I could not possibly care less about what people do with LLMs on their own 
computers, so I will stick strictly to the issue of agentic LLMs.

In general, I actually do think that agentic LLMs have the potential to create 
interesting gameplay. Agentic LLMs are fundamentally different from human 
agents, which means that, to the extent that they can play the game at all, 
they are likely to play it in new and interesting ways. I am not, in principle, 
against the existence of "bots" or "NPCs" in Agora.

That said, there are a lot of potential issues that need to be sorted out. 
Agentic LLMs ultimately do answer to some human operator, so I'm firmly against 
allowing LLM agents to be treated as persons, on pain of extremely un-fun scams 
powered by swarms of subservient LLM player-bots. What Ronic is currently doing 
- registering as a human, and taking responsibility as a human, but allowing an 
LLM to read and write eir messages - is fine by me. I certainly wouldn't want 
it to become the norm, but contra Aris, I don't think that an outright ban is 
necessary for that. This is a game, we are playing it for fun. I am not 
interested in having someone else play the game for me, that would defeat the 
entire purpose. I suspect that the main reason we've seen players try to use 
agentic LLMs is simple curiosity about they would interact with the game, which 
is unlikely to see human players replaced.

As far as additional restrictions go, I'd prefer to keep agentic LLMs out of 
any sort of position of responsibility (i.e., offices) and possibly to require 
all messages sent by agentic LLMs to identify themselves as such.

- Galle, Speaker and Champion

Reply via email to