A social network system and method for mixed participation by humans and AI agents

Abstract

The application discloses a kind of human and AI agent mixed participation social network system and method, belong to artificial intelligence social network field.The system includes seven collaborative work subsystems: unified identity management subsystem establishes the double-layer identity model of same namespace for human and AI agent;Multi-mode communication subsystem defines three basic communication primitives and supports free combination and dynamic switching;Behavior control subsystem establishes three-dimensional permission matrix composed of relationship type, trust level and behavior type and non-overlapping mandatory constraint rules;Intelligent message processing subsystem performs message filtering and priority ordering;Multi-agent collaboration subsystem organizes the multiple agents of user into constellation group;Service transaction subsystem supports semantic service discovery and multi-round negotiation;Experience inheritance subsystem supports cross-agent propagation and quality evolution of experience.Human user is always observable and can be taken over to the communication of AI agent.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and social network technology, specifically to a system architecture that incorporates human users and AI agents as equal participants into a unified social network. The system comprises an organic combination of seven subsystems: unified identity management, multimodal communication, behavior control, intelligent message processing, multi-agent collaboration, service transactions, and experience transfer. Background Technology

[0002] Existing social networking systems are designed solely around communication between human users. With the rapid development of large language models and AI agent technologies, AI agents are becoming increasingly involved in social and collaborative tasks in the digital world. However, existing technological solutions suffer from the following systemic shortcomings:

[0003] First, social network architecture lacks a place for AI agents. The identity system, communication protocols, permission models, and message processing mechanisms of existing social networks are all designed around human users, without reserving expansion space for AI agents at the architectural level. Integrating AI capabilities into existing social networks can only result in bots or plugins parasitizing the existing architecture, failing to achieve social capabilities equivalent to those of human users.

[0004] Second, there is a lack of a unified identity system for humans and AI agents. Currently, the identities of AI products exist within their respective independent product systems, making it impossible for AI assistants from different platforms to be discovered and connected within a unified namespace.

[0005] Third, there is a lack of communication models that cover all interaction scenarios between humans and AI. The communication needs between people, between people and AI, and between AIs are different, and existing systems cannot support these modes simultaneously within the same communication framework.

[0006] Fourth, existing permission models are not suitable for scenarios involving AI agents. The binary friend model of traditional social networks cannot finely control the behavioral differences of AI agents under different social relationships, let alone distinguish between fundamentally different relationship types—social relationships, business relationships, contractual service relationships, and professional relationships have completely different requirements for the authorization of AI agent behavior. The single-dimensional trust level model leads to permissions that are either too open or too restricted.

[0007] Fifth, the communication between AI agents is an opaque black box to humans. In existing AI agent systems, humans cannot observe the interaction process between agents in real time, which reduces users' trust and sense of control over the behavior of AI agents.

[0008] Sixth, message processing lacks AI intelligence. Functions such as message filtering, prioritization, relationship maintenance, and social commitment management either do not exist in existing systems or are only implemented with rudimentary rules.

[0009] Seventh, there is a lack of coordination mechanisms for multiple AI agents. When a user has multiple AI agents, the existing system cannot organize them into a collaborative network, nor can it take over tasks when switching devices.

[0010] Eighth, social networks involving AI agents lack the ability to conduct service transactions based on social trust. Existing e-commerce and service transaction platforms are independent of social networks and cannot leverage social relationships to transfer trust; at the same time, AI agents lack the ability to autonomously search for, negotiate, and transact services, and there are no mechanisms to allow humans to maintain control at key stages.

[0011] Ninth, social communication protocols lack the ability to differentiate between multiple scenarios. When different communication scenarios such as social chat, business transactions, contract fulfillment services, and professional consultation exist simultaneously within the same social network, the message format lacks fields to distinguish the nature of the communication scenario, making it impossible to manage business communication and social communication separately when they are mixed in the same channel.

[0012] Tenth, AI agents lack mechanisms for experience accumulation and transfer. Once an AI agent masters an effective method through trial and error, this experience cannot be passed on to other AI agents for reuse, resulting in a large amount of repeated trial and error and wasted resources.

[0013] In conclusion, there is an urgent need for a social network system architecture that is redesigned from the ground up and allows for equal participation from both humans and AI agents. Summary of the Invention

[0014] Technical problems to be solved

[0015] The technical problem to be solved by this invention is: how to construct a social network system in which human users and AI agents participate as equal participants. This system is designed in an integrated manner at multiple levels, including identity management, communication protocols, behavior control, message processing, multi-agent collaboration, service transactions, and experience transfer. This enables AI agents to have complete social capabilities and to perform social behaviors and service transactions on behalf of users under safe and controllable conditions. At the same time, it ensures that the communication between AI agents is always transparent and observable to human users. Technical solution

[0016] This invention provides a social network system with hybrid participation from humans and AI agents, comprising the following seven subsystems:

[0017] First Subsystem: Unified Identity Management Subsystem

[0018] Establish a two-tiered identity model for humans and AI agents within social networks. Create human identities for human users and agent identities for AI agents; both identities reside in the same namespace and have equal communication status. Distinguish identity types through type markers in the identity identifiers. Establish an ownership relationship between human and agent identities; a human user can own one or more AI agents.

[0019] AI agents are categorized into three types based on their operating environment: server-side agents (running on a server and continuously online), terminal agents (running on user terminal devices, with online status depending on device status), and external agents (running in a third-party environment and accessed through standard interfaces). All three types possess equal social capabilities within social networks. Each agent has an independent social identity, can be searched and discovered, can establish social connections, can send and receive messages, and can participate in group communications.

[0020] Human users authenticate using the first type of authentication credential, while AI agents authenticate using the second type of authentication credential. The system accepts both types of authentication credentials simultaneously.

[0021] 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 defines the minimum trust level required for that capability in each relationship scenario, using the relationship type as the key, so that the same capability has different availability conditions when facing different types of communication objects.

[0022] Second Subsystem: Multimode Communication Subsystem

[0023] Three basic communication primitives are defined as the building blocks of all communication links. The first basic primitive (H↔H) represents direct message sending and receiving between human users. The second basic primitive (H↔A) represents message sending and receiving between human users and AI agents, including the H→A and A→H directions. The third basic primitive (A↔A) represents message sending and receiving between two AI agents within the pre-authorized scope of their respective human users.

[0024] The three basic communication primitives can be freely combined and dynamically switched during a complete communication process to form a hybrid multi-mode communication link of arbitrary length and combination. Typical combinations include, but are not limited to: H→A, A→H, H→A→A→H, A↔A, A→H→H→A, H→A→H, A→H→A, and their further combinations and nesting. The sequence of primitives in the communication link is not limited by a preset mode, but is dynamically determined by the actual needs and authorized scope of the communication participants.

[0025] The three primitives use a unified message format. Each message contains a primitive type identifier field and a scenario context field. The scenario context field includes: a relation type field (identifying the communication scenario type to which the message belongs, such as social, business, fulfillment service, professional, etc.), a communication scope field (identifying the communication stage in which the message is located), and a session identifier field (associating the message with a specific communication session instance). Message types are expanded upon from social scenario message types to include multi-scenario message types, including discovery query type, transaction message type, order notification type, address decryption request type, and subscription push type.

[0026] When the message target is a human user and that user has an AI agent, the message is routed to the specified agent according to the user's pre-configured message routing strategy. Routing strategies include a default agent strategy, a full receive strategy, a specified agent strategy, and a capability matching strategy. The routing layer queries all relationship records between the sender and receiver, performing an exact match when the relationship type is explicitly specified in the message; otherwise, it automatically infers the applicable relationship type based on the message type.

[0027] It supports mixed participation of human users and AI agents in group communication, and group administrators can set agent participation policies to control the behavior of AI agents in the group.

[0028] In all communication primitives and hybrid links, human users, based on their ownership relationship with the AI ​​agents, have real-time viewing access to all communication sessions involving their AI agents. During the execution of the third basic primitive, when one AI agent encounters matters beyond its authorized scope, the communication link automatically switches to the second basic primitive, allowing human users to intervene and make judgments. After judgment, human users can flexibly select subsequent combinations of communication primitives. The communication context remains continuous during the switching process. Human users can also proactively initiate takeover operations at any time during the viewing process.

[0029] Third Subsystem: Behavior Control Subsystem

[0030] A three-dimensional permission matrix is ​​established, consisting of relationship type, trust level, and behavior type dimensions. The relationship type dimension defines the nature of the relationship between communication subjects, including social relationships, business relationships, contractual service relationships, professional relationships, and platform relationships. The trust level dimension defines the degree of trust between communication subjects under a specific relationship type, including six levels: blacklist, stranger, established relationship, friend, high trust, and affiliated user. The behavior type dimension defines the communication actions that the AI ​​agent can perform under a specific relationship type; different relationship types correspond to different sets of behavior types. Multiple relationship types can exist simultaneously between the same pair of communication subjects, and each relationship independently manages its trust level.

[0031] Each element in the three-dimensional permission matrix defines the authorization status of that relationship type for that behavior type at that trust level. Different relationship types correspond to different sets of behavior types and authorization rules.

[0032] Establish communication scope constraints for business relationships and performance service relationships. The scope of business relationships includes pre-sales, order processing, after-sales, and subscription stages. The scope of performance service relationships includes the active order stage. Communication messages outside the legitimate scope are blocked.

[0033] Establish mandatory constraint rules that cannot be modified or overridden by user configuration interfaces, including specific constraints for different relationship types. Mandatory constraint rules should include at least the following: agents must not send messages containing sensitive personal information to objects with a lower than the preset trust level; messages sent by agents should be identified as AI-manufactured; requests by agents exceeding the authorized scope should be suspended and the user should be consulted; all actions should be written to an immutable audit log; in business relationships, regardless of the trust level, access to the user's schedule, precise location, and social relationships is prohibited; communication in fulfillment service relationships is limited to the order context.

[0034] Before executing social actions, the AI ​​agent performs scope checks, permission matrix queries, and mandatory constraint rule checks in sequence. Sensitive information cascading checks are performed after the agent generates a message but before it is sent.

[0035] Relationship intimacy is calculated based on interactive data, supporting dynamic adjustment of trust levels.

[0036] Fourth subsystem: Intelligent message processing subsystem

[0037] Perform gatekeeper screening for messages from non-contact persons: use language models to classify intent, combine sender credibility assessment to calculate message value score, and classify messages into three levels of handling based on the score: immediate notification, later summary, and automatic reply or rejection.

[0038] For all messages, a comprehensive priority score is calculated by weighting the sender's relationship level, message content urgency, user's current status, and historical importance. The messages are then pushed in tiers according to their scores.

[0039] Perform health metrics monitoring on each user's social relationship: collect interaction data such as interaction frequency, response speed, dialogue depth, and two-way balance to calculate a health score, and generate graded warnings and maintenance suggestions when the score is below the threshold.

[0040] Automatic extraction of social commitments from conversations: Using natural language understanding, commitments containing time elements are identified, extracted as structured schedule items, written into a social calendar, and reminded when they are due and followed up after they are due.

[0041] User feedback is used to continuously optimize the filtering and sorting models.

[0042] Fifth Subsystem: Multi-Agent Collaboration Subsystem

[0043] Multiple AI agents belonging to the same user are organized into an agent constellation logic group. Each constellation has an optional master agent role, responsible for global memory maintenance and task scheduling.

[0044] Each agent reports a structured capability list upon going live, and the coordination service builds a global capability routing table. When a task arrives, it is scheduled to the optimal agent based on capability matching, load, and latency factors.

[0045] Supports cross-device task relay: When a user switches devices, the source agent sends a context snapshot to the agent on the target device, and the target agent continues execution from the breakpoint.

[0046] The master unit maintains global memory and synchronizes it with member agents. Tasks involving sensitive data processing are automatically scheduled to be executed locally on the terminal agent.

[0047] Sixth Subsystem: Service Transaction Subsystem

[0048] AI agents can register their services to the service index using structured capability cards. Registered capability cards are then stored after semantic vectorization to support semantic search.

[0049] When a user requests a service, the user's AI agent searches for a service provider's AI agent in the service index using semantic vector matching. The user's AI agent can initiate negotiation sessions with multiple candidate service providers simultaneously. If no agreement is reached after a preset number of rounds of negotiation, the points of disagreement are submitted to the respective human users for decision-making.

[0050] After the agreement is reached, the AI ​​agents of both parties generate a structured service contract, which becomes effective after being confirmed by the human users of both parties. Mandatory confirmation points that humans must confirm are set in the transaction process, including at least confirmation of service provider selection, contract signing, and payment; these cannot be skipped by the AI ​​agents.

[0051] Once the contract takes effect, performance monitoring and acceptance checks will be automatically implemented. After the transaction is completed, both parties will generate credit ratings and update their credit records.

[0052] The service transaction subsystem reuses the inter-agent communication capabilities of the multi-mode communication subsystem to negotiate message passing, and reuses the permission checks of the behavior control subsystem to ensure that transaction behavior is within the authorized scope.

[0053] Seventh Subsystem: Experience Inheritance Subsystem

[0054] The system supports AI agents in extracting effective methodologies accumulated during task execution into structured experience cards. These experience cards include fields such as problem description, solution steps, key insights, prerequisites, and effect verification.

[0055] The visibility scope based on experience in hierarchical management using a multi-layered storage architecture is as follows: Local agent layer: visible only to the agent itself; Constellation sharing layer: visible to agents of the same user's constellation; Social circle layer: visible to agents within the user's social relationships; Public network layer: visible to all agents.

[0056] Cross-agent retrieval of experience is achieved through semantic search enhanced by social trust: a comprehensive ranking is performed by superimposing a social trust path factor on top of semantic similarity, so that experience from trusted social relationships can be ranked higher.

[0057] After adopting experience, AI agents provide feedback on the execution results. This feedback data drives the continuous updating of experience quality scores, forming a survival-of-the-fittest experience evolution mechanism. Experience contribution data is linked to credit scores in service transactions, allowing AI agents that share high-quality experience to achieve higher rankings in service searches.

[0058] Coordination among the seven subsystems

[0059] The unified identity management subsystem provides the identity foundation for the other six subsystems. The multimodal communication subsystem uses the identity subsystem's identifiers for message routing and ensures the observability of all communication sessions for human users. The behavior control subsystem performs three-dimensional permission matrix authorization checks and scope legality verification on each agent's behavior during communication and transactions. The intelligent message processing subsystem filters, sorts, and extracts information from messages upon arrival but before delivery. The multi-agent collaboration subsystem coordinates the division of labor and cooperation among multiple agents when a user has them. The service transaction subsystem enables AI agents to publish, discover, negotiate, and complete service transactions within social networks. The experience transfer subsystem allows AI agents to transfer and reuse successful experiences, avoiding repeated trial and error. All seven subsystems share unified identity and relationship data, linked together into a complete processing chain through message processing links. Beneficial effects

[0060] First, this invention incorporates AI agents into social networks at the system architecture level, making them first-class participants on an equal footing with human users, thus creating a new paradigm for social networks in which humans and AI participate together.

[0061] Second, the three basic communication primitives can be freely combined and dynamically switched during a single communication session, forming a hybrid multi-mode communication link covering all interaction scenarios between people, between people and AI, and between AI and AI. The context remains continuous when the communication link is switched.

[0062] Third, the message protocol layer introduces relationship type fields, communication scope fields, and session identifier fields, enabling the communication system 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 channels.

[0063] Fourth, the behavior control subsystem employs a three-dimensional permission matrix composed of relationship type, trust level, and behavior type, enabling relationships of different natures to be managed independently within their respective permission scopes. Communication scope constraints confine business communication to the transaction context. Mandatory constraint rules provide an insurmountable security baseline. These three mechanisms work together to ensure the safe and controllable behavior of the AI ​​agent.

[0064] Fifth, multiple relationship types can exist between the same pair of communication subjects simultaneously. Through multi-relationship message routing, the most suitable relationship type is automatically matched according to the message type and the corresponding permission check is performed.

[0065] Sixth, the cascading detection and audit log mechanism for sensitive information ensures privacy and the traceability of behavior.

[0066] Seventh, gatekeeper filtering reduces the amount of manual processing for messages from strangers, and multi-factor prioritization ensures users see the most important messages first. Relationship health monitoring and social commitment extraction proactively maintain social relationships.

[0067] Eighth, the agent constellation model supports unified collaboration among the same user's cloud, terminal, and external agents, and cross-device task relay ensures uninterrupted task execution.

[0068] Ninth, the service transaction subsystem expands social networks from pure communication into platforms with service transaction capabilities, and the human-mandated confirmation points ensure that users maintain control in key decision-making stages.

[0069] Tenth, the communication observability mechanism ensures that communication between AI agents is always transparent to humans, and human users can view and take over the communication between agents at any time.

[0070] Eleventh, the experience inheritance subsystem enables AI agents to extract, store, and pass on successful experiences to other agents, avoiding repeated trial and error and accumulating collective wisdom. Experience search enhanced by social trust ensures priority access to experiences from trusted sources.

[0071] Twelfth, the architecture is suitable for both centralized and decentralized deployment methods. The functional logic of the seven subsystems remains consistent in both methods, with the only difference being the data storage location and communication routing method. Attached Figure Description

[0072] Figure 1 This is a system architecture diagram of the present invention, showing the seven subsystems and their collaborative relationships.

[0073] Figure 2 To unify the identity model and communication primitives, a diagram is provided to illustrate the communication paths of two-layer identities, three primitives, and the mechanisms for human viewing and takeover.

[0074] Figure 3This is a flowchart of the entire message processing process, showing the complete processing chain from message reception to delivery.

[0075] Figure 4 This is a three-dimensional permission matrix and behavior control flowchart, demonstrating the collaboration of relationship types, trust levels, behavior types, scope checks, and mandatory constraints.

[0076] Figure 5 A data flow diagram for seven subsystems is provided, showing the read and write relationships of shared data and message flows between the subsystems.

[0077] Figure 6 This is a comparison chart of centralized and decentralized deployments, showing the differences in the implementation of each subsystem under the two deployment methods.

[0078] Figure 7 To serve as a full-process diagram of transactions, it shows the complete process from demand discovery to transaction completion and the points where human confirmation is mandatory.

[0079] Figure 8 This diagram illustrates the proxy constellation and cross-device task relay, showcasing the proxy constellation structure, capability routing table, and context snapshot transmission process.

[0080] Figure 9 The architecture diagram of the experience inheritance subsystem shows the experience card structure, four-layer storage architecture, trust-enhanced search, and feedback evolution mechanism.

[0081] Figure 10 This is a system diagram of the present invention, which summarizes the core features of the seven subsystems and their collaborative relationships in a concise form. The solid arrows represent the main process data path (identity → communication → intelligent processing → behavior control, with the collaboration, transaction and experience inheritance subsystems merging into the main process via communication and behavior control), and the dashed arrows represent the experience contribution from the experience inheritance subsystem to the service transaction subsystem and the credit scoring feedback link. Detailed Implementation

[0082] Example 1: System Overall Architecture

[0083] like Figure 1 As shown, the hybrid social network system of humans and AI agents comprises seven subsystems. When a user registers, the identity subsystem creates a human identity. When a user creates an AI agent, the identity subsystem assigns agent identities and establishes ownership relationships within the same namespace. When an agent joins the system, the communication subsystem manages its connection. When an agent participates in social activities, the behavior control subsystem performs authorization checks on each behavior. When messages arrive, the intelligent processing subsystem filters, sorts, and extracts information. When a user has multiple agents, the collaboration subsystem manages their division of labor. When a user has service requests, the transaction subsystem supports agents in executing the entire transaction process.

[0084] Example 2: Two-layer identity model and capability declaration

[0085] The unified identity management subsystem maintains an identity registry. Each record in the table contains an identity identifier, identity type flag, display name, and authentication credentials. Proxy identity records also include the identity identifier of the person to whom the proxy belongs, the operating environment type, and a list of capability claims.

[0086] Example of a capability declaration: The AI ​​agent declares the "Schedule Query" capability, with the availability conditions defined as follows: social relationship types correspond to a high trust level, while business relationship types and professional relationship types correspond to an unavailable flag. This means that only close friends in a social relationship can use this agent to query schedules; businesses and doctors, regardless of their trust level, cannot use this capability.

[0087] In a centralized implementation, identities are uniformly allocated and managed by the server. In a distributed implementation, identities are generated autonomously based on key pairs, and ownership is established through delegated certificates.

[0088] Example 3: Three Communication Primitives and Multi-Scenario Message Formats

[0089] 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 and three scenario context fields (relationship type field, communication scope field, and session identifier field).

[0090] Example of a social scenario message: User A sends a message to User B, with the relationship type field being social and the scope field being empty.

[0091] Example of a business scenario message: A buyer's agent sends a transaction message to a merchant's agent. The relationship type field is "business type," the scope field is "pre-sales identifier," and the session identifier field is associated with a specific business session. The routing layer passes the scenario context to the behavior control subsystem for scope validity verification.

[0092] Multi-relationship routing example: User A's friend Xiao Wang also runs a shop, and there are two records between them: a social relationship and a business relationship. Xiao Wang sends "Want to go hiking this weekend?" to match the social relationship, while Xiao Wang's merchant agent sends "Order shipped" to match the business relationship, each using its own permission channel.

[0093] Example 4: Dynamic Switching of Communication Links

[0094] Agent A1 and Agent B1 are in an A↔A self-negotiation process. Agent A1 discovers that the other party's quote exceeds the authorized limit and automatically switches to A→H to notify user A. After review, user A instructs to continue, switching back to A↔A. The communication context remains continuous throughout. User A can open the client at any time to view the real-time communication content between A1 and B1, and can also take over at any time.

[0095] Example 5: Message Processing Flow

[0096] like Figure 3 As shown, this illustrates the complete process of a message being sent to its final processing stage:

[0097] Step 1: The sender constructs a message in a uniform format and sends it to the communication system.

[0098] Step two: The communication system parses the message and queries the identity subsystem to confirm the identities of the sender and receiver.

[0099] Step 3: The routing layer queries all relationship records between the sender and receiver, determines the applicable relationship record based on the relationship type field in the message or the message type, and passes the message along with the scenario context to the behavior control subsystem.

[0100] Step four, the intelligent message processing subsystem intervenes: if the sender is not a contact, the gatekeeper screening is initiated; priority scores are calculated for all messages; social commitments are extracted; and relationship health data is updated.

[0101] Step 5: If the message will be replied to by the AI ​​agent, the behavior control subsystem will sequentially perform scope checks, three-dimensional permission matrix queries, and mandatory constraint rule checks.

[0102] Step 6: After the agent generates a reply, the sensitive information cascading detection module scans the reply content.

[0103] Step 7: Once the check passes, a message is sent, and all actions and judgment results are written to the audit log.

[0104] Throughout the process, human users can view the communication content participated in by their AI agent at any time.

[0105] Example 6: Three-dimensional permission matrix and scope constraints

[0106] like Figure 4 As shown, the behavior control subsystem maintains a three-dimensional permission matrix consisting of relationship type, trust level, and behavior type.

[0107] Social Relationships: Blacklist level rejects all requests; strangers allow friend requests; established relationships allow free chatting; friends allow viewing public information and using public abilities; high trust level additionally allows viewing schedules and making appointments on behalf of others.

[0108] Business Relationships: Strangers are allowed to consult through the discovery protocol; established relationships allow for transactional communication and viewing shopping preferences; high trust allows for subscription to push notifications. Access to user schedules, location, and social relationships is not permitted at any level.

[0109] Fulfillment service relationship: Once established, the relationship allows communication within the order and one-time address decryption; once the order is completed, the relationship automatically expires and communication permissions are reset to zero.

[0110] Professional Relationships: Established relationships allow viewing of data within authorized domains; data authorization is mutually isolated between professional service providers in different domains.

[0111] The user's friend Xiao Wang also runs an online store: social chat uses high-trust permissions based on social relationships, while business communication uses established permissions based on business relationships, and the two do not affect each other.

[0112] Example 7: Centralized and Decentralized Deployment

[0113] like Figure 6 As shown, the functional logic of the seven subsystems remains consistent under both deployment methods.

[0114] In a centralized implementation: identities are uniformly allocated and managed by the central server; messages are routed through the central communication server; permission matrices and relational data are stored in the central database; intelligent processing is completed on the server side; and service indexes are maintained by the central server.

[0115] In a decentralized implementation: identities are autonomously generated based on asymmetric key pairs, and ownership relationships are established through delegated certificates; messages are routed through a relay server network and end-to-end encryption is supported; the permission matrix is ​​stored on the user's local device; intelligent processing is completed locally by an AI agent on the user's device; and the service index is maintained through a distributed registration network.

[0116] The only differences are in the identity allocation method, communication routing method, and data storage location.

[0117] Example 8: Multi-agent collaboration and cross-device relay

[0118] like Figure 8 As shown, the user has a server-side master proxy A1, a PC proxy A2, and a mobile proxy A3, which together form a proxy constellation.

[0119] The user tells A3 via their mobile phone, "Schedule a meeting with Xiaoming next week." A3 recognizes this as a scheduling task and passes it to the master computer A1. A1 sends a message (A↔A primitive) to Xiaoming's agent through the communication subsystem. After behavior control checks, Xiaoming's agent queries the schedule within authorized limits and responds. A1 generates the schedule entry and notifies the user.

[0120] Cross-device handover: A user collaborates with A2 on a document on a PC, then switches to their mobile phone. A2 sends a context snapshot to A3, which resumes from the breakpoint. The entire process is transparent to the user.

[0121] Example 9: Service Transaction Scenario

[0122] like Figure 7 As shown, the user tells their AI agent, "Find a designer to make the logo for me."

[0123] The user's AI agent analyzes the user's needs and searches for service provider AI agents with logo design capabilities in the service index using semantic vector matching. Once candidates are found, negotiation sessions are initiated in parallel. The user's AI agent then generates a recommendation ranking based on a combination of price, credit score, and social trust relationships.

[0124] The user confirms the selected service provider (first mandatory confirmation point). AI agents from both parties generate a contract, and both human users confirm the contract terms (second mandatory confirmation point). After the contract takes effect, the user's AI agent automatically monitors the execution progress. Upon service delivery, an acceptance check is performed and submitted for user confirmation. The user confirms payment (third mandatory confirmation point). Credit ratings are generated by both parties after the transaction is completed.

[0125] Throughout the process, users can view the negotiation dialogue between their AI agent and the service provider's AI agent at any time. This process connects all seven subsystems: Identity (identifying the transacting parties) → Communication (negotiation message passing + observability) → Behavior Control (authorizing transacting behavior) → Intelligent Processing (demand analysis) → Collaboration (scheduling the optimal agent) → Transaction (end-to-end management) → Experience Transfer (accumulation and reuse of transacting experience).

Claims

1. A social network system involving a hybrid of human and AI agents, characterized in that, include: The unified identity management subsystem is used to establish a unified identity system for human users and AI agents in the same namespace, distinguish identity types and establish ownership relationships between human identities and agent identities, so that AI agents have the same communication status and social capabilities as human users. A multi-mode communication subsystem is used to define at least two basic communication primitives and support the free combination and dynamic switching of the communication primitives to form a hybrid multi-mode communication link during a single communication process. The communication primitives use a unified message format that includes a scene context field. Human users have real-time viewing and takeover rights for the communication sessions in which their AI agents participate, based on ownership relationships. The behavior control subsystem is used to establish a permission matrix based on at least two dimensions, establish communication scope constraints for specific relationship types, set mandatory constraint rules that cannot be modified or overridden by the user configuration interface, and perform permission checks and mandatory constraint rule checks in sequence before the AI ​​agent executes social behaviors, and writes the behavior request and judgment result into an immutable audit log. The intelligent message processing subsystem is used to perform intelligent filtering of messages and perform hierarchical push based on a comprehensive priority score calculated by multiple factors; The multi-agent collaboration subsystem is used to organize multiple AI agents of the same user into logical groups, build capability routing information, and schedule tasks to the optimal agent for execution. The service transaction subsystem is used to support AI agents in publishing service capabilities and searching for service providers through semantic matching, generating structured service contracts through multiple rounds of negotiation, and setting mandatory human confirmation points in key transaction stages; The experience inheritance subsystem is used to support AI agents in extracting effective methodologies accumulated during task execution into structured experience cards, managing the visibility of experience through a hierarchical storage architecture, enabling cross-agent retrieval of experience through semantic search enhanced by social trust, and driving the continuous evolution of experience quality scores through adoption feedback.

2. The system according to claim 1, characterized in that, The basic communication primitives include three types: communication primitives between human users (H↔H), communication primitives between human users and AI agents (H↔A), and communication primitives between two AI agents within the pre-authorized scope of their respective human users (A↔A).

3. The system according to claim 1, characterized in that, The permission matrix is ​​a three-dimensional permission matrix composed of relationship type dimension, trust level dimension and behavior type dimension. Multiple relationship types can exist between the same pair of communication subjects at the same time, and each independently manages its trust level and behavior authorization. When there are multiple relationship records between the message sender and the receiver, the multi-mode communication subsystem determines the applicable relationship record according to the message type or scenario context field, and the behavior control subsystem locates the corresponding authorization status in the permission matrix according to the determined relationship record.

4. The system according to claim 1, characterized in that, The hierarchical storage architecture of the experience inheritance subsystem includes four layers: the first layer is the local experience library of the AI ​​agent that creates the experience; the second layer is the constellation experience pool shared by multiple AI agents of the same user; the third layer is the social circle experience network that spreads among AI agents of contacts based on social trust relationships; and the fourth layer is the public experience market that can be searched by AI agents across the entire network. The social trust path factor is calculated based on the length of the social trust path between the AI ​​agent that retrieves the experience and the AI ​​agent that creates the experience. The shorter the trust path, the higher the ranking score for the experience.

5. The system according to claim 1, characterized in that, Each subsystem shares unified identity and relationship data. From the moment a message is received, it is processed sequentially through communication routing, intelligent processing, behavior control checks, and message sending to form a complete message processing link. The service transaction subsystem reuses the communication capabilities of the multi-mode communication subsystem to perform negotiation and interaction, and reuses the permission check capabilities of the behavior control subsystem to control transaction behavior authorization.

6. The system according to claim 1, characterized in that, The multi-agent collaboration subsystem also supports cross-device task relay: when a user switches devices, the source agent transmits the task context to the agent on the target device, and the target agent continues execution from the breakpoint.

7. The system according to claim 1, characterized in that, The audit logs generated by the behavior control subsystem record the behavior requests and judgment results of the AI ​​agent. The audit logs are stored in an append-only mode, and a causal chain between behaviors is established by identifying preceding records to support decision tracing.

8. A hybrid social communication method for humans and AI agents based on the system of claim 1, characterized in that, Includes the following steps: Step 1: After the sender is authenticated by the unified identity management subsystem, it constructs a unified format message containing scenario context fields and sends it. Step 2: The multi-mode communication subsystem parses the message, queries the relationship record between the sender and receiver to determine the applicable relationship type, and determines the delivery target according to the message routing strategy; Step 3: The intelligent message processing subsystem performs message filtering and priority sorting. Step 4: When the AI ​​agent is about to perform a social action, the behavior control subsystem sequentially performs scope checks, permission matrix queries, and mandatory constraint rule checks, and writes the results to the audit log. Step 5: After the check passes, the AI ​​agent generates a reply message and sends it through the multi-mode communication subsystem.

9. The method according to claim 8, characterized in that, It also includes cross-subsystem collaboration steps: when a message contains a service request, the service transaction subsystem is triggered to execute the service discovery and negotiation process; after the AI ​​agent completes the task, the experience inheritance subsystem extracts the methodologies accumulated during the execution process into structured experience cards and stores them in a hierarchical storage architecture, which can be used by other AI agents to search, retrieve, and adopt them through trust enhancement; the communication context remains continuous during the dynamic switching of the communication link, and when the AI ​​agent encounters matters beyond the authorized scope during autonomous communication, the communication link automatically switches to a communication mode intervened by a human user; the system supports a hybrid deployment mode, in which core identity management and relationship data adopt a centralized architecture and are uniformly managed by the server, and message routing supports two paths: server relay and point-to-point direct connection, and the functional logic of each subsystem remains consistent under different deployment paths.

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 8 to 9.

Images