Weirdly enough, that's a thought that I'm having in financial trading as far as using AI for idea generation.
At first glance, charting the future possible moves of a chess game is just a huge branching tree, but humans (and engines that don't have the power to fully brute force the game) use filters to trim the tree. Some lines are dead ends, even though they may play out for a while (sacrifice both rooks and the game is over, no need to follow those branches). There is also a sort of heat map and gravity to some of the lines, in that there are likely directions that players will travel in (paths where you don't give away too many pieces, where the king isn't exposed, etc).
Machines can help highlight specific areas where there are branching points that lead in many viable directions (these are the critical decision-making points in a game of chess), that are deceivingly hidden behind lines that look dead for a while.
It would output a sort of heat map, and the search could even be tweaked for certain variables, such as for number crunching complexity (if the opponent is a bit weak there) or pathways into brutal end-game scenarios (if the opponent is weak there).
This is a microcosm for the real world as well. Lines through time have reflexivity and can reinforce each other. A geopolitical situation can reinforce an economic situation which then feeds back into the political situation. Take something like inflation which tends to do that. But when humans normally look at the world, they see in a sort of normal distribution that is oversimplified. It's commonly understood that humans downplay the left and right tail risks (as explained by Taleb), but it's more nuanced than that. It's more like the chess game, in that there are these hot spots of complexity and interesting situations throughout the forward probability distribution.
Some of these hotspots are deceivingly hidden, because only one multiple possible situations unfold do they feed back into each other and create something emergent.
Back to an arena like trading, participants tend to track each possibility line independently of one another, which makes sense because humans are siloed and specialized to some degree. Technology like machine learning has the ability to synthesize this data and spit back out hot spots, just like in the chess example.
The short-sighted conclusion that most will have is to say "Great! Let it give me a list of trades, and then we can back-test it." when I'm pointing out is that there is a lot of value when it comes to idea generation and efficiently mentally traversing the future probability space. Spending your time focusing on interesting places. Maybe a traitor would look at an implied outcome distribution and realize "Hey, I think that this little part of the curve is underpriced. Maybe I should hedge this specific outcome, because I have exposure to the inputs that feed into this underpriced emergent possibility."
Of course, the trading example is also an abstraction from the raw real world, but it's a bit more close to reality than the chess example. Really, I think that this approach to using machine learning as a tool could be applied to many areas. Even more creative areas could potentially benefit from it.