Approaches to artificial intelligence

The good old fashioned approach to artificial intelligence (AI) is based ontology engineering and knowledge base construction. The earliest ontology languages like KIF were dialects dialects of Lisp because of its association with the artificial intelligence community. Ontology languages used today are now more often based upon XML like OWL. Given some domain, all the semantic relations between objects in that domain can be specified to create a semantic network.

The main alternative trend in artificial intelligence right now is to use neural networks. Neural networks have gained a lot more traction under the name deep learning. It is well known that neural networks have surpassed humans at go playing. Combinatorial game theory is essentially a solved problem as humans are generally going to have a hard time against a well trained neural network.

I have great optimism in the future of semantic networks, the modeling of semantic relationships between phenomena can still be greatly improved. In particular, semantic networks are especially useful for modeling symbolic thought. I think even as neural networks are improved semantic networks and ontology engineering are still worth pursuing and can still be improved on their own. Perhaps superhuman artificial general intelligence will be created with some appropriate combination of semantic networks, neural networks, and other methods.

See also: Symbolic vs Connectionist A.I.