Automated empathetic adjustment of decisions in artificial intelligence (AI) models

Abstract

To provide automated empathetic reconciliation of decisions of artificial intelligence (AI) models.SOLUTION: In an embodiment, operations include receiving first information associated with a first person and a first request of the first person to one or more institutions. A set of attributes of the first person is extracted and used to construct a causal model. The causal model represents causal relationships amongst attributes of the set of attributes. For the first person, a utility function associated with each of a plurality of AI models associated with the one or more institutions is determined. The utility function is determined based on the causal model, a first set of empathy criteria associated with the first person, and a second set of empathy criteria associated with each of the one or more institutions. For the utility function, optimin-point information is determined to reconcile a plurality of decisions taken by the plurality of AI models for the first request. The reconciled decision is rendered.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The embodiments discussed in this disclosure relate to automated empathetic adjustment of decisions made by artificial intelligence (AI) models. [Background technology]

[0002] Advances in the fields of machine learning and artificial intelligence (AI) have led to the development of numerous machine learning and AI-based decision-making tools. Such tools may be used by various agencies to make decisions related to evaluating various types of applications associated with individuals. Examples of such applications, though not limited to loan applications, may include loan applications. However, certain machine learning and AI-based tools may be biased in their decision-making. For example, certain tools may be biased based on a person's gender (or age, race, or caste).

[0003] The subject matter claimed in this disclosure is not limited to embodiments that resolve any shortcomings or operate only in the environments described above. Rather, this background art is provided solely to illustrate one exemplary technical field in which some of the embodiments described in this disclosure may be implemented. [Overview of the project]

[0004] According to an aspect of one embodiment, the method may include a set of actions, which may include receiving first information associated with a first person. The first information may further include a first request of the first person to one or more institutions. The set of actions may further include extracting a set of attributes associated with the first person from the received first information. The set of attributes may include at least one of a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. The set of actions may further include constructing a causal model based on the extracted set of attributes associated with the first person. The constructed causal model may represent a set of causal relationships between attributes from the extracted set of attributes. The set of actions may further include determining a utility function for the first person associated with each of several artificial intelligence (AI) models associated with one or more institutions. The determination of the utility function can be based on a constructed causal model, a first set of empathy criteria associated with a first person, and a second set of empathy criteria associated with each of one or more institutions. The set of actions may further include determining the optimism point information of the determined utility function associated with each of the multiple AI models. The determined optimism point information can correspond to a coordinated judgment from multiple judgments made by multiple AI models of one or more institutions for a first request associated with a first person. The set of actions may further include presenting the coordinated judgment.

[0005] The objectives and advantages of the embodiments are realized and achieved by at least the elements, features, and combinations specifically indicated in the claims.

[0006] Both the general description above and the detailed description below are given as examples and are illustrative, not limiting to the claimed invention. [Brief explanation of the drawing]

[0007] Exemplary embodiments are described and illustrated with further specificity and detail through the use of the accompanying drawings. [Figure 1] FIG. 1 is a diagram showing an exemplary environment related to automated empathetic adjustment of artificial intelligence (AI) model judgments. [Figure 2] FIG. 2 is a block diagram showing an exemplary electronic device for automated empathetic adjustment of artificial intelligence (AI) model judgments. [Figure 3] FIG. 3 is a flowchart showing an exemplary method for automated empathetic adjustment of artificial intelligence (AI) model judgments. <000009​​​​​​​​​​​​​​​Some embodiments described in this disclosure relate to methods and systems for automated empathic reconciliation of decisions made by artificial intelligence (AI) models. In this disclosure, first information associated with a first person may be received. The first information may be further associated with first requests made by the first person to one or more institutions. From the received first information, a set of attributes associated with the first person may be extracted. The set of attributes may include at least one of a set of need-related attributes, a set of protected attributes, or a set of outcome-related attributes associated with the first person. A causal model may then be constructed based on the extracted set of attributes associated with the first person. The constructed causal model may represent a set of causal relationships between attributes from the extracted set of attributes. For the first person, a utility function may be determined associated with each of several artificial intelligence (AI) models. The several AI models may be associated with one or more institutions. The determination of the utility function can be based on a constructed causal model, a first set of empathy criteria associated with a first person, and a second set of empathy criteria associated with each of one or more institutions. Furthermore, for the determined utility function associated with each of the multiple AI models, optimism-point information can be determined. The determined optimism-point information can correspond to a refined judgment from multiple judgments made by multiple AI models of one or more institutions in response to a first request associated with a first person. Furthermore, the refined judgment can be rendered.

[0010] According to one or more embodiments of this disclosure, the technical field of automated determination of human requests to one or more agencies can be improved by configuring a computing system such that the computing system may be able to empathetically adjust the decisions made by multiple AI models of one or more agencies regarding a request with minimal bias. The computing system can determine a utility function for each of the multiple AI models that may be used by one or more agencies for decision-making. The utility function may be determined based on a causal model, a first empathy criterion associated with a person, and a second empathy criterion associated with one or more agencies. The causal model may show causal relationships between different attributes associated with a person (e.g., a set of protective attributes). The utility function of a particular AI model may show the benefit that one or more agencies gain when that particular AI model makes a particular decision, e.g., a decision to accept a human request. The utility function may take into account fairness to a person based on a set of protective attributes of the person determined based on the causal model. Furthermore, the utility function can incorporate empathy for a person making a request to one or more institutions (i.e., based on a first set of empathy criteria) and for one or more institutions determining the request based on multiple judgments of the AI ​​models (i.e., based on a second set of empathy criteria). To reconcile multiple judgments of multiple AI models, the computing system can then determine the optimism point information of the utility function determined for each of the multiple AI models. The determined optimism point information can represent a reconciled judgment from multiple judgments made by the multiple AI models regarding a person's request. Thus, multiple judgments made by multiple AI models associated with one or more institutions can be reconciled into a single reconciled judgment on the request. The reconciled judgment can be fair and empathetic towards the person and empathetic towards one or more institutions.

[0011] The system may be configured to receive first information associated with a first person. The first information may further be associated with a first request of the first person to one or more institutions. Examples of one or more institutions, but not limited to, may include one of the following: a bank, an insurance company, an employment organization, an educational institution, a travel-related institution, or a judicial court. The first request of the first person, but not limited to, may include at least one of the following: a loan application, an insurance application, a job application, an admissions application, a visa application, or a judicial petition application. An example of the first information is further provided, for example, in Figure 4.

[0012] The system may further be configured to extract a set of attributes associated with a first person from the received first information. The set of attributes may include, but is not limited to, at least one of the following: a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. The set of needs-related attributes associated with the first person may correspond to the purpose of the first person's first request to one or more institutions. Furthermore, the set of protection attributes associated with the first person may include, but is not limited to, at least one of the following: age, gender, race, caste, salary, nationality, or disability associated with the first person. The set of outcome-related attributes associated with the first person may correspond to one or more outcomes associated with one or more decisions made by the AI ​​model of one or more institutions regarding the first request associated with the first person. Furthermore, the set of outcome-related attributes associated with the first person may also correspond to one or more long-term benefits associated with the first person. The extraction of the attribute set is further illustrated, for example, in Figures 3 and 4.

[0013] The system may further be configured to construct a causal model based on a set of extracted attributes associated with a first person. The constructed causal model can represent a set of causal relationships between attributes from the set of extracted attributes. The construction of the causal model may be based on at least one of user input or a structured learning model. An example of a causal model is provided, for example, in Figure 4.

[0014] The system may further be configured to determine a utility function associated with each of several AI models for a first person. The multiple AI models may be associated with one or more institutions. The determination of the utility function can be based on a constructed causal model, a first set of empathy criteria associated with the first person, and a second set of empathy criteria associated with each of the one or more institutions. The first set of empathy criteria associated with the first person may include, but is not limited to, a first criterion associated with one or more of the sets of protective attributes associated with the first person, a second criterion associated with one or more of the sets of needs-related attributes associated with the first person, and a third criterion associated with one or more of the sets of outcome-related attributes associated with the first person. The second set of empathy criteria associated with each of the one or more institutions may correspond to thresholds related to the cost associated with the acceptance of the first request provided by one or more institutions. The determination of the utility function is further illustrated, for example, in Figure 3.

[0015] The system may further be configured to determine the optimism point information of the determined utility function associated with each of the multiple AI models. The determined optimism point information can correspond to a coordinated decision from multiple decisions made by multiple AI models of one or more institutions to a first request associated with a first person. In this specification, the AI ​​model decision from multiple AI models may be one of acceptance or rejection of the first request of the first person given by the corresponding AI model of one or more institutions. The system may further be configured to present the coordinated decision. The determination of the optimism point information is further illustrated, for example, in Figures 3 and 5.

[0016] Typically, conventional systems may not incorporate empathy in their automated assessment of a person's requests to one or more institutions. Conventional systems may use only attributes that can be directly used to judge a person's request (e.g., loan application) (e.g., income, education level, etc.) without considering the person's needs, such as for higher education, or the consequences associated with accepting the request, such as a better quality of life. Furthermore, conventional systems may have biases regarding certain factors (e.g., a person's protective attributes), such as a person's age, gender, race, caste, salary, nationality, or disability. Thus, conventional systems may not be empathetic or fair to the person when it comes to automated evaluations of a person's requests to one or more institutions. Moreover, conventional systems may evaluate a person's requests based only on factors associated with the person, and not on factors associated with one or more institutions, thereby potentially being unempathetic towards one or more institutions in the decision-making process.

[0017] On the other hand, the disclosed system can empathetically adjust the decisions made by multiple AI models associated with one or more agencies regarding a person's (i.e., the First Person's) request (i.e., the First Request) with minimal bias. It can construct a causal model showing the causal relationships between different attributes associated with the First Person (e.g., a set of protection attributes, a set of needs-related attributes, a set of outcome-related attributes). Furthermore, a utility function can be determined for each of the multiple AI models used by one or more agencies for decision-making. The utility function may be determined based on the causal model, a first empathy criterion associated with the First Person, and a second empathy criterion associated with one or more agencies. The utility function of a particular AI model can show the benefit that one or more agencies gain when that particular AI model makes a particular decision, e.g., a decision to accept the First Person's First Request. Furthermore, the utility function can take into account fairness to the First Person based on a set of the First Person's protection attributes determined based on the causal model. Furthermore, the utility function can also incorporate empathy for a first person based on a first set of empathy criteria (e.g., a second set of criteria associated with a set of needs-related attributes, and a third set of criteria associated with a set of outcome-related attributes). Furthermore, the utility function can also incorporate empathy for one or more institutions based on a second set of empathy criteria (e.g., thresholds associated with the cost of accepting the first request of the first person). The disclosed system can further reconcile multiple judgments of multiple AI models based on determinations of the optimism point information of the utility function determined for each of the multiple AI models. The determined optimism point information can represent a reconciled judgment from multiple judgments made by the multiple AI models regarding the request for the first person. Thus, multiple judgments made by multiple AI models associated with one or more institutions can be reconciled into a single reconciled judgment on the first request. The reconciled judgment can be fair and empathetic towards the first person and can also be empathetic towards one or more institutions.

[0018] Embodiments of this disclosure will be described with reference to the accompanying drawings.

[0019] Figure 1 shows an exemplary environment relating to automated empathetic adjustment of decisions of an artificial intelligence (AI) model, arranged according to at least one embodiment described herein. Referring to Figure 1, Environment 100 is shown. Environment 100 may include an electronic device 102, a database 104, a user end device 106, and a communication network 108. The electronic device 102, the database 104, and the user end device 106 may be communicatively coupled to one another via the communication network 108. Figure 1 further shows a user 110 who is associated with or can operate the electronic device 102 or the user end device 106. Furthermore, first information 112A associated with a first person, first information 112B associated with a second person, and so on, up to first information 112N associated with an Nth person. The first piece of information 112A may include a set of attributes associated with a first person (for example, a set of needs-related attributes 114A, a set of protection attributes 116A, and a set of result-related attributes 118A). The first piece of information 112B may include a set of attributes associated with a second person (for example, a set of needs-related attributes 114B, a set of protection attributes 116B, and a set of result-related attributes 118B). Similarly, the first piece of information 112N may include a set of attributes associated with an Nth person (for example, a set of needs-related attributes 114N, a set of protection attributes 116N, and a set of result-related attributes 118N). The first piece of information 112A associated with a first person, the first piece of information 112B associated with a second person, and so on, up to the first piece of information 112N associated with an Nth person, may be stored in the database 104.

[0020] The electronic device 102 may include appropriate logic, circuits, interfaces, and / or code that can be configured to empathetically determine a person's request made to one or more institutions based on the automatic coordination of decisions of multiple AI models associated with one or more institutions. The electronic device 102 may be configured to receive first information 112A associated with a first person from a database 104. The first information may further be associated with a first request made by the first person to one or more institutions. The decisions of the multiple AI models may be either acceptance or rejection of the first request made by one or more institutions. Examples of one or more institutions may include, but are not limited to, a bank, an insurance company, an employment organization, an educational institution, a travel-related institution, or a judicial court. The first request made by the first person may include, but are not limited to, at least one of a loan application, an insurance application, a job application, an admission application, a visa application, or a judicial petition application. An example of the first information 112A is provided, for example, in Figure 4.

[0021] The electronic device 102 may be configured to extract a set of attributes associated with a first person from the received first information 112A. The set of attributes may include, but is not limited to, at least one of a set of needs-related attributes 114A, a set of protection attributes 116A, or a set of outcome-related attributes 118A associated with the first person. The set of needs-related attributes 114A associated with the first person may correspond to the purpose or needs of the first person's first request to one or more institutions. Furthermore, the set of protection attributes 116A associated with the first person may include, but is not limited to, at least one of age, gender, race, caste, salary, nationality, or physical disability associated with the first person. The set of outcome-related attributes 118A associated with the first person may correspond to one or more outcomes associated with one or more decisions made by an AI model of one or more institutions regarding the first request associated with the first person. Furthermore, the set of result-related attributes 118A associated with the first person can also correspond to one or more long-term benefits associated with the first person. The extraction of the attribute set is further illustrated, for example, in Figures 3 and 4.

[0022] The electronic device 102 may be configured to construct a causal model based on a set of extracted attributes associated with a first person. The constructed causal model can represent a set of causal relationships between attributes from the set of extracted attributes. The construction of the causal model may be based on at least one of user input or a structured learning model. An example of a causal model is provided, for example, in Figure 4.

[0023] The electronic device 102 may be configured to determine a utility function associated with each of several AI models for a first person. The multiple AI models may be associated with one or more institutions. The determination of the utility function may be based on a constructed causal model, a first set of empathy criteria associated with the first person, and a second set of empathy criteria associated with each of the one or more institutions. The first set of empathy criteria associated with the first person may include, but is not limited to, a first criterion associated with one or more of the set of protective attributes 116A associated with the first person, a second criterion associated with one or more of the set of needs-related attributes 114A associated with the first person, and a third criterion associated with one or more of the set of outcome-related attributes 118A associated with the first person. The second set of empathy criteria associated with each of the one or more institutions may correspond to thresholds related to the cost associated with the acceptance of the first person's first request by one or more institutions. The determination of the utility function is further explained, for example, in Figure 3.

[0024] The electronic device 102 may further be configured to determine the optimism point information of the determined utility function associated with each of the multiple AI models. The determined optimism point information can correspond to a coordinated decision from multiple decisions made by multiple AI models of one or more institutions to a first request associated with a first person. In this specification, the AI ​​model decision from multiple AI models may be one of the acceptance or rejection of the first request of the first person given by one or more institutions (or by corresponding AI models of one or more institutions). The electronic device 102 may further be configured to present the coordinated decision. The determination of the optimism point information is further illustrated, for example, in Figures 3 and 5.

[0025] Examples of electronic device 102 may include, but are not limited to, a job engine or machine, a loan valuation engine, a judicial machine, a mobile device, a desktop computer, a laptop, a computer workstation, a computing device, a mainframe machine, a server such as a cloud server, and a group of servers. In one or more embodiments, electronic device 102 may include a user end terminal device and a server communicatively coupled to the user end terminal device. Electronic device 102 may be implemented using hardware including a processor, a microprocessor (for example, one or more operations that perform or control the performance of such operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In some other examples, electronic device 102 may be implemented using a combination of hardware and software.

[0026] Database 104 may include appropriate logic, interfaces, and / or code that can be configured to store first information associated with multiple people. The first information about a person may include information associated with a set of attributes associated with that person. For example, database 104 may store first information 112A associated with a first person, first information 112B associated with a second person, and so on, up to first information 112N associated with an Nth person. Database 104 may be a relational database or a non-relational database. Furthermore, in some cases, database 104 may be stored on a server such as a cloud server, or it may be cached and stored on an electronic device 102. The server of database 104 may be configured to receive queries for first information (e.g., first information 112A) associated with a person (e.g., a first person) from the electronic device 102 via a communication network 108. In response, the database 104 server may be configured to retrieve the queried first information associated with a specific person based on the received query via the communication network 108 and provide it to the electronic device 102. In some embodiments, the database 104 may include multiple servers stored in different locations. Furthermore, or instead, the database 104 may be implemented using hardware including a processor, a microprocessor (e.g., one or more operations, or controlling the execution of such operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In some other examples, the database 104 may be implemented using a combination of hardware and software.

[0027] The user end device 106 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to generate or receive first information (e.g., first information 112A) of a person (e.g., a first person). For example, the user end device 106 may include web client software or email software that allows the user end device 106 to receive first information of a particular person. Furthermore or alternatively, the user end device 106 may include word or text processing software that allows first information of a particular person (e.g., a set of attributes) to be generated based on user input from user 110 or from that person themselves. The user end device 106 may upload the generated or received first information associated with a particular person to electronic device 102 for extraction of a set of attributes associated with that person. Furthermore, the user end device 106 may upload the generated or received first information to database 104 for storage.

[0028] The user end device 106 may further be configured to receive coordinated judgments from multiple AI models on a person's (e.g., a first person's) request (e.g., a first request) from the electronic device 102. The user end device 106 can present the received coordinated judgments on the person's request on the user end device 106's display screen for the user 110. In some embodiments, the user end device 106 can receive a query from the user 110 to determine a person's (e.g., a first person's) request (e.g., a first request) to one or more institutions and provide a judgment on the request (i.e., a coordinated judgment) based on the determination of the person's request. The user end device 106 can further send a query to the electronic device 102 for a decision on the coordinated judgments of the multiple AI models on this person's request. Examples of the user end device 106 may include, but are not limited to, mobile devices, desktop computers, laptops, computer workstations, computing devices, mainframe machines, servers such as cloud servers, and groups of servers. In Figure 1, the user end device 106 is separated from the electronic device 102, but in some embodiments, the user end device 106 may be integrated with the electronic device 102 without departing from the scope of this disclosure.

[0029] The communication network 108 may include a communication medium through which the electronic device 102 can communicate with a server and user end device 106 that can store a database 104. Examples of the communication network 108 include, but are not limited to, the Internet, a cloud network, a Wireless Fidelity (Wi-Fi) network, a Personal Area Network (PAN), a Local Area Network (LAN), and / or a Metropolitan Area Network (MAN). Various devices within the environment 100 may be configured to connect to the communication network 108 according to various wired and wireless communication protocols. Examples of such wired and wireless communication protocols include, but are not limited to, Transmission Control Protocol and Internet Protocol (TCP / IP), User Datagram Protocol (UDP), Hypertext Transfer Protocol (HTTP), File Transfer Protocol (FTP), ZigBee®, EDGE, IEEE 802.11, Light Fidelity (Li-Fi), 802.16, IEEE 802.11s, IEEE 802.11g, Multihop Communication, Wireless Access Point (AP), Device-to-Device Communication, Cellular Communication Protocol, and / or Bluetooth® (BT) Communication Protocol, or at least one of these combinations.

[0030] Figure 1 may be modified, added to, or omitted without departing from the scope of this disclosure. For example, the environment 100 may include more or fewer elements than those illustrated and described in this disclosure. For example, in some embodiments, the environment 100 may include the electronic device 102 but not the database 104 and the user end device 106. Furthermore, in some embodiments, the functionality of the database 104 and the user end device 106 may be incorporated into the electronic device 102 without departing from the scope of this disclosure.

[0031] Figure 2 is a block diagram of an exemplary electronic device for automated empathetic adjustment of decisions of an artificial intelligence (AI) model, arranged according to at least one embodiment described herein. Figure 2 is described in relation to the elements of Figure 1. Referring to Figure 2, a block diagram 200 of a system 202 including an electronic device 102 is shown. The electronic device 102 may include a processor 204, memory 206, persistent data storage 208, input / output (I / O) devices 210, a display screen 212, a network interface 214, and a plurality of artificial intelligence (AI) models 216 and causal models 218.

[0032] The processor 204 may include appropriate logic, circuitry, and / or interfaces that can be configured to execute program instructions associated with different operations to be performed by the electronic device 102. For example, some operations may include receiving first information 112A, extracting a set of attributes, constructing a causal model (e.g., causal model 218), determining a utility function, determining optimism point information, and presenting a coordinated decision. The processor 204 may include any appropriate general-purpose or general-purpose computer, computing entity, or processing device, including various computer hardware or software modules, and may be configured to execute instructions stored in any applicable computer-readable storage medium. For example, the processor 204 may include a microprocessor, microcontroller, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or any other digital or analog circuitry configured to interpret and / or execute program instructions and / or process data.

[0033] Although shown as a single processor in Figure 2, the processor 204 may include any number of processors configured to individually or collectively perform or direct the performance of any number of operations of the electronic device 102 described in this disclosure. Furthermore, one or more processors may reside on one or more different electronic devices, for example, different servers. In some embodiments, the processor 204 may be configured to interpret and / or execute and / or process program instructions stored in memory 206 and / or persistent data storage 208. In some embodiments, the processor 204 may retrieve program instructions from persistent data storage 208 and load them into memory 206. After the program instructions are loaded into memory 206, the processor 204 can execute the program instructions. Some examples of the processor 204 may be a graphics processing unit (GPU), a central processing unit (CPU), a reduced instruction set computer (RISC) processor, an ASIC processor, a composite instruction set computer (CISC) processor, a coprocessor, and / or a combination thereof.

[0034] Memory 206 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to store program instructions executable by the processor 204. In certain embodiments, memory 206 may be configured to store operating system and associated application-specific information. Memory 206 may include a computer-readable storage medium that carries or has stored computer-executable instructions or data structures. Such a computer-readable storage medium may include any available medium that can be accessed by a general-purpose or dedicated computer, such as the processor 204. Such a computer-readable storage medium may include, but is not limited to, random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory devices (e.g., solid-state memory devices), or any other storage medium that can be used to carry or store specific program code in the form of computer-executable instructions or data structures and that can be accessed by a general-purpose or dedicated computer, tangible or non-temporary computer-readable storage medium. The above combinations may also fall within the scope of computer-readable storage media. Computer-executable instructions may include, for example, instructions and data configured to cause the processor 204 to execute a specific operation or group of operations associated with the electronic device 102.

[0035] The persistent data storage device 208 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to store program instructions, operating systems, and / or application-specific information such as logs and application-specific databases, which are executable by the processor 204. The persistent data storage device 208 may also include a computer-readable storage medium that carries or holds the stored computer-executable instructions or data structures. Such a computer-readable storage medium may include any available medium that can be accessed by a general-purpose or dedicated computer, such as the processor 204.

[0036] Such computer-readable storage media may include, but are not limited to, tangible or non-temporary computer-readable storage media, including compact disk read-only memory (CD-ROM) or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices (e.g., hard disk drives (HDDs)), flash memory devices (e.g., solid-state drives (SSDs), secure digital (SD) cards, other solid-state memory devices), or any other storage media that can be used to transport or store specific program code in the form of computer-executable instructions or data structures and that are accessible by a general-purpose or dedicated computer. Combinations of the above may also be included in the scope of computer-readable storage media. Computer-executable instructions may include, for example, instructions and data configured to cause a processor 204 to execute a specific operation or group of operations associated with an electronic device 102.

[0037] In some embodiments, memory 206, persistent data storage device 208, or a combination thereof may store first information (e.g., first information 112A) of a person (e.g., first person), a set of extracted attributes of the person, and a constructed causal model (e.g., causal model 218) received from database 104. Memory 206, persistent data storage device 208, or a combination thereof may further store a plurality of AI models 216, utility functions for each of the plurality of AI models 216, determined optimism point information, a plurality of decisions made by the plurality of AI models 216, and adjusted decisions.

[0038] Each of the multiple AI models 216 may be a system of computational networks or artificial neurons configured in multiple layers as nodes. The multiple layers of each AI model (of the multiple AI models 216) may include an input layer, one or more hidden layers, and an output layer. Each layer of the multiple layers may include one or more nodes (or, for example, artificial neurons represented by circles). The outputs of all nodes in the input layer may be coupled to at least one node in the hidden layer. Similarly, the input of each hidden layer may be coupled to the output of at least one node in the other layers of the AI ​​model. The output of each hidden layer may be coupled to the input of at least one node in the other layers of the AI ​​model. Nodes in the final layer may receive input from at least one hidden layer and output a result. The number of layers and the number of nodes in each layer may be determined from the hyperparameters of the AI ​​model. Such hyperparameters can be set before or during training the AI ​​model on the training dataset.

[0039] Each node in an AI model can correspond to a mathematical function (e.g., a sigmoid function or a rectified linear unit) with a set of tunable parameters during training of the AI ​​model. The parameter set may include, for example, weight parameters, regularization parameters, etc. Each node can use the mathematical function to compute an output based on one or more inputs from nodes in other layers of the AI ​​model (e.g., previous layers). All or some nodes in the AI ​​model can correspond to the same or different mathematical functions. During training of the AI ​​model, one or more parameters of each node in the AI ​​model can be updated based on the AI ​​model's loss function, based on whether the output of the final layer for a given input (from the training dataset) matches the correct result. The above process may be repeated for the same or different inputs until a minimum value of the loss function can be achieved and the training error can be minimized. Several methods for training are known in the field, such as gradient descent, stochastic gradient descent, batch gradient descent, gradient boosting, and metaheuristics.

[0040] Each AI model may include electronic data for execution by a processing device such as a processor 204, such as software programs, code for software programs, libraries, applications, scripts, or other logic or instructions. Each AI model may also include code and routines configured to enable a computing device including the processor 204 to perform one or more tasks, such as making decisions on an individual's requests to one or more agencies. Furthermore, or alternatively, the AI ​​model may be implemented using hardware including a processor, a microprocessor (e.g., one or more operations that perform or control the execution of such operations), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). Alternatively, in some embodiments, the AI ​​model may be implemented using a combination of hardware and software.

[0041] Examples of multiple AI models 216 may include, but are not limited to, deep neural networks (DNNs), convolutional neural networks (CNNs), artificial neural networks (ANNs), fully connected neural networks, deep Bayesian neural networks, and / or combinations of such networks. In some embodiments, multiple AI models 216 may include numerical computation methods that use a dataflow graph. In certain embodiments, multiple AI models 216 may be based on a hybrid architecture of multiple deep neural networks (DNNs). In some embodiments, multiple AI models 216 may include machine learning (ML) models, which may be classifiers or rastering models, that can be trained to identify relationships between inputs, such as features in a training dataset (i.e., first information), and to output labels (i.e., judgments on a person's request made to one or more institutions).

[0042] The causal model 218 may also be a mathematical model represented by a directed acyclic graph of variables, such as a set of attributes associated with a person (e.g., Person 1). The causal model 218 can represent a set of causal relationships between attributes from a set of attributes. The directed acyclic graph of the causal model 218 may include directed edges that go from nodes associated with input variables to nodes associated with output variables that have a causal relationship with the input variables. The input variables (e.g., one or more first attributes from a set of attributes (e.g., gender and age)) may be variables that can influence other variables, such as output variables, through causal relationships. The output variables (e.g., one or more second attributes from a set of attributes (e.g., income)) may be variables that can be influenced by the input variables through causal relationships.

[0043] The I / O device 210 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to receive user input. For example, the I / O device 210 may receive user input to retrieve first information (e.g., first information 112A) associated with a person (e.g., first person). In another example, the I / O device 210 may receive user input to generate first information associated with a person, or to edit an existing dataset containing first information associated with a person, and / or to store generated / edited first information associated with a person. The I / O device 210 may further receive user input that may include instructions for determining empathetically coordinated judgments of multiple AI models 216 in response to a request (e.g., first request) of a person (e.g., first person). The I / O device 210 may further be configured to provide output in response to user input. For example, the I / O device 210 can present a coordinated decision (which may be determined by the electronic device 102) in response to a first request from a first person on the display screen 212. The I / O device 210 may include various input and output devices that can be configured to communicate with the processor 204 and other components, such as the network interface 214. Examples of input devices, but not limited to, include a touchscreen, keyboard, mouse, joystick, and / or microphone. Examples of output devices, but not limited to, include a display (e.g., the display screen 212) and a speaker.

[0044] The display screen 212 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to display a coordinated judgment on a first request of a first person. The display screen 212 may be configured to receive user input from user 110. In such a case, the display screen 212 may also be a touchscreen for receiving user input. The display screen 212 may be implemented by some known technologies, including, but not limited to, liquid crystal displays (LCDs), light-emitting diode (LED) displays, plasma displays, and / or organic LED (OLED) display technologies, and / or other display technologies.

[0045] The network interface 214 may include appropriate logic, circuitry, interfaces, and / or code that can be configured to establish communication between the electronic device 102, the database 104, and the user end device 106 over the communication network 108. The network interface 214 may be implemented by using various known techniques to support wired or wireless communication of the electronic device 102 over the communication network 108. The network interface 214 may include, but is not limited to, an antenna, a radio frequency (RF) transceiver, one or more amplifiers, a tuner, one or more oscillators, a digital signal processor, a coder-decoder (CODEC) chipset, a subscriber identification module (SIM) card, and / or a local buffer.

[0046] Without departing from the scope of this disclosure, the exemplary electronic device 102 may be modified, added to, or omitted. For example, in some embodiments, the exemplary electronic device 102 may include any number of other components that are not expressly shown or described for brevity.

[0047] Figure 3 is a flowchart illustrating an exemplary method for automated empathetic adjustment of decisions of an artificial intelligence (AI) model, according to one embodiment of the present disclosure. Figure 3 is described in relation to the elements of Figures 1 and 2. Referring to Figure 3, flowchart 300 is shown. The method shown in flowchart 300 can be initiated at 302 and may be performed by any suitable system, apparatus, or device, such as the exemplary electronic device 102 in Figure 1 or the processor 204 in Figure 2. Although shown in separate blocks, steps and operations associated with one or more blocks of flowchart 300 may be divided into further blocks, combined into fewer blocks, or removed, depending on the particular implementation.

[0048] In block 302, first information associated with a person (e.g., a first person) may be received. The first information may further be associated with a request made by the first person to one or more institutions (e.g., a first request). Examples of one or more institutions include, but are not limited to, a bank, an insurance institution, an employment organization, an educational institution, a travel-related institution, or a judicial court. The first request of the first person may include, but are not limited to, at least one of a loan application, an insurance application, a job application, an admission application, a visa application, or a judicial petition application. In one embodiment, the processor 204 may be configured to receive first information associated with a first person (e.g., a user 110). As an example, the processor 204 may receive first information 112A associated with a first person. The first information 112A may correspond to a record or row of information associated with a first person stored in the database 104. An example of the first information 112A is provided in Figure 4. For example, as shown in Figure 4, the first information 112A of the first person may include gender as female, income as "5000 USD / month", and work history in months as "3". The processor 204 can extract the first information 112A from the database 104 and store the extracted first information 112A in memory 206 and / or persistent data storage device 208. Alternatively, the processor 204 may obtain the first information 112A based on user input from user 110 or from the first person themselves.

[0049] In block 304, a set of attributes associated with a first person may be extracted from the received first information 112A. In one embodiment, the processor 204 may be configured to extract a set of attributes associated with a first person from the received first information 112A associated with a first person. For example, the first information 112A associated with a first person (i.e., a record or row of information) may include a set of attributes associated with a first person as fields of the record or row of information. The processor 204 can extract a set of attributes from various fields of the record or row of information associated with the first information 112A. The set of attributes associated with a first person may include, but is not limited to, at least one of the following: a set of needs-related attributes 114A, a set of protection attributes 116A, or a set of result-related attributes 118A. The set of needs-related attributes 114A associated with a first person may correspond to the purpose or needs of a first request of a first person to one or more organizations. Furthermore, the set of protective attributes 116A associated with the first person may include, but is not limited to, at least one of age, gender, race, caste, salary, nationality, or disability associated with the first person. The set of outcome-related attributes 118A associated with the first person may correspond to one or more outcomes associated with one or more decisions made by each AI model of one or more agencies for a first request associated with the first person. The decisions of the AI ​​models (i.e., from the multiple AI models 216) may be one of acceptance or rejection of the first request of the first person given by one or more agencies. Each of the one or more agencies may include an AI model for making decisions corresponding to different requests (e.g., first requests) received from different people (e.g., the first person). In some embodiments, different officers or senior employees of one or more agencies may use the multiple AI models 216 to make decisions corresponding to different requests (e.g., first requests) received from different people (e.g., the first person).Furthermore, the set of result-related attributes 118A associated with the first person can also correspond to one or more long-term benefits associated with the first person. An example of a set of attributes associated with the first person is provided, for example, in Figure 4.

[0050] In block 306, a causal model (e.g., causal model 218) may be constructed based on a set of extracted attributes associated with a first person. The constructed causal model 218 can represent a set of causal relationships between attributes from the set of attributes associated with the first person. In one embodiment, the processor 204 may be configured to construct the causal model 218 based on a set of extracted attributes associated with a first person. In one embodiment, the causal model 218 can be represented by a directed acyclic graph of a first set of nodes associated with a set of input variables and a second set of nodes associated with a set of output variables. In this specification, the set of input variables may include one or more first attributes from a set of attributes that can influence one or more second attributes from the set of attributes based on causal relationships with one or more second attributes. Furthermore, the set of output variables may include one or more second attributes from a set of attributes that can be influenced by one or more first attributes from the set of attributes based on causal relationships with one or more second attributes. The directed acyclic graph of causal model 218 may include edges between a first set of nodes and a second set of nodes. Edges in the directed acyclic graph may have an outward direction with respect to the first set of nodes and an inward direction with respect to the second set of nodes.

[0051] In one embodiment, the configuration of the causal model 218 may further be based on at least one of user input or a structured learning model. For example, the user input may be received from a domain knowledge expert. The user input may indicate a set of input variables and a set of output variables from a set of attributes associated with a first person. The user input may further include causal relationships between the set of input variables and the set of output variables. In another scenario, the processor 204 may determine the causal model 218 based on a machine learning model. For example, the processor 204 may use a structured learning model to identify a set of input variables and a set of output variables from first information 112A and determine causal relationships between the set of input variables and the set of output variables. Based on the identified set of input variables, the set of output variables, and the determined causal relationships, the processor 204 can determine the causal model 218. An example of a constructed causal model is provided, for example, in Figure 4.

[0052] In block 308, a utility function can be determined for a first person, associated with each of several AI models 216 associated with one or more institutions. The determination of the utility function can be based on the constructed causal model 218, a first empathy criterion set associated with the first person, and a second empathy criterion set associated with each of the one or more institutions. In one embodiment, for the first person, the processor 204 may be configured to determine a utility function associated with each of several AI models 216 associated with one or more institutions. Based on the causal relationship between the set of input variables and the set of output variables associated with the causal model 218, the processor 204 can extract a set of needs-related attributes 114A, a set of protection attributes 116A, and a set of outcome-related attributes 118A from the set of attributes associated with the first person.

[0053] The first set of empathy criteria associated with the first person may include at least one of the following: a first criterion associated with one or more of the set of protective attributes 116A; a second criterion associated with one or more of the set of needs-related attributes 114A; or a third criterion associated with one or more of the set of outcome-related attributes 118A. In one embodiment, the first criterion may correspond to a counterfactual criterion that can demonstrate a first difference. The first difference (e.g., D1) may be between a first probability (e.g., P1) of the acceptance of a first request by the first person having a first value (e.g., A=0) of the first protective attribute in the set of protective attributes 116A and a second probability (e.g., P2) of the acceptance of a second request by the second person having a second value (e.g., A=1) of the first protective attribute, where the values ​​of the other attributes in the set are assumed to be the same for both the first and second people. The first difference (i.e., D1) can be expressed by equation (1) as follows:

number

[0054] For example, in this specification, Y=1 is an AI model that accepts human requests (e.g., an AI model) i ) can represent the judgment of an AI model (for example, an AI model) that rejects a human request. i) can represent a judgment. Furthermore, A=0 can represent the first value of the first protective attribute, and A=1 can represent the second value of the first protective attribute. For example, if the first protective attribute is a person's gender, A=0 may represent that the person is female, and A=1 may represent that the person is male. Referring to equation (1), in one example, P1 takes into account that the gender of the first person (e.g., the first protective attribute) is female (i.e., A=0) and the AI ​​model (e.g., AI model i P2 may also be the probability that the AI ​​model's decision is to accept the first request of the first person (i.e., Y=1). Furthermore, P2 may also be the probability that the AI ​​model's decision is to accept the second request of the second person (i.e., Y=1), taking into account that the second person's gender (i.e., the same first protective attribute) is male (i.e., A=1). Here, all the remaining attributes of the first person (age, race, salary, nationality, etc.) and the corresponding attributes of the second person may be the same.

[0055] In one embodiment, the second criterion may correspond to a first threshold (e.g., e1(N), where N may represent a need) of a first need-related attribute (e.g., need N) from the set of need-related attributes 114A. In one embodiment, the first difference (e.g., D1) associated with the first criterion may be less than or equal to the first threshold (i.e., e1(N)) corresponding to the second criterion. Thus, the relationship between the first difference (e.g., D1) and the first threshold (i.e., e1(N)) may be expressed by the following equation (2).

number

[0056] Referring to Equation (2), for example, the first difference (i.e., D1) between the first probability of accepting a request from a female person (e.g., P1) and the second probability of accepting a request from a male person (e.g., P2) can be within the range of the first threshold value (i.e., e1(N)) associated with the needs of the person (i.e., Needs N), assuming other attributes are the same for both persons. The aforementioned conditions (e.g., Equation (1) and Equation (2)) can take into account counterfactual fairness and empathy based on the needs of a person (e.g., the first person). The first threshold value (i.e., e1(N)) may vary for different types of needs of the first person. For example, for basic needs such as education or a job that may be important for a person's livelihood, the first threshold value (i.e., e1(N)) may have a lower value. Therefore, the first difference (i.e., D1) between P1 and P2 has a smaller value, thereby ensuring that, in the above example, the judgment of the AI model (e.g., AI model i ) can be fair and empathetic for both male and female persons. For other higher-level needs such as lifestyle needs (e.g., purchasing a luxury car), the first threshold value (i.e., e1(N)) may have a relatively high value. Therefore, the first difference (i.e., D1) may also have a higher value, indicating that the judgment of the AI model (e.g., AI model i ) can be distorted by favoring one gender (e.g., male with a relatively high income) over the other gender (i.e., female with a relatively low income). The utility function (i.e., U i ) associated with the first person for an AI model (e.g., AI model) from multiple AI models may be determined based on the first difference (D1) associated with the first criterion. Therefore, the utility function (i.e., U i ) can take into account counterfactual fairness and empathy based on the needs of the first person. Based on the use of counterfactual fairness, the disclosed electronic device 102 can be more robust than conventional systems (e.g., statistical measurement systems).​​

[0057] In one embodiment, the third criterion may correspond to a benefit criterion that can show a second difference (e.g., D2). The second difference (i.e., D2) may be between the third probability (e.g., P3) of a first result-related attribute (e.g., W=1) in the acceptance of the first request (i.e., Y=1) and the fourth probability (e.g., P4) of a first result-related attribute (i.e., W=1) in the rejection of the first request by the first person (i.e., Y=0). The second difference (i.e., D2) can be expressed by equation (3) as follows:

number

[0058] For example, as discussed above, in this specification, Y=1 is an AI model that accepts human requests (e.g., an AI model) i ) can represent the judgment of an AI model (for example, an AI model) that rejects a human request. i ) can represent a judgment. Furthermore, W=1 can represent the first value of the first outcome-related attribute. For example, if the first outcome-related attribute can be a long-term benefit to a person, then W=1 may represent that the person will secure a job. Referring to equation (3), in one example, P3 may be the probability that a job (e.g., the first outcome-related attribute) can be secured by the first person (e.g., W=1) given that the first person's first request (e.g., an education loan) is accepted (e.g., Y=1). On the other hand, P4 may be the probability that a job (e.g., the same first outcome-related attribute) can be secured by the first person (e.g., W=1) given that the first person's first request (e.g., an education loan) is rejected (e.g., Y=0).

[0059] In one embodiment, a second set of empathy criteria associated with each of one or more agencies may correspond to a second threshold (e.g., e2(c)) associated with a cost (e.g., c) associated with the acceptance of a first request of a first person by one or more agencies. In one embodiment, a second difference (e.g., D2) associated with a third criterion may be less than or equal to a second threshold (e.g., e2(c)) corresponding to the second set of empathy criteria. Thus, the relationship between the second difference (e.g., D2) and the second threshold (i.e., e2(c)) may be expressed by the following equation (4).

number

[0060] Referring to equation (3), for example, the second difference (i.e., D2) between the third probability (i.e., P3) that a person will get the job if the request is accepted and the fourth probability (i.e., P4) that a person will get the job if the request is rejected may be within the range of a second threshold (i.e., e2(c)) related to the cost for one or more agencies to accept the request. The aforementioned conditions (e.g., equations (3) and (4)) can take into account outcome-based empathy for a person (e.g., the first person) and cost-based (or benefit-based) empathy for one or more agencies. For example, the likelihood of a person being hired (i.e., the outcome) may be maximized, while at the same time, the risk (i.e., cost) associated with one or more agencies accepting the request may be minimized. AI models from multiple AI models 216 (e.g., AI models i The utility function associated with the first person (i.e., U) i ) may be determined based on the second difference (D2) associated with the third criterion. Thus, the utility function (i.e., U i ) can take into account outcome-based empathy for the first person and cost-based (or benefit-based) empathy for one or more institutions.

[0061] In one embodiment, for a first person, a utility function (i.e., U) is associated with each of the multiple AI models 216 of one or more institutions. i ) may be determined based on a first difference (i.e., D1) associated with a first criterion and a second difference (i.e., D2) associated with a third criterion. In one embodiment, the utility function (i.e., U i ) may also be the sum of the first difference (i.e., D1) and the second difference (i.e., D2), and can be expressed by the following equations (5) or (6).

number

[0062] In block 310, the determined utility function (U) associated with each of the multiple AI models 216 is shown. i The optimism point information of ) can be determined. In one embodiment, the determined optimism point information can correspond to a coordinated decision from a plurality of decisions made by a plurality of AI models 216 of one or more institutions for a first request associated with a first person. In one embodiment, the processor 204 determines the utility function (U) associated with each of the plurality of AI models 216. i The system may be configured to determine the optimism point information of ). The determination of the optimism point information is further illustrated, for example, in Figure 5.

[0063] In block 312, a refined decision may be presented. In one embodiment, the processor 204 may be configured to present a refined decision associated with a plurality of AI models 216, which may be determined based on the optimism point information. For example, the processor 204 can display the refined decision on the display screen 212 of the electronic device 102. Alternatively, the processor 204 can transmit the refined decision to the user end device 106 and control the user end device 106 to display the transmitted refined decision on the display screen of the user end device 106. The control may then proceed to termination.

[0064] While flowchart 300 is shown as individual operations such as 302, 304, 306, 308, 310, and 312, in a particular embodiment, such individual operations may be further divided into additional operations, combined into fewer operations, or eliminated, depending on the particular implementation, without impairing the essence of the disclosed embodiment.

[0065] Figure 4 shows an exemplary scenario for constructing a causal model from first information associated with a first person, arranged according to at least one embodiment described herein. Figure 4 is described in relation to the elements of Figures 1, 2 and 3. Referring to Figure 4, an exemplary scenario 400 is shown. Scenario 400 includes a dataset 402 associated with multiple people, such as person-1, person-2, and person-3. For example, dataset 402 may include first information 402A associated with person-1, first information 402B associated with person-2, and first information 402C associated with person-3. Furthermore, a causal model 406 that may be constructed based on dataset 402 is shown.

[0066] Consider an example of a bank as one of one or more institutions and person-1 as a first person who may be a customer of the bank. The first request may also be a loan application that person-1 (i.e., first person) may apply for with the bank (i.e., one or more institutions). The first information 402A may be information associated with person-1 that may relate to person-1's loan application to the bank. In one example, a set of person attributes 404 may be attributes associated with person-1's loan application made to the bank. As shown in Figure 4, the set of person attributes 404 may include, but are not limited to, gender, income (e.g., in US dollars per month), work history (in months), how long the person has been a customer of the bank, loan amount (e.g., in US dollars), housing (e.g., real estate owned by the person), age, credit score, deposit account balance (e.g., in US dollars), needs / purpose (e.g., for a loan from the bank), and the person's benefits (e.g., consequences or long-term benefits associated with the acceptance of the loan).

[0067] Processor 204 receives first information 402A associated with person-1 and can extract a set of attributes 404 associated with person-1 from the first information 402A. For example, as shown in Figure 4, person-1 may be a woman who can earn 500 USD / month, has been employed for the past three months, has been a bank customer since 2018, and has applied for a loan of 50,000 USD. Furthermore, person-1 may not own a home, may be 20 years old, have a credit score of 550, have a deposit bank account balance of 1,000 USD, may need a loan for educational purposes, and the benefit of employment may be derived based on loan approval by the bank.

[0068] In one embodiment, the processor 204 may be configured to construct a causal model 406 based on a set of extracted attributes 404 associated with person-1. The causal model 406 can represent a set of causal relationships between attributes from the set of attributes 404 associated with person-1. In one embodiment, the construction of the causal model 406 may further be based on at least one of user input or a structured learning model. For example, user input may be received from a domain knowledge expert. User input may represent a set of input variables and a set of output variables from the set of attributes 404 associated with person-1. User input may further include causal relationships between the set of input variables and the set of output variables. In one embodiment, the causal model 406 can be represented by a directed acyclic graph of a first set of nodes associated with the set of input variables and a second set of nodes associated with the set of output variables. The directed acyclic graph of the causal model 406 may include edges between the first set of nodes and the second set of nodes. Edges in a directed acyclic graph may have an outward direction with respect to a first set of nodes and an inward direction with respect to a second set of nodes.

[0069] For example, as shown in Figure 4, the causal model 406 may include nodes such as the first node 406A, the second node 406B, the third node 406C, the fourth node 406D, the fifth node 406E, the sixth node 406F, and the seventh node 406G. The first node 406A may represent the attribute "gender" of person-1. The second node 406B may represent the attribute "income" of person-1, and the third node 406C may represent the attribute "credit score" of person-1. The fourth node 406D may represent a decision (i.e., an "AI decision") that may be made by one of several AI models 216 used by the bank to determine whether to accept or reject person-1's loan application request. The fifth node 406E may represent the attribute "loan amount" of person-1, and the sixth node 406F may represent the attribute "needs" (i.e., necessity / purpose) of person-1. Furthermore, the seventh node 406G can represent "education," which may be a benefit attribute associated with person-1 (e.g., long-term benefits or outcomes).

[0070] As shown in causal model 406, person-1's "income" (represented by the second node 406B) may be influenced by person-1's "gender" (represented by the first node 406A). For example, typically, a woman may have a lower income than a man in an organization. Furthermore, person-1's "credit score" (represented by the third node 406C) may be influenced by person-1's "income" (represented by the second node 406B). This may be because a person with a higher income has higher creditworthiness and, consequently, a higher credit score. Furthermore, the "AI decision" (represented by the fourth node 406D) may be influenced by person-1's "credit score" (represented by the third node 406C) and person-1's "loan amount" (represented by the fifth node 406E). This may be because a person's loan amount and credit score can directly influence the AI ​​model's decision to accept or reject a person's loan application. On the other hand, both person-1's "needs" (represented by the sixth node 406F) and the "AI judgment" regarding person-1's loan request (represented by the fourth node 406D) can directly influence person-1's outcome or long-term benefit (i.e., "education" represented by the seventh node 406G).

[0071] It should be noted that Scenario 400 shown in Figure 4, concerning Person-1's loan application to a bank, is presented merely as an example and should not be construed as limiting the scope of this disclosure. Similarly, Processor 204 may be configured to construct causal models based on different sets of attributes extracted from the same or different information associated with persons who may have applied for different requests (such as insurance applications, job applications, admission applications, visa applications, or judicial petition applications) to different institutions (such as insurance institutions, employment organizations, educational institutions, travel-related institutions, or judicial courts, but not limited to these).

[0072] Figure 5 is a flowchart illustrating an exemplary method for determining the optimism-point information for the utility function of an artificial intelligence (AI) model, arranged according to at least one embodiment described herein. Figure 5 is described in relation to the elements of Figures 1, 2, 3, and 4. Referring to Figure 5, flowchart 500 is shown. The method shown in flowchart 500 can be initiated at 502 and may be performed by any suitable system, apparatus, or device, such as the exemplary electronic device 102 in Figure 1 or the processor 204 in Figure 2. Although shown in separate blocks, steps and operations associated with one or more blocks of flowchart 500 may be divided into further blocks, combined into fewer blocks, or removed, depending on the particular implementation.

[0073] In block 502, the first utility value associated with the first decision of each of the multiple AI models 216 is determined by the determined utility function (i.e., U) associated with each of the multiple AI models 216. i ) can be estimated based on a determined utility function (i.e., U) associated with each of the multiple AI models 216. In one embodiment, the processor 204 estimates the utility function (i.e., U) associated with each of the multiple AI models 216. i It may be configured to estimate a first utility value associated with each of the decisions of the multiple AI models 216 (i.e., the first decision) based on the utility function (i.e., U i The decision of the AI ​​model is further explained, for example, in Figure 3 (at 308). i The first decision of ) may also be a decision to accept a first request associated with a first person. For such a first decision to accept a first request, the processor 204 uses equation (5) or (6) to determine the AI ​​model (i.e., the AI ​​model) as further explained in Figure 3 (in 308). i The first utility value associated with (for example, U 1 i) can be determined. In one embodiment, the processor 204 can determine probabilities P1, P2, P3, and P4 based on first information (e.g., first information 112A, first information 112B, ..., and first information 112N) associated with a plurality of persons (e.g., person 1, person 2, ..., and person N) stored in the database 104. For example, the processor 204 may determine probabilities P1, P2, P3, and P4 based on a history of decisions made by a plurality of AI models 216 in response to requests received from a plurality of persons, and first information associated with each of the plurality of persons. In one embodiment, the processor 204 can determine a first threshold (i.e., e1(N)) corresponding to a second criterion and a second threshold (i.e., e2(c)) associated with the cost or benefit for one or more agencies to heuristically accept the first request.

[0074] In block 504, a second utility value associated with each counterfactual decision of the multiple AI models 216 is used to construct the causal model 218 and the utility function (U) associated with each of the multiple AI models 216. i ) can be estimated based on the constructed causal model 218 and the utility function (U) associated with each of the multiple AI models 216. In one embodiment, the processor 204 uses the constructed causal model 218 and the utility function (U) associated with each of the multiple AI models 216. i It may be configured to estimate a second utility value (i.e., associated with each of the counterfactual judgments of the multiple AI models 216) based on ). For example, an AI model (e.g., an AI model i If the first decision of the AI ​​model is to accept the first request of the first person, then the AI ​​model (for example, the AI ​​model) i A counterfactual judgment of ) could be a judgment rejecting the first request of the first person. With respect to such a counterfactual judgment of rejecting the first request, processor 204 uses an AI model (i.e., an AI model) based on the use of a parameterized predictor such as a neural network model or a logistic regression model. iA second utility value associated with (for example, U 2 i ) can be determined. The parameterized predictor can be used to determine one or more probability values, such as probabilities P1 and P2. Based on the loss function (e.g., squared loss or log probability) and the training dataset, the empirical loss of the parameterized predictor can be minimized in terms of parameters. The loss function of the parameterized predictor can be expressed in terms of random variables corresponding to a conditional distribution of background variables (e.g., latent variables). The conditional distribution of background variables may be determined based on the causal model 218. In one embodiment, the loss function can be minimized analytically or based on a Markov chain Monte Carlo process.

[0075] In block 506, a value function (e.g., V) associated with a set of agreed-upon decisions among each of the multiple AI models 216 corresponds to an estimated first utility value (e.g., U) associated with each of the multiple AI models 216. 1 i ) and the estimated second utility value (e.g., U 2 i ) may be determined based on the following: In one embodiment, the processor 204 determines the estimated first utility value (e.g., U determined in 502) associated with each of the plurality of AI models 216. 1 i ) and the estimated second utility value (e.g., U determined in 504) 2 i The AI ​​model may be configured to determine a value function (e.g., V) (i.e., associated with a set of decisions agreed upon among each of the multiple AI models 216) based on the above. For example, the processor 204 can determine a matrix of utility values ​​(e.g., a first utility value and a second utility value) associated with each of the multiple AI models 216. i The utility value associated with the first person's first request is the AI ​​model (i.e., the AI ​​model) iThis can represent the gain or economic benefit associated with the decision made by ). An example of a matrix of determined utility values ​​for a first AI model (e.g., associated with a first institution) and a second AI model (e.g., associated with a second institution) from multiple AI models 216 is provided in Table 1 below. [Table 1]

[0076] For example, referring to Table 1, for rejection decisions for the first AI model and rejection decisions for the second AI model, the utility value of the first AI model may be "0" and the utility value of the second AI model may be "10". Furthermore, for rejection decisions for the first AI model and acceptance decisions for the second AI model, the utility value of the first AI model may be "100" and the utility value of the second AI model may be "50". Furthermore, for acceptance decisions for the first AI model and rejection decisions for the second AI model, the utility values ​​of the first AI model and the second AI model may be "40" and "10", respectively. Similarly, for acceptance decisions for both the first and second AI models, the utility values ​​of the first AI model and the second AI model may be "95" and "80", respectively. Note that the data provided in Table 1 may be considered experimental data only and should not be interpreted as limiting this disclosure.

[0077] The processor 204 can determine a value function from a determined matrix of utility values ​​associated with each of the multiple AI models 216, based on game theory methods. The value function can correspond to the minimum utility value of each of the multiple AI models 216 for a particular agreement of judgments among the multiple AI models 216. For example, an agreement among multiple AI models 216 (e.g., the first AI model and the second AI model) regarding a set of judgments of the multiple AI models 216 can be modeled as a non-cooperative game when the players or actors associated with the game cannot form an alliance, or as a game where the agreement among players may need to be self-enforcing (e.g., through a trustworthy threat). Referring to Table 1, an example of a value function that can be determined for each of the first and second AI models, given the utility value matrices of the first and second AI models, is shown in Table 2 below. [Table 2]

[0078] For example, referring to Tables 1 and 2, if we consider the case where both the first AI model and the second AI model agree to accept the first request of the first person, the minimum utility of the first AI model may be "95". This is because the second AI model cannot deviate from the agreement to accept the first request, because rejecting the first request may result in lower utility for the second AI model (i.e., utility may decrease from "80" to "10"). Furthermore, the minimum utility for the second AI model may be "50". This is because the first AI model can deviate from the agreement to accept the first request, because rejecting the first request may result in higher utility for the first AI model (i.e., utility may increase from "95" to "100"). Therefore, the value functions of the "accept-accept" decision sets for the first and second AI models could be "95" and "50". Note that the data provided in Table 2 can only be considered experimental data and should not be interpreted as limiting this disclosure.

[0079] In another example, Table 3 below provides a matrix of utility values ​​determined for a first AI model (e.g., of the first institution), a second AI model (e.g., of the second institution), and a third AI model (e.g., of the third institution) from a plurality of AI models 216. [Table 3]

[0080] Please note that the data provided in Table 3 may be considered experimental data only and should not be interpreted as limiting this disclosure. Referring to Table 3, Table 4 shows examples of value functions that can be determined for each of the first, second, and third AI models, for the utility value matrices of the first, second, and third AI models. [Table 4]

[0081] Please note that the data provided in Table 4 may be considered experimental data only and should not be interpreted as limiting this disclosure.

[0082] In block 508, the optimism point information of the determined utility function associated with each of the multiple AI models 216 may be determined based on the determined value function. In one embodiment, the processor 204 may be configured to determine the optimism point information of the determined utility function (i.e., associated with each of the multiple AI models 216) based on the determined value function (determined in 506). For example, (ΔX i ,u i ) i∈N Consider this as a non-cooperative game involving N players (e.g., N=1,2,...n), where each player may be one of several AI models 216. In one embodiment, the non-cooperative game may be a mixed expansion. In this specification, ΔX i This is a finite set of actions X i This can be a set of probability distributions for a given value,

number

number

number

number

number

number

[0083] Optimin point (i.e., agreement)

number

number

[0084] In this specification, the i-th player (i.e., the AI ​​model) i ) value v i(p) can correspond to the worst-case payoff that the i-th player can receive based on an agreement p among the N players. Players in a non-cooperative game are not bound by the agreement and may break it based on the choice of another unagreeable decision. Therefore, v i (p) can correspond to the least utility of the i-th player when all players abide by the contract, or when one or more players deviate from the agreement for a better utility return.

[0085] For example, referring to Tables 1 and 2, processor 204 may determine the optimism point information as "100,50" based on the maximum combined utility across all agreements regarding the first and second AI models. Based on the optimism point information, processor 204 may determine a coordinated decision. For example, the set of decisions "reject-accept" may be determined as the coordinated decision for the first and second AI models. In another example, referring to Tables 3 and 4, processor 204 may determine the optimism point information as "80,60,100". The set of decisions "reject-accept-reject" or "reject-accept-accept" may be determined as the coordinated decision for the first, second, and third AI models. Control may proceed to termination. Therefore, the disclosed electronic device 102 can be configured to determine a coordinated decision from multiple decisions made by AI models 216 of one or more institutions, even if different AI models provide conflicting decisions regarding a first request associated with a first person.

[0086] While flowchart 500 is shown as individual operations such as 502, 504, 506, and 508, in certain embodiments such individual operations may be further divided into additional operations, combined into fewer operations, or eliminated, depending on the particular implementation, without impairing the essence of the disclosed embodiments.

[0087] The disclosed electronic device 102 can empathetically adjust, with minimal bias, the decisions made by multiple AI models 216 associated with one or more agencies in response to a first request of a first person. The electronic device 102 can construct a causal model 218 that shows the causal relationships between different attributes associated with the first person (e.g., a set of protection attributes 116A, a set of needs-related attributes 114A, and a set of outcome-related attributes 118A). Furthermore, the electronic device 102 can determine a utility function for each of the multiple AI models 216 used by one or more agencies for decision-making. The utility function may be determined based on the causal model 218, a first empathy criterion associated with the first person, and a second empathy criterion associated with one or more agencies. The utility function of a particular AI model can show the benefit that one or more agencies gain when that particular AI model makes a particular decision, for example, a decision to accept the first request of the first person. Furthermore, the utility function can take into account fairness to the first person based on a set of protective attributes 116A of the first person, determined based on the causal model 218. Furthermore, the utility function can incorporate empathy for the first person based on a first set of empathy criteria (e.g., a second criterion associated with a set of needs-related attributes 114A, and a third criterion associated with a set of outcome-related attributes 118A). Furthermore, the utility function can also incorporate empathy for one or more institutions based on a second set of empathy criteria (e.g., thresholds associated with the cost or benefit of accepting the first person's first request). The disclosed electronic device 102 can adjust multiple decisions of multiple AI models 216 based on determinations of optimism point information of the utility function determined for each of the multiple AI models 216. The determined optimism point information can represent the adjusted decision from multiple decisions made by the multiple AI models regarding the first request of the first person. Furthermore, the adjusted decisions indicated by the determined Optimin point information may be in the favor of all stakeholders affected by multiple AI models 216 (i.e., customers, officers, regulators, and government authorities associated with one or more institutions).Therefore, multiple judgments made by multiple AI models 216 associated with one or more institutions can be combined into a single, coordinated judgment for the first request. This coordinated judgment may be fair and empathetic towards the first person, and may also be empathetic towards one or more institutions.

[0088] Various embodiments of this disclosure can provide one or more non-temporary computer-readable storage media configured to store instructions causing a system (such as an exemplary electronic device 102) to perform an action in response to being executed. The action may include receiving first information associated with a first person. The first information may further include a first request of the first person to one or more agencies. The action may further include extracting a set of attributes associated with the first person from the received first information. The set of attributes may include at least one of a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. The action may further include constructing a causal model based on the extracted set of attributes associated with the first person. The constructed causal model may represent a set of causal relationships between attributes from the extracted set of attributes. The action may further include determining a utility function for the first person associated with each of several artificial intelligence (AI) models associated with one or more agencies. The determination of the utility function can be based on a constructed causal model, a first set of empathy criteria associated with a first person, and a second set of empathy criteria associated with each of one or more institutions. The operation may further include determining the optimism point information of the determined utility function associated with each of the multiple AI models. The determined optimism point information can correspond to a coordinated judgment from multiple judgments made by multiple AI models of one or more institutions for a first request associated with a first person. The operation may further include presenting the coordinated judgment.

[0089] When used in this disclosure, the terms “module” or “component” may refer to a specific hardware implementation configured to perform the actions of a module or component, and / or a software object or software routine that can be stored and / or executed on general-purpose hardware of a computing system (e.g., computer-readable media, processing devices, etc.). In some embodiments, different components, modules, engines, and services described in this disclosure may be implemented as objects or processes that run on a computing system (e.g., as separate threads). While some of the systems and methods described in this disclosure are generally described as being implemented in software (stored and / or executed on general-purpose hardware), specific hardware implementations, or combinations of software and specific hardware implementations, are also possible and intended. In this description, “computation entity” may be any computing system as previously defined in this disclosure, or any combination of modules or modulates that operate on a computing system.

[0090] In this disclosure, the terms used in particular in the claims set forth separately (for example, in the text of the claims set forth separately) are generally intended to be "open" terms (for example, the term "contains" should be interpreted as "contains but not limited to," the term "has" should be interpreted as "has at least," and the term "includes" should be interpreted as "contains but not limited to," etc.).

[0091] Furthermore, if a specific number of introduced claim descriptions are intended, such intention is explicitly stated in the claim; if such statement is absent, such intention does not exist. For example, for the sake of understanding, the claims listed below may include the use of the introductory phrases “at least one” and “one or more” to introduce claim descriptions. However, the use of such phrases should not be seen as implicitly indicating that the introduction of a claim description by the indefinite article “a” or “an” limits any particular claim containing such introduced claim descriptions to embodiments containing only one such description, even when the same claim includes the introductory phrase “one or more” or “at least one” and an indefinite article such as “a” (for example, “a” and / or “an” should be interpreted as meaning “at least one” or “one or more”). The same applies to the use of definite articles used to introduce claim descriptions.

[0092] Furthermore, even when a claim description introducing a specific number is explicitly stated, a person skilled in the art will recognize that such description should be interpreted to mean at least the number stated (for example, the statement “two descriptions” without other modifiers means at least two descriptions, or two or more descriptions). Moreover, in examples where a similar provision is used, such as “at least one of A, B, and C, etc.” or “one or more of A, B, and C, etc.”, such structure is generally intended to include A alone, B alone, C alone, A and B together, A and C together, B and C together, or A, B, and C together, etc.

[0093] Furthermore, any separate word or phrase presenting two or more alternative terms, whether in the specification, claims, or drawings, should be understood to include the possibility of containing one of the terms, either one of the terms, or both. For example, the phrase "A or B" should be understood to include the possibility of "A" or "B" or "A and B".

[0094] All examples and conditional language described herein are intended for educational purposes to assist readers in understanding the disclosure and the concepts to which the inventors have contributed to advancing the art, and should be considered as not being limited to the examples and conditions described herein. Although embodiments of this disclosure have been described in detail, various changes, substitutions, and modifications can be made without departing from the spirit and scope of this disclosure.

[0095] The following note is left regarding the above embodiment. (Note 1) A method executed by a processor, A step of receiving first information associated with a first person, wherein the first information is further associated with a first request of the first person to one or more agencies, Steps include: extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person; Steps include: constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes; The first step is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with the one or more institutions mentioned above for the first person, The determination of the utility function is based on the constructed causal model, a first set of empathy criteria associated with the first person, and a second set of empathy criteria associated with each of the one or more institutions, and is a step-by-step process. The steps include determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person, and The step of presenting the adjusted judgment, A method that includes this. (Note 2) The aforementioned one or more institutions include one of the following: a banking institution, an insurance institution, an employment organization, an educational institution, a travel-related institution, or a judicial court. The first request of the first person includes at least one of the following: a loan application, an insurance application, a job application, an admissions application, a visa application, or a legal petition application. The method described in Appendix 1. (Note 3) The method according to Appendix 1, wherein the AI ​​model's determination from the plurality of AI models is one of the acceptance or rejection of the first request of the first person given by one or more of the institutions. (Note 4) The set of needs-related attributes associated with the first person corresponds to the purpose of the first request of the first person to the one or more organizations, as described in Appendix 1. (Note 5) The method according to Appendix 1, wherein the set of protective attributes associated with the first person includes at least one of age, gender, race, caste, salary, nationality, or physical disability associated with the first person. (Note 6) The method according to Appendix 1, wherein the set of result-related attributes associated with the first person corresponds to one or more results associated with one or more decisions made by the AI ​​model of one or more institutions for the first request associated with the first person. (Note 7) The set of result-related attributes associated with the first person corresponds to one or more long-term benefits associated with the first person, as described in Appendix 1. (Note 8) The construction of the causal model is further based on the method described in Appendix 1, wherein the causal model is based on at least one of user input or a structured learning model. (Note 9) The first set of empathy criteria associated with the first person is: A first criterion associated with one or more of the set of protective attributes associated with the first person, A second criterion associated with one or more of the set of needs-related attributes associated with the first person, or A third criterion associated with one or more of the set of result-related attributes associated with the first person. The method described in Appendix 1, comprising at least one of the following. (Note 10) The method according to Appendix 9, wherein the first criterion corresponds to a counterfactual criterion showing a first difference between the first probability of the acceptance of the first request of the first person having a first value for the first protective attribute of the set of protective attributes and the second probability of the acceptance of the second request of the second person having a second value for the first protective attribute, and the values ​​of the other attributes of the set of attributes are assumed to be the same for both the first person and the second person. (Note 11) The method described in Appendix 9, wherein the second criterion corresponds to a first threshold for the first needs-related attribute from the set of needs-related attributes. (Note 12) The first difference associated with the first criterion is less than or equal to the first threshold corresponding to the second criterion, The utility function of the AI ​​model from the plurality of AI models associated with the first person is determined based on the first difference associated with the first criterion. The method described in Appendix 11. (Note 13) The method according to Appendix 9, wherein the third criterion corresponds to a benefit criterion that shows a second difference between a third probability of a first result-related attribute having a first value in the set of result-related attributes of the first person in the acceptance of the first person's first request and a fourth probability of a first value of the first result-related attribute in the rejection of the first person's first request. (Note 14) The method according to Appendix 9, wherein the second set of empathy criteria associated with each of the one or more agencies corresponds to a second threshold relating to the cost associated with the acceptance of the first request of the first person provided by the one or more agencies. (Note 15) The second difference associated with the third criterion is less than or equal to the second threshold corresponding to the second set of empathy criteria, The utility function of the AI ​​model associated with the first person from the plurality of AI models is determined based on the second difference associated with the third criterion. The method described in Appendix 14. (Note 16) A step of estimating a first utility value associated with a first decision of each of the multiple AI models, based on the determined utility function associated with each of the multiple AI models, A step of estimating a second utility value associated with the counterfactual judgment of each of the multiple AI models, based on the constructed causal model and the utility function associated with each of the multiple AI models, The steps include determining a value function associated with a set of decisions agreed upon among each of the multiple AI models, based on the estimated first utility value and the estimated second utility value associated with each of the multiple AI models, The steps include determining the optimism point information of the determined utility function associated with each of the plurality of AI models based on the determined value function, The method described in Appendix 1, further including the method described in Appendix 1. (Note 17) One or more non-temporary computer-readable storage media configured to store instructions that cause an electronic device to perform an action in response to being executed, wherein the action is: Receiving first information associated with a first person, the first information being further associated with a first request of the first person to one or more institutions, The process involves extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of the following: a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. This involves constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes. The first person is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with the one or more institutions mentioned above. The determination of the utility function is based on the constructed causal model, the first set of empathy criteria associated with the first person, and the second set of empathy criteria associated with each of the one or more institutions. The process involves determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person. To present the aforementioned adjusted judgment, One or more non-temporary computer-readable storage media, including [the specified text]. (Note 18) The aforementioned operation is, Based on the determined utility function associated with each of the plurality of AI models, estimate a first utility value associated with the first decision of each of the plurality of AI models, Based on the constructed causal model and the utility function associated with each of the multiple AI models, a second utility value associated with the counterfactual judgment of each of the multiple AI models is estimated. Based on the estimated first utility value and the estimated second utility value associated with each of the multiple AI models, a value function is determined that is associated with a set of decisions agreed upon among each of the multiple AI models. Based on the determined value function, the optimism point information of the determined utility function associated with each of the plurality of AI models is determined, One or more non-temporary computer-readable storage media as described in Appendix 17, further including the above. (Note 19) It is an electronic device, Memory for storing instructions, The process includes a processor coupled to the memory and executing the instructions to execute the process, and the process is Receiving first information associated with a first person, the first information being further associated with a first request of the first person to one or more institutions, The process involves extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of the following: a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. This involves constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes. The first person is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with the one or more institutions mentioned above. The determination of the utility function is based on the constructed causal model, the first set of empathy criteria associated with the first person, and the second set of empathy criteria associated with each of the one or more institutions. The process involves determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person. To present the aforementioned adjusted judgment, Electronic devices, including those mentioned above. (Note 20) The aforementioned process, Based on the determined utility function associated with each of the plurality of AI models, estimate a first utility value associated with the first decision of each of the plurality of AI models, Based on the constructed causal model and the utility function associated with each of the multiple AI models, a second utility value associated with the counterfactual judgment of each of the multiple AI models is estimated. Based on the estimated first utility value and the estimated second utility value associated with each of the multiple AI models, a value function is determined that is associated with a set of decisions agreed upon among each of the multiple AI models. Based on the determined value function, the optimism point information of the determined utility function associated with each of the plurality of AI models is determined, The electronic devices described in Appendix 19, further including the electronic devices described in Appendix 19.

Claims

1. A method executed by a processor, A step of receiving first information associated with a first person, wherein the first information is further associated with a first request of the first person to one or more institutions. The step of extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. Steps include: constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes; The first step is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with one or more institutions, for the first person, The determination of the utility function is based on the constructed causal model, a first set of empathy criteria associated with the first person, and a second set of empathy criteria associated with each of the one or more institutions, and is a step-by-step process. The steps include determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person, and The step of presenting the adjusted judgment, A method that includes this.

2. The aforementioned one or more institutions include one of the following: a banking institution, an insurance institution, an employment organization, an educational institution, a travel-related institution, or a judicial court. The first request of the first person includes at least one of the following: a loan application, an insurance application, an employment application, an admissions application, a visa application, or a legal petition application. The method according to claim 1.

3. The method according to claim 1, wherein the determination of the AI ​​models from the plurality of AI models is one of the acceptance or rejection of the first request of the first person given by the one or more institutions.

4. The method according to claim 1, wherein the set of needs-related attributes associated with the first person corresponds to the purpose of the first request of the first person to one or more institutions.

5. The method according to claim 1, wherein the set of protective attributes associated with the first person includes at least one of age, gender, race, caste, salary, nationality, or physical disability associated with the first person.

6. The method according to claim 1, wherein the set of result-related attributes associated with the first person corresponds to one or more results associated with one or more decisions made by the AI ​​models of one or more institutions for the first request associated with the first person.

7. The method according to claim 1, wherein the set of result-related attributes associated with the first person corresponds to one or more benefits associated with the first person.

8. The method according to claim 1, wherein the construction of the causal model is further based on at least one of user input or a structured learning model.

9. The first set of empathy criteria associated with the first person is: A first criterion associated with one or more of the set of protective attributes associated with the first person, A second criterion associated with one or more of the set of needs-related attributes associated with the first person, or A third criterion associated with one or more of the set of result-related attributes associated with the first person. The method according to claim 1, comprising at least one of the following.

10. The method according to claim 9, wherein the first criterion corresponds to a counterfactual criterion showing a first difference between the first probability of the acceptance of the first request of the first person having a first value for the first protective attribute of the set of protective attributes and the second probability of the acceptance of the second request of the second person having a second value for the first protective attribute, and the values ​​of the other attributes of the set of attributes are assumed to be the same for both the first person and the second person.

11. The method according to claim 9, wherein the second criterion corresponds to a first threshold of the first needs-related attribute from the set of needs-related attributes.

12. The first difference associated with the first criterion is less than or equal to the first threshold corresponding to the second criterion, The utility function of the AI ​​model associated with the first person from the plurality of AI models is determined based on the first difference associated with the first criterion. The method according to claim 11.

13. The method according to claim 9, wherein the third criterion corresponds to a benefit criterion that expresses a second difference between a third probability of a first value of the first result-related attribute in the set of result-related attributes of the first person in the acceptance of the first person's first request and a fourth probability of a first value of the first result-related attribute in the rejection of the first person's first request.

14. The method according to claim 9, wherein the second set of empathy criteria associated with each of the one or more agencies corresponds to a second threshold relating to the cost associated with the acceptance of the first request of the first person provided by the one or more agencies.

15. The second difference associated with the third criterion is less than or equal to the second threshold corresponding to the second set of empathy criteria, The utility function of the AI ​​model associated with the first person from the plurality of AI models is determined based on the second difference associated with the third criterion. The method according to claim 14.

16. A step of estimating a first utility value associated with a first decision of each of the plurality of AI models, based on the determined utility function associated with each of the plurality of AI models, A step of estimating a second utility value associated with the counterfactual judgment of each of the multiple AI models, based on the constructed causal model and the utility function associated with each of the multiple AI models, The steps include determining a value function associated with a set of decisions agreed upon among each of the multiple AI models, based on the estimated first utility value and the estimated second utility value associated with each of the multiple AI models, The steps include determining the optimism point information of the determined utility function associated with each of the plurality of AI models based on the determined value function, The method according to claim 1, further comprising:

17. A computer program that causes an electronic device to perform an action, wherein the action is: Receiving first information associated with a first person, the first information being further associated with a first request of the first person to one or more institutions, The process involves extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of the following: a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. This involves constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes. The first person is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with one or more institutions, The determination of the utility function is based on the constructed causal model, the first set of empathy criteria associated with the first person, and the second set of empathy criteria associated with each of the one or more institutions. The process involves determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person. To present the aforementioned adjusted judgment, A computer program that includes [this].

18. The aforementioned operation is, Based on the determined utility function associated with each of the plurality of AI models, estimate a first utility value associated with the first decision of each of the plurality of AI models, Based on the constructed causal model and the utility function associated with each of the multiple AI models, a second utility value associated with the counterfactual judgment of each of the multiple AI models is estimated. Based on the estimated first utility value and the estimated second utility value associated with each of the plurality of AI models, a value function is determined that is associated with a set of decisions agreed upon among each of the plurality of AI models. Based on the determined value function, the optimism point information of the determined utility function associated with each of the plurality of AI models is determined. The computer program according to claim 17, further comprising:

19. It is an electronic device, Memory for storing instructions, The process includes a processor coupled to the memory and executing the instructions to execute the process, and the process is Receiving first information associated with a first person, the first information being further associated with a first request of the first person to one or more institutions, The process involves extracting a set of attributes associated with the first person from the received first information, wherein the set of attributes includes at least one of the following: a set of needs-related attributes, a set of protection attributes, or a set of outcome-related attributes associated with the first person. This involves constructing a causal model based on the extracted set of attributes associated with the first person, wherein the constructed causal model represents a set of causal relationships between attributes from the extracted set of attributes. The first person is to determine the utility function associated with each of the multiple artificial intelligence (AI) models associated with one or more institutions, The determination of the utility function is based on the constructed causal model, the first set of empathy criteria associated with the first person, and the second set of empathy criteria associated with each of the one or more institutions. The process involves determining the optimism point information of the determined utility function associated with each of the plurality of AI models, wherein the determined optimism point information corresponds to a coordinated judgment from a plurality of judgments made by the plurality of AI models of one or more institutions in response to the first request associated with the first person. To present the aforementioned adjusted judgment, Electronic devices, including those mentioned above.

20. The aforementioned process, Based on the determined utility function associated with each of the plurality of AI models, estimate a first utility value associated with the first decision of each of the plurality of AI models, Based on the constructed causal model and the utility function associated with each of the multiple AI models, a second utility value associated with the counterfactual judgment of each of the multiple AI models is estimated. Based on the estimated first utility value and the estimated second utility value associated with each of the plurality of AI models, a value function is determined that is associated with a set of decisions agreed upon among each of the plurality of AI models. Based on the determined value function, the optimism point information of the determined utility function associated with each of the plurality of AI models is determined. The electronic device according to claim 19, further comprising:

Images