& opinion


Although I’ve been assured, while talking to the two heads of CCNR, Michael O’Shea and Phil Husbands, that a Matrix-style robot takeover is not on the cards, a movie of a line of robots is making me feel slightly uneasy. As we watch three boxes on wheels doing a geriatric conga across a lab floor, I want to ask have you never seen Terminator? Do you really want to be visited by an angry John Connor and his seven foot tall Austrian cyborg?’

The reason for my paranoia is that they have organised themselves into a moving line - one robot taking the role of leader and the other two following - without any prompting or sophisticated communication between them; only four infra red sensors each. When they reach a wall, they swap roles, and start in the opposite direction. They can do this because their brains have evolved that way; no engineer with a blank piece of paper could have done a better job.

So, how do you go about evolving robot brains? Imagine a load of points connected up randomly with wires. The way in which they are connected can be represented as a string of information (see the first part of the picture) in much the same way as you can represent genes. Create thousands of these and you have an initial artificial gene pool. You have to test each one for the ability to perform a certain task; say move towards an object that looks like a box on wheels (there are far too many genes to really do this, so it’s all simulated on a computer). Kill off the most useless - a sort of un-natural natural selection - and chop up and mix all the genes that are left. You then have to mutate a small amount of the information, swapping zeros for ones and vice versa for about 4 bits in 500 it is interesting that this is absolutely essential to the evolutionary process - it can only be explained if you think of evolution as a ‘fitness landscape’ with hills valleys and plateaux, rather than any misleading notion of a linear march towards perfection. The mutations can be thought of as the gene pool feeling around the local area for any nearby hills). Repeat this for about 10,000 generations and the robots are ready.

The power of this process is such that it can be used for all sorts of things from designing computers to aircraft wings; anything where there are too many changeable parts (like robot brains), or where you have too much chaos (like the flow of air over a wing). One area in CCNR is concerned with evolving circuits in a special type of silicone without modelling it on a computer first. This has given birth to parts of circuits that seem to have no point, but turn out to be essential when removed. The process can feel its way round properties of the silicone that are unknown to the experimenters!

CCNR works at the cross-over between biology, psychology, electronics and IT. Taking inspiration from biological systems, it strives to create ‘simple autonomous intelligent machines’ which can be used in inhospitable places, such as nuclear reactors or the surface of Mars (part of the funding comes from the British Space Centre). There are teams looking at a diverse range of areas from; how insects can learn and re-learn landmarks to guide them through changing landscapes; to modelling the brain as an ever changing chemical machine; and as already indicated, evolutionary theory.

Knowledge works both ways; modeling natural systems brings invaluable insights into the workings of them. Why have animals never evolved wheels? How do insects not get lost? How has the complex interplay of chemicals and electric pulses that are our brains come about? These are things that CCNR is interested in.

But what about the enslavement of the human race? Phil Husbands reassures me that the shuffling procession on the screen ‘can only be described as life in a very philosophically pedantic way’, and that the next step is to ‘make the robots move things, learn and re-learn landmarks - interact with their surroundings’. I wonder if the robots see it that way.