|
COMPUTING
|
|
Unless you've been living under a rock for the past 40 years (or in Yorkshire) you may have noticed a computer revolution. We've all got one - well, obviously (otherwise you wouldn't be reading this) - but I'm actually referring to the widespread use of the pocket calculator.
We have seen advancing technology scale down computers the size of steam engines to fit into our pockets, large reels of copper wire replaced by floppy disks and huge vacuum tubes reduced to micro-chips.
We can only wonder what may come next.
Well the square-eyed brain-bods believe the answer may lie in neural networks.
This is where scientists get computers to think like humans by using - oddly enough - a neural network.
So this means computers will soon learn to be compulsive liars, totally selfish and make other people's lives a misery - and could eventually be elected.
What is a neural net?
The neural net is a complex circuit, made of nodes and links - designed to mimic the neurons and dendrites within the human brain.
Input (from light sensors, heat detectors etc) is passed into several nodes at the top level of nodes. Each of these nodes is linked to some or all of the other nodes in the following level, and passes on the data it recieves (in an altered form) down the chain. This is repeated onwards until one node, at the end of the net, outputs a single value which is returned to the user.
How does it work?
Example data is fed into the input end of the net. The numerical values are passed down the circuit, and at each node a simple calculation takes place - the node sums up all the inputs that are passed into it and outputs a pulse of electricity only if the sum exceeds a certain value (the Threshold Value).
The data filters through to the output end, and if the result was an incorrect one, the programmer alters the threshold value for one or some of the nodes and re-tests the net's response until it produces correct results of all the example data.
Alas, the complexity of the human brain is phenominal. It has in excess of 10 thousand million million neurons, each connected to about 1000 others in, what usually appears to be, a random way.
A whole human brain would need a gigantic silicon equivalent and would take almost forever to program.
Only simple neural nets, performing single tasks are availiable at the moment.
Personally, though, I have to agree with my AI lecturer, who believes that computers will never become totally human-like. This is mainly because we, ourselves, don't even know how our own brains work, so how could we ever program this into a computer?
However, despite this, neural networks is by far the best ideology in AI.
Researchers sit and learn the neural network to recognise objects - just like we are by our parents - so at least appears to mimic the human learning process.
Nearly all other approaches to AI require alot of human intelligence to create the program that the computer works on, meaning that the output from the computer is not all that different from what was programmed in.
At the moment most AI is more artifical than intelligent.
If you're into neurology, then why not take a look at this page on logic programming, or at future fantastic's web site at the BBC, or the Aston Uni. neural network researchers - we are one of the leading researchers in that field.
Or if you're really not interested in modern doppelganger, then why get into the swing of things at Disney's Toy Story site. Or maybe just take a peep at the latest games
from E.A., Gremlin, Sega or Nintendo.
