Miércoles 30 de Mayo de 2007, Ip nº 193

Let's dance
Por John Crace

Want to know how culture develops, or where humour and the arts spring from?

If you've got a lively imagination, the flashing lights could almost be eyes, the dome a head and the wheels legs. But even then you'd be struggling to call it cute. It looks pretty much what it is: an off-the-shelf, £350 "e-puck" robot, complete with circuit board, sensors, loudspeaker and microphone. It's not even a very clever robot; with just tens of neurons it's a great deal dimmer than the average insect.
Somehow, it's all a bit disappointing. You don't necessarily expect anything cuddly with a smiley face, but it would be nice if it had a few more anthropomorphic qualities. Because - if all goes to plan - 60 of these bog-standard robots could tell us as much about the processes and mechanisms of the emergence of culture and cultural behaviour as several decades of conventional evolutionary anthropology.

It sounds daft, but it's not quite so ridiculous as it seems, because we actually know surprisingly little about how culture develops. We understand a great deal about the emergence of social, cooperative behaviour - hunting, genetic diversification etc - but not very much about the development of cultural artefacts, such as dance, art and humour. We can theorise from archaeological finds, but the science is necessarily inexact as no one has ever been around to record the birth of proto-culture. Until now.

The idea came out of an Engineering and Physical Sciences Research Council (EPSRC) sandpit - they don't do blue skies any more - week, where a few dozen academics from a whole range of subjects were locked up in a hotel outside Reading and told to think about "emergence". Six months later, a multi-disciplinary team, led by roboticist Professor Alan Winfield at the University of the West of England, has been awarded a £750,000 grant to put a group of six or so robots into individual robot villages and to observe their interactions - both in real-time and in computer evolutionary time (a genetic algorithm can skip through the evolution of hundreds of generations in a matter of minutes) - for signs of cultural behaviour.

Winfield is quick to throw in a caveat. "The behaviours that emerge and evolve will not be human but decidedly robotic. We do not expect these artificial memes to have any meaning in a human cultural context - they will only be meaningful within the closed context of this artificial society. One of our key challenges in this research will be to identify and interpret these patterns of behaviour as evidence for an emerging robot culture and to see whether this new understanding may shed some light on how culture emerges and whether this has any implications for human, animal or artificial societies. In a sense we will be using robots like a microscope to study the evolution of culture."

This is easier said than done. While there may be perfect fidelity of observation - the robots will be on camera 24/7 - interpreting what their sounds and movements mean is rather more tricky. "It is a controversial area," Winfield admits, "but our team includes a social scientist, a philosopher, a biologist and a cultural theorist, and together we get a broad spectrum of answers. And there will inevitably be disagreements about what behaviours are social and what are cultural, but it's not just us who will be observing the robots: schools and universities will be able to watch them via the internet and their feedback will be welcome."

How, then, will the robots interact and why should they do anything interesting at all? Might they just not avoid each other for four years? "They won't do nothing," says Winfield, "because they will be programmed with some basic survival instincts that force them to do something. But otherwise anything is possible and you can't make predictions. Will they need sleep? Maybe. If so, we might introduce circadian rhythms. Will some robots prove dominant and bully others? Again, maybe. At present, everything is still very much up in the air."

But Winfield is confident something interesting will happen. "In an experiment conducted last year," he explains, "scientists in Belgium and Switzerland gave two robots a collaborative task that could be done with or without communication. The evolution of communication was not programmed into the robots, so there was no inevitability they would talk to one another. But when the genetic algorithms were run, the researchers discovered that on some occasions the robots did communicate. Somehow, they evolved to develop a language they both understood. Why? Because they could."

Dividing the robots into separate villages that only occasionally interact with each other should also help to ensure the greatest diversity of behaviour, but Winfield is under no illusions about the scale of what he is taking on - not just in the design of the experiment but in the potential it has for attracting a lot of flak from other scientists. "There have been suggestions the whole project should take place purely as a computer programme," he says, "but I believe it's important to use real robots. Although each robot is nominally the same and will be programmed in the same way, the reality is that their component parts will vary slightly from one to the other; so as the robots evolve generationally, the possibility of mutation becomes much more likely."

Winfield is happy to call this "adventure" research. "It's right out there," he smiles, "but why shouldn't we take a few risks?" Nor is he that bothered about failure. After all, while some colleagues may be having a few laughs behind his back, think of the fun if the project actually worked.


  01/05/2007. The Guardian.