Rodney Brooks is trying to build a human. He tells Kathryn Jackson what it is like to play God
Thirty years ago an Australian boy was given two books: The How and Why Book of Electricity and The How and Why Book of Robots and Electrical Brains. Fascinated, he dreamt of building a computer. It was the early 1960s and he had never seen one. Yet with wires, indicators and other electrical components stripped from obsolete telephone switchboard units, he finally succeeded, aged 12, in building a noughts-and-crosses machine. Two years later, using metal ice-cream tubs and scavenged electrical components, he set out to build a robotic tortoise.
He never got the tortoise to work. Today, as director of the Artificial Intelligence Laboratory at the Massachusetts Institute of Technology, Rodney A. Brooks, 44, is working on a much more complicated and controversial project: "We're building a human being," he says.
Brooks is working on the "most ambitious robotic project anywhere in the world", says Phil Husbands of the Sussex Centre for Computational Neuroscience and Robotics. "And it is one of the most important. Even if it fails, we are bound to learn a lot from it."
Cog and Kismet are the two most advanced humanoids yet developed. Cog (short for cognitive) is a massive robot, developed by a team headed by Brooks. Kismet, a baby humanoid, has been developed, nurtured and maintained by Cynthia Breazeal, one of Brooks's PhD students. To meet either is an unsettling experience.
Taller than 6'4", Cog is an imposing presence of metal and movement. Despite being just a head, arms and a torso fixed to its base, Cog swivels what body it has with uncanny grace. It can play with a doll or a Slinky, and sees with the aid of inner-eye cameras. Most importantly, says Brooks, Cog was designed to interact with humans. And, sure enough, when you walk into the lab, you cannot stop yourself from trying to get Cog's attention.
Kismet is smaller, but more engaging. Breazeal's aim in designing Kismet was to create a robot that interacted emotionally with people. "That's why I made Kismet so adorable, with big blue eyes and big floppy ears that move."
These robots are only two of the menagerie of artificial creatures in Brooks's lab. Started in 1959 by Marvin Minsky and John McCarthy, the lab has a long tradition of experimentation, culminating, most recently, in a wide range of robots and applications of artificial intelligence - everything from new tools to assist image-based surgery to an army of microbots that can do the dangerous work of land-mine detection. "I wrote a paper called 'Fast, Cheap and Out of Control' which suggested that space exploration could best be done by hordes of tiny robots," says Brooks. "Nasa (the American space agency) eventually sent one very large robot to Mars, based on our work."
About 15 years ago, Brooks rejected the dominant, top-down, theory of artificial intelligence. "Until the 1980s, everyone working in AI focused on reasoning and planning," says Brooks. "Their robots could operate only within a fixed environment. They would sit, plan, compute, move for 15 minutes and then start the whole process over again."
The AI establishment was trying to duplicate complicated forms of intelligence that had evolved over millions of years. It was, in effect, starting at the end of the evolutionary process. "I wanted to recapitulate phylogeny in a single robot. This meant developing a layered approach, and it meant that any intelligent robot had to be embodied (in their environment) - just like all thinking things- and had to perceive and react to the dynamic world around it."
Soon after Brooks joined the AI Lab at MIT in 1984, he started working on the first of his behaviour-based robotic insects. "I constructed the robots from the bottom up, in layers. First, there was a simple control system, based on a set of sensors that allowed them to avoid obstacles as they moved around. The second layer had them seeking out a target. But if there was an obstacle, the lower-level system took priority: they would avoid the obstacle instead of heading straight to their goal." On this model, intelligence is not the starting point; rather, intelligence emerges as a creature reacts flexibly to an unpredictable world. Robots could be designed without the increasingly complex computations needed for the top-down approach.
It took Brooks ten years to build a robot based on an insect. He intended then to build a series of creatures: first a reptile, then a small mammal and so on, all based on the layering of the more complex over the simple, and all behaviour-based. "But I saw my own mortality getting in the way," says Brooks. "So I decided to jump straight to a human being." Hence, Cog.
With Cog, of course, we come to the big question: is it human or does it just look human? Brooks has little time for philosophical critics, such as John Searle, who say that robots lack "consciousness" - the human awareness of everything, smells, colours, taste - and so cannot be intelligent. "These critics believe in an elixir of life. But what could that be? My one worry is that there is some aspect of living systems that we're missing, but my hope is that it is something simple. If I had my own version of the elixir of life, it would be an equation."
For Brooks, a robot is human not because of what is going on inside it, but because of how we respond to it: "I will have achieved my goal when we treat humanoids as our equals in all ways - legally, socially and in our hearts."
Once a dreamer, always a dreamer.