MSFT Microsoft Corporation

By Phred Dvorak 

One of the hottest topics in the tech world these days is the race to develop and deploy artificial intelligence. U.S. companies like Alphabet Inc. have made waves with machines that beat humans at complex games of strategy. China has vowed to become an AI leader by 2030 and has made huge strides in areas like facial recognition.

Key to much of AI development is data. That's because AI has made big leaps in recent years in machine learning, where computers learn by picking out patterns in masses of data, and deep learning, a technique in machine learning that is inspired by the activity of layers of neurons in the brain.

In many of these areas of AI, the more data that is fed into the computers, the better the results. And China, experts say, has a lot of data -- from video captured by surveillance cameras to information about shoppers' buying habits. Taking the lead in AI can confer advantages in everything from the businesses that power economic growth to the military.

So, who's ahead now, and how likely are they to stay there?

The Wall Street Journal asked two experts: Oren Etzioni, chief executive of the Allen Institute for Artificial Intelligence in Seattle, and Tsuhan Chen, chief scientist for AI Singapore, a national program to foster artificial-intelligence research, and a deputy president at the National University of Singapore. The interviews were conducted separately and edited together.

The race to AI

WSJ: Who leads in AI now?

DR. ETZIONI: Virtually everyone agrees that the U.S. is pre-eminent. In the commercial sector, the U.S. still clearly dominates -- with Google and Microsoft and Amazon and others -- although the distance has been shrinking. In the academic realm, we still lead in measures like number of papers, number of citations, number of key leaders.

DR. CHEN: What everyone talks about as AI these days is actually a branch of AI which we refer to as machine learning, or deep learning. Lately, all the advances in computing -- big data, faster computers -- make deep learning better and better.

Already, China is dominating in machine learning. They are leveraging the availability of data. There are definitely more people to crunch data, and there are more people providing data.

WSJ: What are the U.S.'s strengths in AI?

DR. ETZIONI: We have this history and tradition of openness. The U.S. is still an incredibly attractive place for a lot of people. A huge part of developing AI is having the smartest people, the most innovative, creative, successful.

If we succeed in keeping the United States and our university system and our companies as a place where the best and the brightest from all over the world want to come, then I think we're going to substantially win the talent wars.

DR. CHEN: Being allowed to be creative is what the American system offers.

WSJ: What about China?

DR. CHEN: The China advantage is about implementation. You don't really need too many innovative ideas, you just need lots and lots of computation and lots of data to get what you want.

China has a more structured government and the population all complies. That makes AI deployment a lot easier.

If you talk about using surveillance cameras to track people, in the U.S. that would be a lot more challenging to deploy. In China, that would be easier.

China is more effective in deploying these systems because of the availability of data. In the U.S., there are all these regulations or policies guarding people in terms of privacy and security. In China, they have data that's not protected by these privacy or security concerns.

DR. ETZIONI: Some key variables in AI development are data, where the Chinese have an advantage because they're less focused on privacy, and they have more data.

A second dimension is government support. The Chinese have made a point to have tremendous investment in AI at all levels. The U.S. is also making substantial investment through Darpa [Defense Advanced Research Projects Agency, the arm of the U.S. Department of Defense that oversees research in emerging technologies], through our venture-capital ecosystem, through the investments that Google, Microsoft and others are making.

The weak spots

WSJ: Are there limits to China's data advantage?

DR. ETZIONI: When it comes to building the next generation of AI systems, there are many data sets of different kinds. And, in many of these, China does not have an advantage.

If we want to build better autonomous vehicles, we need to train AI with pictures from the road and labels on them. I believe the U.S. is the leader in such data sets.

In natural-language processing, there are longstanding advantages in terms of English data sets that have been created and curated over many years. I think we're at least equal and probably ahead there.

Where China has more of the lead is, for example, medical data. Here we have so many strictures around releasing that data. And that's less of an issue in China.

You really have to go case by case and consider what's going on there. Data is the fuel that drives this AI revolution. But it's not like there's one kind of fuel.

Another question is: What phase in the development of AI are we in? A tremendous amount of innovation needs to happen in order to get from where AI is today -- which is a promising technology with increasing applications -- to where AI can be in the future, where the techniques can be far easier to deploy, can use far less data and can be far more resilient to errors than they are today.

There's still huge numbers of fundamental research problems, which is why I emphasize our university system and that whole ecosystem as a U.S. advantage.

DR. CHEN: Machine learning benefits from more and more data. But there are also subsets of AI that will continue to require innovation.

One is "explainable AI." This thing called deep learning, that benefits from more and more data, honestly we don't know why it works. We need to know why it works, so when it fails we'll know how to fix it. This cannot be addressed by just more data.

Another is what we refer to as "sense-making." The ultimate aim of AI is to have computers that will make sense of everything and do what's appropriate. That's the next level of AI, which machine learning is not able to do yet.

This AI-powered machine that we've built -- it can drive our cars, make investment decisions for us. But it really doesn't make sense of things. It just picks up patterns in the data.

Take autonomous driving: How is it able to avoid people? Because it's seen data before that says if you see people, you should steer away. However, it's not able to infer that a person on a bicycle is something you should also avoid.

Deep learning is recognizing patterns in data. Just because you are able to recognize patterns in Chinese consumer data, you won't necessarily be able to apply that to consumers in the U.S.

If 10 years from now we have machines that can make sense of things, then whatever works well in China may also do well in the U.S. and the rest of the world. Right now it doesn't.

China has a lot of tech companies coming to Singapore now for exactly this reason: If you train your machines and computers using data only in China, then it works only in China.

But in Singapore, everything is multiculture, multilanguage, multi this, multi that. So, Singapore's advantage is that very variety of data.

If you can make your facial-recognition system work in Singapore, if you can make your speech recognition work in Singapore, then it will work anywhere else in the world.

In China, facial recognition is pervasive. Yet these facial-recognition systems that work well because of the data there, actually do not work well anywhere else in the world.

Ms. Dvorak is a senior Asia correspondent at The Wall Street Journal. She can be reached at phred.dvorak@wsj.com.

(END) Dow Jones Newswires

November 12, 2018 11:30 ET (16:30 GMT)

Copyright (c) 2018 Dow Jones & Company, Inc.