Today was the first day of the Artificial Life conference in Kyoto. I took the obligatory photo out of my window, then joined my colleague at breakfast.
Then off to find the conference venue, and join the registration queue (at essentially midnight UK time, but I felt fine, having slept reasonably well). Queuing also involved saying hi to many friends and colleagues I hadn't seen since last year. This first day was dedicated to workshops. I went to an excellent Neuroevolution tutorial in the morning. The lunch queue was long, but moved quickly, as pre-packed food boxes were being briskly handed out.
In the afternoon I went to a workshop on Generative ALife with a wide range of talks. I was quite fascinated by the various talks using LLMs to implement the very simple agents in Minsky's Society of Mind. Now, I'm an LLM mega-skeptic: these stochastic parrots are most definitely not "intelligent". However, the whole point of SoM is that intelligence emerges from lots of unintelligent agents communicating. These researchers are using LLMs as SoM's internal agents, communicating via natural language. There are agents controlling the "mouth", which talks to/at people through a robot avatar; (these conversations still have typical "hallucination" issues), other agents are attached to the "ears", "eyes", etc, and more are just internal agents, making up the whole internal society. In addition to the transcript of the robot/people conversations, the researchers can also get transcripts of the internal conversations between the internal agents, giving insight into how the agents are interacting to give the resulting output. I found this an interesting novel use of these devices.
Then a bunch of us went out for dinner.
A great first day!
No comments:
Post a Comment