General challenge for site members

[Johnh]

I wrote a tic-tac-toe program for a TI-58 in 1979. It'd beat you every time if it went first, unless you made the correct play to force a draw. And if you went first, it would always draw. But that's not even a micro-version of AI, it'was simply a programmed code to achieve a deterministic outcome.

A couple of years before, I had made a vehicle that would hunt around the floor looking for bright things. If it hit something, it would remember which side it hit, back up a little (unless it reversed into something else), and then steer around the object to carry on its journey, at which point it forgot what it had hit last time. So it had three memory neurons, to learn and remember if it had just hit something on the right, or on the left, or at the back. All discreet transistors, just because....

Back to calculators, and to qualify as a version of AI, we'd need to define what AI is? Is it a system that can learn through experience, and use that knoledge to produce steadily improved outcomes in some task, by whatever criteria is set for it? Statistical functions could be a basis for this, whereby a steadily accumulating set of data could lead to better and better predictions about future data. eg, a set of numbers between 2 and 12, which are actually generated by throwing two dice, but we dont know that. Experience of a string of such values will let it learn the likely probability of each outcome. And if betting with some odds were presented to it, it will learn when to play and when to hold.