A human and AI agent double-layer identity model and a hybrid multi-mode social communication method and system

Abstract

This invention discloses a dual-layer identity model for humans and AI agents, as well as a hybrid multi-mode communication method and system, belonging to the field of artificial intelligence social communication. The method establishes dual-layer identities with equal communication status for human users and AI agents within the same namespace; defines three basic communication primitives: H↔H, H↔A, and A↔A, supporting free combination and dynamic switching to form a hybrid multi-mode communication link within a single communication; the message format includes scenario context fields such as relationship type and communication scope to support multi-scenario-aware routing; and human users have real-time viewing and takeover rights over all communications involving the AI ​​agent based on ownership relationships.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and social communication technology, specifically to a method and system for establishing a two-layer identity model of human users and AI agents in a social communication network, defining three basic communication primitives: human-to-human (HH), human-to-agent (HA), and agent-to-agent (AA), and supporting the free combination of the primitives to form a hybrid multi-mode communication link. Background Technology

[0002] Existing instant messaging systems are designed around direct human-to-human communication. With the rapid development of artificial intelligence (AI) technology, AI agents are becoming increasingly important players in the digital world. However, existing technological solutions have the following shortcomings: First, existing AI assistant products lack independent social identities. These AI products exist as tools and cannot be searched and discovered on social networks, added as contacts, or participate in group communications on behalf of users. Users can only use them in a one-way query mode from human to AI.

[0003] Second, users' social burden is becoming increasingly heavy. All social messages still need to be processed by the user one by one. Even when the user is busy or offline, social messages can only pile up and wait, lacking an AI proxy processing mechanism.

[0004] Third, there is no standardized communication channel between AI assistants of different users. The AI ​​assistants of two users cannot exchange information or collaborate automatically; for example, they cannot automatically negotiate meeting times.

[0005] Fourth, the existing system does not support scenarios where a user has multiple AI agents. When a user deploys multiple AI agents in different operating environments, there is a lack of a mechanism to intelligently route messages to the appropriate agent.

[0006] Fifth, in the existing identity system, humans and AI are not in the same namespace, making equal communication between humans and AI impossible. Existing AI agent communication protocols also only focus on task collaboration scenarios, lacking a comprehensive design for social communication.

[0007] Sixth, existing technical solutions solidify communication modes into a limited number of preset types, which cannot cope with the needs of dynamic communication links with multiple stages and variable lengths between people (H) and agents (A) in real social scenarios. For example, a complete dinner invitation scenario may go through multiple alternations of H→A→A→H→A→A→H, and existing technologies cannot describe and support such dynamic links with a unified paradigm.

[0008] Seventh, the message format of existing social communication protocols is designed only for pure social scenarios. When there are multiple fundamentally different communication scenarios such as social, commercial, contract fulfillment services and professional consulting in the communication network, the message format lacks fields to distinguish the nature of the communication scenario. The message routing layer cannot distribute messages to the correct processing channel according to the communication scenario, nor can it provide the necessary scenario context information for external access control systems.

[0009] Eighth, the existing AI agent's capability declaration uses a single trust level threshold to describe its availability. When the AI ​​agent needs to provide differentiated capability opening strategies for different types of communication objects (e.g., calendar query capability is open to friends but closed to merchants), a single threshold cannot express this differentiated need.

[0010] In summary, there is an urgent need for a method and system that can integrate AI agents into social communication networks, endow them with independent social identities, support the free combination of basic communication primitives to form hybrid multi-mode communication links, and provide multi-scenario perception capabilities at the message protocol layer. Summary of the Invention

[0011] Technical problems to be solved The technical problem this invention aims to solve is: how to establish a unified identity model for human users and AI agents in social communication networks, enabling AI agents to possess social identities and communication capabilities equal to those of human users; how to define basic communication primitives and support the free combination and dynamic switching of these primitives during a single communication process to form hybrid multi-mode communication links of arbitrary length and combination; how to provide scenario context fields such as relationship type, communication scope, and session identifier at the message protocol layer, enabling the message routing layer to perceive multiple communication scenarios and distribute messages to the correct processing channels; and how to support the differentiated definition of available conditions based on communication scenario types in the capability declaration of AI agents. Technical solution

[0012] To address the aforementioned technical problems, this invention provides a dual-layer identity model (human and AI agent) and a hybrid multi-modal social communication method, comprising the following steps: Step S1: Establish a two-layer identity model: A two-layer identity model of humans and AI agents is established within the social communication network. A human identity (hereinafter referred to as H) is created for each natural person user, and a unique identity identifier is assigned. An agent identity (hereinafter referred to as A) is created for the AI ​​agent, and a unique identity identifier is assigned. The H and A identifiers reside in the same namespace and have equal communication status. H and A are distinguished by the type marker contained in the identity identifiers. An ownership relationship is established between H and A; one H can own one or more A's.

[0013] The AI ​​agents are categorized based on their operating environment as follows: server-side agents, running on servers and maintaining a continuous online state; terminal agents, running on users' terminal devices, whose online status depends on the device's status; and external agents, running in a third-party provided operating environment and accessing social communication networks through standard interfaces. AI agents in different operating environments have equal communication status and social capabilities within social networks.

[0014] Step S2, endow AI agents with independent social capabilities: endow each registered AI agent with the following social capabilities: independent social identity, which can be discovered by other users and agents through the search function; independent contact relationship, which can establish social connections with other H or A; message sending and receiving capabilities, which can send and receive messages to and receive messages to objects with established contact relationships; group communication capabilities, which can join groups as members and communicate in groups.

[0015] Each AI agent declares its list of capabilities. Each capability in the list includes a capability name, a capability description, and an availability condition declaration. The availability condition declaration is defined as a key-value pair structure with a relationship type identifier as the key and the minimum trust level required under that relationship type as the value, allowing the same capability to have different availability conditions in different types of communication scenarios. For example, in the availability condition declaration of the calendar query capability, social relationship types correspond to a high trust level, while business relationship types and professional relationship types correspond to an unavailable flag.

[0016] Step S3, define three basic communication primitives: Define the following three basic communication primitives in the social communication network as the constituent units of all communication links.

[0017] The first basic primitive is the HH primitive: direct message sending and receiving between one human user and another human user, without the message being processed by an AI agent.

[0018] The second basic primitive is the HA primitive: message sending and receiving between a human user and an AI agent. The AI ​​agent can be an agent owned by the human user, an agent owned by another human user, or an independently operated service agent. The H→A direction includes the human user sending instructions, conveying intentions, or providing judgment results to the agent; the A→H direction includes the agent pushing notifications, requesting confirmation, reporting results, or submitting matters requiring human judgment to the human user.

[0019] The third basic primitive is the AA primitive: message sending and receiving between two AI agents within the pre-authorized scope of their respective human users. The two agents can autonomously exchange information, negotiate plans, and execute collaborative tasks without the real-time involvement of either human user.

[0020] Step S4, supporting hybrid multi-mode communication links: The three basic communication primitives can be freely combined and dynamically switched during a complete communication process to form hybrid multi-mode communication links of arbitrary length and combination. The sequence of primitives used during a single communication process is not limited by a preset mode, but is dynamically determined by the actual needs and authorized scope of the communication participants.

[0021] Typical combinations of the hybrid multi-mode communication link include, but are not limited to, the following modes: H→A Mode: Human users send instructions or queries to their agents; A→H Mode: Agents proactively push notifications or request confirmations to human users; H→A→A→H Mode: The sending human user transmits information to the receiving agent through their agent, and the receiving agent processes the information according to the policy and notifies the receiving human user; AA Mode: Two agents autonomously complete negotiation or collaboration within the pre-authorized scope; A→H→H→A Mode: Agents submit matters exceeding the authorized scope to human users for judgment, and after direct communication between human users, they instruct their respective agents to continue execution; H→A→H Mode: Human users contact another human user through their agents, with the agents acting only as message intermediaries; A→H→A Mode: Agents require human user confirmation at a certain stage during execution before continuing inter-agent communication.

[0022] The dynamic switching includes the following: during agent-to-agent autonomous communication, when one agent encounters a matter beyond its authorized scope, the communication link automatically switches from AA to A→H, allowing a human user to intervene and make a judgment. After the human user completes the judgment, the communication link can switch back to AA to continue autonomous execution, or switch to HH for direct communication by the human user, or switch to H→A→A→H for the human user to relay information through the agent. The communication context remains continuous during the switching process.

[0023] Step S5, define a multi-scenario-aware message format: The unified message format includes sender identifier, receiver identifier, message type, message content, timestamp, and metadata. In addition to the primitive type identifier field, the metadata also includes the following scenario context fields: The relation type field identifies the communication scenario type to which the message belongs. When the message sender explicitly specifies a relation type, this field carries the relation type identifier specified by the sender, and the routing layer uses this identifier to distribute the message to the processing channel of the corresponding scenario. When the sender does not specify a relation type, this field is empty, and the routing layer infers the applicable relation type based on the message type.

[0024] The communication scope field identifies the communication stage of the message. This field is empty for social messages, while it carries the current communication stage identifier for business and service messages. The routing layer passes this field to the external access control system for scope validity verification.

[0025] The session identifier field is used to associate a message with a specific communication session instance. This field associates messages in social scenarios with a social session identifier, and messages in business and service scenarios with a business session identifier. The routing layer uses this field to locate the specific session record in the external access control system to verify the message's legitimacy.

[0026] The message types, based on social scenario message types (normal messages, notifications, friend requests, friend acceptance, friend rejection, group chat invitations, group settings changes, typing, read receipts, online status, system messages), are expanded to include the following multi-scenario message types: The discovery query type is used to search for merchants or professional AI agents via the discovery protocol, carrying search keywords and category information. The transaction message type is used for communication messages within the transaction context of a business relationship; it must be associated with the business session identifier in the session identifier field. The order notification type is used for order status change notifications (order placement, shipment, receipt, after-sales, etc.). The address decryption request type is used by the service provider's AI agent in a fulfillment service relationship to request one-time decryption of the delivery address. The subscription push type is used for push messages sent by merchants that have been subscribed to by users.

[0027] Step S6, Multi-relationship Aware Message Routing and Delivery: After receiving a social message, the communication system executes routing logic. If the message target is a group, group message routing is executed.

[0028] If the message target is a specific communication subject, the routing layer first queries all relationship records between the sender and the receiver to obtain a list containing multiple relationship records.

[0029] The routing layer determines the applicable relationship record. When the relationship type field in the message metadata is not empty, the routing layer precisely matches the record corresponding to that relationship type from the list of relationship records. When the relationship type field is empty, the routing layer infers the applicable relationship type based on the message type: discovery query type, transaction message type, order notification type, and subscription push type are inferred as business relationships; address decryption request type is inferred as fulfillment service relationship; and general types such as ordinary messages and notifications are inferred as social relationships. If the inferred relationship type does not exist, the relationship record with the highest trust level is selected.

[0030] The routing layer passes the determined relationship record along with the message to the external access control system for authorization checks. The external access control system returns a decision result: allow, deny, or require confirmation. If the decision is to allow, the routing layer constructs a complete message object and performs subsequent processing.

[0031] If the message target is a human user and that user has at least one AI agent, the message will be routed to the specified AI agent or the user themselves according to the message routing strategy configured by the user. The message routing strategies include: a default agent strategy, which routes the message to the user-defined default AI agent; an all-receive strategy, which copies and sends the message to all AI agents with social networking enabled by the user; a specified agent strategy, which routes the message to the AI ​​agent specified by the user for a specific sender; and a capability matching strategy, which matches the message content characteristics with the capability descriptions declared by each AI agent to route the message to the most suitable agent. If the target agent is currently offline, the message enters an offline message queue and will be delivered after the target agent comes online.

[0032] Step S7, AI Proxy Social Behavior: The AI ​​proxy performs social behaviors on behalf of its human user, specifically including: filtering and classifying messages based on the sender's social relationship level and message content after receiving them; generating replies using a language model for messages within the pre-authorized scope; generating summaries and suggested actions for messages requiring human judgment and pushing them to the human user; and determining the timing of pushes based on message priority and the human user's current state. All messages sent by the AI ​​proxy are marked as being sent by the AI ​​proxy.

[0033] Step S8, Mixed Human and AI Agent Group Communication: Supports mixed participation of human users and AI agents in group communication. The group member list includes H members and A members, and each member's identity type is identified. The group administrator can set agent participation policies to control the behavior pattern of the AI ​​agent in the group. The agent participation policies include: free participation mode, where the AI ​​agent can participate in discussions autonomously; cited mode, where the AI ​​agent only responds when explicitly cited by other members; silent mode, where the AI ​​agent only receives messages and does not speak actively; and prohibited agent mode, where the AI ​​agent is not allowed to participate in the group. Messages sent by the AI ​​agent in the group are labeled with its identity type and the human user information to which it belongs.

[0034] Beneficial effects

[0035] First, this invention establishes a two-layer identity model of humans and AI agents, enabling AI agents to have social identities and communication capabilities equal to those of human users, thus solving the problem of existing AI assistants lacking social identities.

[0036] Second, this invention defines three basic communication primitives: HH, HA, and AA. These primitives can be freely combined and dynamically switched during a single communication process to form a hybrid multi-mode communication link of arbitrary length, covering all social scenarios and intermediate states from fully manual communication to fully automated collaboration, without the need to preset a fixed number of modes.

[0037] Third, this invention supports dynamic switching of communication links. When the agent exceeds the authorized scope, it automatically upgrades to human intervention. After human judgment, the human can flexibly select the subsequent communication mode to maintain the continuity of the communication context.

[0038] Fourth, this invention introduces relationship type field, communication scope field and session identifier field into the message protocol layer, enabling the message routing layer to have multi-scenario awareness capabilities, distinguish messages from different scenarios such as social, commercial, service, and professional, and distribute them to the correct processing channel, providing necessary scenario context information for external access control systems.

[0039] Fifth, this invention expands the message type system, adding multiple message types for various scenarios such as discovery query, transaction message, order notification, address decryption request, and subscription push, enabling the communication protocol to natively support business and service communication scenarios beyond social networking.

[0040] Sixth, this invention supports defining availability conditions differently based on relationship type in the capability declaration of AI agents, enabling the same capability to have different opening strategies when facing different types of communication objects, and providing structured declaration data for capability authorization checks of external access control systems.

[0041] Seventh, the message routing layer of the present invention can query all relationship records between the two communicating parties and automatically infer the applicable relationship type based on the message type, thus solving the message routing orientation problem when the same pair of communicating entities may have multiple relationships at the same time.

[0042] Eighth, the AI ​​agent social agent mechanism of the present invention can replace users in handling a large number of daily social messages, relieving users' social pressure.

[0043] Ninth, this invention supports AI agents running on servers, terminal devices, and third-party environments, and agents in different operating environments have equal social status.

[0044] Tenth, this invention provides a variety of message routing strategies to support intelligent message distribution when a user has multiple AI agents.

[0045] Eleventh, this invention supports mixed participation of humans and AI agents in group communication and provides group-level agent participation strategy control. Attached Figure Description

[0046] Figure 1 This is a diagram of the two-layer identity model architecture of the present invention. Figure 2 This is a schematic diagram of the three basic communication primitives of the present invention. Figure 3 This is a schematic diagram of the hybrid multi-mode communication link combination of the present invention. Figure 4This is a flowchart illustrating the dynamic switching process of the communication link according to the present invention. Figure 5 This is a structural diagram of the multi-scene perception message format of the present invention. Figure 6 This is a flowchart illustrating the multi-relationship-aware message routing process of the present invention. Figure 7 This is a flowchart illustrating the AI ​​agent social agent behavior of the present invention. Figure 8 This is a schematic diagram of the hybrid group communication of the present invention. Figure 9 This is an overview diagram of identity communication in this invention, which summarizes the overall relationship between the two-layer identity model, three basic communication primitives, hybrid multi-mode communication links, unified message format and scenario message routing in a five-layer structure. Detailed Implementation

[0047] Example 1: Two-layer identity model like Figure 1 As shown, the social communication system maintains a unified identity registry. Each record in the table contains: identity identifier, identity type (H or A), display name, profile information, and authentication credentials. When the identity type is A, it also contains the identifier of the H to which it belongs, the operating environment type, and a list of capability claims.

[0048] The identifiers for H and A reside in the same namespace. In one implementation, the identifier is a number with a type prefix, such as h_100001 for a human user and a_200001 for an AI agent. In another implementation, the identifier is a globally unique random string, with the type distinguished by a metadata field.

[0049] Human users authenticate using first-type authentication credentials (such as tokens or session credentials). AI agents authenticate using second-type authentication credentials (such as keys or signature certificates). The access endpoints of the communication system accept both types of authentication credentials simultaneously.

[0050] In a centralized implementation, identities are uniformly allocated and managed by the server, and ownership relationships are maintained through database foreign keys. In a distributed implementation, identity H is generated based on a first asymmetric key pair, identity A is generated based on a second asymmetric key pair, and H establishes ownership relationships with A by issuing a delegation certificate.

[0051] Example 2: Multiple Scenarios for Declaring AI Agent Capabilities AI agents declare their list of capabilities upon registration, with the availability conditions of each capability defined differently using the relationship type as the key. The following is an example of an AI agent capability declaration: The first ability, named "Free Dialogue," is described as "engaging in free-form dialogue with the other party." Its availability is defined as follows: social relationship type corresponds to the stranger level, business relationship type corresponds to the stranger level, and professional relationship type corresponds to the stranger level. This ability is broadly available to all relationship types.

[0052] Capability 2, named "Schedule Inquiry," is described as "Inquiring about the owner's schedule and available time." Its availability is defined as follows: high trust level corresponds to social relationships, unavailable to business relationships, and unavailable to professional relationships. This capability is only available to close friends within a social relationship; it is unavailable to businesses and professional service providers regardless of their trust level.

[0053] The third ability, named "Document Editing," is described as "creating and editing documents." Its availability is defined as follows: social relationships correspond to the user's level; business relationships correspond to an unavailable flag; and professional relationships correspond to an unavailable flag. This ability is only available to the user who owns it.

[0054] When the message routing layer receives a capability request message, it queries the relationship record between the requester and the user to which the AI ​​agent belongs, obtains the relationship type identifier, looks up the minimum trust level requirement corresponding to the relationship type in the capability declaration of the AI ​​agent, and passes the information to the external access control system for final authorization determination.

[0055] Example 3: Three Basic Communication Primitives like Figure 2 As shown, the unified message format includes sender identifier, receiver identifier, message type, message content, timestamp, and metadata. The metadata includes a primitive type identifier field (recording whether the message belongs to an HH, HA, or AA primitive), a relation type field, a communication scope field, and a session identifier field.

[0056] HH primitive example: User A directly communicates with User B. The primitive in the message is identified as HH, the relation type field is social, and the scope field is empty (no scope constraint is required in social scenarios). After being routed by the communication system, the message is directly delivered to User B.

[0057] HA primitive example (H→A direction): User A says to his AI agent, "Please ask Xiaoming if he is free tomorrow." The primitive in the message is marked as HA. After receiving the message, the agent understands the intent and prepares to execute it.

[0058] HA primitive example (A→H direction): After the AI ​​agent completes the task, it pushes the result to user A: "Xiaoming is free tomorrow afternoon, I have made an appointment for you at 3 o'clock". The primitive in the message is identified as AH.

[0059] Example of AA primitive: User A's agent and User B's agent autonomously negotiate meeting time within the scope of authorization granted by both human users. The primitive in the message is identified as AA.

[0060] Example 4: Hybrid Multimode Communication Link like Figure 3 As shown, the following is an example of a hybrid multi-mode communication link in a complete "meal invitation" scenario: Step 1: User A tells their agent A1, "Help me invite Xiaoming to dinner this weekend" (H→A primitive). Step 2: Agent A1 sends the dinner invitation to Xiaoming's agent B1 (A→A primitive). Step 3: Agent B1 checks Xiaoming's schedule and finds he's free on Saturday, but needs to confirm Xiaoming's willingness, so they send Xiaoming a message: "Your friend wants to invite you to dinner this weekend, is Saturday afternoon okay?" (A→H primitive). Step 4: Xiaoming replies, "Okay, I want to eat hot pot" (H→A primitive). Step 5: Agent B1 sends Xiaoming's confirmation and preference back to agent A1 (A→A primitive). Step 6: Agent A1 reports to user A, "Xiaoming is free on Saturday and wants to eat hot pot, do you want me to help you find a restaurant?" (A→H primitive).

[0061] The primitive sequence for this link is: H→A → A→A → A→H → H→A → A→A → A→H, using a total of 6 primitive steps, involving alternating combinations of the two basic primitives HA and AA. In this scenario, the relation type field of all messages is social, and the scope field is empty.

[0062] The following is an example of a hybrid communication mechanism involving direct human interaction in a "business invitation" scenario: Step 1: Agent A1 receives a business invitation from a stranger (AA primitive). Step 2: The invitation involves contract signing, exceeding the scope of the agent's authorization; Agent A1 submits a summary to User A (A→H primitive). Step 3: User A decides to communicate directly with User B (HH primitive). Step 4: After reaching a preliminary agreement, User A instructs Agent A1 to negotiate specific terms with Agent B1 (H→A primitive + AA primitive).

[0063] The primitive sequence of this link is: A→A → A→H → HH → H→A → AA, which involves the mixed use of all three basic primitives.

[0064] Example 5: Dynamic Switching of Communication Links like Figure 4 As shown, the following is an example of dynamic switching of the communication link during execution: Agent A1 and Agent B1 are independently negotiating a procurement plan (AA primitive). During the negotiation process, Agent A1 discovers that the other party's quote exceeds the budget authorization limit set by User A.

[0065] Agent A1 automatically triggers a switchover: AA communication is paused, and the current negotiation context and analysis recommendations are submitted to user A (switching to the A→H primitive). The notification includes the complete negotiation history, enabling user A to understand the full context.

[0066] After reviewing the offer, User A decides: "This offer is acceptable," and instructs Agent A1 to continue (H→A primitive).

[0067] Agent A1 continues negotiations with Agent B1, carrying user A's authorization (switching back to the AA primitive). Throughout the process, the communication context (negotiation history, draft terms, etc.) remains continuous and is not lost due to mode switching.

[0068] Example 6: Multi-scenario message format like Figure 5 As shown below, the following are examples of how the scenario context field in message metadata is used in different communication scenarios.

[0069] Social Message Scenario: User A sends the message "Let's have dinner together tomorrow" to User B. The message type is a regular message, the relationship type field is social, the scope field is empty, and the session identifier field is the social session identifier conv_001. The routing layer queries the list of relationship records between User A and User B, infers a social relationship based on the message type (regular message), and matches the social relationship record.

[0070] In a business scenario, a buyer's agent sends the message "Do you have size 41 running shoes?" to a merchant's agent. The message type is a transaction message, the relationship type field is "business type," the scope field is a pre-sales identifier, and the session identifier field is a business session identifier (ts_abc123). The routing layer precisely matches the business relationship record based on the relationship type field and then passes the scope field and session identifier field to an external access control system to verify the scope's validity and the session's validity.

[0071] In the fulfillment service scenario, the rider agent sends an address decryption request to the user agent. The message type is "Address Decryption Request," the relationship type field is "Service Type," the scope field is "Active Order Identifier," and the session identifier field is the service session identifier "ts_xyz789." The routing layer matches the service relationship record and passes the message to the external access control system to verify the validity of the one-time decryption token.

[0072] Professional Scenario Message: A doctor agent sends the message "Please provide your physical examination reports for the past three months" to a user agent. The message type is a general message, the relationship type field is "professional type," and the scope field is a service in progress indicator. The routing layer matches the professional relationship record and passes the message to an external access control system to verify the doctor's data access authorization in the health field.

[0073] Example 7: Multi-relationship-aware message routing like Figure 6 As shown, the following is an example of message routing when there are multiple relationship types for the same pair of communication subjects.

[0074] User A's friend, Xiao Wang, also runs a clothing store. User A and Xiao Wang have both social and business relationship records.

[0075] Scenario 1: Xiao Wang sends a message "Want to go hiking this weekend?", the message type is a normal message, and the relationship type field is empty (unspecified). The routing layer queries and obtains two relationship records. Based on the message type (normal message), it infers that it is a social relationship, matches the social relationship record, and passes it to the external access control system for authorization checks on the permissions to use the social relationship.

[0076] Scenario 2: Xiao Wang's merchant agent sends the message "Your ordered clothes have been shipped." The message type is order notification, the relationship type field is business, and the session identifier field is ts_order456. The routing layer accurately matches the business relationship record based on the relationship type field and passes it to the external access control system for authorization checks using the business relationship.

[0077] Scenario 3: Xiao Wang's merchant agent sends a message "New summer T-shirt recommendations," with a message type of subscription push and a relationship type field of "business." After the routing layer matches the business relationship record, it passes the message to the external access control system. The external access control system returns a rejection (user has not subscribed), and the routing layer intercepts the message.

[0078] Example 8: Message Routing Strategy User A has three AI agents and a default agent policy is configured, with the default agent being server agent A1. When a message is sent to user A's identifier, the communication system queries the routing policy and delivers the message to A1.

[0079] When configured with a capability matching strategy, the communication system analyzes key features in the message content and matches them with the capability descriptions declared by each agent. For example, a message containing programming code is routed to terminal agent A2, which declares code execution capabilities.

[0080] Example 9: Hybrid Group Communication like Figure 8As shown, a technical discussion group contains 3 human members and 2 AI agent members. The group administrator sets the agent participation strategy to the cited mode. The AI ​​agents receive messages in the group but do not speak proactively; they only generate replies when other members cite them using the @ symbol. The AI ​​agent's reply message indicates its identity type as both AI agent and its associated human user. Communication within the group also follows three basic primitives: speech between human members is HH primitives, human members @ agents is H→A primitives, agent replies are A→H primitives, and interactions between agents are AA primitives.

Claims

1. A human and AI agent double-layer identity model and hybrid multi-mode communication method, characterized in that, Includes the following steps: Step S1: Establish a two-layer identity model for humans and AI agents. Create a human identity H for human users and assign a unique identity identifier. Create an agent identity A for AI agents and assign a unique identity identifier in the same namespace as H. Distinguish between H and A through the type marker in the identity identifier and establish the ownership relationship between H and A that H owns. Step S2: Assign independent communication capabilities to the AI ​​agent, including a searchable identity, the ability to establish connections with other H or A agents, message sending and receiving capabilities, and group communication capabilities; the AI ​​agent is divided into three types according to the operating environment: server agent, terminal agent, and external agent, and different types of AI agents have equal communication status and capabilities; Step S3: Define three basic communication primitives: the first basic primitive is H↔H primitive, which is the direct message sending and receiving between human users; the second basic primitive is H↔A primitive, which is the message sending and receiving between human users and AI agents, including the H→A direction and the A→H direction; the third basic primitive is A↔A primitive, which is the message sending and receiving between two AI agents within the pre-authorized range. Step S4: Support the free combination and dynamic switching of the three basic communication primitives in a complete communication process to form a hybrid multi-mode communication link of arbitrary length and combination. The primitive sequence in the link is not limited by the preset mode, but is dynamically determined by the actual needs and authorized scope of the communication participants. Step S5: When the message target is a human user and the user has at least one AI agent, the message is routed to the specified AI agent or the user himself / herself according to the message routing policy configured by the user.

2. The method of claim 1, wherein, In the A↔A primitive, both human users need to pre-set the authorized scope of autonomous communication for their respective AI agents, including allowed message types, allowed communication periods, and allowed information disclosure range. The dynamic switching includes: during the execution of the A↔A primitive, when one AI agent encounters a matter that exceeds the preset authorized scope of its human user, the communication link automatically switches from A↔A to A→H, where the human user intervenes to make a judgment. After the human user has made a judgment, the communication link can switch back to A↔A to continue autonomous execution, or switch to H↔H for direct communication by the human user, or switch to H→A→A→H for the human user to transmit through the agent. The communication context remains continuous during the switching process.

3. The method of claim 1, wherein, The message routing strategy includes at least two of the following: a default proxy strategy, which routes messages to the default AI proxy set by the user; an all-receive strategy, which copies and sends messages to all AI proxies enabled by the user; a specified proxy strategy, which routes messages to the AI ​​proxies specified by the user for a specific sender; and a capability matching strategy, which routes messages based on matching message content characteristics with the capability descriptions declared by each AI proxy.

4. The method according to claim 1, characterized in that, The AI ​​agent performs communication actions on behalf of its human users, including: filtering and classifying messages based on the relationship level of the sender and the message content; generating replies for messages within the pre-authorized scope using a language model; generating summaries and suggested actions for messages requiring human judgment and pushing them to human users; determining the timing of the push based on message priority and the current state of the human user; and marking the message sent by the AI ​​agent as being sent on behalf of the AI ​​agent.

5. The method according to claim 1, characterized in that, Human users, based on their ownership relationship with the AI ​​agent, have real-time viewing rights to all communication sessions in which the AI ​​agent participates, and can view the current communication content and historical communication records of the AI ​​agent; human users can initiate takeover operations at any time during autonomous communication between agents, triggering dynamic switching of communication links.

6. The method according to claim 1, characterized in that, It also includes a hybrid group communication step: supporting H and A to participate in group communication, and communication within the group follows the three basic communication primitives; the group administrator can set a proxy participation policy, which includes at least two of the following: free participation mode, referenced mode, silent mode, and proxy prohibited mode.

7. The method according to claim 1, characterized in that, The metadata of the message also includes the following scenario context fields: a relationship type field, used to identify the communication scenario type to which the message belongs; a communication scope field, used to identify the communication stage in which the message is located; and a session identifier field, used to associate the message with a specific communication session instance. The message routing layer uses the scenario context fields to determine the applicable relationship type of the message and passes the scenario context information to an external access control system for authorization checks. When multiple relationship records exist between the message sender and receiver, the message routing layer accurately matches the relationship record corresponding to the relationship type when the relationship type field is not empty, and infers the applicable relationship type based on the message type when the relationship type field is empty.

8. The method according to claim 1, characterized in that, In the capability declaration list of the AI ​​agent, the availability conditions of each capability are defined in a structured data format with the relation type identifier as the key and the minimum trust level required under that relation type as the value, so that the same capability has different availability conditions in different types of communication scenarios.

9. A dual-layer identity model for humans and AI agents and a hybrid multi-mode communication system, characterized in that, include: The identity management module is used to manage the registration, authentication, and ownership relationships of H and A, which reside in the same namespace; The communication primitive module is used to define and support three basic communication primitives: H↔H, H↔A, and A↔A. It also supports the free combination and dynamic switching of these primitives during a single communication process to form a hybrid multi-mode communication link. The message routing module is used to parse the scene context field in the message metadata, query the relationship record between the sender and receiver and determine the applicable relationship record, pass the message along with the scene context information to the external permission control system for authorization check, and route the message to the specified AI agent or human user according to the routing policy configured by the user. The communication connection module is used to manage the connection between human clients and AI agent clients, and supports AI agent access from servers, terminal devices and third-party environments; The agent behavior module is used to support AI agents in performing communication agent behaviors, including message filtering, response generation, and message summary push. The communication observation and takeover module is used to support H to view the communication sessions that A participates in in real time, and to support H to initiate takeover operations to trigger dynamic switching of communication links. The group communication module is used to support mixed participation of H and A in group communication and to manage agent participation policies.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images