[andagainagain]

This makes sense from all sorts of different perspectives. Heuristics have, over and over again, proved to be faster than manual logic for basically everything. And, if the heuristic is well trained, it can often be equal to or better in statistical results.

The limit, I think, is bounded by two things that come off the top of my head. First is the problems with heuristic fail states. The second is in the statistical portion in those results.

Heuristics fail sometimes spectacularly. This is how you get a question like "what is the the most effective way to play Tetris" with a result like "drop all the pieces on top of each other and then pause the game forever".

And when heuristics fail in a more subtle way, they can quietly do all sorts of things like remove trust, discourage workers, reduce quality, etc etc.

Statistics, meanwhile, can hide a lot of what is actually going on. And sometimes this can cause long term problems, that data (with a good scientific process behind it) would reveal.