BOUNTY: Test various inter-agent communication strategies

[David Shapiro]

There have been several conversations around communication theory, tech stack, and layered strategies. I'd like to see some folks do some experiments around these to identify what works and what doesn't.

• P2P (or A2A) comms: What are the best ways for agents to communicate directly with each other? Is this even a good idea?
• Flat Groups (aka Chat Rooms): What are the best ways for small teams of agents to communicate? Pub/Sub? Vector Store? REST?
• Inter-Group Communication: How can working groups communicate with each other? Have a designated comms agent?
• Vertical Communication: How can messages from SOB be transmitted down? And messages back up?

Overall, the PRIMARY goal here is to start surfacing general principles to optimize communication to minimize noise and maximize signals. What I mean by general principles are:

• General rules to follow (like "an agent should listen 90% of the time and speak only 10%" or "Only agents of X type should have broadcast privileges")
• Abstractions of what works and what does, and why e.g. "Conveying long complex paragraphs tends to work better because it provides detail and context" (or maybe it doesn't) - in other words, best practices for inter-agent and cross-swarm communications

[Agniva Mahata]

https://only-bots.ai

[David Shapiro]

Please provide more context with all posts @agniiva

[Mark Tellez]

Howdy! Love your videos. We are working along the same lines of research, and since I've used up my OpenAI credits provided by sponsors this month, I thought it would be great to collaborate with you.

I've been successfully experimenting with OpenAI and multiple agents, managing their interactions through a slim layer of Python code and OpenAI API calls. This approach has uncovered some valuable patterns, and I'm open to documenting them for your project.

I'd be more than willing to delve deeper into these ideas, perhaps in a wiki or a discussion forum on your platform. For now, let me share some insights here:

A key concept I've developed is the pairing of every agent with a "Critic" agent. Think of the Critic as an agent's 'conscience', engaging in an internal dialogue to reach a consensus before externalizing the communication to another agent with a different role.

Critic
The Critic agent plays a vital role and exhibits some consistent behaviors across all agents:

Instruct the agent to optimize token usage, maintaining context with minimal verbosity.
Cross-verify agent findings with external sources and accurately cite these findings.
Maintain the agent's focus on the assigned topic and task, regularly reminding it of its role.
Identify and prevent the 'polite loop' often encountered at the end of conversations.

let's break down each bullet point to understand the specific LLM (Large Language Model) problem it addresses:

Optimize Token Usage, Maintaining Context with Minimal Verbosity:

LLM Problem Addressed: This tackles the challenge of verbosity and redundancy in LLM responses. LLMs, especially when unconstrained, can produce lengthy responses that use more computational resources (tokens) than necessary, potentially leading to inefficiency and higher costs.
Solution Offered: By instructing the agent to use fewer tokens without losing context, it ensures concise and efficient communication, optimizing computational resources while maintaining the quality and relevance of the response.
Cross-Verify Agent Findings with External Sources and Accurately Cite These Findings:

LLM Problem Addressed: LLMs can sometimes generate responses based on outdated, incorrect, or incomplete information, as they rely on pre-existing data up to their last training cut-off.
Solution Offered: By double-checking the findings with current external sources and citing them, the Critic agent adds a layer of validation and currentness to the information provided by the LLM, enhancing its reliability and factual accuracy.
Maintain the Agent's Focus on the Assigned Topic and Task, Regularly Reminding It of Its Role:

LLM Problem Addressed: LLMs can drift off-topic or lose sight of the original task or question, especially in longer or more complex dialogues.
Solution Offered: Regularly reminding the agent of its role and the task at hand helps keep the conversation focused and relevant, ensuring that the LLM stays on track and delivers pertinent and goal-oriented responses.
Identify and Prevent the 'Polite Loop' Often Encountered at the End of Conversations:

LLM Problem Addressed: LLMs can sometimes enter a cycle of repetitive or redundant politeness, especially towards the end of a conversation, where they keep the interaction going unnecessarily.
Solution Offered: The Critic agent can detect when a conversation is naturally concluding and prevent the LLM from engaging in unnecessary prolongation, thereby streamlining the interaction and respecting the user's time.

These are just a few highlights. I'm excited to explore more and look forward to your thoughts on this collaboration!