I agree that is vague. But there are plenty of RL algorithms that learn with essentially no domain knowledge, e.g., MuZero[1] which doesn't even use a model. Th

I agree that is vague. But there are plenty of RL algorithms that learn with essentially no domain knowledge, e.g., MuZero[1] which doesn't even use a model. That doesn't make it AGI -- it only masters a single game at a time and we currently don't know how to transfer that knowledge to other domains. [1] https://arxiv.org/abs/1911.08265 https://arxiv.org/abs/1911.08265