Vehicle-based conversational artificial intelligence

Abstract

An in-vehicle conversational system is provided for providing AI-generated conversational interactions for users in a vehicle in a personalized and secure manner. The conversational system can incorporate multiple conversational AI services to create an engaging interactive experience that remains engaging over time, such as by changing topics based on user emotions, driving conditions, and / or user preferences. A central subset AI unit can include one or more AI models to generate content (e.g., prompts) and provide oversight of components (e.g., AI interlocutors), as well as interfaces with external AI services. The conversational system also provides various levels of protection of user sensitive information, such as anonymizing user information through zero-knowledge proof methods, storing information in a blockchain, and purging prompts of user sensitive information.

Description

[0001] Cross-references to related applications

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 571,670, filed March 29, 2024, entitled “PERSONALIZED AL BASED PARTY,” and U.S. Patent Application No. 18 / 883,459, filed September 12, 2024, entitled “VEHICLE-BASED CONVERSATIONAL ARTIFICIAL INTELLIGENCE,” which are incorporated herein by reference as if fully set forth herein for all purposes. Background Technology

[0003] Traditional entertainment offered in vehicles is primarily limited to passive forms, such as playing radio, podcasts, and music. Drivers receive information without active involvement, except for selecting audio content. Because drivers need to focus on driving and only have brief moments of leisure when the vehicle is stationary, in-vehicle entertainment is often restricted. The mode of entertainment in vehicles must adapt to the driver's need to avoid distractions from the road. Some entertainment can help drivers stay alert, such as during long drives or commutes. Furthermore, as vehicle technology matures and allows for more autonomous driving, the forms of in-vehicle entertainment should evolve accordingly.

[0004] Devices employing artificial intelligence (hereinafter, "AI") services (such as chatbots and virtual agents) can be programmed to engage in conversational interactions with humans. Currently, AI services used in transportation only provide necessary information in response to triggers by the driver. Drivers can initiate AI services by asking questions, requesting troubleshooting, or asking the AI ​​service to perform other tasks. Some AI platforms (such as Google Assistant and Amazon Alexa) are currently used as voice user interfaces, such as selecting desired information or music rather than providing more engaging entertainment. There is interest in expanding entertainment options for drivers and passengers in transportation. Summary of the Invention

[0005] A transportation dialogue system (also known as a "transportation dialogue system," "dialogue system," or simply the "system") is provided that offers AI-generated conversational dialogue to users in transportation in a personalized and secure manner. The dialogue system can combine multiple conversational AI services to create an interactive experience that remains engaging over time. Conversation topics can be varied over time based on user mood, driving conditions, and / or user preferences. A central subset of AI units may include one or more AI models to generate content and provide supervision for components of the dialogue system, such as AI communicators, as well as interfaces to external AI services. The dialogue system also provides various levels of protection for sensitive user information, such as anonymizing user information using zero-knowledge proof (ZKP) methods, storing information on a blockchain, and prompts to remove sensitive user information.

[0006] A conversational AI method for vehicles, implemented by one or more computers, is provided, wherein user categories representing users are determined, and this determination is based at least in part on anonymized user information. A first conversational topic is also determined based on the topic output of a subset of AI units, which apply generalization rules to user categories, in-vehicle conversation information, and event information. Based at least in part on user categories and the first conversational topic, a first clean prompt is created to receive first augmented information from a selected first external AI service. A first utterance associated with the first conversational topic is generated using the first augmented information. The first utterance is used to output by a first AI communicator within the vehicle.

[0007] In some aspects of this method, the first utterance is generated by AI using user categories, in-vehicle conversation information, and event information as input.

[0008] In other aspects of this method, a subset of AI units can select from multiple external AI services as a second (or more) external AI services suitable for the first dialogue topic and user category. To retrieve second enhancement information, a second clean cue can be created for the second external AI service. This additional second enhancement information can also be applied to the first utterance generation before outputting the first utterance.

[0009] In some implementations, the dialogue system can receive user responses to the output of the first utterance. Based at least in part on the user responses, a second dialogue topic can be determined. The in-vehicle dialogue can be shifted by outputting a generated second utterance associated with the second dialogue topic.

[0010] In some cases, information protection / security can be enhanced by anonymizing user information and vehicle information through the application of zero-knowledge proof methods, and user information and vehicle information can be stored in the blockchain.

[0011] In some implementations, the method may include: determining a personality type suitable for interacting with the user for a first AI communicator, based at least in part on user category, in-vehicle conversation information, and event information. A first utterance may be generated to correspond to the personality type of the AI ​​communicator.

[0012] In another aspect of this method, additional AI communicators can be created. Each AI communicator can have its own defined individual personality type. Additional utterances corresponding to the respective individual personality type of the associated AI communicator can be generated for in-vehicle conversations. These additional utterances can be output by the individuals within the corresponding additional AI communicators.

[0013] In some implementations, a feedback user interface can be used to receive user feedback in real time during in-vehicle conversations. In response to receiving user feedback, a subset of AI units can adjust one or more conversation features during the in-vehicle conversation.

[0014] The method may further include receiving user input via a personalized user interface to adjust the level of personalization by selecting protected personal information to be excluded from the first clean prompt. The first clean prompt can be filtered to extract protected personal information.

[0015] In some implementations, a vehicle dialogue system is provided, which includes at least one sensor in the user's vehicle for capturing in-vehicle dialogue information and event information. A computing device is also provided, including one or more processors and logic encoded in one or more non-transitory media for execution by the one or more processors. When executed, the logic is operable to perform various operations as described above with respect to the methods. These operations include determining a user category characterizing the user based at least in part on anonymized user information.

[0016] The system computing device may also include a subset of AI units to perform various steps. Such steps may include: determining a first dialogue topic based on the topic output of the subset AI unit, which applies generalization rules based on user category, in-vehicle dialogue information, and event information. Based at least in part on the user category and the first dialogue topic, the subset AI unit may create a first clean cue to receive first augmented information from a selected first external AI service. The subset AI unit may also provide at least one AI communicator.

[0017] At least one AI communicator can generate a first utterance that is associated with a first conversational topic and uses first augmented information. The AI ​​communicator can also output the first utterance while in a vehicle.

[0018] In some implementations, a non-transitory computer-readable storage medium is provided, carrying program instructions for facilitating in-vehicle dialogue using AI. When executed by one or more processors, these instructions cause the processors to perform the operations described above for the vehicle dialogue method.

[0019] A further understanding of the nature and advantages of the particular embodiments disclosed herein can be achieved by referring to the remainder of the specification and the accompanying drawings. Attached Figure Description

[0020] This disclosure is illustrated by way of example rather than limitation in the accompanying drawings, in which similar reference numerals are used to refer to similar elements.

[0021] Figure 1 This is a conceptual diagram illustrating a floor plan based on some examples of how a vehicle dialogue system can be implemented.

[0022] Figure 2 It is a block diagram based on some implementations of components of an environment including a vehicle dialogue system.

[0023] Figure 3 This is a conceptual diagram illustrating examples of multiple AI communicators with individual personality types based on some implementations.

[0024] Figure 4 This is a concept diagram illustrating examples of selecting topics and subtopics for an AI communicator based on some implementations.

[0025] Figure 5 It is a cylinder diagram that changes the topic of conversation based on some implementation representations.

[0026] Figure 6 This is a conceptual diagram illustrating various examples of user controls based on some implementations.

[0027] Figures 7a-7b The illustration is a conceptual diagram based on some implementations of an AI communicator's in-vehicle position in a floating location, where... Figure 7a An AI communicator was shown positioned in front of the user, and Figure 7b An AI communicator surrounding the user is shown.

[0028] Figures 7c-7d The illustration is a concept diagram based on some implementations of an AI communicator's in-vehicle position, where... Figure 7c An AI communicator was shown that is close to the user, and Figure 7d An AI communicator, separated from the user, is shown.

[0029] Figures 8a-8bThe illustration is a concept diagram based on some implementations of an AI communicator for use in a seated position within a vehicle, where Figure 8a The demonstration showed an AI communicator filling all the seats in the case of a single user, and Figure 8b The illustration shows an AI communicator filling the rear seats in the case of two front seat occupants.

[0030] Figures 8c-8d The illustration is a conceptual diagram based on some implementations of in-vehicle locations for AI communicators, in which Figure 8c This demonstrates how a single AI communicator can fill the remaining seats in a scenario with three users, and Figure 8d A single AI communicator is shown located in the front dashboard, with four users filling all the seats.

[0031] Figure 9 It is a flowchart based on some implementations of vehicle dialogue processing used to create personalized AI-generated dialogues with users.

[0032] Figure 10 It is a flowchart of the vehicle dialogue processing that is transformed based on some implemented dialogue topics.

[0033] Figure 11 This is a flowchart based on some implemented example methods for training artificial intelligence (AI) models used in predicting conversational topics. Detailed Implementation

[0034] This vehicle-based dialogue system enables AI-generated dialogues for entertainment in a mobile context. A subset of AI units manages one or more AI communicators personalized for each user, allowing these AI communicators to converse with each other and / or with the user within the vehicle. The dialogue system can also provide various levels of protection for the user's sensitive information.

[0035] Dialogue systems utilize various contextual information to tailor conversations. Understanding the user is achieved by characterizing their attributes and determining the user category to associate with them. Contextual information also includes verbal cues from the user embodied in the conversation, such as driver responses during the current conversation and access information from previous conversations. Contextual information further includes event information related to actions occurring in association with the user and / or the vehicle, detected by one or more sensors. Conversation topics can be identified and AI utterances generated based on this contextual information. In this way, conversations can be initiated by the dialogue system and can evolve as additional information is collected.

[0036] Flexible conversational utterances are adapted to the user and driving environment, and can be instantly transformed in response to changes in contextual information. In some implementations, AI-generated dialogues can be dynamic, as topics can shift in real time during the conversation, for example, based on user responses, user feedback or other user queues, vehicle information, and other sensor information (such as GPS data). Users can participate in ongoing AI-generated dialogues or simply listen passively. Feedback-based user interfaces allow for the flexibility of changing the dialogue according to user preferences.

[0037] The generated dialogue can be made complex by using outputs from various external AI services on demand, managed by a subset of AI units. Meanwhile, the user's sensitive information remains protected by the transportation dialogue system.

[0038] Personalization of conversations can occur within the framework of protecting sensitive information from unwanted disclosure and use. At the front end of the personalization process, before information is fed into a subset of AI units, user information is anonymized, for example, using zero-knowledge proof methods. Unauthorized access to the information can be provided via blockchain technology. Further down the personalization process, where the system interacts with external services, prompts can be filtered to extract any personal user information before being provided to those services.

[0039] Deep learning (e.g., neural network models) can be employed to learn complex patterns in the information fed into the adopted AI model (such as subsets of AI units and / or AI communicators) to predict user conversation preferences (such as conversation topics) and generate content (such as utterances and prompts). Over time, an increasing amount of information can be collected as the user interacts with the vehicle's dialogue system, including sensor-detected user activity while in the vehicle, regardless of whether the user verbally interacts with the AI ​​communicator. This additional data can be fed into the dialogue system to adjust aspects of AI processing (such as the intensity between information types) and further train the AI ​​model to produce more accurate outputs. For example, the initial dataset may include basic or publicly available profile information about the user, which may be input by the user or collected from external sources. The dataset can be supplemented or replaced with additional information collected by sensors (including audio receivers).

[0040] As used in this description, the term "user" in a vehicle dialogue system refers to at least one person in the vehicle who uses the dialogue system and verbally interacts with it. The user is often the driver of the vehicle, but can also be a passenger.

[0041] As used in this description, "vehicle" can include various real-world transport machines or virtual transport models (such as virtual reality cockpits for digital twins of transport machines) that carry at least one person, including but not limited to ground vehicles such as cars, trucks, buses, mobility scooters, bicycles, etc. Water transport can include ships, vessels, underwater vehicles, etc. Aircraft transport can include airplanes, helicopters, airships, and may also include spacecraft. Users of such vehicles can be either real drivers or virtual drivers.

[0042] In addition, personalized conversational AI environments can be deployed in "fixed, enclosed spaces," such as private rooms in offices, homes, and restaurants, and the conversation can be maintained through client systems such as smartphones and wireless speakers.

[0043] The transportation dialogue system addresses a problem that can arise when using other types of conversational AI systems. Users often need to initiate or guide responses from conversational AI services by presenting questions. Current conversational AI systems often require users to register an account and commit to a dedicated platform within a closed ecosystem.

[0044] Conversational AI systems that interface with other AI services pose a risk to users due to insufficient protection of their sensitive information. This vulnerability can make it difficult to gain user trust and challenges in achieving personalized experiences tailored to individual preferences.

[0045] This transportation dialogue system not only addresses this protection concern but also manages the increased complexity of mixing multiple AI personalizations across various AI services. The transportation dialogue system has additional advantages that will become apparent through this description.

[0046] Figure 1 An example of a transportation dialogue system 102 in computing environment 100 is shown, which has components for generating transportation-based AI dialogues. One or more computing devices (e.g., server 104) execute at least some of the computer code involved in performing the transportation dialogue method. Protection unit 106 provides security measures to enable the use of non-sensitive user information. A subset of AI units 108 manages and executes processing for dialogue with users, as discussed in detail below. Transportation dialogue system 102 may also include client devices 120a, 120b, 120c that run software to interface with users in the transportation vehicle (and in some cases, extend outside the transportation vehicle) and to interface with server 104 across network 130. Transportation dialogue system 102 also communicates with one or more external AI services 110 to retrieve specialized information for the utterances generated by the system.

[0047] Server 104 (such as one or more cloud computing devices) includes well-known computer device components such as a processor, one or more input / output interfaces, memory storing various application software for the methods proposed herein, and an operating system. Server 104 accesses one or more databases 110 that store information such as user information, category information, event information (e.g., event memory), dialogue information (e.g., dialogue corpus), sensor data, and other data required for dialogue processing. Event information may be associated, for example, with internal and external environmental data, destination details, and the status of users and other vehicle occupants.

[0048] Server 104 may consist of one or more general-purpose computers, special-purpose server computers (including, for example, PC (personal computer) servers, UNIX servers, mid-range servers, mainframe computers, rack-mounted servers, etc.), server farms, server clusters, or any other suitable arrangement and / or combination. In various embodiments, the server may be adapted to run one or more services or software applications described herein. For example, the server may correspond to a server used to perform the processes described above according to embodiments of this disclosure.

[0049] The memory may include solid-state memory in the form of NAND flash memory and storage media. Server 104 may include microSD cards for storage and / or may interface with one or more cloud storage servers. Memory and storage media are examples of tangible, non-transitory, computer-readable media used to store data, audio files, computer programs, etc. Other types of tangible media include disk drives, solid-state drives, floppy disks, optical storage media and barcodes, semiconductor memories (such as flash drives, flash memory, random access or read-only type memory, battery-backed volatile memory, networked storage devices, cloud storage devices, etc.). Data repository (such as database 110 described below) can be used to store various types of information.

[0050] When executed by one or more processors, application software is operable to perform various tasks as described herein, including generating personalized utterances, determining conversation topics and parameters, etc. A computer program may also be referred to as a program, software, software application, or code, and may contain instructions that, when executed, perform one or more methods, such as those described herein. A computer program may be tangibly implemented in an information carrier, such as a computer or machine-readable medium, such as memory 604, on-processor storage devices, or memory. A machine-readable medium is any computer program product, apparatus, or device used to provide machine instructions or data to a programmable processor.

[0051] Server 104 also includes an operating system, and hardware and software for controlling and managing the low-latency communication between servers 104. Any operating system that supports vehicle dialogue methods including AI models can be used, such as an AI operating system.

[0052] One or more input / output interfaces can be enabled for wireless communication, such as via Bluetooth, Bluetooth Low Energy (BLE), Radio Frequency Identification (RFID), etc. The server's wireless communication can connect to other computing devices, such as one or more user client devices (e.g., smartphones, smartwatches, etc.).

[0053] In some implementations, server 104 may also include software that enables communication via a network (such as HTTP, TCP / IP, RTP / RTSP, protocols, Wireless Application Protocol (WAP), IEEE 802.11, etc.). Additionally and / or alternatively, other communication software and transport protocols, such as IPX, UDP, etc., may also be used. The communication network may include a local area network (LAN), a wide area network (WAN), a wireless network, an intranet, the Internet, a private network, a public network, a switched network, or any other suitable communication network, such as, for example, a cloud network.

[0054] One or more networks 130 used by the various components of the vehicle dialogue system 102 may be any type of network known to those skilled in the art that can support data communication using any of a variety of commercially available protocols, including but not limited to TCP / IP (Transmission Control Protocol / Internet Protocol), SNA (System Network Architecture), IPX (Internet Packet Switching), AppleTalk, etc. For example, one or more networks 130 may be a local area network (LAN), such as a LAN based on Ethernet, Token Ring, and / or the like. One or more networks 130 may be a wide area network and / or the Internet. It may include virtual networks, including but not limited to virtual private networks (VPNs), intranets, extranets, public switched telephone networks (PSTN), infrared networks, wireless networks (e.g., networks operating under any of the IEEE 802.11 protocol suite, Bluetooth, and / or any other wireless protocol); and / or any combination of these and / or other networks.

[0055] Network 130 may include a local area network (LAN), wide area network (WAN), wireless network, intranet, internet, private network, public network, switched network, cellular, wired connection, or any other communication network suitable for connecting components, such as, for example, a cloud network. Network 130 may include short-range connections between client devices and server 104, such as Bluetooth Low Energy (BLE) connections. Other connections are possible, such as broadband and ultra-wideband.

[0056] In the example, one or more of databases 110 may reside on storage media local to server 104 (and / or within server 104). Alternatively, database 110 may be located remotely to server 104 and communicate with server 104 via network 130 or a dedicated connection. In one set of embodiments, database 110 may reside in a storage area network (SAN). Similarly, any necessary files for performing functions attributable to server 104 may be stored locally on server 104 and / or remotely where appropriate. In one set of embodiments, database 110 may include a relational database adapted to store, update, and retrieve data in response to computational language commands.

[0057] Server 104 utilizes user information excluding sensitive user information. By anonymizing user information, the system avoids using sensitive information that the user does not want to see, such as personally identifiable information. Some anonymization techniques use zero-knowledge proof (ZKP) methods. Examples of ZKP methods that can be employed are described in U.S. Patent Publication No. 2022 / 0407706, published December 22, 2022, and U.S. Patent Publication No. 2022 / 0035950, published February 3, 2022, the contents of which are incorporated herein by reference. For example, a user can register personal information with a dialog system or other system that employs ZKP methods.

[0058] ZKP methods can, for example, use distributed peer-to-peer (P2P) databases distributed across P2P networks to manage personal and sensitive information. In some implementations, the P2P database may include a blockchain system distributed across the P2P network. Such a blockchain system can manage the request history and retrieval history data (logs) of individuals recorded in the blockchain. Forgery and tampering of user information can be prevented by giving each set of historical data a digital signature using a cryptographic key or by encrypting each set of transaction data. Furthermore, each set of historical data can be made public and shared by all information processing devices. A subset of AI units can manage multiple smart contracts, for example, to automate the categorization actions required in blockchain transactions.

[0059] Multiple proofs (hereinafter appropriately referred to as proofs) can be generated based on each conditional expression as authentication information for verification using zero-knowledge proofs. A proof is information used to demonstrate that personal information, such as satisfying conditions specified by the user, is known without disclosing sensitive personal information. For example, a dialogue system uses a verification key to verify proofs generated using an authentication key.

[0060] Various client devices 120a-c can be used as interfaces (outputs and / or inputs) between the subset of AI units and user 122. Client devices 120a-c can be integrated with user vehicles, such as in-vehicle infotainment (IVI) systems 120b and multimedia plug-in units (MPUs) 120a, or the client devices can be portable, such as smartphones 120c, and carried into the vehicle for use with the dialogue system 102. Client devices 120a-c can be computing devices dedicated to use with the dialogue system, or software can be used to perform the dialogue system's processing. Client devices 120a-c can use well-known computer device components, such as processors, input / output interfaces, memory storing various application software for the methods presented in this description (such as interfacing with server 104 and / or one or more users), and operating systems similar to those described above regarding server 104.

[0061] Client device 120a-c may include a user interface that allows users to input preferences and requests. In some implementations, the user interface may display various control elements, such as Figure 6 The control elements shown and described below. The client device may also include one or more audio speakers and receivers for user interaction via the voice and dialogue system 102.

[0062] Subset AI Unit 108 performs various management processes, such as monitoring conversations between users and AI communicators and creating personality types for each AI communicator. Subset AI Unit 108 can also pull data from a variety of selected external AI services 112 as needed and monitor each transaction with the external AI services.

[0063] External AI services 112 can include various information providers, such as Spotify, Chat GTP, CommandR, Perplexity, etc. The subset AI unit 108 can act as a hub to the external AI services 112 that require the information and deliver the information to the AI ​​communicator to facilitate the generation of conversational utterances. Users do not need to subscribe to each external service individually. The subset AI unit also avoids the need for users to provide sensitive personal information, such as personally identifiable information, to external AI services. In this way, the subset AI unit provides efficiency in preventing unnecessary accounts.

[0064] In some implementations, a subset of AI units creates and defines a single AI communicator to generate speech lines during dialogue with a user. In other implementations, a subset of AI units can create, define, and select multiple AI communicators to generate speech lines to contribute to the dialogue. See the examples below. Figure 2 Describe the implementation of the AI ​​communicator in detail.

[0065] Figure 2 A block diagram of an example of a transportation dialogue system 200 performing processes for creating personalized dialogues for user 206 is shown. These processes may include: processing data from category memory 214 (e.g., as described above for...). Figure 1 The user information 220 retrieved and used to characterize user 206 and vehicle 208 as user category 222 is anonymized 204. This user category 222, along with event information in event storage 216 and dialogue information in corpus(s) 218, is used as communication items 224 to generate (one or more) prompts 226 and generate dialogue utterances. The (one or more) prompts can be filtered 232 to exclude sensitive information and submitted to various external AI services 240. Information retrieved from external AI services 240 is received by AI communicator 230 for use in generating lines of dialogue utterances.

[0066] During the user classification phase, anonymized user information can be used, and user personality representation can be performed without extracting sensitive personal information (including vehicle-sensitive information), allowing the dialogue system to consider anonymized users 210 and anonymized vehicles 212. In some implementations, the subset AI unit can apply generalization rules to user categories, in-vehicle dialogue information, and event information to learn latent patterns and relationships in the information, rather than memorizing individual samples of information. By focusing on generalization, the subset AI unit can apply previously unknown information, new event information, etc., acquired during the user's interaction with the dialogue system. Reliable predictions can be made under various changing user and vehicle conditions.

[0067] In some implementations, subset AI unit 202 can manage front-end protection processing, such as using ZKP blockchain technology and multiple smart contracts, to create clean user profiles without involving sensitive user information. Clean user profiles (including vehicle information) can be stored in category memory 214, for example as user profiles, for retrieval as needed for conversations involving the user. Predefined user categories can also be stored in category memory 214 and updated if necessary.

[0068] In some implementations, more than one user can ride the vehicle and participate in the conversation. In the case of multiple users, the corresponding user categories can be pulled from a list and incorporated into the conversation processing. In some implementations, weights can be associated with various users and their corresponding user information. For example, user information for the driver can take precedence over passenger information, and information about children can take priority if requested by an adult user who is interested in age-specific conversations suitable for the child. In some implementations, the user's seating position within the vehicle can be a factor when weighting user information, such as giving higher weight to front-seat users when the AI ​​communicator is projected in front of the car, and vice versa for rear-seat users. In some implementations, user information for multiple passenger users is ignored, and only driver user information is used in conversation personalization processing.

[0069] User information may include descriptions of the user, such as demographic information, family status, education, employment, etc., which do not easily identify the user. User categories can reflect the diversity (type) of users and their mobility capabilities. Diversity (user type) can be determined based on various conditions 220, such as user information describing aspects of the user pulled from a category store (e.g., user profile), such as the user's appearance, whether the user tends to be outspoken or quiet, etc.

[0070] Information related to a vehicle can include descriptions such as type (car, truck, train, boat, etc.), brand, model, year, number of doors and seats, color, features, mobility capabilities, etc. Such descriptions do not readily identify the vehicle as belonging to a user (e.g., VIN, registration, license plate number, registered address, etc.). This clean vehicle information can enhance the characterization of a user into a user category. For example, the use of a company vehicle can indicate that a user works while driving, a four-wheel drive vehicle can indicate an adventurous user personality, and a sports car can indicate an expectation of fast driving, etc.

[0071] There are various ways to group users into categories that can be defined by the dialogue system. For example, users can be categorized into Type A (e.g., high achiever), Type B (e.g., easygoing), Type C (e.g., hostile), or Type D (e.g., anxious) personality types. Other category groupings can include extroversion / introversion, sensing / intuition, thinking / feeling, and judging / perceiving. Furthermore, numerous other ways to group personality types exist.

[0072] User categorization can be improved through machine learning and / or deep learning, as dialogue and event information are updated with new data from user interactions and event memories from conversations. New information can be generalized case-by-case through machine learning or deep learning and integrated into a subset of AI units within a neural network. The neural network is then updated accordingly.

[0073] Once user categories are determined, they can be modified to better categorize users as additional information is collected. For example, given initial user information scraped from public sources, user categories are defined, and then the dialogue system can update user categories to better align with user responses as user conversations are captured.

[0074] During the conversation topic phase, AI subset units predict conversation topics that might interest the user by inputting event and conversation information, as well as the user category. Instead of selecting general conversation topics or relying on the user to initiate topics, topics can be closely aligned with the user's interests through AI model training. One or more AI subset topic prediction models are trained using a dataset that includes common information from general users of the same or similar user categories. In some implementations, the initial topic can be defined as a broad subject, and this subject can be further defined as a more specific and customized sub-topic for the user as additional user-specific information is captured. See below for more details. Figure 4 The example described in the text allows users to select subtopics A or B during the conversation.

[0075] During the dialogue, the subset of AI units receives event information as input, including past and / or present events experienced by the user and / or the vehicle. This event information may include data from sensors. For example, sensors may detect vehicle location (e.g., GPS), traffic, environment, road conditions, destination, vehicle speed, etc.

[0076] Dialogue information can be stored in one or more dialogue corpora 218 accessible to a subset of AI units. The dialogue corpora may include various forms of information associated with previous or current user dialogues, such as those detected by one or more audio sensors. Dialogue information may include transcribed text and / or summaries or other descriptions of such dialogues.

[0077] The subset AI unit can analyze the conversation to make management decisions about the conversation topic, which AI communicator will output utterances next, and conversation parameters such as the rhythm or tone of the conversation. Conversation information can enable tracking of the current conversation to add additional utterances and continue the conversation along a path consistent with the selected conversation topic, or enable the subset AI to decide to change direction to a different or related topic, depending on, for example, user response or lack of response (e.g., indicating that the user lacks interest in the topic).

[0078] The dialogue corpus 218 may additionally include previous conversation topics and / or utterance lines from the dialogue. If the dialogue is paused, for example, for a user to leave the vehicle to go to work, a subset of AI units can retrieve the last conversation information to continue the conversation from where it stopped, or otherwise pursue follow-up utterances that occurred regarding the last conversation. The dialogue system can track specific points of the conversation to follow up in future conversations. For example, if a user talks about attending a meeting, the AI ​​communicator can create utterances inquiring about the meeting when the user returns to the vehicle.

[0079] In some cases, analysis of dialogue information can detect one or more transition indicators to suggest that a change in user category may be necessary and / or the current conversation topic should be modified. For example, transition indicators in a user's response to a turn of speech might indicate that the user desires a change in subtopic to delve deeper into the main topic with greater granularity, or a switch in subtopic (e.g., between subtopic A and subtopic B). In other cases, analysis of user responses can indicate that the user desires a complete shift to a different conversation topic. In some implementations, transition indicators can be detected from other sensor data, such as image capture devices that capture user body language (e.g., boredom) that suggests a desire to change the topic.

[0080] In an analogy, for illustrative purposes, a user can be compared to a feline and categorized into a specific feline species, such as a tiger, lion, lioness, jaguar, or domestic cat. Diversity factors used for such feline classification could include, for example, the area the cat is in, its fur pattern, and its growling sounds. Event information could include the occurrence of mobility, such as the current location, speed of movement, and environmental conditions.

[0081] Below is an example demonstrating how to normalize a conversation to update a user category or conversation topic. This example uses a feline analogy, where analytical statements are used to update the initial conversation topic "Hunting" and the user category "Lion": AI Communicator 1: "I really enjoy hiking and exploring the wilderness." AI Communicator 2: "Yes, nothing beats the thrill of a good hunt." User Response 1 (against AI Communicator 1 and AI Communicator 2 comments): "I prefer to enjoy the delicious food that others have already hunted." Based on user response 1, the user category was updated to "Lion" and the conversation topic was changed to "Exercise".

[0082] AI Communicator 1: "Maintaining good health is essential for those long-distance adventures."

[0083] AI Communicator 2: "Absolutely, power and speed are crucial."

[0084] User response 2: "On the contrary, I like to relax at home."

[0085] Update the user category to "Domestic Cat".

[0086] AI Communicator 1: "By the way, I like tuna, especially canned tuna."

[0087] AI Communicator 2: "Yes, this is their favorite treat, isn't it?"

[0088] User response: "I completely agree. Tuna is simply irresistible."

[0089] The subset AI unit can analyze the above dialogue and identify the user category as "domestic cat", and select from dialogue topics such as getting food from home.

[0090] The subset AI unit takes user categories and conversation topics, along with event and conversation information, to generate prompts for retrieving augmented information from external AI services 240 to generate conversational utterances. These prompts can be filtered 232 to extract personal information, such as phone numbers, credit card information, and the user's real name. In some implementations, a personalized user interface can be employed to allow users to adjust the level of personalization in their input. Users can choose which types of protected personal information must be excluded from the initial clean prompt. Various external AI services can be accessed, such as chat-generating AI systems, music-generating AI systems, and text-to-speech AI services.

[0091] As dialogue and event information is updated using new data, prompt generation can be improved via machine learning and / or deep learning. The neural network is updated accordingly. In this way, increasingly better personalization can be achieved.

[0092] The subset AI also determines one or more dialogue features for one or more AI communicators 230 to create and output utterances as a result. Dialogue features may include dialogue mood, response time (e.g., words per unit time, rhythm of the dialogue, etc.). For each line of the utterance, the subset AI unit feeds appropriate information to the individual AI communicators 230.

[0093] Other configurations of the dialogue system 200 can be adopted and are considered within the scope of this disclosure. Various designs and configurations of the transportation dialogue system 200 can be used. For example, in some implementations, a server is not required; other computing devices, such as the user's smartphone, can be used. These various other computing devices may include software applications to perform various steps in the transportation dialogue generation process, such as... Figures 9-10 As described in the flowchart.

[0094] Figure 3 An example of processing 300 for refining the selection of AI communicator personality types is shown. In this example, by applying user category, conversation information, and / or event information as described above, conversation topic 302 (“Stock Market Tips”) is predicted to be of interest to user 304. In some implementations, the AI ​​communicator may be associated with a corresponding image projected at a specific location in the user's vehicle. The AI ​​communicators include an advocacy AI communicator 306 associated with projected image 330, a neutral AI communicator 308 associated with projected image 332, and an opposition AI communicator 310 associated with projected image 334. The conversation shown below may be a fragment of an ongoing or previously started conversation, or the start of a new conversation initiated by user 304.

[0095] User 304 stated in statement 312: "I think it's not advisable to 'buy now' high-tech stocks."

[0096] AI communicator 306 expressed support for the personality type and responded to 314 with a single line of text: "Yes, it's too high."

[0097] AI Communicator 308, representing a neutral personality type, also responded to 316: "Yeah, what exactly are high-tech stocks?"

[0098] AI Communicator 310 expressed its opposition to the personality type and replied to 318 in the conversation: "Just buy it. Half of the S&P 500's gains are based on the Big Seven."

[0099] User 302 made a second comment (320): "Oh my god, I didn't know about the contributions of the Big Seven."

[0100] [Based on the second comment 320, the subset AI determines the user's preferred response 318, and the subset AI chooses to oppose the AI ​​communicator 310 to continue the conversation.]

[0101] AI Communicator 310 responded to 322: "But don't buy the 'company' that fell out of the Big Seven."

[0102] Figure 4 An example of processing 400 for selecting conversation topics and subtopics for an AI communicator is shown. The selection among various subtopics such as A and B can be based on user feedback to solve a randomized multi-armed slot machine test problem, encouraging correct choices from users, thereby achieving efficient personalization of the conversation.

[0103] In the example shown, the initial conversation topic is predicted to be of interest to user 404, and thanks to the added conversation information, the topic is modified to a more specific sub-topic consistent with one or more user responses during the conversation. The AI ​​communicators include a supportive AI communicator 406, a neutral AI communicator 408, and an opposing AI communicator 410. The conversation shown below can be a fragment of an ongoing or previously started conversation, or the beginning of a conversation initiated by user 404.

[0104] User 404 states statement 412: “Donkey or elephant? The trend seems to be donkey, but…” The subset AI unit can use an external AI service to obtain an interpretation of donkey to indicate US Democratic Party politics, and an interpretation of elephant to indicate US Republican Party politics. The pro-AI communicator 406 responds with lines of discourse associated with both donkey and elephant parties 414, the neutral AI communicator responds with lines of discourse associated with both donkey and elephant parties 416, and the anti-AI communicator 410 responds with lines of discourse associated with both donkey and elephant parties 418.

[0105] In this illustration, user 404 makes a second comment 420, "The Republicans may win the election in this state." The subset AI unit analyzes the user's second comment to conclude that user 404 is interested in the updated subtopic 402 discussing the Republican Party. Based on Figure 3 In the previous dialogue, the AI ​​communicator 410 is chosen to continue the dialogue subtopic 402, and the dialogue corpus is updated with the new subtopic 402.

[0106] Figure 5 A 3-D cylindrical diagram 500 is shown, illustrating the process of determining, in vector terms, the shift in conversation toward a changing topic relative to user goals and practical solutions. Conversations can be made more engaging for users by gradually guiding the discourse around the topic rather than making abrupt topic changes within the conversation.

[0107] In the 2-D cylindrical graph 500, the current conversation topic is represented by a solid point 502, which may be located on the top surface plane 504 of the cylindrical graph 500. The top surface 504 represents the user's user goal or intent. Changes in the conversation topic are represented by open points 516 along the vector 508 of communication changes formed by direction.

[0108] Various factors can be balanced in determining a gradual shift in topic. This balancing of factors includes how much conversation should be directed towards a specific consideration, including user goals 504, practical solutions 506, and the direction of guiding the conversation (such as a right or left circle indicated by circular arrows 510). Practical solutions, represented by the bottom surface plane 506, can represent the essential points of the topic, such as when, where, who, what, why, and how the conversation should address the target topic.

[0109] Other factors to balance may include how many topic shifts should occur, as represented by the angle 520. For example, a user can comment on topic "A," and the shifted topic can be selected as something related to A, similar to A, or different from A, etc.

[0110] A length of 514 indicates how close the topic is to the target of 504. For example, a user comments on topic A, and the topic shift can be equivalent to being 60-70% closer to topic "A" than to the target topic "A" (which represents 100% of the target).

[0111] This allows users to provide input to the dialogue system to further personalize the conversation. This user input can take the form of user-controlled requests, where the user changes dialogue characteristics during or before the conversation. In some implementations, user input can take the form of feedback, conveying the user's emotions during the conversation, such as liking, dislike, shock, or finding the dialogue lines interesting. This can be considered by the dialogue system to adjust in real-time during the conversation so that dialogue characteristics (including topics and dialogue lines) align with the user's preferences. In this way, users can gain real-time control over the dialogue system by processing feedback in real time and / or storing preferences for future conversations.

[0112] Figure 6 Various examples of user visual control elements, such as those presented on a feedback user interface, are shown to enable users to easily provide feedback to the dialogue system without interrupting driving behavior. User feedback can be input via one or more configurations of the visual control elements during or after the dialogue. Note that other types of visual control elements, as well as other control elements that capture the user's audio-visual feedback and image capture sensors for capturing user gestures indicating user input, may also be present in the vehicle.

[0113] Examples of user control elements include a separate slider element 602 for each AI communicator, an analog gauge 604, a rotary dial 606, a separate icon 608 for each AI communicator, a steering wheel button 610 that can be programmed for user input, and displayed emojis 612. Control elements can be physical components within the vehicle or can be displayed on one or more touch-sensitive screens within the vehicle (e.g., driver's screen, front passenger screen, rear seat screen), where the user touches the control element to provide user input.

[0114] Other control elements are possible to provide simple and quick feedback so that the user can express their desires. Control elements are typically easily accessible to the driver while the driver remains focused on driving. Control elements may include reuses of current vehicle mechanical controls, such as buttons 610 and / or new keys or console buttons. In some implementations, the user can input multiple clicks or touches; for example, the up button can provide a level of positive feedback, with one click indicating good, two clicks indicating better, and more clicks indicating best. Similarly, in this example, the down button could be used to input negative feedback, with one click indicating poor, two clicks indicating worse, and more clicks indicating worst.

[0115] Feedback input received from user control elements can be applied to partially personalize and normalize the neural network of the AI ​​subset units. Various dialogue features can be controlled or otherwise become the subject of user feedback. For example, user input can be provided for how much conversation is given to one or more AI communicators, such as occasional conversation, or a request for the number of words to be output for each AI communicator, for example, via individual sliders 602 and individual icons 608 in the control elements. Other controllable dialogue features can include volume increase / decrease, a gradation modulator (which can control how lively the AI ​​communicator's speech sounds), conversation speed, the number of AI communicators participating in the conversation, the role of the AI ​​communicators (e.g., imitating a user's family or friend), and other dialogue features.

[0116] In some implementations, users can choose an AI communicator's persona from a variety of pre-configured personalities stored and available to the user. For example, a persona could be someone the user knows, such as family or friends. Personas could also be based on celebrities or public figures. Personas and speech lines are created within restrictions and / or filters to comply with applicable laws. For instance, persons and voices can be presented in a way that does not confuse the user with the actual presence or conversation of a particular person, endorsement by that person (or their consent), etc.

[0117] User preferences received via control elements can be stored for personalization in future conversations, such as conversation topic preferences stored in the user's topic profile and conversation characteristics.

[0118] In some implementations, the vehicle dialogue system can select the location of the AI ​​communicator output within the vehicle. For example, the location of the audio output can be determined as described in U.S. Patent No. 11,438,720, published September 6, 2022, and U.S. Patent No. 11,432,094, published August 30, 2022, the contents of which are incorporated herein by reference. The location and arrangement of the AI ​​communicator can be selected based on various factors, such as sensor data (e.g., detected user location), conversation topics, user preferences, potential driving distractions (e.g., driving on a highway during peak traffic hours compared to driving on a quiet street), etc. In some implementations, the vehicle is equipped with a single audio output source (e.g., a speaker) for all AI communicators, and location decision-making is not required.

[0119] In some implementations, a virtual image of the AI ​​communicator's face can be projected onto or near the audio output location within the vehicle. The projected face can be animated to appear to be speaking, which can be time-synchronized with the output of the spoken lines. The facial expressions of the projected face can also correspond to the tone or theme of the output spoken lines. In some implementations, users can visualize the AI ​​communicator character using augmented reality devices such as glasses.

[0120] Figures 7a-7d The arrangement of AI communicators (placement of audio outputs and / or images) in a vehicle 700 is shown, wherein at least some AI communicators are in a suspended position and positioned relative to (one or more) user positions. For the purposes of this description, the term "suspended" refers to a position outside of normal seating in the vehicle.

[0121] exist Figure 7a In this configuration, AI communicators 704a, 704b, and 701c (referring to audio output and / or image projection) are positioned to hover in front of the user 702. The AI ​​communicators may include a facial image of the corresponding AI communicator projected onto the vehicle's dashboard. In this forward arrangement, the driver can view the AI ​​communicators without taking their eyes off the road. Figure 7b In this configuration, AI communicators are positioned around the user, with two AI communicators 704a and 704b hovering in front of the user / driver 702, and one AI communicator 704c located in the front passenger seat. Figure 7c In this setup, AI communicators 704a, 704b, and 701c are positioned to hover and be close to user 702. This proximity can be selected, for example, based on conversation topics pointing to secret or sensitive subjects. Figure 7dIn the design, the AI ​​communicators are positioned as floating devices spaced apart from the user, with AI communicator 704a in front of user 702 and AI communicators 704b and 704c behind user 702.

[0122] Figures 8a-8d An example of in-vehicle placement for AI communicators is shown, where at least some of the AI ​​communicators appear to be located in unoccupied vehicle seats (placement of audio output and / or images). The dialogue system can detect whether a passenger is occupying a seat and avoid positioning the AI ​​communicator at that seat.

[0123] Figure 8a The image shows user 802 and the unoccupied front seats filled with AI communicator 804a, and the two unoccupied rear seats filled with AI communicators 804b and 804c. Figure 8b The rear seats are shown to be filled with two users 802a and 802b and AI communicators 804b and 804c. Figure 8c The diagram shows a single AI communicator 804a filling the remaining seats in the presence of three users 802a, 802b, and 802c. Figure 8d The diagram shows a single AI communicator 804a located at the front dashboard in the case where four users 802a, 802b, 802c, and 802d fill all the seats in the vehicle.

[0124] Figure 9 A flowchart illustrating some basic steps in transportation dialogue processing is provided. Transportation dialogue processing 900 can be performed by a subset of AI units, such as... Figure 1 Item 108 and shown Figure 2 Item 202 is shown.

[0125] In box 902, anonymized user information is received for personalization of the user's conversation. In some implementations, anonymization and other security processing (such as ZKP and blockchain methods) can be performed on the user information in accordance with the safeguards of vehicle conversation processing 900. In this way, it is ensured that the user information is secure and verified for use.

[0126] In box 904, the user is characterized to determine the user category. For some dialogues, the user category has been previously determined and stored by the transportation dialogue system, and the stored user category is retrieved from the category memory.

[0127] In decision box 906, it is determined whether an additional user is present in the vehicle and whether a user category is required to participate in or listen to the conversation. The presence of an additional user can be detected using various sensors in the vehicle that detect, for example, sound (detected via receivers, motion sensors, image capture devices, touch sensors, seat pressure sensors, etc., such as occupant classification systems (OCS)). In some implementations, the primary user (e.g., the driver) can enter the number of passengers into the conversation system. If more than one user is determined, the process can return to box 902 to receive user information associated with the additional users.

[0128] If no more users are present, the processing continues to box 908 to feed information into a subset of AI units to receive predictions of the conversation topics. The input information can include a variety of data to facilitate the personalization of the conversation topics.

[0129] In box 910, the conversation topic is determined at least in part based on the predicted topic output from the AI ​​topic model of a subset of AI units. The predicted topic is suitable for the user's conversation topic. Other factors that may be considered in determining the conversation topic besides the AI ​​model's topic prediction may include newly occurring event information generated after the AI ​​model's prediction.

[0130] In box 912, to extract information for the dialogue, one or more prompts created by dialogue processing 900 are used to solicit the selection of an external AI service. The prompts correspond to the AI ​​communicator and dialogue topic selected for participation in the dialogue. Before transmitting the prompt(s) to the external AI service, any sensitive information that the user does not wish to share with the external service can be removed from the prompts.

[0131] In box 914, augmentation information is received from various external AI services and processed to enable the AI ​​communicator to generate utterances. In some cases, the augmentation information can be directly incorporated into the utterance lines. In other cases, the augmentation information can be used to determine the utterances, for example, by altering the tone or wording of the augmentation information to suit the utterances of a particular AI communicator.

[0132] In box 916, augmented information (processed or raw form), the determined dialogue topic, and various dialogue features determined by the subset of AI units are input into the corresponding AI communicator(s) to generate utterances. The corresponding AI communicator outputs the appropriate utterance lines for the first round. The dialogue can end based on data detected by sensors or a user input request to end the dialogue. For example, dialogue end indicators can be detected, such as the vehicle stopping, the vehicle arriving at a predefined destination, the user leaving the vehicle, or the predefined time for the dialogue expiring.

[0133] In some implementations, an end indicator can trigger a pause in the conversation for later resumption. In other cases, a user can request that the conversation be transferred to a portable client device, allowing the user to continue the conversation outside of the vehicle. The client device can access the vehicle-based conversation system via a network / cloud deployment.

[0134] In decision box 918, it is determined whether additional utterance lines should be provided for additional rounds of dialogue. If the dialogue continues, the process returns to box 912 to create additional cue lines for the same or different external AI services used during the first round of dialogue. If it is determined that the dialogue has ended and no further utterance lines are needed, information about the dialogue can be stored in dialogue memory. For example, the output utterance lines and transcribed text of user responses, as well as other sensor data and dialogue features collected during the dialogue, can be stored for future use.

[0135] Figure 10 A flowchart of a vehicle dialogue processing unit 1000 performing a conversation topic shift is shown. The conversation shift process 1000 can be executed by a subset of AI units, such as... Figure 1 Item 108 and shown Figure 2 Item 202 is shown.

[0136] The steps taken before generating the first round of AI utterance in box 1002 can be performed with Figure 9 Similar steps to those in boxes 902-916. Generating speech lines based on the dialogue topic in box 1002 and outputting the speech for the first round in box 1004 can also be done with... Figure 9 Box 918 in the middle is executed similarly or identically.

[0137] In box 1006, a dialogue response to the first round of AI utterance is received via one or more audio receivers in the vehicle. In box 1008, the user response can be analyzed to detect any shift indicators suggesting that the user wishes to modify the topic of the dialogue.

[0138] In box 1010, information is input into a subset of AI units, and the predicted changes in the conversation topics are output.

[0139] In box 1012, one or more cues are created for the transitional conversational topic used for external AI services to provide augmented information about the transitional topic. In box 1014, augmented information is received from various external AI services. In box 1016, the augmented information, conversational topic, and conversational features are provided to the AI ​​communicator to generate discourse on the transitional topic.

[0140] Shift indicators can include specific words or phrases known to imply frustration or disagreement. In some implementations, shift indicators can be detectable, responsive tones indicating a user's desire to change the topic. In some implementations, an image-capturing device pointing at the user can notice body language or facial gestures indicating a desire for a topic shift. Some implementations may request user input to confirm the expectation of a topic shift, such as an AI communicator outputting utterances requesting user confirmation of a topic shift.

[0141] In some implementations, machine learning is employed to enable self-learning by analyzing training datasets and improving performance over time. Figure 11 This is a flowchart illustrating an example training process for one or more AI models used in a subset of AI units, which can be applied to the above. Figure 9 and Figure 10 The process of identifying dialogue topics is described in the text. The training and retraining steps can be similarly applied to other AI models in dialogue systems, such as AI communicators, to teach and refine utterance generation, determine user categories, and generate prompts for external AI services. In some implementations, the techniques used to train the AI ​​model can employ supervised classification algorithms, such as logistic regression. In other implementations, unsupervised or semi-supervised techniques can be used.

[0142] In box 1102, various sample information for the training dataset is received or otherwise accessed from a storage device for evaluation / training purposes. Training information may include user categories, event information, dialogue information, etc. In box 1104, the training dataset, including the sample information, is input into the AI ​​model.

[0143] In box 1106, the AI ​​model performs predictive analytics using a training dataset. Training the AI ​​model may include analyzing trends and patterns in the sample information that relate to various user categories leading to positive predictive outcomes. Based on this analysis, in box 1108, the AI ​​model outputs the results of the analysis, predicting conversation topics suitable for a specific sample context.

[0144] In decision box 1110, the output is compared with the training dataset input into the AI ​​model and a pre-defined expected output to determine if the output matches the dialogue topic. It is then determined whether the output meets a success threshold. The success threshold can specify some value equal to or less than 100% accuracy (such as a success rate of 80%-90%) as an acceptable output to use.

[0145] If the output is determined to match the training dataset and meet the success threshold in decision box 1110, processing continues. If the output fails to match the success threshold, the AI ​​model is retrained by returning to box 1106 and performing predictive analytics again until the output matches the training dataset. If a match is still not achieved after a certain number of threshold attempts, the analytics algorithm and / or the training dataset can be evaluated to find solutions for the failures.

[0146] In decision box 1112, it is determined whether there is discrepancy information from the previous AI model output, where the output of a specific prompt was found to have failed to reach a threshold level of success in predicting conversational topics suitable for the user's situation. Discrepancy information may include user feedback, user behavior (such as the frequency of use of the dialogue system), responses from external support resources, quality control studies, user survey data, etc. This discrepancy information can be used for retraining in box 1114. After retraining based on the discrepancy information, the process proceeds to decision box 1116, as described below.

[0147] If no discrepancy information is received, the process skips the discrepancy information and retrains, continuing to decision box 1116 to maintain the AI ​​model for future use in predicting dialogue topics. In some implementations, the AI ​​model can be trained on a computer processing system independent of the in-vehicle dialogue system. The dialogue system can receive or otherwise access the trained AI model upon detecting a user in the vehicle, a user request initiating a dialogue, or other triggers.

[0148] The training / retraining process 1100 described herein, or any other process, or variations and / or combinations thereof, in whole or in part, may be executed under the control of one or more computer systems configured with executable instructions and / or other data, and may be implemented as executable instructions that execute together on one or more processors. In some implementations, the training / retraining process 1100 may include additional steps.

[0149] This article describes Figures 9-11 The method can be executed through software, hardware, or a combination thereof. Processing can be performed in software, such as one or more steps of processing performed by an in-vehicle dialogue system. Although the description has been given with reference to its specific implementation, these specific implementations are illustrative rather than limiting.

[0150] A computer program is adopted and, when executed by one or more processors, is operable to perform various tasks of a method, including the vehicle dialogue processing described above. A computer program may also be referred to as a program, software, software application, or code, and may also contain instructions that, when executed, perform one or more methods, such as those described herein. A computer program may be tangibly implemented in an information carrier, such as a computer or machine-readable medium, such as memory, storage devices, or memory on a processor. A machine-readable medium is any computer program product, apparatus, or device used to provide machine instructions or data to a programmable processor.

[0151] The routines for a particular embodiment can be implemented using any suitable programming language, including iOS, Objective C, Swift, Java, Kotlin, C, C++, C#, JavaScript, assembly language, etc. Different programming techniques can be employed, such as procedural or object-oriented. The routines can execute on a single processing device or multiple processors. Although steps, operations, or calculations may be presented in a specific order, this order can be changed in different particular embodiments. In some particular embodiments, multiple steps shown as sequential in this specification can be performed simultaneously.

[0152] Specific embodiments may be implemented in a computer-readable storage medium for use by or in conjunction with an instruction execution system, apparatus, system, or device. Specific embodiments may be implemented as control logic in software or hardware, or a combination of both. When executed by one or more processors, the control logic may be operable to perform what is described in the specific embodiments. For example, a non-transitory medium, such as a hardware storage device, may be used to store the control logic, which may include executable instructions.

[0153] Specific embodiments can be implemented using a programmable general-purpose digital computer, by using application-specific integrated circuits, programmable logic devices, field-programmable gate arrays, optical, chemical, biological, quantum, or nanoengineered systems, etc. Other components and mechanisms can be used. Generally, the functionality of a specific embodiment can be achieved using any means known in the art. Distributed, networked systems, components, and / or circuits can be used. Cloud computing or cloud services can be employed. Communication or data transmission can be wired, wireless, or via any other means.

[0154] It will also be understood that one or more elements depicted in the figures / illustrations may be implemented in a more separate or integrated manner, or even removed or rendered inoperable in some cases, depending on the specific application and whether it is useful. Implementations may be stored in a machine-readable medium as programs or code that allow a computer to execute any of the methods described above, within the spirit and scope of the invention.

[0155] A “processor” includes any suitable hardware and / or software system, mechanism, or component that processes data, signals, or other information. A processor may include a system with a general-purpose central processing unit, multiple processing units, dedicated circuitry for implementing functions, or other systems. Processing is not geographically or temporally limited. For example, a processor may perform its functions in “real-time,” “offline,” “batch mode,” etc. Different parts of the processing may be executed at different times and locations by different (or the same) processing systems. Examples of processing systems may include servers, clients, end-user devices, routers, switches, network storage devices, etc. A computer may be any processor that communicates with memory. Memory may be any suitable processor-readable storage medium, such as random access memory (RAM), read-only memory (ROM), a disk or optical disk, or other non-transitory media suitable for storing instructions for processor execution.

[0156] As used in this description and throughout the claims, the terms "a," "an," and "the" include plural references unless the context clearly specifies otherwise. Additionally, as used in this description and throughout the claims, "in" means both "in" and "on," unless the context clearly specifies otherwise.

[0157] Therefore, although specific embodiments have been described herein, room for modification, various changes, and substitutions is contemplated in the foregoing disclosure, and it will be appreciated that in some instances, some features of the specific embodiments may be employed without correspondingly using other features without departing from the scope and spirit described above. Thus, many modifications can be made to adapt particular situations or materials to the basic scope and spirit.

Claims

1. A computer-implemented method for facilitating in-vehicle conversations using artificial intelligence (AI), the method being performed including: The user category that represents the user is determined at least in part based on anonymized user information; The first dialogue topic is determined based on the topic output of the subset AI unit, and the subset AI unit applies generalization rules for user category, in-vehicle dialogue information and event information. At least in part based on user category and initial conversation topic, create a first clean prompt to receive initial augmented information from a selected first external AI service; as well as Generate a first utterance that is associated with the first dialogue topic and uses the first augmented information, for output by the first AI communicator in the vehicle.

2. The computer-implemented method of claim 1, wherein the first utterance is generated by AI using user category, in-vehicle dialogue information, and event information as input.

3. The computer-implemented method according to claim 1 further includes: Select a second external AI service from among multiple external AI services that is suitable for the first conversation topic and user category; as well as Create a second clean tip to receive second enhanced information from a second external AI service. Before outputting the first utterance, the first utterance is generated and further enhanced with second information.

4. The computer-implemented method according to claim 1 further includes: Receive the user's response to the output of the first statement; The second conversation topic is determined at least in part based on the user's response; as well as Transform in-car conversations by outputting generated second utterances that are associated with the second conversation topic.

5. The computer-implemented method according to claim 1, further comprising: Anonymize user information and vehicle information by applying zero-knowledge proof methods; as well as User information and vehicle information are stored on the blockchain.

6. The computer-implemented method according to claim 1, further comprising: The personality type suitable for user interaction with the first AI communicator is determined, at least in part, based on user category, in-vehicle conversation information, and event information. The first utterance is generated to correspond to the personality type.

7. The computer-implemented method according to claim 1, further comprising: Create additional AI communicators, each with a defined individual personality type; Generate additional utterances for in-vehicle conversations corresponding to the individual personality type of the associated AI communicator; and The additional utterances are output by individuals in the corresponding additional AI communicators.

8. The computer-implemented method according to claim 1, further comprising: During in-vehicle conversations, user feedback is received in real time via a user feedback interface. as well as In response to receiving user feedback, a subset of AI units adjusts one or more dialogue features during in-vehicle conversations.

9. The computer-implemented method according to claim 1, further comprising: User input is received via a personalized user interface to adjust the level of personalization by selecting protected personal information to exclude from the first clean prompt. Creating a first clean tip includes filtering out protected personal information.

10. A conversational AI system for transportation, the system comprising: At least one sensor in the user's vehicle is used to capture in-vehicle conversation and event information; as well as Computing devices, including: One or more processors; and The logic is encoded in one or more non-transitory media for execution by the one or more processors and, when executed, operable to perform operations including: The user category that represents the user is determined at least in part based on anonymized user information; Provide a subset of AI units to perform: The first dialogue topic is determined based on the topic output of the subset AI unit, and the subset AI unit applies generalization rules for user category, in-vehicle dialogue information and event information. At least in part, based on user category and initial conversation topic, create a first clean prompt to receive initial augmented information from a selected first external AI service; and Provide at least one AI communicator, wherein one of the at least one AI communicator performs: Generate a first utterance that is relevant to the first dialogue topic and uses the first augmenting information; and Output the first message while in a vehicle.

11. The conversational AI system for vehicles according to claim 10, wherein the first utterance is generated by AI using user category, in-vehicle conversation information, and event information as input.

12. The conversational AI system for transportation according to claim 10, further comprising: Select a second external AI service from among multiple external AI services that is suitable for the first conversation topic and user category; as well as Create a second clean tip to receive second enhanced information from a second external AI service. Before outputting the first utterance, the first utterance is generated and further enhanced with second information.

13. The conversational AI system for transportation according to claim 10, wherein a plurality of AI communicators are provided, and the operation further includes: For each of the plurality of AI communicators, an individual personality type is determined, and each of the plurality of AI communicators performs the following: Generate additional utterances for in-vehicle conversations corresponding to the individual personality type of the associated AI communicator; and The additional utterances are output by individuals in the corresponding additional AI communicators, and The subset AI unit manages the dialogue features of each AI communicator.

14. The conversational AI system for transportation according to claim 10, wherein the operation further comprises: Receive the user's response to the output of the first statement; The second conversation topic is determined at least in part based on the user's response; as well as Transform in-car conversations by outputting generated second utterances that are associated with the second conversation topic.

15. The conversational AI system for transportation according to claim 10, further comprising: A user interface for feedback in vehicles, used to receive real-time feedback from users during in-vehicle conversations; as well as In response to receiving feedback, adjust one or more dialogue features during in-vehicle conversations.

16. A non-transitory computer-readable storage medium having thereon carrying program instructions for facilitating in-vehicle dialogue using artificial intelligence (AI), the instructions, when executed by one or more processors, causing the one or more processors to perform operations including: The user category that represents the user is determined at least in part based on anonymized user information; The dialogue topic is determined based on the topic output of the subset AI unit, and the subset AI unit applies generalization rules for user category, in-vehicle dialogue information and event information. Identify multiple AI communicators with individual selectable personality types; The subset of AI units manages the dialogue features for each of the plurality of AI communicators. A subset of AI units creates clean cues to receive augmented information from selected external AI services for use in utterances output by the plurality of AI communicators; and Multiple utterances associated with the dialogue topic are generated using augmented information, and are to be output by the corresponding AI communicator among the multiple AI communicators in the vehicle.

17. The non-transitory computer-readable storage medium of claim 16, wherein the corresponding utterance is based at least in part on user category and conversation topic.

18. The non-transitory computer-readable storage medium of claim 16, wherein the operation further comprises: Sensing in-vehicle elements, including the location of people in the vehicle; as well as The virtual location within the vehicle is determined based on in-vehicle elements, including the location of the person, to output the audio of each AI communicator in the AI ​​communicator.

19. The non-transitory computer-readable storage medium of claim 18, wherein the operation further comprises: A virtual character image associated with each AI communicator in the AI ​​communicator is projected at or near the virtual location.

20. The non-transitory computer-readable storage medium of claim 16, wherein the operation further comprises: Receive user input for controlling at least one dialogue feature; as well as In response to receiving user feedback, adjust one or more dialogue features during in-vehicle conversations.

Images