Today, I read an article from the MIT Technology Review, called "AIs PR Problem". The article is about how Artifical Intelligence (AI) is being portrayed by the movies, and what is really happening in the world - and i can only back the findings.
Currently, the state-of-the-art methods, being worked on at universities, are nothing more, than ways to search through trees, very fast (its simplified, I know, but stay with me). It is not far from how human brains work, but it still doesn't do the trick. Take a spoon for an excample. When you look at it, you instinctively know, that you usually use it for eating, but you can also use it for other things. You also know, that you use the spoon with your hands (or feet, if you're taught it). This is deeply encoded into your brain, and you do it, without thinking about it.
Walking, is another example. When you walk, you're actually constantly in a state of falling, and you move one foot in front of the other, to prevent you from falling over. So how is it, that babies can learn to walk, when we cannot teach a computer to do it properly? It all boils down, to the definition of a computer. A computer is really good at moving bits. A bit is either a one or a zero. Think of it as a switch - the light is either on or off. The way we try to teach a robot to walk, is that we set up a series of rules, and then program it to the computer. These rules mimic the physics involved in walking. However, as we can only turn stuff on or off, it will be impossible for a computer to mimic human walking, with the methods we currently use.
These methods, are really good at doing doing things that are inherently logic. Chess, go, and most of the games in existence, are based on logic. And this is why computers are so good at them. And this is also why they are good at beating humans at them. But to evolve from a "on/off" paradigm, to something "on/partially on/partially off/off" is a huge step. Quantum computing, is a leap in that direction. Researchers are working on methods that evolve the bits, to assume more values than "on/off" - but what programming methods can we use?
At the department i'm visiting currently, they are really good at doing motion detection, image analysis and so forth. And they're constantly striving to make the methods better. The problem still remains however, we're still developing ways to search through a logic tree. A person is able to watch a game of sports, and (assuming you have an idea of the sport), figure out what it is about. A computer wouldn't be able to do so. And this is where we need to head, for true AI to be developed.
I think AI should be defined in terms of creativity. To evolve your thoughts, you should be able to think "creatively". And this is where i think we should redefine the "Turing Test". If a computer is able to "evolve" on thoughts, it can be defined as intelligent. We need to teach computers to think, rather than do tedious tasks for us. That's when we will evolve!
We're not there yet, and i don't think we're close! But the important part, is that we're going! If you have to eat an elephant, you have to do it one small bite at a time.
Anyway, its saturday night, i've had a few glasses of wine - but its an interesting discussion, nevertheless! Actually, i think its a whole new field of scienc,e "intelligence science".
Do contact me if you want to discuss the matter - my mail is firstname.lastname@example.org. I'll reply as fast as possible.
All the best,