The First AI Winter:
It was caused by a couple of major factors. The first and foremost was the lack of computer power. The examples that were put forward were "toy" solutions that really did not appeal to the investors at all. This set off a period of infighting about Natural Language Processing (NLP) in the AI community which scared off the investors for a long time.
The Second AI Winter:
It was caused by the fact that the AI portion of an application had to run on a detached expensive LISP machine that really wasn't that much more powerful that newer computers. To compound the situation, the AI systems were brittle and expensive to create and maintain. The isolation and expense was too much for users and investor alike kicking off an even longer AI winter.
What's Different Now?
- AI now has a great NLP base to leverage and we will soon be seeing NLP services for front ending legacy transactions and systems.
- AI is data focused now; not as rule and algorithm focused. This means as the data changes, the AI can change right long with the data. This reduces costs of maintenance.
- Computing capability is strong enough and getting stronger to take on even more ambitious uses. We see much parallelism coming on soon in Quantum Computing that is necessary for more ambitious uses of AI.
- AI is working hard to assist people instead of displacing them except for dangerous and low level boring work.
- AI can consume large amounts of data from vastly different location and give advice to people
There is no sure thing with this generation of AI, but if we all make AI easy and helpful as it takes on more challenging tasks, I think we can avoid another AI winter. The economic and computing power issues seem to be under control. It is now our job to make AI interface well with people to make them feel secure. All that stuff about AI taking the world over and turning on people sounds good for a fiction book. Do no harm otherwise ethics and law will likely have to change.