This life: the artificial one

The quest to create life-like creatures out of silicon chips, wires and wheels is well under way and may one day culminate in the creation of a virtual human. Kam Patel reports

Last week artificial life researchers showed off their latest man-made creations at a conference in Brighton. Among them, a walking eight-legged creature, a robot that can climb stairs, a pet that likes its stomach patted and gets grumpy if you bother it too much. Are these and other Alife creatures evidence that life is not necessarily a property of organic chemistry, but rather a function of how matter is organised? Is there any reason why living things cannot be constructed out of silicon chips, wires and wheels?

For philosopher and conference delegate Mike Wheeler, the work of Alife researchers poses, in a very stark way, the question "What is life?". "If you were to ask me, 'have we succeeded in creating life-like creatures, simulations and robots?', I think the answer is yes. These artificial creatures exhibit many of the behaviours we see in animals".

Many people think that artificial, in the context of artificial life, means "not quite real". But Wheeler, of Oxford University, says that to researchers, "artificial life means the creation and exploration of life-like processes in artificial media, not biological stuff, but using, for instance, silicon and computer simulations."

There is no consensus in science or philosophy about what counts as being alive. The question could be posed by asking for a list of properties believed essential to living systems. Being a product of reproduction, being able to reproduce and adapt to environments are among the prime conditions many intellectuals insist have to be fulfiled before something can be classed as biologically alive. Wheeler believes some of these conditions are already satisfied by Alife creatures. But there is a long way to go. Metabolism is an example of a key feature of biological systems not yet incorporated into Alife creatures. "For some of us, there has to be some complex internal, energy-budgeting system that maintains the body. I don't see it yet in Alife but I do not rule it out in the future."

The philosophical problem posed by Alife is reflected by the scope of the discipline. Dave Cliff, lecturer in computer science and artificial intelligence at Sussex University, says: "If you say you are interested in life-like phenomena, that includes everything from self-reproducing chemical chains like DNA, or the evolution of single species right up to social phenomena such as the emergence of economies in advanced nations and the cultural transmission of information. All those features could, in principle, be subsumed within the field of artificial life."

Much of the work in this fast-moving field revolves around trying to understand how complex systems and behaviour emerge from large collections of very simple elements. The extraordinary feats of the human brain, for instance, are made possible by tiny nerve cells firing and interacting.

Mathematical modelling of such elements has been crucial to advances in Alife, while the falling cost of computing power has enabled sophisticated simulation of artificial life forms. Cliff says that experiments which now require massive supercomputers are likely to be feasible in five to ten years using off-the-shelf systems. But he stresses that Alife is not just about building interesting robots, programmes and silicon chips. There are serious endeavours in the field, including attempts to further understand facets of theoretical biology: "There is this desperate problem that life has only evolved on earth once and it has taken four billion years. It is not plausible to send a space ship to another planet and observe that for billions of years. So what we are trying to do is build theoretical models, simulated on computer, which can track the progress of evolution under different circumstances." Such models could make it possible to glean what the implications for evolution on earth might have been if, for instance, earth had just one land mass instead of five continents, or if there were less sunlight.

This artificial evolution, on computer, of effective, reliable systems which can then be tried out in the real world, is throwing up radical ideas. Sussex University's Inman Harvey, co-ordinator of last week's conference, cites the development of the walking eight-legged creature by his colleague Nick Jakobi. Typically Alife researchers will evolve robot behaviour on computer before loading it into a real robot for testing. Often they fail. Jakobi, who is interested in how good simulations need to be to cope with the real world, has devised a radical theory which proposes that very little data need be given to the evolving program. Instead of specifying, in detail, what the eight legs of his creature should be doing in space and time in order to walk, Jakobi provided the simulation with only key information.

Harvey explains: "It is a crude simulation of what the legs would be doing if the robot were successfully walking. The trick is to make sure the simulation doesn't provide some other route for walking which would fail in the real world. Typically, with artificial evolution, if you make a mistake in a programme, the simulation will misuse and abuse it." He cites the classic case of researchers who evolved on computer the design of an aircraft with the task of flying as long as possible on a tank of fuel. The team left the simulation running overnight and came back the following morning to find that the fittest planes were flying forever. Backwards. "The simulation blindly followed the rules. By flying backwards, fuel was pumped into the tank from nowhere and the tank filled to maximum all the time. According to the simulation, it was a brilliant solution but it is not something you can sell to airlines," says Harvey.

It is not just academics who are showing interest in Alife. Firms with interests ranging from entertainment to pharmaceuticals and telecoms are also looking at what it has to offer. Cambridge-based CyberLife, for instance, wanted to make its games characters more life-like. The firm, which has research links with Sussex and MIT, discovered it could build sophisticated artificial organisms. Last year it released its acclaimed Creatures software package which generates an artificial environment consisting of land and water, occupied by vehicles, food and creatures who are exposed to hostile elements including viruses. A host of techniques from the world of artificial intelligence, including neural networks that mimic the operation of real nerve cells, are embodied by the creatures to provide a rich set of responses. If the creature walks too near a flame chemicals are released in its bloodstream, affecting its central nervous system. The creature has to reason what it must do to reduce the pain. It can also experience hunger and boredom and suffer from disease. It even has a sex drive. Anil Malhotra, CyberLife's business development manager says: "They can sense the world, reason, learn and therefore adapt to changes. And all the different characteristics - how their biochemistry interacts, how the neural network forms, learns and activates - all of that is specified by a digital DNA. The genome of offspring is a unique product of crossover of the genes of the parent creatures." It all amounts, he claims, to an open-ended mechanism for evolution to take place.

Described as the world's first mass-market artificial life experiment, Creatures has spawned interest worldwide. Malhotra says: "We have sold 300,000 copies and there are over 100 individual websites where people exchange information on the package. Someone has built a software program to artificially edit genes. And somebody in Germany decided to categorise and name all the different diseases the creatures can catch - the list is available on the web."

CyberLife's aim is to create "completely integrated, holistic life-forms". Malhotra says: "By the year 2020 we want to have developed a virtual human. It would have intelligence, language, reasoning, conversation, the lot. It will be something on screen or whatever devices we have at the time and you will find it as comfortable and natural interacting with the form as you would with a human."

For a philosopher, the possibility of life-like Alife forms poses interesting ethical questions. Mike Wheeler says: "If we were satisfied at some point in the future that we really could not get a better definition of life than something already proposed, and decided these creatures met that definition, that there really wasn't any interesting difference between them and biological forms, then that would raise a host of moral questions. Whatever moral rights we grant to animals, for example, we would have to grant to these creatures."

The question "what is life?" is not being asked enough in philosophy, Wheeler says. He hopes the emergence of Alife will inspire philosophers to think more about it. "You cannot extrapolate directly to humans - but we have always learned a bit about ourselves by looking at other creatures and I don't see why Alife should be any different."

* How Gary beat Vinnie, Eric and Alan at football

Gary's first movewas a dazzling spin with the ball, which he drove into the goal. Unfortunately it was his own goal. Cantona, one up without trying, displayed his namesake's penchant for idle philosophy: rather than expend effort, he sat against a wall while Gary scored more random goals. The result:a three-all draw.

The first Auton-omous Robotics Football Tournament displayed the programming skills of Artificial Life researchers - and demonstrated the problems posed by even simple tasks. Maintaining direction, for example, is tricky for a robot on a symmetrical pitch.

Gary, entered by Queen Mary and Westfield College researchers, was programmed to play like Gary Lineker. QMW also entered Vinny, a more brutal player. Rushing around the metre-long pitch in search of anything to kick, Vinny intimidated his opponent more often than the ball.

Although resear-chers tried creative programming, different techniques gave broadly similar results. "There's not much difference between the clever approach and just grabbing the ball and heading for goal," said Gary's programmer, Andy Rix.

Tom Smith of Sussex University abandoned AL principles in order to make Cantona respond at all. "The original program, based on biological techniques, was designed using evolutionary computation: it evolved to find better and better solutions, but too slowly. In the end I went for basic reactions - find the ball, avoid theother robot, move towards the goal."

Even such simple tactics failed towin Cantona many games. Vinny developed an even closer attachment to his opponent, and Alan (despite a Newcastle black-and-white strip) never hit form. The way was open for Gary who, luckily, faced his opponent's goal more often than his own, to win.