As far as low-tech tic-tac-toe "AIs" go, this XKCD map is my favorite:

https://xkcd.com/832/

I too built one of these in my youth (when I was 8 or so); I found it a pleasant inspiration and inspired me trying to do it for checkers (which, needless to say, had too many states, and failed). Are there other such mechanical examples of machine learning with which to inspire young people?

I wonder if it's possible for two instances of Menace to get into a local optimum that is different from the global optimum strategy.