Life with the creature feature – Jack Schofield charts the rise of the robots
The creation of artificial life will be one of the themes of the 21st century; it’s already helping to drive the toys and games market, from CyberLife’s Creatures to Sony’s AIBO robot dog.
A generation of children is keeping “virtual pets” in PCs, while Creatures (www.creaturelab.com) allows them to explore breeding and genetic manipulation using Norns.
Steve Grand, British designer of Creatures, says: “Probably some of the best artificial life scientists in the world are 13-year-old girls, because of what they’ve done with Creatures.”
Toys allow companies to test advanced technologies that may not be ready for hard commercial use, such as speech recognition. Talking Max2, a cartoon parrot, obeys verbal commands and engages in conversations.
Soon designers will be able to simulate more sophisticated life forms. Mindscape recently launched Babyz as the natural progression from Dogz and Catz. Other companies, like Dave Morris’s AIdeaLabs in Texas, have already tried “virtual girlfriends”.
The problem is making the simulation varied and believable enough to keep the player’s interest. The creatures must have the ability to learn, and to reward their teachers.
When they become smarter, artificial life constructs will become part of computer interfaces. Microsoft Agent technology is already being used experimentally on dozens of websites.
There is a huge potential market for such AI (artificial intelligence) constructs. Software robots will be intelligent opponents for games players. They’ll scour the Web for information, go comparison shopping and eventually answer the phone and your e-mail.
But some are worried about what happens when they become too smart. Professor Hugo de Garis (www.hip. atr.co.jp/~degaris), a “brain builder” based in Japan, believes that a battle for “species dominance” will unfold. “Humanity will have to decide whether it wants to build artilects (artificial intellects), which are coming simply due to the rise of the technologies. It’s the Terminator scenarioE…”
Will bots and robots still be our pets, like the Robo Kitten that De Garis is developing? Or will we be theirs?
Surely not. After all, AI research has been going on for decades without making an impact outside a few specialised areas. Why should that change now?
“Computer power,” replies professor Hans Moravec, director of the Robotics Institute (www.frc.ri.cmu.deu/ ~hpm) at Carnegie Mellon University in Pittsburgh. “Once it was like having a rubber band engine to power a 747: it didn’t matter how good your aerodynamics were. That’s changed.”
In the 1990s micro-computers increased in power from less than one MIPS (the ability to perform a million instructions per second) to 100 MIPS, and they’re now approaching 1 000 MIPS. In fact, computer power is increasing so rapidly that Moravec is turning down speaking engagements to concentrate on developing navigation systems for things like guard robots and cleaning machines.
“I’m rushing to capitalise on decades of preparation before further increases in computer power make the problem so easy everyone is doing it,” Moravec says.
“The business plan says that with 1E000 MIPS we can build that kind of thing. We’ll have 1E000 MIPS microcontrollers in about three years, and we’ll go commercial then.”
Moravec compares the computer power available for AI with animal life forms. The one MIPS machine – think of a 1985 personal computer – is as smart as a nematode worm. The new 1E000 MIPS systems have advanced to insect level, approaching a spider. Moravec thinks he’ll be able to apply the power of a guppy to a robot vacuum cleaner in about 2005.
After more thousand-fold increases, it’s possible to imagine machines as smart as monkeys, or men, perhaps in about 2040.
“I strongly believe that the development [of human-level robots] can be done incrementally, by evolution,” says Moravec. “I look for a series of slightly more sophisticated machines, each capable of earning a living.”
Grand has also decided to go into robotics, forming a new company: CyberLife Research (www.cyberlife- research.com). “I want to make machines that think,” says Grand. “I’m trying to develop a general-purpose brain on a chip. I’m not looking for applications right now, but I’m thinking of building a model glider that can learn to fly.”
Grand is sceptical about the field he’s just entered – “it took evolution millions of years for good reasons” – but he agrees that 40 years is a reasonable time-scale for brain building. “We should beat chimpanzee level.”
De Garis thinks he’s leading the way, with his CBM or CAM-Brain Machine (CAM stands for Cellular Automata Machine).
The idea, in essence, is to build machines that create brain-like structures called neural networks, generating them automatically. De Garis says the CBM “can evolve a neural net circuit module of some 1E000 neurons in a second or so”. The aim is to have an artificial brain with a billion neurons by 2001, though 75-million neurons now looks the likely maximum. “Evolutionary engineers” then need to turn this raw material into a working brain. De Garis says “proof of concept” should arrive soon.
“In the short term, the products will have very beneficial effects: they’ll be wonderful,” says de Garis. “But eventually they take over their own development; humans lose control.”
Grand is more optimistic. “They’re just another life form; there are millions of other species out there that haven’t `taken over the world’. People are bound to be resistant at first, but then they’re also scared of impersonal technology. A-life can make machines more friendly. People will soon learn to love them.”
ENDS