Artificial Intelligence - what is that?

Well, for starter, it is the holly grain of all computer science professors, going back to the fifties, when everybody thought that something like the A* will solve every problem in the world.
With time, people understood that it is not the case and that artificial intelligence is more than optimal planning, or finding efficiently the best path in the graph - it is more about noise and data, a lot of data, heuristic decisions, which can be dynamically overridden due to some online sensory signals, etc. Machine learning algorithms stood up from all the others and suggested that you can learn whatever you need given a large set of data. Well, the curse of dimensionality struck then... evolutionary algorithms were good, but there was no time to wait for the answer. Neural networks were also a great solution, but who the hell can build them correctly for a generic problem?

So what is artificial intelligence, and will the human race ever be able to say proudly - we have built it!?

Thinking that artificial intelligence is mimicking the human activity is really naive, on my opinion. We do not want to create an automatic human being... First of all, it takes 16-20 years for the natural process of nature to create a grown up man or women, and that is a process which matured for a long time. What are the chances that we will be able to create such a thing and how much time will it take for us? Moreover, why do we need a machine which has the abilities of a human being, we have a lot of that kind of resource on our planet already. Well, of course, robots will be able to drive, to rescue, to take any risk needed to satisfy our human needs. But I think that this is a wrong path to go... If we build machines with human abilities, why should we treat those machines differently? I hear: "Robots rights!". Why wouldn't they have the same rights as we humans do? And here you go - we are in the scenario of "Terminator", or "The Matrix", where machines win the war and we have no future. Not good.

So what is AI if not building a humanoid robot? I personally think that machines need to do whatever they are good at. Are they good at chatting or talking? May be driving a car? I do not think so... Machines are good at processing automatically large amounts of data, doing it fast and reliably, without a need in cup of tea or a nap, and presenting some interesting insights when they finish. That is really practical AI for me. If we build a machine which can process Terra byte of data in a single second and discover interesting clues about it, that would be a real achievement. Humans never will be able to do that, they even can't read one page of a simple text in a second - this kind of stuff also can be put under the hood of AI. If we define "intelligence" as the ability to process Tera byte of data in one second and learn from it something, suddenly humans become not as intelligent as even the most simple machine learning algorithms.

However there is one aspect of humans which I think is interesting to AI and computer science in general - their brains. I guess that you will not be surprised, if I tell you that this part of our body remains a mystery even now days. A lot has been learned and more to come. Each time that the scientists discovery something related to our brains, they practically discover additional puzzles which need to be solved in order to really understand how those powerful machines work. So if we could create an artificial brain, a powerful and fault tolerant, which does not need to sleep or eat, or at least use that wisdom in our machine learning algorithms, that would be nice. Restricted Boltzmann Machines are very simplistic versions of our brain, they are actually very restricted neural networks, which can be learned using a relatively simple algorithm - Contrastive Divergence. So that is an interesting direction.