If you were trying to program an AI system, what language would you use?

 

The canonical answer seems to be Lisp/Scheme, although any functional programming weenie would probably tout their language of choice simply because they believe functional voodoo is perfect for everything.

 

Long ago there was Prolog, the "natural language of artificial intelligence," which has since fallen into disuse.

 

Nowadays, I'd say Erlang. But not necessarily as a functional weenie.

 

The brain uses a shared nothing message passing architecture to achieve massive concurrency. Neurons are discrete and don't have anything like shared memory, such as the type multiple threads (the canonical concurrency mechanism) inside your computer do.

 

Neurons are much like little processes, which store state, mutate it depending on the messages they receive (or don't receive) in realtime, and use the combination of the incoming (or expected) messages and internal state to generate output.

 

Numenta, perhaps the only people on this planet working on a high level computer model of the mammalian neocortex (BlueBrain is working on an extremely low level model in comparison) have devised a simulation for the neocortical hierarchy, which operates as a shared nothing process architecture, except the level of granularity is raised to that of a neocortical column which is the fundamental building block of the mammalian neocortex. Neocortical columns, much like neurons, are a shared nothing architecture which uses message passing, and likewise so is the NuPIC software (although that's written in C++)

 

So, all that said, what's the best language for Artificial Intelligence? Well, ostensibly one that uses a shared nothing message passing architecture.

 

So, Erlang!

Machine Code.