Eric Schmidt: I Was Skeptical About AI at First
Grow Your Business, Not Your Inbox
Artificial intelligence research should be done publicly and for the public good, according to Alphabet Executive Chairman Eric Schmidt, who admitted to not initially recognizing the importance of AI.
In a speech here at RSA, Schmidt argued that AI research "should be done in the open," lest the work encourage paranoia and suspicion between countries. "The research should be done for the benefit of the many, not the few, and should apply to every human," he said.
Diversity of though is also important, he stressed. "More people looking at it means a better outcome," he said. "Our approach [at Google] is to do this in the open and model this behavior," he said.
Schmidt did voice concern that AI could be weaponized. What if attackers tampered with the data used to train the algorithms that power a medical neural net designed to diagnose illness based on images? The AI would make incorrect and potentially dangerous decisions.
The hacker at home
During his speech, Schmidt was challenged by moderator Gideon Lewis-Kraus about what role the garage inventor could play in the development of modern AI. It's one thing to release these tools to the public, but only companies as large as Google can train such models.
Schmidt acknowledged Google's lead with data, but said there is plenty of room for contributions to be made to the field of AI development. A tinkerer, he said, could always make a more efficient algorithm that requires less data to be trained. "There are plenty of reservoirs of public data available to the garage tinkerers," said Schmidt.
A little history
Schmidt began his talk by making clear that he was previously skeptical of the value of AI. "When I was a Ph.D. student, AI was all the rage," said Schmidt. He resisted it because the field hit a major slump in the 1980s as experts moved to more promising areas of research -- an AI winter, Schmidt called it.
"I, being prejudiced from the years of winter, when I saw the results of [computer] vision and speech, I said 'oh, you know it won't scale,'" Schmidt told the crowd. "I have been proven completely wrong."
Indeed, Google has long employed neural networks at many levels, from algorithms that identify pictures in Google images, aided by millions of Google users, to the underlying mechanisms of Google's ad technology. Data centers, meanwhile, were designed by some of the smartest humans on the planet, but AI analysis identified improvements that increased power efficiency by 15 percent, Schmidt said.
Still, it hasn't all been mind-blowing. When neural networks discerned the existence of cats from YouTube videos, for example, "I was very upset," said Schmidt. "The event that launched the quest for general AI, you'd think they'd discover set theory. Instead, they discovered cats on YouTube."