Memristors could unlock world of AI

HOUGHTON – Memristors should someday be exponentially more powerful than microchips, said Michigan Tech engineering Professor Yun Hang Hu, who has developed what appears be the most successful memristor to date. Not only that, they won’t be particularly expensive to make.

Eventually he expects to see them in all sorts of computing devices, right down to phones. That will require differently programmed non-binary computers, though, to take advantage of all that memory, so he expects the first applications to take place at the most complex end of computing, simulating things like quantum mechanics that have a huge number of variables.

Memristors could also be the key to the first successful artificial intelligence, he said.

“They could make computer thinking maybe simulate the human brain,” Hu said. “The human brain creates new knowledge. Computers can’t create new knowledge. If you can simulate the human brain, you can create knowledge.”

This could eventually lead to medical applications, actually helping a brain-damaged person think, he said. Or, a computer could be tasked with problems that require both human-like creativity and vast stores of data to solve.

But just like in science fiction films – think “I, Robot” or “Terminator” – there are also risks to artificial intelligence and computers that can think for themselves, possibly beyond human control.

“If it’s driving a car, or running a life-support system, it’s a concern,” Hu acknowledged. “You should have a way to control the computer, a physical off switch.”