By some definitions of AI, a light switch IS AI. That is my point. AI is so broadly defined, and applied, that it is a useless term.

Deep Blue, Alpha, matters not. These systems play chess, because they were set up to play chess by humans. They can not of their own volition suddenly decide to not play chess, but to play something else they were not designed for. The neural nets are trained on a specific task. They make decisions based on that training, and that task, and the task inputs. It is still basically algorithmic, where the algorithms have built-in modifiable parameters that can be real-time adjusted within their limits. It is a long way from mimicking neurons. It mimics what some human theorist THOUGHT neurons performed like. But it is still a programed algorithm that comes from a human mind, just that it is on a different technological platform than a binary computing device. It is an example of a machine being able to fine-tune a system output in real time based on feedback inputs.

The intelligence has not evolved, the human capacity to create algorithms and devices to apply those algorithms in more novel and complex ways has evolved. It is human thinking that has evolved, not the ‘artificial intelligence’ per say.

You are very, very wrong about the ‘no one knows how these neural networks work’. This statement is a perfect example of the hype behind AI. They are not hard to understand, and their functionality is not hard to grasp, as long as one can get around the bug-a-boo that they are not digital or Boolean devices. They do not follow truth tables or traditional truth table logic. But it is perfectly understood how they make decisions. We are, however, in the very rudimentary state when it comes to graphically or diagrammatically or schematically or even mathematically depicting how they work - the iconography, symbology, terminology has not yet developed comprehensively.

The ‘nets’ have absolutely no idea what is ‘winning’ or ‘losing’. or ‘reward’ or ‘punishment’. Those are human concepts that have been anthropomorphically applied to inanimate devices. What it is in reality is some form of feedback circuit (human intervention or automated) that drives the system closer or further away from the desired state -‘desired’ as determined by the human operator. We did this many decades ago, even before digital computers, using analog potentiometers and electrical meters. Musicians do this all the time when they ‘fine tune’ their instruments. We have just gotten better and better at automating it and applying it to more complex situations. Some chess moves result in a better melody, others result in a more noisy sound. The instrument - the chess playing device - is simply fine tuned by repeated performances to produce the best sound, as we humans have determined ‘best sound’ to be.

Living neurons, on the other hand, are still not completely understood, nor do we understand exactly how neurons make decisions. The best guess is that they use quantum effects, but that is only based on the fact that we are discovering more and more that life itself is based on quantum effects - photosynthesis for example, or the methods birds use for navigation across continents. But living neurons have nothing in common with these ‘neural nets’ except that a picture of one was used as some conceptual pattern or intellectual starting point that triggered some ideas in the mind of a very creative person. Like seeing a bird fly triggered the idea that maybe humans can fly. But neural networks have as much in common with living neurons as airplanes have in common with how birds fly.

But in general, what we call AI is still nothing more than humans setting up machines to automate the application of the algorithms our human minds think of in the first place. Just a more complex, complicated, light switch - some device that allows us to automate the process of connecting the light to a power source, without having to connect the wires every time we want to use it.