For decades books and films have debated whether androids of the future would dream of electric sheep. Now science is beginning to embrace the idea that robots could love them too.
Brian Cox, the broadcaster and scientist, is the latest figure to suggest that our own mental abilities could easily be surpassed by machine intelligence in the future. “There is nothing special about human brains. They operate according to the laws of physics,” Professor Cox told The Times. “In a sufficiently complex computer, I don’t see any reason why you couldn’t build AI.”
His comments follow rapid advances in the fields of machine learning, which raise the prospect of a blurring of the boundaries between humans and machines and a future in which we could not only work alongside robots but even socialise with them.
At such a point, even the term “artificial intelligence” will become outdated, he suggests. “It will just be intelligence.”
Professor Cox is not the first to hint that sentient machines could be just over the horizon. In June, Ray Kurzweil, director of engineering at Google, predicted that within 15 years computer intelligence would be sufficient for people to have emotional relationships with their operating systems, as portrayed in the film Her. Humour, romance and “being sexy” will not be beyond the grasp of computers for much longer, Mr Kurzweil claims.
Advertisement
Professor Cox was intrigued by the portrayal of human-computer relationships in the film, in which Scarlett Johansson plays a super-intelligent computer operating system. “I like the idea that we’re staring at computers so much, we’re becoming dehumanised and, actually, the computer is the most human character in the film,” he said.
He stopped short of forecasting when such advances might become a reality, but cited the EU’s Human Brain Project, an attempt to model the entire human brain inside a computer, as proof of the field’s ambition.
“In principle it’s possible,” he said, speaking before a lecture this week at the Society of Biology in London. “It might be terrifically complicated in practice. There’s often scepticism about things that are difficult, but it doesn’t mean you can’t do it.”
It may not yet be possible to share a joke or flirt with your smartphone, but in the past few years computers have begun to display abilities that were until recently considered uniquely human. A decade ago, image recognition software struggled to distinguish a dog from a horse. Now Microsoft has a programme that can not only accurately identify dogs, but also the breed.
Google has developed self-driving cars and a European project, MyCopter, is under way to develop a helicopter that can fly itself without need for human intervention. There have been major advances in “natural language”, the idea that you could type, or speak, conversationally with your computer and get meaningful responses.
Advertisement
Richard Socher, of Stanford University, has developed a program called NaSent that has a crude ability to recognise human sentiment in speech. He trained the software by giving it thousands of sentences taken from the film review website Rotten Tomatoes, which gives both verbal reviews and scores.
Within certain limited contexts, computers were “already there” in terms of understanding language, he said. “If you ask your Android phone: ‘What’s the weather like in San Francisco?’ it will often give you the right answer already.” But many of the subtleties of language, such as jokes, were still totally lost on a computer, he conceded.
“Sarcasm is also hard to detect because it can be so dependent on the knowledge of the person or situation,” he said. “‘Yeah, this is a great movie’, could be meant literally but could also be meant sarcastically if said by somebody who you clearly know hates this kind of movie.”