I agree with this. Even today when we just recently saw what applied statistics can do (gpt, AlphaFold, AlphaGo,...), I think the problem fits the application nicely and we can go beyond what would be possible by "manual" research.<p>Comparably, conventionally developing an AI for blackjack would almost be trivial. (sorry for the blackjack fans, and yes, the AI might still help you with body language)<p>But that is still more statistics than intelligence. In fact a lot of that is required to reduce a strategy an AI has developed into general rules that isn't dependent on training data or derived information anymore.