temperature, blood glucose, oxygen saturation, weight, respiratory rate, electrocardiogram data, sleep time, activity level, etc. as action data from wearable terminals, automatically acquiring step count, fall detection, sleep time, activity level, heart rate variability, location information, emergency notification button press history, etc. as information entered by nursing staff or family members, and uniformly manages nursing records, logs, messages from family members, changes in the individual's recent condition or emotions, the individual's new hopes or anxieties, nursing staff opinions, comments from medical professionals, etc.For example, when caregivers input daily care records from the terminal, when family members update a person's values ​​or wishes via smartphone, when a wearable device detects a fall at night, when the latest medical information is automatically retrieved from medical institutions via API linkage, when a person expresses a desire for a natural process rather than aggressive treatment via voice input, or when a family member inputs a wish for end-of-life care at home, all this information is accumulated in real time on the server and dynamically updated based on changes in circumstances or new inputs. Specifically, this system operates in a distributed computing environment. The nursing object information management department is configured as a distributed database system. This department uses a natural language processing model to convert various input text data and time-series numerical values ​​into embedding tensors in a high-dimensional vector space. The department stores the converted embedding tensors in a vector database. The situation analysis department has a multimodal neural network. This department combines time-series vital sign data and text embedding tensors as input features. The nursing decision support department has a large-scale language model and a reinforcement learning agent. The simulation department executes a Monte Carlo tree search algorithm. This interface unit communicates with the terminal using an asynchronous communication protocol. This data linkage unit normalizes raw data acquired from external systems. This security management unit uses advanced encryption algorithms to protect data. Through these structures, this system transcends simple information digitization, enabling comprehensive semantic analysis of heterogeneous and mixed data. This system uses vector operations in high-dimensional space to detect changes in the state of care recipients with speed and accuracy exceeding the limits of human cognition. This system brings about the technological benefits of improved efficiency in data management and increased accuracy in anomaly detection in nursing settings. This system functions as a real-time decision support foundation in specific application areas such as elderly care facilities or home healthcare networks.

[0081] The Situation Analysis Department uses generative AI to comprehensively analyze medical or nursing records, lifestyle data, recent health monitoring data, communication records from family members or caregivers, and information from the medical field stored in the patient information management department. When input from the terminal, such as "fell in the toilet," "persistent fever," "choked while eating," "wandering at night," "abnormal values ​​detected by vital signs sensors," "family contacted saying the patient's condition has changed," "caregiver reported a decline in the patient's level of consciousness," "medication error occurred," "frequent bowel movement problems," or "new bedsores have developed," and the fall detection sensor is activated, the system immediately refers to past fall history, the presence or absence of osteoporosis, the stability of vital signs, and the patient's "desire to avoid..." The system automatically assesses urgency based on the values ​​of "no hospitalization required," family contact information, and current care plans, estimating the risk of fractures or head injuries. In cases of fever, it comprehensively evaluates the risk of pre-existing conditions or infections, medication status, dehydration risk, family wishes, and coordination with medical institutions to derive necessary responses. In cases of nighttime wandering, it analyzes the progression of dementia, past wandering history, family hopes, facility safety measures, and GPS-based location information to determine the increased risk of wandering or the necessity of additional measures. In cases of aspiration during eating, it comprehensively analyzes past aspiration history, swallowing function assessments, dietary patterns, nurse or speech-auditory therapist opinions, family wishes, and coordination with medical institutions to determine the necessity of emergency treatment or medical attention. Specifically, this situation analysis unit executes a multimodal large-scale language model based on the Transformer architecture. It receives text information, time-series vital sign data, and sensor signals as input data. The unit tokenizes the text information and converts it into 512-dimensional word embedding vectors. This situation analysis unit inputs time-series vital sign data into a one-dimensional convolutional neural network to extract local feature maps. It then converts sensor signals into frequency domain spectrograms. This unit combines these heterogeneous data to generate a comprehensive feature tensor. For example, the input data to this unit includes the embedding vector of the text "I fell in the toilet," a time-series array of heart rates over the past 24 hours, and a waveform tensor from a 3-axis accelerometer. The unit inputs this comprehensive feature tensor into a self-attention mechanism to calculate the hidden correlations between the data. The unit outputs an urgency score, a risk probability distribution, and a category label for the recommended action via the softmax function of the output layer. For example, the output might be an urgency score of 0.85, a fracture risk probability of 72%, and the recommended action of immediate contact with a medical institution. If the output urgency score exceeds a pre-set threshold, the unit sends an interruption request to the subsequent nursing decision support unit.Unlike existing rule-based judgments that rely on human experience, this condition analysis unit uses nonlinear feature representations in a high-dimensional space to assess the condition. This allows it to identify complex causal relationships between minute, instantaneous changes in vital signs and past medical history. This condition analysis unit effectively reduces anomaly detection latency to the millisecond level and significantly lowers the false positive rate. In specific applications such as remote monitoring systems and intelligent nursing facilities, this condition analysis unit helps prevent fatal incidents.

[0082] Based on the analysis results obtained from the Situation Analysis Department or the data from the Patient Information Management Department, the Nursing Decision Support Department executes a generative AI-based ethical decision-making algorithm. This algorithm comprehensively considers the patient's values ​​or desires, family wishes, nursing plans, the situation at the medical facility, and social ethical regulations to generate the optimal nursing policy or procedures. It provides specific options, such as immediately contacting family members and prioritizing observation within the facility in case of a fall; contacting medical institutions and ensuring adequate hydration and rest in case of fever; providing emergency treatment and seeking medical attention as needed in case of aspiration during eating; ensuring safety and reporting the situation to family members when the patient wanders at night; enhancing home care when the patient does not wish to be hospitalized; adjusting medication to prioritize pain relief; avoiding specific treatments for religious reasons; and rapid medical coordination and nursing care in the event of pressure ulcers. The nursing decision support unit (NDSU) includes options such as reassessing the care plan, providing immediate reporting to doctors in case of medication errors and proposing countermeasures to prevent recurrence, strengthening bowel care and explaining to family members in case of frequent bowel problems, etc. The NDSU provides highly transparent explanations of the reasons or justifications for these options in natural language, automatically generating explanations based on the situation. For example, if the patient has registered their desire to avoid hospitalization, it is recommended to work with family members and conduct observation at home or in the facility; if the risk of fracture or head injury due to a fall is high, it is recommended to seek immediate medical attention; if fever is confirmed, the risk of dehydration or infection is considered, and it is recommended to contact a medical institution and provide hydration; if frequent nighttime wanderings occur, it is recommended to carry a GPS device or strengthen the nighttime care system; if new pressure sores have developed, it is recommended to contact a medical institution and reassess the care plan, etc. Specifically, this NDSU possesses a decision neural network optimized through reinforcement learning. The NDSU receives urgency scores and risk probability distributions output from the situation analysis unit as input. The NDSU further combines value vectors obtained from the patient information management unit and care plan tensors as input features. This nursing decision support department feeds this input data into a multilayer perceptron to calculate expected reward values ​​for multiple action candidates. The department uses a multi-objective optimization function, quantified by the suitability of the individual's wishes, medical safety, and reduction of burden on the family, as the reward function. The department selects the action candidate with the highest calculated expected reward value. The department maps the selected action candidate into the cue space of a large-scale language model. This large-scale language model performs autoregressive decoding to generate explanatory text based on natural language. For example, the input to this nursing decision support department includes an urgency score of 0.6, a value vector indicating a preference for avoiding hospitalization, and a family intention vector indicating a desire for home care.This Nursing Decision Support Unit outputs the home observation data as an action tag and the generated text "Considering the individual's intention to avoid hospitalization and the stability of vital signs, home observation is recommended." This unit then sends the output action tag and generated text to the Interface Unit, triggering on-screen rendering. This unit transforms the subjective human judgment process into a mathematical optimization problem in a multi-dimensional vector space. Thus, it handles complex and intertwined ethical constraints and medical requirements without contradiction. This unit delivers the technical advantage of rapidly deriving optimal solutions while ensuring transparency in decision-making and minimizing the consumption of computer resources. In specific application areas such as accessing nursing stations or community integrated support centers, this unit functions as a business support tool for nursing managers.

[0083] The simulation department uses generative AI to virtually simulate various emergency situations or nursing and medical scenarios. These scenarios include: falling in the toilet at night; high fever; choking while eating; wandering at night; family members being unreachable from afar; limited hospital access during a novel infectious disease outbreak; needing to receive care in a shelter during a disaster; a cluster infection in a nursing facility; wishing to die at home; worsening bedsores; and repeated medication errors. For each scenario, natural language prompts are provided to guide users, their families, and caregivers through the simulation, offering insights into the decision-making outcomes and impacts based on the individual's values, desires, and care plan. This allows users, their families, and caregivers to experience multiple scenarios beforehand, preparing for or reassessing decisions. The simulation results automatically update pre-instructions, personal information, and care plans. Specifically, the simulation department executes a probabilistic state transition model based on a Markov decision process. The simulation department sets the current state vector of the care recipient as the initial value. It uses a random number generator to probabilistically generate multiple possible future events. This simulation unit uses a recurrent neural network to predict state changes at the occurrence of various events in a time-series manner. The predicted future state tensor is input into the nursing decision support unit to calculate virtual response strategies. The simulation unit uses a Monte Carlo tree search algorithm to rapidly evaluate thousands of scenario branches on parallel computing nodes. For example, the input to this simulation unit includes the current health status vector, environmental variable tensors, and a 30-day simulation period. The simulation unit outputs the probability score of each scenario, the predicted value of the final health status, and a recommended list of pre-emptive preparations. The simulation unit extracts scenarios where the output predicted values ​​exceed a specific risk threshold and automatically updates the flag in the patient information management unit's database. This simulation unit is not simply a statistical summary of past data, but rather simulates dynamic state transitions in a virtual space within the computer. Therefore, this simulation unit pre-validates the system's response to catastrophic situations that cannot be tested in the real world. This simulation unit effectively improves the accuracy of predictions for unknown risks and enhances the robustness of the entire system. This simulation unit is utilized as a data-driven predictive tool in specific application areas such as business continuity planning for nursing facilities and end-of-life planning for individuals.

[0084] The user interface, installed on the terminal, provides a natural language dialogue interface for two-way communication with the user, their family, or caregivers. When users input questions such as whether hospitalization is necessary due to a fall, what preventative measures can be taken to avoid a recurrence, what treatment strategies would be available if they were diagnosed with dementia, what preparations are needed if they wish to die at home, what procedures are required for admission to a nursing facility, what countermeasures are needed for frequent nighttime wanderings, how to deal with worsening bedsores, or what preventative measures can be taken to avoid recurrence of repeated medication errors, generative AI provides easily understandable explanations of medical and nursing options or legal procedures. This allows users, their families, or caregivers to input their wishes or desires in a conversational format. The input is immediately reflected in the patient information management department, dynamically updating pre-instructions or care plans. Specifically, this user interface incorporates a natural language processing engine and a speech recognition module. It converts analog speech signals acquired from the terminal's microphone into digital waveform data. The user interface then uses a Hidden Markov Model and a deep learning-based acoustic model to convert the digital waveform data into text strings. This interface unit tokenizes and shapes text strings into a prompt format that can be input into a large-scale language model. It then sends the shaped prompts to the server-side nursing decision support unit. The unit uses a speech synthesis engine to convert the response text received from the server into a speech signal, which is then output through a speaker. For example, input to this unit might include speech waveform data for the question "What are the countermeasures in cases of frequent nighttime wanderings?" and the user's identification ID. Output from this unit is structured data describing the countermeasures and a rendering tree for screen display. The unit extracts text representing new hopes or values ​​input by the user and calls the update API of the nursing object information management unit to rewrite the database records. Unlike existing static graphical user interfaces, this unit performs dynamic dialogue state management that maintains context in memory. This allows even elderly people unfamiliar with computers or busy caregivers to intuitively operate complex systems. This unit reduces input error rates and significantly shortens processing time required for data input. It enables seamless human-machine interaction in specific application areas such as the control interfaces for smart speakers or nursing robots.

[0085] The Data Linkage Department integrates with external health management services, electronic medical record systems, nursing record systems, and wearable terminals via APIs to automatically acquire data such as blood pressure, pulse, temperature, steps, sleep duration, fall history, medical records, medication reminders, rehabilitation progress, nursing records, electrocardiogram data, activity levels, GPS location information, and emergency notification history. This data is then provided to the patient information management department or the status analysis department. Specifically, this Data Linkage Department functions as middleware managing multiple asynchronous communication threads. It periodically sends requests to endpoints in external systems. It converts raw data in various formats received from external systems into a standardized data model for the system. It performs data cleaning, imputing missing values ​​and filtering outliers. It applies timestamp synchronization and standardized sampling rates to time-series data. Finally, it temporarily stores the processed data as a multidimensional tensor in a memory buffer. For example, the input to this data linkage unit includes 100Hz electrocardiogram waveform data sent from a wearable terminal and text-formatted prescription data obtained from an electronic medical record system. As output, this data linkage unit outputs a normalized heart rate variability feature vector and an embedded tensor of the medication code. This data linkage unit asynchronously transmits the output tensor data to the status analysis unit via a message queue system. This data linkage unit uses an internal automatic conversion algorithm to absorb inconsistencies in data formats between heterogeneous systems. Therefore, this data linkage unit completely eliminates the need for manual data transcription by humans. This data linkage unit effectively minimizes latency in data integration processing and ensures the real-time performance of the entire system. This data linkage unit functions as a data center in specific application areas such as medical information linkage networks or personal health record integration foundations.

[0086] The Security Management Department is responsible for the security and privacy protection of the entire system. It possesses functions such as encrypted communication, multi-factor authentication, access control, audit log management, data backup, information leakage detection, usage history recording, terminal authentication, and access control for each user. Appropriate information disclosure scopes are set according to the user and their family, caregivers, medical professionals, and other relevant personnel. An individual's medical history and values ​​are only accessible to the individual and their family, and information is only disclosed to legal representatives and medical professionals within necessary limits, achieving strict information management. Specifically, the Security Management Department has authentication and authorization modules based on a zero-trust architecture. The department uses end-to-end encryption with Transport Layer Security (TLS) for all data communication applications within the system. The department encrypts personal information and medical data stored in the database using Advanced Encryption Standard (ALS) algorithms during data storage. Upon receiving an access request from a user, the department verifies a digital certificate based on public-key cryptography. The department also performs multi-factor authentication using biometric data or one-time passwords. The department dynamically determines access permissions for requested resources based on a role-based access control list. For example, the input to this security management unit includes an access request data packet and session token sent from the caregiver's terminal, as well as the data ID of the requested object. The output of this security management unit represents a Boolean value indicating allowed access and a session key used for decryption. If access is denied, this security management unit sends an alert to the intrusion detection system and temporarily blocks communication from that IP address. Unlike existing perimeter-based security, this security management unit verifies all transactions on a micro-segment basis. This minimizes the risk of information leakage due to internal crimes or the lateral movement of malware. This security management unit delivers a significant technological boost to the protection of confidential data and achieves secure data management that meets compliance requirements. This security management unit provides a robust foundation for specific application areas such as cloud systems in healthcare institutions or administrative platforms that handle highly confidential personal information.

[0087] Specific examples of prompts read by the generative AI when implementing this invention include: "An 85-year-old woman, requiring care level 2, with a history of hypertension and osteoporosis, was detected to have fallen during nighttime toilet care. She wishes to avoid hospitalization. She has two previous fall experiences, neither of which were serious. Her current vital signs are stable. Her family lives far away. Regarding this fall, considering her values ​​and her family's wishes, explain the best coping strategy and its rationale in natural language." "The patient has a fever and a history of diabetes and heart disease. She does not wish for aggressive treatment. Her family wishes to seek medical attention. What is the best coping strategy? Explain the options and their rationale." "The patient wanders at night, suspected of dementia progression. Considering past wandering history, family wishes, and facility safety measures, propose future countermeasures." "An aspiration occurred during eating. Based on swallowing function assessment or past aspiration history, explain the emergency treatment and future care plan reassessment plan," etc. Specifically, this system has a template engine for dynamically generating these prompts. This template engine inserts structured data extracted from the nursing object information management department into placeholders in natural language. The system converts the generated prompt text into a tokenized sequence suitable for the input layer of a large-scale language model. During tokenization, the system uses a byte-pair encoding algorithm to efficiently segment unknown words or technical terms into subwords. The system maps the tokenized sequence to a 512-dimensional embedding vector space and assigns positional encoding. For example, the input data to this system is a data object containing key-value pairs of information such as age 85, gender female, nursing care requirement level 2, history of hypertension, and event of nighttime fall. The system outputs a natural language prompt string incorporating this information, along with its converted numerical tensor. The system inputs the output numerical tensor into the encoder block of the Transformer model, enabling it to perform context parsing based on a self-attention mechanism. This system does not use a simple combination of fixed strings but rather a dynamic prompt construction algorithm that considers contextual dependencies. This allows large-scale language models to accurately capture subtle nuances of the situation and suppress unfounded output. The system achieves the technical effect of improving the accuracy of inferences and the reliability of generated responses. This system enables high-quality text generation in specific application areas such as medical diagnostic support systems or customer support AI with complex conditional branches.

[0088] (Implementation steps) Step 1: Collection and registration of patient information. Nursing staff, family members, and medical professionals use terminals to collect the patient's medical history (hypertension, diabetes, heart disease, stroke history, osteoporosis, presence or absence of dementia, past hospitalization history, surgical history, allergy information, infectious disease history, etc.), nursing history (fall history, aspiration history, frequency of nighttime toileting care, frequency of feeding care, rehabilitation records, bedsore history, bowel movement disorders, medication errors, refusal of care, daytime activity level, recreational participation, etc.), lifestyle habits (exercise habits, diet, smoking and drinking habits, sleep patterns, interests and hobbies, frequency of outings, social participation, etc.), values ​​(desire to avoid hospitalization, desire to avoid life-prolonging treatment, prioritizing pain relief, religious considerations, emphasis on meetings with family members, desire to die at home, desire for active treatment, desire for a natural process, designation of a proxy decision-maker, etc.), and family's intentions (living far away, designation of a proxy decision-maker, participation in nursing policies, disagreements among family members, and other related information). The system manages a wide range of patient information, including: information on reducing mental stress and prioritizing quality of life; legal information (adult guardians, arbitrary guardianship agreements, executors of wills, medical consent forms, prior instructions, declarations of dignified death, designation of acting decision-makers, methods of medical expense payment, and utilization of long-term care insurance); care plans (bathing frequency, rehabilitation frequency, medication management, dietary content, excretion care, support for outings, recreational participation, nursing visits, doctor visits, short-term care, day care, and night care systems); vital sign data (blood pressure, pulse, temperature, blood glucose, oxygen saturation, weight, respiratory rate, ECG data, sleep duration, and activity level); motion data from wearable devices (steps, fall detection, sleep duration, activity level, heart rate variability, location information, and emergency notification button press history); and input information from caregivers or family members (care records, logs, communication items, changes in the individual's status or emotions, new hopes or anxieties, staff opinions, and comments from medical professionals). This data is stored in the patient information management department on the server. Specifically, in this step, the system executes a distributed data collection agent on each terminal. This data collection agent asynchronously acquires text data input via the user interface and sensor data received wirelessly from wearable devices. The agent temporarily buffers the raw data in local memory and uses a data compression algorithm to reduce the payload size. The agent then sends the compressed data to the client information management department on the server side via a secure communication channel. The client information management department decompresses the received data and performs data pattern verification. It applies natural language processing to the text data, extracting keywords and linking entities. Finally, it applies a denoising filter to the time-series vital sign data and calculates feature vectors.For example, the input to this system includes the text "Loss appetite today" entered by a caregiver and a heart rate array at 1-minute intervals from a wearable device. The system outputs the extracted entity label "Loss appetite" and a vector of statistical features including the mean or variance of the heart rate. The system inserts the structured output data and feature vector into corresponding tables in a relational database and a vector database. This system constructs a data pipeline that seamlessly integrates manual data input from humans with automated collection from IoT devices. This prevents data loss and ensures the system remains up-to-date. The system improves the comprehensiveness of data collection and the quality of input data in subsequent analytical processing. This system functions as a unified information management foundation in specific application areas such as large-scale nursing facility networks or wide-area collaborative systems for home healthcare.

[0089] Step 2: Detection and Data Linkage of Condition Changes. The system automatically detects abnormal or routine changes such as falls, fever, abnormal vital signs, wandering, aspiration, medication errors, bowel movement problems, and pressure sores from wearable terminals, nursing record systems, electronic medical record systems, and health management services. These changes are manually input by caregivers or family members via the terminal. The data linkage department provides this information in real-time to the patient information management department and the condition analysis department. Specifically, in this step, the data linkage department executes an event-driven stream processing engine. This stream processing engine stores time-series data continuously flowing in from multiple data sources in a window buffer in memory. It uses a sliding window approach to apply real-time summarization and pattern matching processing to the streaming data within a certain time range. The engine utilizes machine learning models based on isolated forests or autoencoders as anomaly detection algorithms. It calculates whether the input data points deviate statistically significantly from the normal distribution. For example, the input to this stream processing engine might include continuous tensor data from a 3-axis accelerometer sent per second from the wearable terminal. This stream processing engine outputs anomaly scores for each data point, along with event trigger signals for detected anomalies. When the anomaly score exceeds a pre-set threshold, the engine immediately sends an interrupt request to the condition analysis unit and transmits a relevant data snapshot. Instead of periodic data verification based on batch processing, this system performs continuous stream analysis in memory. This allows the system to detect critical events such as falls or sudden changes in vital signs with millisecond-level latency. The system maximizes the real-time performance of anomaly detection, preventing the escalation of illness due to delayed responses. This system functions as a life-safety monitoring system in specific application areas such as patient monitoring in intensive care units and care services for elderly people living alone.

[0090] Step 3: Generative AI-based Situation Analysis and Decision Support Prompt Generation. The Situation Analysis Department comprehensively references medical records, nursing records, lifestyle habits, values, family intentions, legal information, nursing plans, vital sign data, action data, and nursing records stored in the Patient Information Management Department to generate prompts for situation analysis and decision support using generative AI. Specific examples of prompts could include: "An 85-year-old female, requiring a nursing level of 2, with a history of hypertension and osteoporosis, was detected to have fallen during nighttime toilet care. She intends to avoid hospitalization. She has two previous fall records, neither of which resulted in serious injury. Her current vital signs are stable. Her family lives far away. Regarding this fall, considering her values ​​and her family's intentions, explain the best coping strategy and its rationale in natural language." Another example is: "The patient has a fever and a history of diabetes and heart disease. She does not wish for aggressive treatment. Her family wants her to seek medical attention. What is the best coping strategy? Explain the options and their rationale." Specifically, in this step, the situation analysis department executes a knowledge graph-based context extraction module. This module, triggered by a detected anomalous event, queries the database of the patient information management department. It explores the graph database, extracting nodes related to the patient's medical history, values, and recent vital signs. The extracted node group is then integrated to construct a multi-dimensional context tensor. The situation analysis department inputs this constructed context tensor into a Transformer model for prompt generation. This Transformer model uses an attention mechanism to weight information most relevant to the current anomalous event. For example, the input to the situation analysis department includes a nighttime fall as an anomalous event and attribute vectors extracted from the graph database, such as osteoporosis or a desire to avoid hospitalization. The situation analysis department outputs a natural language prompt string that logically constructs this information, along with a tag embedding matrix representing its internal structure. This output prompt string is then passed to the input interface of the large-scale language model in the subsequent nursing decision support department. This system does not employ simple template filling; instead, it utilizes a dynamic context-building algorithm based on knowledge graphs. This allows the system to filter out only the truly needed information from massive amounts of personal data, identifying the current situation and providing suggestions to the AI. This results in the system overcoming the limitations of the context window in large-scale language models, improving the accuracy of inference and processing efficiency. The system achieves a high level of situational understanding in specific application areas such as emergency triage and transport of patients with complex medical histories or personalized medical consultation systems.

[0091] Step 4: Prompts and explanations of optimal nursing policies and procedures based on generative AI. The Nursing Decision Support Department executes an ethical decision-making algorithm based on generative AI, comprehensively considering the individual's values ​​or wishes, family's intentions, nursing plans, the situation at the medical site, and social ethical regulations, to generate optimal nursing policies or procedures. These include prompts for immediate contact with family and prioritizing observation within the facility in case of falls; contacting medical institutions and ensuring thorough hydration and rest in case of fever; providing first aid and seeking medical attention as needed in case of aspiration during eating; ensuring safety and reporting the situation to family when the individual wanders around at night; enhancing home care if the individual does not wish to be hospitalized; adjusting medication to prioritize pain relief; avoiding specific treatments for religious reasons; rapid medical intervention and reassessment of the nursing plan in case of pressure ulcers; immediate reporting to the doctor and proposals for measures to prevent recurrence in case of medication errors; and enhancing excretion care and explaining to family members in case of frequent excretion problems. The rationale or justification for each of these options is explained transparently in natural language. Specific examples of prompts could include: "Since you have registered your intention to avoid hospitalization, it is recommended that you coordinate with your family and undergo observation at home or in a facility," "In cases where there is a high risk of fracture or head injury due to a fall, it is recommended that you seek immediate medical attention," and "Due to the confirmed fever, considering the risk of dehydration or infection, it is recommended that you contact a medical facility and replenish fluids." Specifically, in this step, the Nursing Decision Support Department executes a finely tuned large-scale language model tailored to medical and nursing specializations. This large-scale language model consists of a deep neural network with billions of parameters. The Nursing Decision Support Department tokenizes the prompt string generated in step 3 and feeds it into the model's input layer. The large-scale language model computes a hidden state vector for the context of the input using an encoder and decoder architecture. The large-scale language model uses a beam search algorithm to autoregressively generate the sequence of words with the highest probability. The Nursing Decision Support Department applies a rule-based safety filter to the generated text to verify whether it contains medically inappropriate instructions. For example, the input to the Nursing Decision Support Department might be a prompt tensor containing conditions such as fall detection, stable vital signs, and the desire to avoid hospitalization. As output, this Nursing Decision Support Department outputs the action tag "Recommendation for observation within the facility" along with natural language text explaining its medical and ethical basis. The Department packages the output action tag and explanatory text as structured data and sends it to the Interface Department. This system performs complex trade-offs assessments, using human experts as tacit knowledge, as an optimization problem of probability distributions in a high-dimensional space on a computer. This system thus instantly derives an objective decision that balances individual values ​​with medical appropriateness. This system achieves the technical effect of eliminating biases from individualized judgments and pursuing individualized optimization while standardizing the quality of nursing care.This system complements expert insights in specific application areas such as automation of nighttime standby response in nursing facilities or educational support systems for new nursing staff.

[0092] Step 5: Dialogue and Response to User, Family Members, and Caregivers. Through the interface, the user, their family member, or caregiver confirms the options or explanations provided by the generative AI on the terminal, and can input additional questions or express their wishes using natural language. For example, regarding questions such as "Does this fall require hospitalization?", "What are the preventative measures to avoid recurrence?", "What would be the treatment plan if I were diagnosed with dementia?", "What preparations should I make if I wish to face death at home?", "What are the countermeasures for frequent nighttime wanderings?", "What should I do if my bedsores worsen?", and "What are the preventative measures to avoid recurrence of repeated medication errors?", the generative AI provides easily understandable explanations of medical and nursing options or legal procedures. The input is immediately reflected in the patient information management department, dynamically updating the pre-arranged instructions or care plan. Specifically, in this step, the interface executes a stateful dialogue management module. This dialogue management module maintains a series of interactions with the user as a session in memory, preserving contextual continuity. The interface receives the natural language question text sent from the user's terminal and inputs it into the intent classification neural network. This intent classification neural network determines the intent of the question and outputs a corresponding intent vector. The interface combines this intent vector with a tensor of past dialogue history and inputs it into a language model for response generation. For example, the input to this interface includes the text string "Does this fall require hospitalization?" from the user and the current conversation state vector. The output from this interface is the response text "Hospitalization is not required because vital signs are stable," along with an emotional score estimating the user's anxiety level. When the user inputs a new intent, the interface uses an information extraction algorithm to structure its content, automatically generating and executing a database update query for the nursing care information management department. This system is not a simple question-and-answer chatbot, but a dynamic dialogue system that understands context and updates its state. This allows users to interact with complex systems through natural dialogue. The system lowers the psychological barrier for users and accurately reflects potential hopes or values ​​within the system. This system enables smooth information sharing in specific application areas such as patient-doctor communication support tools or status reporting systems to distant family members.

[0093] Step 6: Scenario Simulation and Pre-event Preparation Based on Generative AI. The simulation department uses generative AI to virtually simulate various scenarios, including falling in the toilet at night, developing a high fever, accidentally swallowing food, wandering around at night, being unable to contact family members who are far away, having limited hospital access during a novel infectious disease outbreak, needing to receive care in a shelter during a disaster, experiencing a cluster infection in a nursing facility, wishing to die at home, worsening bedsores, and repeated medication errors. For each scenario, natural language prompts are provided to guide users, their families, and caregivers to consider the decision-making outcomes and impacts based on their values, desires, and care plans. This allows users, their families, and caregivers to experience multiple situations in advance, prepare for or reassess decisions, and automatically update pre-event instructions, personal information, and care plans based on the simulation results. Specific examples of prompts could include: "In the case of a person falling in the toilet at night, consider past fall history, presence of osteoporosis, family's wishes, and care plan, and explain the best response and its rationale," or "In the case of a high fever, consider pre-existing conditions, family's hopes, and coordination with medical institutions, and explain the response plan and its rationale." Specifically, in this step, the simulation department executes a parameterized scenario generation engine. This engine reads the current health status vector and environmental variable tensor of the patient as initial conditions. Based on statistical distributions extracted from a past event database, the simulation department constructs a probabilistic model of future possible events. Using the Markov chain Monte Carlo method, the simulation department generates thousands of virtual time-series scenarios on a parallel computing cluster. In each scenario, the simulation department calls the inference model of the nursing decision support department to calculate the best response under the virtual situation. For example, the input to the simulation department includes the current level of care required, a vector of pre-existing conditions, and the virtual event of a high fever. This simulation unit outputs recommended coping strategies, predicted probabilities of health status deterioration, and a list of necessary pre-emptive preparations for simulated events. It then summarizes the simulation results, generates revised preventative care plans for high-risk scenarios, and feeds this information back to the patient information management department. This system does not perform predictions based on static rules, but rather dynamic future predictions that combine probabilistic simulations and deep learning model inferences. This allows for a comprehensive assessment of various risk scenarios that exceed human imagination. The system enables a shift from reactive to preventative care and mathematically reduces the incidence of major incidents. In specific application areas such as long-term care insurance benefit optimization simulations and facility operation risk management systems, this system supports data-driven strategic planning.

[0094] Step 7: Security Management and Access Control. The Security Management Department is responsible for the security and privacy protection of the entire system. It possesses functions such as encrypted communication, multi-factor authentication, access control, audit log management, data backup, information leakage detection, access history recording, terminal authentication, and access control for each user. Appropriate information disclosure scopes are set according to the user and their family, caregivers, medical professionals, and other relevant personnel. An individual's medical history and values ​​are only accessible to the individual and their family, and information is only disclosed to legal representatives and medical professionals within necessary limits, achieving strict information management. Specifically, in this step, the Security Management Department executes a dynamic policy evaluation engine. This policy evaluation engine functions as a proxy to intercept all data access requests within the system. The Security Management Department generates a context vector containing the IP address and device fingerprint of the access request source, the user's role, and the confidentiality level of the requested data. The Security Management Department inputs the context vector into a machine learning-based anomaly detection model to score the legitimacy of the access request in real time. The Security Management Department only allows queries to be executed on the encrypted database if the score meets a threshold. For example, the input to the Security Management Department may include access request packets from an external network and authentication tokens. This security management department outputs access permission flags and transaction records for auditing. These transaction records are asynchronously written to a tamper-proof distributed ledger. Unlike existing methods that rely on static password authentication, this system employs a zero-trust model that evaluates the context of access in multiple dimensions. This effectively blocks unauthorized access attempts by attackers who have stolen legitimate users' authentication information. This system maximizes security strength and ensures system reliability when handling highly confidential medical information. It provides a secure data flow foundation for specific applications such as wide-area medical information networks or national-level health information management platforms.

[0095] (specific use case) For example, imagine a patient admitted to an elderly care facility with a care need level of 2. This patient has pre-existing conditions including hypertension and osteoporosis, and has a history of two falls. Their values, such as "avoiding hospitalization," "prioritizing pain relief," and "valuing meetings with family," are registered in their pre-admission instructions. Family members live far away, making daily meetings difficult, but the patient is willing to actively participate in important decisions. Wearable devices in the facility continuously monitor the patient's walking status and vital signs (blood pressure, pulse, temperature, steps, sleep duration, etc.). Caregivers input daily care records or the patient's condition, the status of dietary or excretory assistance, rehabilitation progress, and the availability of nighttime toilet assistance into the device. One night, a fall detection sensor activates on the way to the toilet, and the patient is found slumped on the floor. The caregiver inputs "Fall detected during nighttime toilet assistance; patient is conscious, vital signs are stable, no external injury" into the device, and the latest vital sign data from the wearable device is automatically updated. The situation analysis department comprehensively considers an individual's past fall history, presence of osteoporosis, vital sign data, family wishes, care plans, the individual's "hospitalization avoidance" values, nighttime activity patterns, and recent rehabilitation records to generate the following prompts using generative AI. Specifically, in this particular use case, the system operates in an edge cloud architecture where multiple sensor nodes collaborate with a central server. The wearable terminal in this facility functions as an edge device equipped with a 3-axis accelerometer and an optical heart rate sensor. The wearable terminal processes the raw data acquired at a sampling rate of 100 times per second on its local microcontroller, extracting feature vectors. The wearable terminal continuously streams the extracted feature vectors to the server via a low-power wide-area network. The situation analysis department on the server buffers the received streaming data into an in-memory database and continuously executes anomaly detection algorithms. The situation analysis department detects abrupt changes in acceleration data and spatial displacement, and sets fall event flags. For example, the input to this situation analysis unit includes time-series arrays of acceleration tensors and heart rates from a wearable terminal, as well as text data indicating the patient is conscious and without injury, input by a caregiver from the terminal. As output, the situation analysis unit outputs a definitive signal of the fall event and a context matrix integrating relevant personal data. The situation analysis unit then inputs the output context matrix into a natural language generation module to dynamically construct query prompts for a large-scale language model. This system integrates physical signals from hardware sensors and semantic text information from human input in the same vector space. This allows the system to correct sensor false detections through human input, significantly improving the accuracy of situation recognition. The system achieves the technical effect of minimizing the delay from data collection to prompt generation, enabling rapid initial response in emergencies. This system enables advanced situation monitoring in specific applications such as fall prevention systems in smart nursing facilities or hospitals.

[0096] Example prompt: "An elderly woman with a nursing need level of 2, whose past medical conditions include hypertension and osteoporosis, was detected to have fallen during nighttime toilet assistance. She intends to avoid hospitalization, prioritizes pain relief, and values ​​meetings with her family. She has a history of two falls, neither of which resulted in serious injury. Her current vital signs are stable, and she has no external injuries. Her family lives far away. Regarding this fall, considering her values ​​and her family's wishes, explain the best coping strategy and its rationale in natural language." Specifically, this system performs encoding processing to convert the prompt string into a numerical representation that can be processed by a large-scale language model. Referring to a pre-learned vocabulary dictionary, the system maps each word and phrase within the prompt to an array of token IDs. The system converts the token ID array into a continuous embedding vector in a high-dimensional space. The system further adds a positional encoding vector representing the positional information of each token within the sequence. For example, the input to this system is the entire prompt string described above. The system outputs a floating-point tensor with the sequence length and a dimension of 1024. This system inputs the output tensor into the self-attention mechanism block of the Transformer architecture. It doesn't simply pass strings; instead, it maps the semantics of natural language to a mathematical vector space. This enables large-scale language models to quantitatively calculate complex relationships between concepts such as a nursing need degree of 2, hypertension, and hospitalization avoidance. The system achieves the technical effect of accurately conveying the intent of prompt words into the model and maximizing the accuracy of inferences. In specific applications such as medical text parsing and automated response systems with complex conditional specifications, this system serves as the foundation for natural language interfaces.

[0097] The generative AI receives the prompt and first assesses the risk of fracture or head injury. If vital signs are stable and no injury is found, it proposes "Contact family immediately and prioritize observation within the facility." If the individual reports pain, it prioritizes pain relief and suggests contacting a doctor to discuss analgesics if necessary. If abnormal vital signs or head injury are suspected, it recommends "Emergency transport." The AI ​​outputs statements such as "Since the individual has registered their intention to avoid hospitalization, it is recommended to contact family and conduct observation within the facility," "In cases of high risk of fracture or head injury due to a fall, it is recommended to seek immediate medical attention," and "In cases of severe pain, contact a doctor to discuss analgesics," providing the basis or reasoning for each option in natural language. Specifically, this generative AI performs inference operations using a multi-layered Transformer decoder block based on the input prompt tensor. Through a self-attention mechanism, this generative AI calculates the attention weight between the current state of stable vital signs or no injury within the prompt and the value vector of avoiding hospitalization or relieving pain. This generative AI, based on computed weights, outputs a probability distribution for multiple action candidates from the Softmax layer. It selects the action candidate with the highest probability and autoregressively generates word tokens to describe each candidate. For example, the input to this generative AI is the embedding tensor of the cue words and the model's internal parameters. The output of this generative AI is a probability score for each action candidate and a token list of generated natural language text. The output token list is converted into a string using a decoder and sent to the interface. This generative AI treats human intuition as a probabilistic optimization process in a high-dimensional space using billions of parameters. It logically reconciles opposing conditions to derive the optimal solution. This generative AI achieves the technical effect of not black-boxing the decision-making process and generating highly transparent outputs with accompanying evidence. In specific applications such as the automatic generation of clinical pathways in the medical field or supporting the decision-making of nursing policies with ethical dilemmas, this generative AI simulates the expert's thought process.

[0098] Furthermore, when family members input questions such as "Does this fall require hospitalization?", "What are the future measures to prevent recurrence?", and "What safety measures should be taken for nighttime toilet access?" into the terminal, the generative AI returns specific explanations such as "No external injuries were found in this fall, and vital signs are stable, therefore hospitalization is not required. As measures to prevent recurrence, enhanced lighting for nighttime toilet access, the installation of fall prevention mats, the use of walking aids, and the strengthening of nighttime care systems are effective." Specifically, this system processes the question text input from the family member's terminal through a dialogue management module. This dialogue management module combines the context vector of the immediate preceding fall event with the embedding vector of the new question text. The system inputs the combined vector into a large-scale language model, enabling it to perform context-abiding response generation. This large-scale language model references an internal knowledge base and facility environment data obtained from the patient information management department. For example, the input to this system includes the question text "What safety measures should be taken for nighttime toilet access?" and a facility equipment information tensor. This system outputs structured data containing specific countermeasures such as enhanced lighting or fall prevention mat installation, along with text explaining this structured data in natural language. The system sends the output response text to the family member's terminal, displaying it on the screen in a chat-like format. This system is not a single-response question; instead, it performs multi-turn dialogue processing internally, linking past events with the current situation. This eliminates the need for the user to explain the background of the situation each time, achieving natural communication. This system delivers a significant improvement in information retrieval efficiency and reduces the user's cognitive load. This system functions as a foundation for interactive information provision in specific application areas such as remote nursing consultation systems or personalized healthcare assistants.

[0099] Furthermore, the system uses generative AI to simulate various scenarios such as "nighttime falls," "high fever," "aspiration of food," "nighttime wandering," "bedsores," and "medication errors." This allows for pre-emptive virtual experiences of scenarios like "the procedure for choosing emergency transport after a nighttime fall," "the response to contacting medical institutions when experiencing a high fever," "emergency treatment and contacting family members in case of aspiration of food," "risk assessment and additional countermeasures in cases of frequent nighttime wandering," "medical coordination and reassessment of care plans in cases of worsening bedsores," and "prevention of recurrence in cases of medication errors." This allows individuals, family members, and caregivers to be well-prepared and reach a consensus in anticipation of emergencies. Specifically, the system utilizes a distributed computing cluster to process these diverse scenarios in parallel. The simulation unit defines each scenario as an independent computational task and assigns it to multiple computing nodes. Each computing node sets the initial condition tensor for each scenario and performs state transition simulations based on Markov decision processes. At each time step, the computing node calls the inference API of a large-scale language model to calculate the optimal course of action under the virtual conditions. This simulation unit aggregates computational results from all nodes to construct a result tree for each scenario. For example, input to this simulation unit includes the baseline health status vector of the patient and parameters for the virtual event of medication error recurrence. As output, this simulation unit outputs the probability distribution of the impact of medication error on health and an effectiveness score for relapse prevention measures. The simulation unit sends the output result tree to the interface unit, generating a dashboard that allows users to interactively explore the scenario. This system simulates medical or nursing decisions at high speed in a secure sandbox environment on a computer, where trial and error are not permitted in the real world. This allows the system to pre-validate its responsiveness to low-probability but high-impact black swan events. This system leverages data-driven optimization to improve preparedness for unknown risks and enhance the robustness of care plans. This system functions as a highly predictive and analytical platform in specific applications such as training simulators for healthcare professionals or stress testing systems for facility operations.

[0100] In this way, the system of the present invention comprehensively manages diverse information about the patients, uses generative AI to analyze the situation in real time, suggests the best nursing policies or procedures, explains the basis or rationale in natural language, supports convincing decisions based on the patient's, family's, and caregiver's trust, and enables pre-preparation or dynamic updates of nursing plans through scenario simulation. Specifically, the system runs on a distributed computing environment. The system integrates and manages diverse information about the patients as a multidimensional tensor. The data integration unit acquires time-series arrays of vital signs, natural language text of electronic medical records, and numerical vectors of environmental sensors. The data integration unit preprocesses the acquired data through normalization and tokenization. The system converts the preprocessed data into a 512-dimensional input vector. The system inputs the input vector into a large-scale language model. This large-scale language model employs a Transformer architecture with a self-attention mechanism. This large-scale language model computes nonlinear correlations of the input data in a high-dimensional feature space. This computation is a purely mathematical matrix operation, different from human mental activity. The large-scale language model receives, for example, a time-series array representing heart rate variations over the past 72 hours as input data. This large-scale language model takes as input data, for example, a text token column representing past fall history. This large-scale language model takes as input data, for example, a numerical vector representing current room temperature and humidity. Based on this input data, this large-scale language model infers the optimal nursing policy. As a result of the inference, this large-scale language model outputs a probability distribution representing the recommendation level of each nursing step. As a result of the inference, this large-scale language model outputs an attention weight score as a basis. As a result of the inference, this large-scale language model outputs a token column of explanatory text based on natural language. This output data, for example, includes a probability value of 0.95 for the nursing step of immediate hydration. This output data, for example, includes a probability value of 0.05 for the nursing step of emergency evacuation. This output data, for example, includes a contribution score of 0.88 for abnormal heart rate values. This system performs threshold determination processing on the output probability distribution. This system extracts nursing steps with probability values ​​exceeding a predetermined threshold. This system dynamically updates the nursing plan database using the extracted nursing steps. This system generates explanatory text based on natural language based on the attention weight score. This system sends the generated explanatory text to the display module of the terminal. This system features a scenario simulation department. This department uses the Monte Carlo method to generate multiple virtual future state vectors. These vectors are then fed back into a large-scale language model. The system uses this re-input processing to predict future changes in care plans. The predictions are then rendered as a time-series graph. These processes significantly reduce data processing latency. The system performs multivariate analysis that surpasses the limits of human cognition. This results in a dramatic improvement in the accuracy of care policy decisions.This large-scale language model is pre-learned using a large-scale medical and nursing corpus. The learning process updates the network's weight parameters in a manner that minimizes the cross-entropy error function. The learning process employs the gradient descent algorithm. Through these technical structures, the system achieves not merely operational efficiency improvements, but rather improvements in the computer technology itself. The system employs a hybrid architecture combining specific rule-based processing and machine learning models. This prevents black-box operations and ensures transparency in inference. The system demonstrates exceptional technical effectiveness in the specific use case of real-time decision support in medical settings or nursing facilities.

[0101] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0102] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0103] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0104] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0105] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0106] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0107] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0108] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0109] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0110] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0111] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0112] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0113] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0114] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0115] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0116] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0117] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0118] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0119] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0120] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0121] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0122] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0123] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0124] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0125] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0126] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0127] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0128] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0129] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0130] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0131] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0132] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0133] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0134] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0135] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0136] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0137] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0138] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0139] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0140] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0141] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0142] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0143] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0144] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0145] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0146] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0147] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0148] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0149] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0150] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0151] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0152] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0153] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0154] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0155] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0156] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0157] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0158] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0159] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0160] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The robot's emotions can be determined by the emotion-specific model 59. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0161] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0162] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0163] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0164] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0165] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0166] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, nearby sentiment values ​​are similar to each other. Figure 10 Examples show how emotions such as "peace of mind," "stability," and "reassurance" can result in similar emotional values.

[0167] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0168] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0169] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0170] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0171] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0172] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0173] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0174] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0175] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0176] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0177] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0178] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0179] In addition, the following notes further explain the above explanation.

[0180] (Note 1) A system characterized by comprising a patient information management department, a condition analysis department, and a nursing decision support department. The patient information management department has the following functions: acquiring various personal data from external service or nursing site terminals, including the patient's medical history, nursing history, lifestyle habits, values, family intentions, legal information, nursing plans, pre-arranged instructions, vital sign data, action data from wearable terminals, and input information from nursing staff or family members; comprehensively and dynamically managing this data; and updating it in real time based on changes in the condition or new inputs. The condition analysis department uses generative AI to analyze data during emergencies or routine situations. The nursing care department analyzes data on patient conditions such as falls, fever, abnormal vital signs, changes in behavior, nursing records, and reports from family members or staff to determine the urgency or necessary response. The nursing decision support department, based on information obtained from the patient information management department and the condition analysis department, comprehensively considers the patient's values ​​or desires, family intentions, nursing plans, the medical environment, and social ethical regulations. It uses a generative AI-based ethical decision-making algorithm to propose the best nursing policy or procedure, and transparently explains the basis or rationale for the option in natural language, supporting decisions based on the understanding and trust of nursing staff or family members.

[0181] (Note 2) As described in Appendix 1, the system is characterized in that the nursing decision support unit, through a natural language dialogue interface based on generative AI, explains in an easy-to-understand way to the patient, their family, and nursing staff the complex options, nursing procedures, and legal formalities involved in emergency or routine nursing decisions such as falls or fever. This allows the patient, their family, or nursing staff to input their wishes or desires in a dialogue format, and the input is immediately reflected in the patient information management unit to dynamically update the pre-instruction document or nursing plan. When the situation analysis unit detects new changes in the situation, the nursing decision support unit re-proposes the best option through generative AI.

[0182] (Note 3) As described in Appendix 1, the system is characterized in that the nursing decision support unit virtually simulates various emergency situations or nursing and medical scenarios through generative AI, and provides natural language prompts to the patient, their family, and nursing staff regarding the decision-making results and impacts of each scenario based on the patient's values ​​or hopes and the nursing plan. This allows the patient, their family, and nursing staff to prepare for or reassess decisions while experiencing multiple situations in advance, and automatically updates the pre-instruction documents or personal information and nursing plans based on the simulation results.

Claims

1. A system, characterized in that, The system comprises a Personal Information Management Department, a Situation Analysis Department, and a Decision Support Department. The Personal Information Management Department has the following functions: acquiring various personal data from external services or communication terminals, including the user's medical history, lifestyle habits, values, family relationships, legal information, and pre-instruction documents; comprehensively and dynamically managing this data and updating it in real time according to changes in the situation; the Situation Analysis Department uses generative AI to analyze the user's health status, activity data, communication history, and information from the medical scene during emergencies to determine the urgency or necessary response; and the Decision Support Department, based on information obtained from the Personal Information Management Department and the Situation Analysis Department, comprehensively considers the user's values ​​or wishes, the family's intentions, the situation at the medical scene, and social ethical regulations, using a generative AI-based ethical decision-making algorithm to propose the best option, and transparently explaining the basis or reasoning for the option in natural language, supporting decisions based on the understanding and trust of the user or their family.

2. The system as described in claim 1, characterized in that, The decision support department uses a generative AI-based natural language dialogue interface to explain complex options or legal documents related to medical care and nursing in an easy-to-understand way to the user and their family. This allows the user or their family to input their wishes or desires in a dialogue format, and the input is immediately reflected in the personal information management department to dynamically update the pre-instruction. When the situation analysis department detects new changes in the situation, the decision support department uses generative AI to propose the best option again.

3. The system as described in claim 1, characterized in that, The decision support unit uses generative AI to virtually simulate various emergency situations or medical and nursing scenarios, and provides users and their families with natural language prompts for the decision-making outcomes and their impact based on their values ​​or desired outcomes in each scenario. This allows users or their families to prepare for or reassess decisions while experiencing multiple situations in advance, and automatically updates pre-instructions or personal information based on the simulation results.

Images