This has been on my mind as well -- my oldest child has just become a bookworm and I'm intending to get all my kids iPads this year if they can be more than just for games and movies.

While strong AI would be nice, well crafted content alone could take one very far. Then add in what Khan Academy is doing for tracking learning and you might be 90% there.

Kindles (the eInk ones, not the Fire) might be a good push in the "for reading" direction. You really can't do much else with them, but they're really great for reading, as well as super light, and have amazing battery life.

I got a Kindle Fire, and reading on it is just not the same as with the 'real' Kindles - you always have 'other things' like email and Facebook lurking just a few clicks away. And the battery life is something you have to keep in mind, and of course an LCD screen just isn't as nice for reading.

I thought the primer was driven by an actor in a booth, not an AI. Should I re-read?

Yeah. The primer decided all of the content, the actor was for emoting instead of poor voice synth, iirc.

And you could nearly do it now. Certainly, actors on demands already exist, just not in real time.

Combine automatic text generation (a long-time research topic) with some other existing tech and we could be 90% to the primer.

I don't think text generation is realistic at this point, nor will it be in the near future. It's hard enough for a human to figure out the right thing to say to a child, let alone design an algorithm to do that. When I read Stephenson's book I thought that good speech generation was a much smaller problem to overcome (Indeed, it was dropped for the version of the Primer given to the cast-off girls) than content generation. I remember feeling that, were the Primer built today, the on-demand talent required would be guided improvisational interaction rather than mere acting.

However, I could see a Watson like search engine being a possible alternative. The child could carry around a monitoring device that would seed the search with the events of the child's day. If the child struggles in class, gets beat up by a bully, etc. the engine could then set programming from an existing library of stories, movies, etc. that is immediately relevant to what the child is experiencing. Entertainment for children is often designed to teach lessons. This would just time the lessons to have maximum impact. Teaching children is a bit like training dogs in that timing is massively important. Catch a dog in the act of peeing on the rug and you can have a far greater impact than if you don't notice the mess until after it's dried!

AI cannot presently compete with humans when it comes to content generation, but the above device could be with a child night and day throughout their schooling, which is something few, if any, parents can do.