This month, Scott Beaumont, head of Google in Greater China, spoke at Peking University, his second such talk at a Chinese university in half a year.
On both occasions, Beaumont spoke not about search—which, of course, still isn’t available in China—but on AlphaGo and Google’s explorations in AI.
We sat down with Beaumont at Google’s Beijing office to talk with him about just where Google is headed, and what it plans for China.
In May, China’s Go champion, Ke Jie, squared off against Google’s AlphaGo, in what was likely the biggest event for Google in China since its ouster from the country in 2010. But even before the match was held, AlphaGo was making waves, thanks to its prior victory in 2016 against Lee Sedol.
Many Chinese entrepreneurs working in AI have noted that it was with the arrival of AlphaGo on the scene that investors first began to understand what they were going on about. Even though they may have been toiling away for years, it was AlphaGo that sparked public and investor interest.
Many see AlphaGo’s matches against humans as adversarial, but looking back at its research and development and its eventual victory, Beaumont believes that it actually marks a kind of cooperation between humans and machines.
In the course of AlphaGo’s matches, its engineers developed a finer-grained working model for deep learning algorithms, and made breakthroughs in hardware computing power. And AlphaGo’s uncanny playing style proved to be an entirely new approach in a millennia-old game.
In 2016, when AlphaGo faced off against Lee Sedol, it used 50 of Google’s TPUs, examining 50 steps every second and 100,000 board configurations. But this year, as AlphaGo met Ke Jie, it needed only one TPU.
But as impressive as all of that is, the key point here is that in this process AI has found its way into the public consciousness, such that even people on the street are now wondering how it will affect their lives, even as investors contemplate its potential, entrepreneurs consider its applications, and governments study how it might solve problems that have been unsolvable before now.
It has even affected Google’s old China rival. Baidu earlier this year declared that it was no longer a search company, but an AI company.
Yet the reason that so many companies, academic institutions, and developers have poured into AI is not just AlphaGo.
If it’s to be traced back to its source, the new wave of AI didn’t begin with AlphaGo, but in 2009, when Google sought to use the latest successes in deep learning research for its translation service.
In the years that followed, deep learning found its way into almost all of Google’s other products. But even so, that wasn’t enough to lead ordinary people to feel like AI had arrived.
What really set off the AI boom was Google putting AI into the hands of independent developers.
In his talk, Beaumont noted several use cases of AI that have already been realized, in everything from monitoring manatee activity via satellite imagery to medical diagnosis of specific ailments. One project employed 54 doctors and 128,175 annotated images of suspected cases in combination with a 26 layer convolutional neural network. Eventually, the network was able to achieve a diagnosis accuracy rate even greater than that of the doctors themselves.
This neural network is now helping to diagnose nascent eye problems in diabetics earlier, faster, and more conveniently, allowing for better preventitive care that can stave off blindness.
And then there’s the cucumber sorting machine. A Japanese student named Makoto worked with nothing more than TensorFlow and a Raspberry Pi in his spare time to develop a cucumber sorting machine for his parents to use on their farm. The machine he created can assess length, shape, color, texture, and other features to sort cucumbers into nine different grades.
During the harvest season, Makoto’s parents would ordinarily spend hours a day sorting their produce. Now, they leave it to the machine. It’s the most niche of niche applications, and something that, as Beaumont says, Google would never have thought of doing, yet it has meant a clear and dramatic difference for this one family.
These cases might seem to have little if anything to do with Google’s core business, and medical, farming, and environmental protection applications are all a long way from Google Translate. But once Google open sourced TensorFlow at the end of 2015, people began putting it to use in ways that even Google could never have imagined.
In short order, other tech companies followed suit. Facebook, Microsoft, and Baidu all open sourced their own AI frameworks.
Now, when you use a news recommendation engine or an AR app, or listen to background music on a Chinese internet TV show, or view product suggestions on Tmall or JD.com, or even get certain kinds of medical tests, you are reaping the benefits of AI tools that were only just created a few short years ago.
AI still has the feel of something science fictional about it, but in a sense, it’s already all around us, especially in China. And that has everything to do with Google’s future in the country.
Last December, Beaumont observed that, despite popular perception, Google hasn’t left China. And if there is any doubt about the company’s commitment to staying in China, just consider:
Although Beaumont would not say exactly how many people Google currently employs in its dual Beijing and Shanghai offices, he did note that at least two thirds of them are still engineers, actively working in development.
These Google programmers are responsible for, on one hand, localizing Google’s AI and mobile products for China, and on the other, developing products that Google will roll out worldwide. Thus, Google has unexpectedly become a gateway for sharing Chinese culture with the world.
But ultimately, Google still has its sights set on AI, envisioning China not merely as a market, but as a testing ground and development site.
According to Google’s official data, on the popular PyPi repository for Python, TensorFlow has been downloaded more than 900,000 times, with 15% of that demand coming from China.
“Some of the key characteristics of what makes a good home for exploration in AI: you need first-class computer science; you need access to data; you need the ability to prototype quickly. And clearly those three boxes China ticks very strongly,” Beaumont said.
Even more important may be that, as Beaumont observed, China’s government seems more positive about AI than governments in the West, offering “a favorable framework within which to operate.” It is more willing to use AI for addressing problems that previously required so much investment of human attention and effort, such as environmental protection, enhanced medical care, energy conservation, to say nothing of some as yet undiscovered problems.
All of that means that Google has reason to stick around, and leverage the advantages of China’s local industry for its global product development efforts.
So maybe don’t bother asking if or when Google search will return to China. Because after all, Google is no longer a search company.