IKEA furniture and the limits of AI

COMPUTERS have already proved better than people at playing chess and diagnosing diseases. But now a group of artificial-intelligence researchers in Singapore have managed to teach industrial robots to assemble an IKEA chair—for the first time uniting the worlds of Allen keys and Alan Turing. Now that machines have mastered one of the most baffling ways of spending a Saturday afternoon, can it be long before AIs rise up and enslave human beings in the silicon mines?

The research also holds a serious message. It highlights a deep truth about the limitations of automation. Machines excel at the sorts of abstract, cognitive tasks that, to people, signify intelligence—complex board games, say, or differential calculus. But they struggle with physical jobs, such as navigating a cluttered room, which are so simple that they hardly seem to count as intelligence at all. The IKEAbots are a case in point. It took a pair of them, pre-programmed by humans, more than 20 minutes to assemble a chair that a person could knock together in a fraction of the time (see article).

AI researchers call that observation Moravec’s paradox, and have known about it for decades. It does not seem to be the sort of problem that could be cured with a bit more research. Instead, it seems to be a fundamental truth: physical dexterity is computationally harder than playing Go. That humans do not grasp this is a side-effect of evolution. Natural selection has had billions of years to attack the problem of manipulating the physical world, to the point where it feels effortless. Chess, by contrast, is less than 2,000 years old. People find it hard because their brains are not wired for it.

That is something to bear in mind when thinking about the much-hyped effects of AI and automation, especially as AI moves out of the abstract world of data and information and into the real world of things you can drop on your foot. On April 13th Elon Musk, the boss of Tesla, an electric-car firm, said that the production problems which have dogged his company’s high-tech factory were partly the result of an overreliance on robots and automation. “Humans are underrated,” he tweeted. Lots of jobs have physical aspects that robots struggle with. Machines may soon be able to drive delivery vans, for instance. But, at least for now, they could well fail to carry a parcel to a flat at the top of a flight of slippery stairs, especially if the garden was patrolled by a dangerous dog.

Not such a silly Billy

Today’s AI systems are limited in other ways, too. They are pattern-recognition engines, trained on thousands of examples in the hope that the rules they infer will continue to apply in the wider world. But they apply those rules blindly, without a human-like understanding of what they are doing or an ability to improvise a solution on the spot. Makers of self-driving cars, for instance, worry constantly about how their machines will perform in “edge cases”—complicated and unusual situations that cannot be foreseen during training.

Calibrating excitement about AI is tricky. Researchers complain that great progress is quickly forgotten: as soon as a computer can do something, it ceases to count as “AI”. But those same researchers also tend to be more cautious about the future than many pundits. There is no reason, in principle, why a computer could not one day do everything a human can and more. But that will be the work of decades at least. Furniture-assembly helps explain why.