Re: matching AI agents to protocols

Thank you Dave
for the quick response
Let me articulate the thinking a bit more in a paper, then hopefully you
can review/give feedback on the table in context

P




On Mon, Jun 30, 2025 at 6:33 PM Dave Raggett <dsr@w3.org> wrote:

> It would be helpful if you provided further information as it isn’t very
> clear right now.
>
> For instance, an agent that uses facts and rules wouldn’t use MCP which is
> a protocol for agents implemented with generative AI.  My current work on
> extending chunks & rules to swarms of agents uses chunks as the medium of
> communication, hiding the underlying protocols.  Chunks & rules isn’t logic
> based.
>
> I suspect you are focusing on generative AI based agents, where the agent
> exploits pre-training and reinforcement learning with human feedback to
> determine its behaviour.
>
> There is a lot more work needed to advance beyond generative AI. I have
> sketched out some ideas in my slides on sentient AI, e.g. continual
> learning based upon continual prediction, episodic memory and the role of
> type 2 cognition during learning.  This will have a big impact on agents.
>
> On 30 Jun 2025, at 05:32, Paola Di Maio <paola.dimaio@gmail.com> wrote:
>
> Dave, and everyone
>
> several years ago, when the AI KR CG and this CG were started
> Dave hinted in a post that AI would be largely agent based.
>
> I recalled  that prediction as I started to work on AI agents, from a KR
> point of view (categorization of AI agents, ontology driver agent modelling
> etc) and to develop categories of protocols
>
> In trying to create logical to ontological schemas to help us capture what
> is going on, I would welcome CogAI input/evaluation
> on the following table:
>
>
> <image.png>
>
>
> Dave Raggett <dsr@w3.org>
>
>
>
>

Received on Monday, 30 June 2025 13:24:47 UTC