Why are neural networks considered to be artificial intelligence?

Question

Why are we now considering neural networks to be artificial intelligence?

Answer 1

Why are we now considering neural networks to be artificial intelligence?

"We" aren't. It is generally due to reporting by media sources that simplify science and technology news.

The definition of AI is somewhat fluid, and also contentious at times, but in research and scientific circles it has not changed to the degree that AI=NN.

What has happened is that research into neural networks has produced some real advances in the last decade. These advances have taken research-only issues such as very basic computer vision, and made them good enough to turn into technology products that can be used in the real world on commodity hardware.

These are game-changing technology advances, and they use neural networks internally. Research and development using neural networks is still turning out new and improved ideas, so has become a very popular area to learn.

A lot of research using neural networks is also research into AI. Aspects such as computer vision, natural language processing, control of autonomous agents are generally considered parts of AI. This has been simplified in reporting, and used by hyped-up marketing campaigns to label pretty much any product with a NN in it as "Artificial Intelligence". When often it is more correctly statistics or Data Science. Data Science is another term which has been somewhat abused by media and technology companies - the main difference between use of AI and Data Science is that Data Science was a new term, so did not clash with pre-existing uses of it.

The rest of AI as a subject and area of study has not gone away. Some of it may well use neural networks as part of a toolkit to build or study things. But not all of it, and even with the use of NNs, the AI part is not necessarily the neural network.

Answer 2

AI is not only about neural networks.

Formal proof assistants (like Coq, or Frama-C) are in some circles considered as AI. Projects like DECODER have an AI flavor.

Symbolic AI systems (like RefPerSys) and more generally expert systems are advocated in AI books like Artificial Beings, the conscience of a conscious machine. That book (by Pitrat) don't mention much neural networks.

Neural networks are good for some problems (e.g. computer vision), but less adequate for other problems.

Autonomous robots (like Mars Pathfinder) don't use only neural networks.

Natural language processing is a subpart of AI, and is not only using a neural network approach.