Artificial Intelligence Is Getting Smarter
A CENTRAL-heating boiler chatting with the dishwasher while the washing machine makes small talk with visitors?
A scientific revolution as profound as the invention of printing, or even the development of writing or numbers, promises to transform the domestic scene and force a reappraisal of our mental abilities, according to one of the world's leading researchers in artificial intelligence.
The idea that "intelligence" can truly be a property of anything other than a human being is deeply controversial. Philosophers battle over the uniqueness of the human mind and consciousness, but Prof. Donald Michie, who heads the Turing Institute in Glasgow, believes machines capable of language and learning will soon be developed that will challenge and extend human knowledge.
Professor Michie's early career brushed with that of Alan Turing, the founding father of artificial intelligence, who worked on Britain's code-cracking effort during World War II. Turing devised a test which lies at the root of the philosophical arguments. It said that if a computer program can perform in such a way that an expert cannot distinguish its performance from that of a human with a certain mental ability, the program can be said to have that ability.
The arguments have so far been more theoretical than real. Computers, although capable of impressive skills, are easily caught out by many aspects of human intelligence. Most present programs use deductive reasoning, working from the general to the particular. The snag is that all the rules have to be worked out first before giving them to the machine to interpret.
But Michie, formerly at Edinburgh University, where he established one of the leading artificial intelligence research groups, sees inductive reasoning, going from the particular to the general, as the key to machine intelligence. The idea is to give the machine examples and let it work out the underlying rules that govern what is going on. Only inductive systems have the ability to learn.
Michie carried his expertise to the Turing Institute, which he set up jointly with Jim Alty of the University of Strathclyde six years ago to bridge the gap between industry and academia in artificial-intelligence research. His group is among the pioneers developing such "expert" systems.
A harbinger of things to come, developed by several of the main computer companies, is an "agent a cartoon figure with whom one can converse and get help, advice, guidance, and information about what is going on in the operating system, soon to be available on computer screens. But Michie believes gradual improvement in the agents' skills is more likely to undermine than answer the philosophical issues.
"If eventually we really cannot tell the difference between the agent and a [human] colleague, we are going to have no more interest in arguing whether the program is really conscious than I am concerned with the question of whether my graduate students are really conscious," he says.
Other software developments are finding hidden depths to human intelligence, opening up a world that has previously been hidden. Psychologists are finding growing evidence that mental activity is something like an iceberg with only the tip open to conscious introspection, Michie says.
Michie's group has shown in the laboratory, and is beginning to confirm with pilots in flight simulators, that the rules buried in the brain governing the behavior of trained experts, to which they have no conscious access, can now be revealed. A machine using new inductive learning techniques can browse through the behavior records of a skilled human carrying out an expert task and automatically reconstruct a model of how he or she does it.
A key observation is that the model routinely outperforms the human from which it was derived. Michie believes such a system points the way to providing intelligent, highly skilled automatic backup. Machines may not only provide insight into hidden areas of human intelligence, but they also promise to make original contributions to human knowledge we would never otherwise have been able to gain.
Michie believes the software revolution will ultimately spread to robots that will talk with us and each other. It was quickly realized that two or more robots would be needed to carry out many potential tasks, but current efforts to develop communication systems based on computer networking models are overcomplicated, he argues. "You don't need to carry out brain surgery to link up communications between two humans," he says. "A fairly complex task like getting a sofa downstairs may only involve fairly simple communications such as 'up a bit,down a bit,' 'left a little.' Humans have perception-driven intelligence, and we are already at the stage of demonstrating in the laboratory very effective robots with speech synthesizers and speech recognizers that can talk to each other and cooperate on tasks. Such an approach also has the advantage that, if there were a hangup, the supervisor could listen in to find out why," he says.
Michie is convinced that speech-based solutions will eventually be adopted, leading to a domestic revolution in the next century. "Washing machines will talk," he says. "They will tell you when they are ready for a load, or that the washing is not quite dry. The deluxe model may also be skilled in small talk," he adds. Likewise domestic robots helping with the housework will talk with their companions and other household items about what they're doing and what's next."
The machines are bound to acquire personality in the eyes of the family in the way their cats and dogs do, but the extra property of mental transparency and articulate communication will bring a new dimension to mental life, Michie says. "This will be a domestic environment in which children will grow up so different from that of today that it is hard to envisage it."
