Will machines ever be as smart as humans? Intel CTO Justin Rattner thinks that someday, they might.
The notion of a technological "singularity," a time when machines match and surpass human intellect, has been popularized by thinkers such as inventor and author Raymond Kurzweil, who commonly cites Moore's Law in his arguments about the exponential growth of technology.
Rattner's views on the singularity are sought after, given that he is CTO of the world's biggest chipmaker and the head of Intel Labs, the company's primary research arm. In a recent interview with Network World, Rattner said he has "tried to sidestep the question of when [the singularity] might occur," but says machine intelligence is constantly increasing due to laws of accelerating returns, "of which Moore's Law is perhaps the best example."
"There will be a surprising amount of machines that do exhibit human-like capabilities," Rattner said. "Not to the extent of what humans can do today, but in an increasing number of areas these machines will show more and more human-like intelligence, particularly in the perceptual tasks. So yeah, at some point, assuming all kinds of advances and breakthroughs, it's not inconceivable we'll reach a point that machines do match human intelligence."
Already, scientists are working on placing neural sensors and chips into the brain, allowing people to control prosthetic limbs with their own thoughts. This is likely to become a "relatively routine procedure" in a few years, Rattner said.
Rattner said that while many commentators are preoccupied with the far-off singularity, he concerns himself more on how laws of accelerating returns "are real" and could lead to amazing advances in technology, including augmentation of the human body.
"Assuming that interface technology progresses in an accelerating way, the possibilities of augmenting human intelligence with machine intelligence become increasingly real and more diverse," Rattner said.
Rattner's views are also held in high regard in the world of supercomputing, of course, and he will deliver the opening address at the SC supercomputing conference in Portland, Ore. in November. Nearly 80% of the world's 500 fastest supercomputers use Intel processors.
The world's first petaflop machines, capable of performing one thousand trillion calculations per second, came online just last year. But Rattner says the supercomputing industry is already looking forward to the era of the exaflop – 1,000 times faster than a petaflop.
Rattner says the fundamental technologies behind a future exaflop machine could be demonstrated by the middle of next decade, and – depending on government investment – the first exaflop machines could become operational in the second half of the decade.
But this still depends on overcoming limitations in today's computing architectures.
"Now that we've achieved petascale computing, there's all this interest in getting the next factor of 1,000," Rattner said. "But we can't get there with today's technology, largely because of power considerations. You'd need a 500-megawatt nuclear power station to run the thing."
The industry will have to move that number down to something practical, perhaps tens of megawatts, Rattner said. But the work is just getting started.
"We've got a lot of really big engineering challenges," Rattner said. "Today, we just don't know how to get there."