AGI defined as transformative, at least human-level, but potentially superhuman AI, capable of doing at least any cognitive task humans can do, at least at human level.
Related to: https://manifold.markets/MarioCannistra/will-we-get-agi-before-2027
The criterion to resolve this will be very hard to define. Whatever is write on the description doesn’t explain anything. What is considered “human-level”? Cognitive tests? Chat-GPT does a lot of human level cognition already.
AGI is artificial consciousness, capable of adaptation and self awareness in the same way of humans.
@DavidBolin Chat-GPT is already doing a lot of work that you can hire someone to do. I agree that doesn’t make it AGI though.
@drcat Human level means that it can do every cognitive task most humans can do, at least as well as humans.
A good proxy for that is if it can do every job humans can do. If it can't, it's not AGI.
That ignores jobs that require a body of course, that's why I specify cognitive tasks, but it would be trivial for an AGI to design a robot like Atlas, and have someone build it, since humans can do it, and then with that it can do pretty much every job.
@drcat A lot of work is not all work. It means it still lacks something that humans have. When it can do everything humans can do, that's AGI.
@MarioCannistra So if you have a robot that can mimic every human task and pass any cognitive test, but it has to be programmed to do it by a human, is that AGI?
@drcat Depends what you mean by "programmed".
Humans don't have to be programmed, but you have to teach them to do things they don't know, which I guess you could consider a form of programming in natural language. If the AI is at that level, it's AGI.
GPT-4 fails at this, but not by much. You can teach it things, but in many tasks it's not as good as a human (while in others is superhuman).
@Tomoffer I think it updates mostly on general sentiment based on all news regarding AI advancements, not anything in particular.