> With a threshold when the computer starts modeling itself
That's a fascinating comment
It would be great to have a list of systems where this already happened. With most neural nets, since they don't ever perceive the effects of their outputs on the world (only the effects of the last version perhaps, say GPT4 seeing/learning about GPT3 output and its effects on the internet), but for some reinforcement learning agents some level of self-awareness might have already occurred.
That's a fascinating comment
It would be great to have a list of systems where this already happened. With most neural nets, since they don't ever perceive the effects of their outputs on the world (only the effects of the last version perhaps, say GPT4 seeing/learning about GPT3 output and its effects on the internet), but for some reinforcement learning agents some level of self-awareness might have already occurred.