China Has Caught Up To U.S. In AI, Says AI Expert Kai-Fu Lee

Any credible list of influential books about tech from the last decade would include AI Superpowers: China, Silicon Valley and the New World Order by Kai-Fu Lee. Considered the world’s foremost authority on artificial intelligence, Taipei-born Lee got an early start, writing a pioneering speech-recognition program while a student at Carnegie Mellon in the 1980s. He later had a career in China and the U.S. at Apple, Microsoft, Silicon Graphics and Google, where he was president of Google China. Now based in Beijing, Lee runs a venture capital firm called Sinovation, which focuses on AI investments. The interview with Lee took place (virtually) in early October.

Forbes Asia: AI Superpowers made you a global business star. Why did you write the book? 

Kai-Fu Lee: China has—thanks to data, AI, and the entrepreneur ecosystem—rapidly evolved from a copycat into a true innovator. It currently co-leads artificial intelligence with the United States. When AI Superpowers came out in 2018, I think it was a bit surprising to people.

What has surprised you most in the three years since you wrote AI Superpowers?   

Probably, the big surprise is that AI innovation is even faster than I thought. By that I mean the speed at which AI innovation goes from research to products. Everyone knew that was going faster, but it is even faster now. As an example, computer vision algorithms and convolutional neural networks took 30 years from the first research paper to eventually becoming a pervasively used technology. Contrast that with the more recent natural language breakthroughs. Google published an academic paper on the subject, and in just two years just about every commercial company has it in their products. This makes us even more confident that the quick adoption of AI into all kinds of domains and countries will reach commercial applications much faster than just about any other technology we can imagine.

What are China’s natural advantages in AI? 

Let’s use TikTok as an example. It became a runaway success and has proven to be uncopiable by top American companies. TikTok is a great example of China’s natural advantage. The company has leveraged huge amounts of data in China to develop an interface that shows you videos that become attractive and even addictive for you to use. And then, gathering all that data, TikTok uses it to bootstrap in the U.S. and other countries, and create similar experiences for different audiences. AI lets TikTok deliver a targeted experience with each individual and thus gathers data for constant iteration. TikTok shows that if you have a good product, iterate the product, use the AI to get users more interested, you can grow more geographies. It has become a global phenomenon. That is a typical Chinese story. Arguably, there are no breakthrough AI technologies in TikTok. It is excellent execution, lots of data, iteration and aggressive growth. I think that is the formula that led to the Chinese consumer AI success.

You’ve long specialized in investing in AI. What is your investment philosophy?

Think of three steps. First is knowing the importance of data. Second is knowing, if you possess data, you have a great opportunity to apply AI and generate a lot of value. Thirdly, observe how AI technologies are penetrating quickly. My conclusion from these three points is that the possessor of some AI technology is not nearly as valuable as the possessor of data. That [insight] has given Sinovation an idea for a new fund. Instead of seeking rocket science AI companies, we seek companies that have data. Data that is well structured, large in quantity, connects to a business metric and ideally is proprietary. Our target companies do not have AI yet, but they have a large enough revenue stream. They are not startup companies anymore, but they are not so big. We seek companies with values of half billion or a billion dollars and we go in and help them realize the value of their data that, if applied to AI, would very quickly give them an additional upside in terms of their bottom line and market cap. 

Are you a consultant or investor?

A bit of both. The companies we seek do not pay very much for consulting projects. We want the upside when the company valuation goes from half a billion to $5 billion. 

In which industries?

We helped a financial data firm increase usage by 300%. This was very low-hanging fruit. We are also investing in some unusual hardware companies, companies that have strong mechanical capabilities and products, but do not have computer vision. We invested in, for example, a forklift company that makes great forklifts, but they do not have autonomy. Now, we cannot build autonomy, because we’re not vehicle engineers, but we can consult them to help them either acquire the right company, or partner with a university. So, we’ve discovered this “AI value-add” is quite valuable and we can apply it in ways that investors usually do. So, when they make money, we make money. Our first few deals have been very encouraging, very exciting.

Is your compensation all equity?

We are directly compensated only at “cost plus.” But we are not trying to make money with our consultation. We make money when their market cap goes up and that goes up usually because, in the case of a lender, we reduced the default rate about 14%. Now they can operate with fewer loan officers to save further money and to the extent we improve their bottom line by 20%, their market cap will go up. 

Much has been written about ethics in AI. What are your thoughts?

An engineer who is unaware or does not care about other people could make some very dangerous products. A lot of this has to do with the lack of education in the engineers in not knowing that they have a responsibility to make sure the training datasets are plentiful and well-balanced among gender, racial, etc. A good solution to address this problem would be education embedded at an early level so that the engineers are aware that they are not just writing some code and getting some numbers out. We have to improve, or we will have too much regulation. I am optimistic we will address these.

Can AI help transform education?

Yes. China actually started with online education about five years ago, even longer. We funded the company called VIPKid, which has about a million Chinese kids learning English. China has too few teachers, so online education became a solution. I think Chinese AI education is really blossoming and the data is growing, which leads to highly targeted education. We now see AI rendering virtual teachers as cartoon characters like dancing bears, to teach, quiz and engage them. There are measurable results that show that the cartoon characters teaching math produces kids who not only score higher but are more engaged and ask more questions.