The Event Buffer of OpenNARS

[Pei Wang]

New conceptual design at https://cis.temple.edu/tagit/publications/TAGIT-TR-22.pdf

Virus-free.www.avast.com

[Artem Andreenko]

Dr. Wang,

Thank you for sharing this conceptual design and the link to the paper! I greatly appreciate your work and the ongoing efforts to enhance OpenNARS.

I have a question regarding a part of the paper: you mentioned under Communication Channels that "A channel of this type connects the system to an intelligent system that can cooperate with NARS."

Would it be a good continuation of this work to explore connecting multiple OpenNARS instances over such communication channels? This seems like a promising approach to achieve better reaction times and potentially lower energy consumption through distributed cooperation.

Artem

--
You received this message because you are subscribed to the Google Groups "open-nars" group.
To unsubscribe from this group and stop receiving emails from it, send an email to open-nars+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/open-nars/CAK0i%3DhjSLeSdOaV5r4PMGq92VTUEysH07Y5S594544sp0q3aKg%40mail.gmail.com.

[Clément Michaud]

Great article Pei Wang and Bowen Xu. I skimmed through it. Thanks for sharing.

Can you elaborate on what this new design tries to solve in one sentence and also what are the results of using such design on a simple task compared to the previous design? Is there any improvement or regression?

Thanks,

Clément

[Pei Wang]

Hi Artem,

Exactly. Multi-NARS cooperation is a major motivation for this new architecture, and some preliminary explorations are under consideration. Given the resource restriction, it is not a high-priority task of the Temple team, which is busy making the new architecture work.

Given the adaptive nature of NARS, this cooperation will be very different from the conventional multi-agent system and will show interesting properties.

Regards,

Pei

[Pei Wang]

Hi  Clément,

In one sentence, the idea is to improve inference performance on sensorimotor experience, in parallel with language-based experience. The results haven't come yet. ;-)

Regards,

Pei

[Artem Andreenko]

It looks like the multi-stream/buffer approach is becoming popular to increase agent capabilities. 

For instance, real time agent Moshi from Kyutai is based on a multi-modal transformer modeled as a set of audio token streams, including the user's input, agent's output, and the agent's internal monologue.

https://x.com/kyutai_labs/status/1836427396959932492?s=46&t=xprbz829MoBWne2UXtINdA

To view this discussion visit https://groups.google.com/d/msgid/open-nars/93e7626e-8f6e-465e-ba44-7279cffb2757n%40googlegroups.com.