The next big thing in tech is going to be AI. That’s what a lot of people in Silicon Valley has been saying lately. Add one more to the list: former Google CEO Eric Schmidt. “To me the biggest platforms will be the ones that will be driven by artificial intelligence,” he told an audience gathered at Bell Labs, in Murray Hill, N.J., last month.
Schmidt, now executive chairman of Google’s parent company Alphabet, was a speaker at a conference celebrating the life and work of Claude Shannon, the legendary Bell Labs engineer whose ideas revolutionized modern communications. In a wide-ranging talk on innovation, Schmidt, who as a summer student at Bell Labs in the 1970s was part of the group that worked on Unix and C, discussed the importance of taking positions “contrary to the common orthodoxy,” deplored excessive government regulation, and offered product ideas that anyone could use to “create a startup right now” (he suggested attendees skip the coffee break).
But the highlight of the talk were his bold predictions for AI. Schmidt said new machine learning tools will help businesses and individuals solve a wide range of problems by finding patterns in vast amounts of data. “One of the problems we have in life is that most of the problems that we face are complex emergent systems,” he explained, adding that the latest crop of neural networks and other AI techniques offer the best way of tackling such systems.
“But our ultimate goal is actually AI-assisted science,” he said, before listing a number of areas that he believes artificial intelligence will help advance: “disease diagnosis, climate modeling, drug discovery, macroeconomics, particle physics, material sciences, theorem proving, and protein folding.”
THREE RULES ABOUT AI
Schmidt noted that many people are wary of artificial intelligence, and because of that he’s come up with three rules about AI:
“The first is that AI should benefit the many, not the few. This should not be an elite technology. It should not only be available to rich countries, rich individuals. It should be available to everybody.
The second is that the research in this area, which is heating up—this is the hottest area in computer science right now—should be done in the open. I’m not saying open source. I’m saying open as opposed to [done] in military labs. I’m very concerned that some government will say, ‘classify this stuff,’ and put it in some super secret lab, not because it won’t be useful for military purposes, which is a separate discussion, but because we all benefit from the development of these algorithms and the impact that they’re going to have.
And the third is that, systems that people are building need to be under positive control in the sense that we need to know what they’re doing, we need to sort of understand it for obvious control reasons.”
He described how Alphabet is already using AI to power services like Google Translate and Google Photos (“If you type the word ‘hugs’ it will show photos of people hugging each other. It’s extraordinary”). He also mentioned a product—it wasn’t clear which Alphabet company is developing it—that uses machine learning to diagnose an eye disease called diabetic retinopathy with 99 percent accuracy while human doctors achieve 90 percent. “How did we do that? We see more eyes. We train better,” he said. “We saw a million eyes, where the average ophthalmologist trains against 10,000 eyes.”
Another project Schmidt discussed was AlphaGo, the AI system built by Google DeepMind that defeated Go master Lee Sedol earlier this year. “Why does this matter? It’s only a game,” he said. “Because this is the beginning of intuition.” According to Schmidt, one area where intuition plays an important role is science. “Science is about ‘I have an idea, I’m going to try this, and how do I know? I just know. I spent years thinking about this,’ ” he explained. “And there’s evidence that we can now apply this to many, many problems, but definitely science.”
A problem like drug discovery, for example, requires “a high degree of intuition,” Schmidt said. It’s a costly and time-consuming process in which chemists are continually mixing and testing compounds. “So a little bit of knowledge would allow us to automate much of that,” he said. “Now, this doesn’t replace the chemists. It just gives them better judgment.”
For Schmidt, AI will help accelerate research, improve products and services, and free people from dangerous jobs. He’s convinced it will even save lives, as self-driving cars from Google and other companies become more popular.
ERIC AND NOT ERIC
Schmidt also discussed more mundane AI applications. He said he wants an AI assistant capable of helping him with daily tasks. He dubbed it Not Eric. “There’s Eric and Not Eric. And Eric is me, and Not Eric is this assistant, which is personal and I control. It just makes me smarter.”
What exactly would an AI assistant like that do for people? “At a minimum, reply to our email,” he said as the audience nodded approvingly. “But perhaps do much, much more.” In particular, he wants his AI to “make me cool in front of my daughters” by telling him what to say and how to act. “That’s a really useful function. If that problem is solved, we’ll declare success.”
During the Q&A, one attendee asked Schmidt about the Singularity, which many people see as a threat to humanity. He argued that such fears are exaggerated and, in his view, based on a “misreading of where we are in science.” Despite today’s huge advances in AI, he said, we’re still far from a scenario in which robots with volition decide to destroy the human race.
“Everybody here has gone to too many movies.”