Artificial Intelligence
Or, the eternal struggle between right and wrong. And stuff
Artificial Intelligence is, in its broadest sense, the science of making computers think, reason, or act like people. It's a pretty big field though, and incoporates all kinds of things. Philosophy, robotics, computer science, cognitive science, problem solving... all sorts of nonsense.
Here's my problem. An overwhelming amount of my university degree - some 75% or so - leans towards a particular school of thought. Time and time again, it seems to teach that the mind is a sort of calculator. It's a massive database of rules and memories and things. By applying rules at certain times, life happens. So when thinking about a problem - from what to have for breakfast in the morning, to which move to make in a game of chess - all we do is manipulate certain rules until we reach some resolution. It's as if theres a certain amount of pre-set things that the brain can do, and those things are selected one by one. Computationally, this is massively complex. Writing these rules for even the most simple problem is tedious, and complicated, and slow. It creates a very specialised system, useful only for one small problem domain. Clearly, this doesnt reflect the human brain. I mean, our mind is one big general system. How many millions of rules would be required to get through a day? A ridiculous amount: and if just one rule was missing then the whole system would break down.
The mind is not a computer. It isn't built like a computer; it doesn't work like a computer. It doesn't make irriating whiny sounds, or go 'bleep'. Consciousness is more than symbol manipulation and logical deduction. And yet, time and time again, we come back to some sort of rule-based model of the mind. And it's fundamentally wrong. In trying to tie the mind down to a set of symbols and functions that can be understood, science is forgetting what the mind actually is. It's important to remember that the brain is soft, and squashy, and not very computer like. It's a sort of mess of concepts and ideas and memories and thoughts. It's a swirling, moving, organic soup of life. Trying to force that soup down into a fixed set of numbers is worthless. It's like trying to pin the soup to a wall - it stops becoming soup, and becomes a sort of stain.
So, AI must learn to be less... scientific. There is a movement in AI which aims to do this. Rather than trying to make machines that can think like people, theres a movement trying to make machines act like people. Kinda subtle, no? Probabl a bit inaccurate too, but it this is what I mean: Attempt to create behavior based systems, which use systems similiar to our own to imitate our behavior, before attempting to simular our thoughts. Or better to say, insect and animal behavior. Based on neurons in biological brains, it is possible to create an artificial neuron. This neuron will take some input, and give a certain output based on it. By networking a whole system of artificial neurons, fantastic things can happen.
It's like creating a whole bunch of stupid little guys, and giving each of them a couple of cards - maybe one yellow, and one red. Now these guys are stupid, so they can't do much. You say to each 'if you get a red card, give a yellow one to your neighbour'. If you get enough of these guys together, you can get some spectacular communication. Consider an artificial aye, with 100 input neurons (those are little guys who hold up a card based on what they actually see).Each of those guys passes their cards along, and somewhere, one guy collects them up. And this guy, the output neuron, can decide whether he is looking at a yellow object, or a red object.
Where these networks get really interesting is in learning. It's impossible to predefine what each neuron should output for every given input. So you make each neuron a bit flexible, so that it can change its output. If the network as a whole seems to be outputting the right kind of things, you leave it is alone. But if the outputs arent quite right, you adjust each neuron slightly. After alot of tweaking and testing, the system will have learned how to reliable distinguish a yellow object from a red object. Or, less trivially, a friend from a foe; or a picture of a man from a picture of a woman.
This is a fantastically rough and flexible science. The programmer sets up the framework for a brain - a couple of signposts here and there. Then the brain is trained a little, and it learns, all by itself, what it is supposed to do. And hey presto, a working AI system! It's like magic - it's hard to say precisely how it works. In fact, thats irrelevant. But the system has evolved of its own accord. It's wonderfully flexible. With the right frames and the right sorts of training, a network can do all sorts of things. Now, the technology isn't all there just yet. Currently, such networks are only useful for fairly primitive behaviors. Reactions to sensor inputs and the like. High level reasoning and problem solving aren't possible. Just yet, at least.
Based on these networks, a philosophy to AI has emerged. Life isn't complex and mathematical. When walking down the road, or catching a ball, we do not subconciously churn through massively complex maths in order to retain balance and co-ordination. Rather, we do such 'calculations' on the fly. When we see a ball come towards us, we stick a hand out in roughly the right place to catch it. As the ball comes closer, we move our hand closer and closer, until its caught. It's about rules of thumb, and fine-tuning. As we get better at it, we can remember and learn from experience where things are likely to go. This makes it easier to put our hand in roughly the right area to catch the ball, requiring less fine tuning. The same approach is often used in robotics. Instead of plotting co-ordinates and precise movements for a robotic arm - move 6degrees left, 2 milimetres up, one centimetre forward, etc - it is possible to move in a general direction, and then adjust as you get closer. SO instead you move the arm forwards a lot, left a bit, up a tiny bit, and catch. Simple, intuitive, and felxible.
However, rather than focusing on these interesting, flexible and creative approaches to AI, Sussex Uni is determined to do it the hard way. The slow, pedantic, overly precise way. Most of the work we do is based on research that was carried out in the 80's. It's all old news, bad news, or irrelevant news. It leaves me with my own, unique problem. How do I learn about science which seems to be old, wrong, and abusrdly difficult? It's all too artificial, to be honest.
Oh, theres another quandry to it all. What if some things are better left unknown? Do we really want to know everything about our brains? Without even thinking about rampaging hordes or stomping, angry robots; or genetically modified crazy people; maybe it's just not natural to know that much about how life works?