There are 27 bones in the human hand, but almost no one thinks about that when they pick up a water bottle. Ryan Adams does.
Adams, a professor of computer science, is leading research into mechanical intelligence. If artificial intelligence mimics cognition, the mechanical version acts like the autonomic system that drives walking, grasping, and breathing.
“Intelligence is more than just the brain,” he said.
It is not only a matter of computation. Organisms’ physical bodies are optimized for certain behaviors, which reduces the burden on the nervous system, said Adams.
“An example is grasping for primates. The design of your hand makes it easy to control,” he said. “Fish bodies really want to swim. It makes it easy for the brain.”
Over millions of years, these physical systems evolved in tandem with the neurological systems that control them. In Adams’ lab, researchers deploy AI techniques to speed up this process. In one, they use a technique called gradient descent algorithms to allow AI systems to solve problems by trying many potential solutions. As the AI solves a single problem, it also develops the ability to solve other, similar problems.
“We can use things like gradients to get better at solving a problem,” Adams said. “Can we use them to get better at mechanical things?”
