Sutton actually argues that we do not train on data, we train on experiences. We try things and see what works when/where and formulate views based on that. But

Sutton actually argues that we do not train on data, we train on experiences. We try things and see what works when/where and formulate views based on that. But I agree with your later point about training such a way is hugely limiting, a limit not faced by humans