“Why is so much AI focused on analysis,” asked Hod Lipson, “and so little on synthesis?” Speaking to a packed audience at MIT’s EmTech 2015 event in Cambridge, MA, Lipson, professor of engineering at Columbia University, addressed how we should rethink the way we construct robots. As director of the Creative Machines Lab, Lipson’s focus has been coming up with new ways to make “machines that create–and machines that are creative.”
Up to this point, Lipson said, most artificial intelligence concentrates on taking data (in areas like the stock market or weather), distilling it, and making predictions. It asks, “How will you behave? What will you buy?” Lipson said. In the case of autonomous cars, the machines are tasked with “whether to turn, or when to accelerate.”
But this type of data is all “convergent AI, analytical Ai–all about coming to a decision.”
“The kind of intelligence we don’t talk about,” said Lipson, “is divergent thinking. You start off with a small idea, a seed, a principle, and you expand it.”
It’s interesting, Lipson noted, that while innovation is so broadly praised in our society, we put little emphasis on it when it comes to robots. “We crave it badly,” Lipson said. “Not a day goes by where I don’t see some kind of conference about fostering creativity. And if you ask almost any parent to talk about how bright their child is, they talk about creativity.”
“It’s very human,” Lipson said, “unique.”
But when applied to machines, “analysis is much easier.” If you look at an education in engineering, for example, “there’s a lot of focus on analysis, predictions, computation,” Lipson said. “But synthesis, design, is very difficult to teach.”
“Many courses on design are still taught the way they were 200 years ago–by apprenticeship,” Lipson said. “And if we can’t teach [creativity] to people, it’s very difficult to teach to computers.”
However, if we move beyond human creativity, said Lipson, “there’s a much better process of creativity–the mother of creators–evolution.”
Evolution, he said, “started with a seed a billion years ago, and by iterative expansion and selection, has created incredible things all around us. Evolution dwarfs even the best team of engineers.”
“As fascinating as these robots are, they’re designed by people. By engineers sitting at a desk, deciding how pieces should go together,” he said. “Often, we humans are the bottleneck. We can’t create such a rich diversity as evolution can.”
Is there a way to harness that process of evolution? Lipson asked.
The answer, Lipson believes, is in “creating software to breed it. Few people can design something so complex–we don’t have tools or imagination.” Where we should focus, Lipson said, is on the “synergy between computers and people.” It’s in the partnership between human and machine where we can come up with advanced solutions. He gives the example of humans distinguishing shapes that conform with a certain concept and machines building on this to create similar shapes.
“Computers can tease out the ideas from a person,” Lipson said, “and gradually generate new ideas.”