In the movies, smart robots at our beck and call are a commonplace. So why is it taking so long to get there in reality? When will it be just as natural to have a pet robot as to have a pet animal? And what can we learn about biological life when we design artificial life?
Pablo Picasso is said to have remarked “Computers are useless. They can only give answers.” Perhaps it’s as simple as that. Computers, machines, robots — they’re only products of human hands. They’re quick at calculating and efficient at remembering. But it’s still we humans who call the shots.
Professor Peter Gärdenfors is a philosopher. He’s interested in both the human mind and artificial intelligence:
“If we managed to build a really good robot it would be an interesting test problem for us," he says. "A thinking robot needs memory, perception, knowledge, the ability to learn, the capacity to communicate and some kind of ability to plan ahead — things we humans deal with every day.”
So what’s the hardest problem to solve? Why aren’t we there already?
“The industrial robots we have aren’t very independent, nor are they flexible," Gärdenfors says. "They do precisely what they’ve been programmed to do and nothing else. Thinking robots would be more like us, and they’ve proved to be enormously hard to construct. We’re humbled by nature, where evolution has been working for millions of years. We simply don’t understand the whole evolutionary picture.”
Understanding what a robot must learn touches upon the very concept of "understanding," according to Gärdenfors:
“We’ve been developing computers for decades now but we can’t yet use them to understand things — machines are bad at that," he says. "On the other hand, they can visualize and they can be used to simulate phenomena. Computers have become true tools, in teaching for example.”
So getting computers to understand, to learn on their own, has proved to be much more difficult than we thought.
“And remember that human beings and animals are motivated to learn," Gärdenfors says. "Evolution has provided us with subconscious structures, complicated reward systems in the brain that cause us to choose one action rather than another. Robots don’t have motivation.
“Another important thing is the ability to plan. This is where humans are unique in the animal world. If we want to teach a machine to see the consequences of its actions it must first have some model of the world around it.”
So building a thinking robot is more complicated than we first thought. Perhaps it ought to build itself instead, if it’s to resemble us in intelligence? This naturally raises a host of ethical problems.
How do we feel about the prospect of being surrounded by independent, thinking machines? How far has the research come?
“It actually seems most logical to get robots to imitate our own development, for example learning through play," Gärdenfors says. "We’ve made some progress by getting infantoids — "child robots" — to interact with real children.”
But he points out that research is in its infancy and there’s a lot to be done before robots possess both self-consciousness and emotions.
“Heaven preserve us from psychopathic robots!”