Computer games, done properly, could be a good teacher of rationality. If you fail to observe the laws of nature that happen to apply in the game and act accordingly, you die / can't get to the goal. Also, debugging programs has been good for me: plenty of times I could not see an error in my algorithm, but knew there had to be one (modulo system bugs), and so I learned what does and does not constitute valid argument that way. Put these ideas together (games where you have to program things that work in the game universe) and it'd be interesting. Some AI competitions work that way, but we could probably build an iPad app for kids with the right ideas. For example, I'm not talking about teaching basic or some terrible "easy" language. Why not base the programming language of the game off some other Turing-complete system, like, say, cellular automata. The automata rules could change from time to time to keep them guessing. The game could be a sequence of problems: fill in these squares correctly in order to get the desired result and move to the next level. Actually, maybe Markov algorithms would be better. Or a version of the untyped lambda calculus (http://worrydream.com/AlligatorEggs/). Having a good baseline programming "world" is an important part of getting them to think about the right kinds of abstractions.