Robots are doing it for themselves: Computers are being allowed to develop their own intelligence, but there is still a long way to go before we can create artificial life. Jeremy Taylor reports

In spite of more than a century of biological research and evolutionary theorising since Darwin's day, we still understand precious little about life on Earth. We don't know how life got started; we don't know exactly how and why life went multicellular; and we don't know how intelligent behaviour and communication evolved.

Could artificial life provide some answers? The problem is that there are no other natural living systems in the universe with which to compare our own. The radio telescopes eavesdropping on the stars have not found life elsewhere and, despite the best efforts of Steven Spielberg, nobody has come to call. The hope that we could create parallel life forms on Earth that would illuminate the natural world gave rise to the scientific study of artificial life.

The Fourth Artificial Life Conference was held in Boston earlier this summer and an amazing electronic and mechanical menagerie was on display there: three-dimensional computer animations of fish that learn to swim, shoal and fight by evolving their nervous systems in a 'virtual' ocean; immune systems to protect computers against software viruses; herds of tiny robots that work together like ants to get the job done.

Chris Langton, the founder and chief guru of the artificial life movement, told the conference: 'We need to locate life as we know it within the wider realms of life as it could be. To compare what could have happened with what we think happened.'

He believes the best way to understand any form of life is to take an holistic approach rather than the scientific one which breaks it down into bits for analysis. He chooses to equate life with information, and to create digital forms of life that might parallel the real thing. This means creating a 'state of nature' in the computer, seeding it with digital 'organisms', and allowing them to evolve freely. The speed of computation can then compress billions of years of evolutionary time. It is, in effect, using the computer as a Tardis.

Talk like that prompted the British evolutionary theorist John Maynard-Smith to remark, shrewdly, a couple of years ago: 'The problem is that we stand to replace a form of life about which we know very little, with forms of life about which we know even less.'

He need not have worried, because playing God is proving to be a long, hard slog. However, one experiment has captured the imagination of many, and is about to be expanded.

Tom Ray, a biologist from the University of Delaware, created an evolving digital world called Tierra. Mr Ray's 'organisms' were tiny strings of computer code much simpler than, though analogous to, molecules of ribonucleic acid. The electronic organisms each contained the instructions needed to copy or replicate themselves, and they competed with one another to replicate, using CPU (central processing unit) time as a finite 'energy' source.

Evolution occurred through mutations in their instruction code. Tierra has become famous because of the extraordinarily life-like strategies that evolved among Mr Ray's computer 'bugs'. Parasites developed that used the copying instructions of other organisms, thereby cutting the size of those organisms and their replication time. 'Hyper-parasites' evolved to counter the parasites, and so on.

But the simulation was not completely life-like. The problem was that because 'fitness' - a crucial concept in evolution - was defined as the efficiency with which the bugs copied themselves, Tierra tended to evolve towards populations of bugs with small 'genomes' (sets of genes) that were good at copying themselves.

Tierra mitigated against the development of complexity. Life, as we know it, went the other way. Ray now wants to restructure Tierra to create a digital analogue of the Cambrian explosion of life- forms that occurred on Earth some 600 million years ago and that signalled the appearance of multicellular organisms. No single computer can provide an environment large or complex enough, so Tierra is going international.

Ray hopes to spread his digital organisms across the Internet, so that they can use the spare CPU cycles of thousands of computers already interlinked for e-mail purposes.

In this 'digital biodiversity reserve', Ray hopes that collaboration will emerge between different digital entities, creating the equivalent of multicellular organisms. These organisms might then get smart and evolve to respond to temporary gluts of CPU time anywhere on the globe; for instance, when programmers end their working day. They might constantly migrate around the planet, following the night. In this way, the system will be capable of rewarding complex behaviour. Digital naturalists, Ray hopes, will explore this digital jungle, noting changes in its ecology, diversity, behaviour, and morphology. Occasionally, as in the agricultural domestication of wild maize, wild digital organisms might exhibit behaviour that could be commercially usable.

Further up the artificial tree of life, the main concern is the building of brains and behaviour. In Boston, the conference floor was teeming. Robots scurried about collecting pucks into piles; staggered around like drunken arthropods; or clasped bystanders affectionately by the ankles. It stretched the imagination somewhat to realise that none of them had been programmed.

None the less, as the conference host, Rodney Brooks, remarked, they all exhibit 'intelligence without reason'. In none of the robots was there a brain on board with any sophisticated representation of the outside world. The 'brain' processors know nothing more than the fuzzy raw data that stream from the robot's sensors. Small processors whose entire remit is to lift a robot leg engage with others with a similarly blinkered view of the world. Yet remarkably coherent, purposeful behaviour emerges from this humble orchestra. This 'bottom- up' view of how intelligence emerges from many simple processes interacting with one other is already showing more promise than traditional artificial intelligence, where lumbering super-computers crammed with expert knowledge fail miserably when they try to negotiate a changing and ambiguous environment.

Towards the end of the conference a slick duo from Nasa attempted to wow the assembly with project Phoenix, which aims to comb the nearer reaches of the Milky Way for giveaway signals of extraterrestrial intelligence. Ladbroke is giving odds of 14-1 against anyone coming up with reliable evidence of little green men before the end of the millennium. The odds are comparatively short because they have been overwhelmed by takers. What are the odds against the artificial life movement coming up with something more interesting? The Japanese seem to be doing the betting: more than a quarter of those attending the conference were from Japan and two of the movement's brightest stars have already left for Kyoto: Tom Ray and Hugo de Garis. De Garis's speciality is an artificial brain consisting of a billion neurons which the Japanese hope to build using nanotechnology. Reality or delusion? Place your bets.

(Photograph omitted)