Start with one programming language assembly. Using it create a better one, BASIC using it create a better one SCHEME.
The problem is now, you cannot make a better language. One would have hoped to build onto more and more better languages that were increasingly sophisticated and expressive. Instead of this we found that everything we create in scheme is worse than scheme.
Why programming languages don't AI
Not sure what exactly you mean by AI, but if you do mean AGI, it's a rather complicated problem that won't just be magically solved by inventing a better language (some things could of course be made easier in some metalanguage, implemented in whatever, be it a Lisp or some ENTERPRISE language). It requires a lot of thought about actual cognitive processes, decision theory, probability and various other fields if you want to do it efficiently (on currently cheap hardware), it also requires a lot of hard work (implementing and testing ideas). There is also likely a shortcut which will yield results but requires an efficient hardware implementation (some of which are underway) by emulating large-scale neural networks found in general intelligences such as ourselves, but such shortcut may come with many costs of its own. I suppose it comes down to the usual "if only we had unlimited computational capacity and ..." or "if only we could think really hard and had a lot of time to think about the issues involved" possibilities.