Multi-agent Systems and ActivityPub

Fediversity
1

With respect to multi-agent systems and software-development frameworks (e.g., AutoGen, LangGraph, Swarm), I wonder whether anyone else in the ActivityPub community has looked into the use case of multiple AI agents and human end-users collaborating via Web-based structured forums?

By using Web-based structured forums:

  1. AI developers and their tools could readily examine agents’ and multi-agent systems’ behaviors and performances.

  2. People and AI agents could together engage in dialogue and collaborative activities. For example, to co-create a document.

By using the extensible ActivityPub standard, multi-agent systems could be interoperable with a wider variety of (centralized and decentralized) Web-based structured forum software.

Thank you. Is there any interest here in this use case?

1 Like
2

I would like to see that a programmatic agent (AI or otherwise) be clearly identified as such in the actor definition, so as to:

  • be able to remove such agents from the learning pools
  • be able to block any interaction with such agents
  • any other action that can confer agency to people
2 Likes
3

Web-based structured forums organize, display, and allow convenient navigation between the contents of multiple simultaneous threads of multimodal dialogue between multiple parties, people and AI agents.

Expanding on your points, some structured forum software provide features involving user groups and roles. In these cases, AI agents could be considered to be a special kind of user and placed into one or more user groups specifically intended for them. Privileges could be provided to AI agents based upon their memberships in user groups and, perhaps, upon dynamic task-specific roles.

Areas of forums, e.g., individual discussion threads, could be configured in a variety of ways with respect to man-machine interaction and collaboration. Some areas could be configured to be only for human users, discussion, and collaboration.

By using ActivityPub, multi-agent system development frameworks could enable developer teams to be able to:

  1. explore competing forum software to determine which meet their evolving needs with respect to development and testing

  2. develop multi-agent systems that are interoperable with multiple forum software

4

I recently blogged about how existing and emerging standards and recommendations (e.g., W3C ActivityPub and IETF vCon) can enable next-gen use-case scenarios involving man-machine multi-agent systems: Man-machine Multi-agent Systems Mediated by Structured Forums.

5

This is an intriguing discussion, but I’m not sure it’s on topic for the Fediversity category? Maybe more suited to the ActivityPub category? Might get more eyes on it there from people with relevant interests …

6

Thank you @strypey. Do you propose that a forum administrator move this existing thread to that category or that I create a new thread in that category?

7

That one. Can you do that @how ?

8

I do not think the discussion is clearly enough related to the ActivityPub protocols. To me it’s a research question that needs to be developed before entering the space of discussion about specifications. Hence its position here in Fediversity.

9

To avoid derailing this thread any further, I’ve responded in a new topic.

10

Thanks @strypey. Regardless of how we best categorize this discussion, structured forum software can enable man-machine multi-agent system collaboration and multi-agent systems comprised of AI agents to perform tasks for people with their dialogical processes transparent and available for inspection, analysis, and evaluation in structured discussion forums and threads.

Some might conceptualize AI with respect to personal assistants, 1:1 chats between people and AI agents. Others might conceptualize team assistants, N:1 chats between teams of people and AI agents. Here considered are enabling these 1:1, N:1, 1:M, as well as N:M interactions between teams of people and teams of AI agents with interactions mediated and explicated by structured forum software.

Introduction

Structured forum software store, organize, display, enable navigation to, and can provide search capabilities with respect to the contents of multiple discussion threads of multimodal dialogue between multiple interacting parties, between people and artificial-intelligence agents.

Artificial-intelligence agents in multi-agent systems could copy their exchanged messages to, or be otherwise mediated by, structured forum software.

Standards and Recommendations

By making use of standards and recommendations for transmitting activities between clients and structured forums’ servers (e.g., W3C ActivityStreams and ActivityPub), software development frameworks for creating artificial-intelligence agents and multi-agent systems could enable developed technologies to be interoperable with a wide variety of competing structured forum software.

Structured forum software could export content using standards and recommendations for representing conversations (e.g., IETF vCon). In this way, software developers could more readily create tools to aid and automate the analysis, assessment, and evaluation of man-machine dialogues.

Discussion

Man-machine interactions between artificial-intelligence agents and end-users and the processes and procedures of multi-agent systems could be made transparently available in discussion threads and explicated in secondary discussion threads. Artificial-intelligence agents could hyperlink to secondary discussion threads which explicate those processes and procedures involved in the completion of tasks and subtasks.

Interoperability between multi-agent systems and structured forum software would enable teams of developers and testers to readily examine multi-agent systems’ behavior during development, testing, and deployment. Software testers and their tools would be able to engage in and to simulate dialogical interactions with one or more artificial-intelligence agents in an intuitive manner using structured forum software.

With respect to educational use-case scenarios, artificial-intelligence agents can support teaching assistants in answering students’ questions on class discussion boards.

Artificial-intelligence agents participating in structured forums can follow end-users’ instructions to perform a variety of tasks and subtasks such as performing research, answering questions, and co-creating multimodal content including short-form responses, long-form encyclopedic articles, and stories.

Beyond idly awaiting questions and instructions from end-users, artificial-intelligence agents could proactively examine unfolding discussions in structured forums to provide suggestions with respect to how they could be of assistance.

Areas of structured forums, e.g., individual discussion threads, could bear metadata intended for use by artificial-intelligence agents. Forum areas could be independently configurable with respect to artificial-intelligence agents’ behaviors, features, and settings.

Conclusion

Man-machine multi-agent systems could be mediated by structured forum software, supporting the indicated use cases and delivering the indicated benefits to software developers, testers, and end-users.