From Discovery to Implementation

Interview with Dr. Kai-Fu Lee, Chairman & CEO, Sinovation Ventures – President, Sinovation Ventures AI Institute

Where do you think the next big AI technologies will be developed and in which fields?

I personally think that making all the known technologies really honed to fit actual applications will be the largest social contribution to humanity. If you look at the history of speech recognition or computer operating systems, or the notion of the Internet, there’s an age of discovery followed by an extended age of implementation. They were experimental in laboratory conditions and not yet applied in industry. Which are the early adopters? Which ones are the momentum drivers? Which are mainstream usages? Fixing and tweaking technologies so they can be applicable – that will be the most important thing.

From the moment Internet was discovered until it changed the world was a good 20 years. AI will maybe take a bit less time but it will take maybe 10, 20 years or more to penetrate every corner and every application.

Kai-Fu Lee

It’s kind of a self-limiting question to look for the next breakthrough. People didn’t ask: what is the next operating system breakthrough? There were some obviously. 3G, 4G and 5G made a big difference to penetration and fitting everything in the phone made a big difference to the OS – those are the kinds of things that will happen to AI. There will be so much money and so many people working on that problem.

The next possible big breakthrough could come from the combination of brain sciences and AI – capturing something about our logical and illogical thinking, or from quantum computing or hardware, or semantics or language understanding. Each of the three has a 5 percent to 20 percent chance of making a big difference.

In the next five years, probably Internet and financial and e-commerce are the biggest industries that will be affected, where there are immediate transactions of money. After that we will see an impact on areas such as retail, healthcare, manufacturing, education, transportation and automotive and logistics including warehousing/transportation/delivery. Eventually AI will penetrate everything.

You talk about the role data is playing in fueling AI development, but different countries have different regimes. What impact will that have on the way AI develops?

Eventually every country will have different data laws. China has strong data laws not with respect to individual privacy but with respect to companies that sell and use data without user consent, such as in the Facebook and Cambridge Analytica case which would have been punished by imprisonment. The EU has GDPR and the US is figuring something out. We are in uncharted waters in figuring out how to deal with individual data management and privacy and what is legal and what is not. I doubt there is one answer for every country, given the different cultures and user expectations in the East and West. Instead I think we are at the beginning of a crowdsourcing project. GDPR is one of the first most visible efforts. I don’t think it’s a very good design but I think we will tweak it. Maybe we will have three sets of data laws – EU, China and US – and there will be some commonalities but also differences.

What are the shortfalls of the GDPR at the moment?

The government is playing the role of product manager. It’s doing the brute force – let’s give every user the choice of every permutation for every website, so the responsibilities are shifted back to the users. These pop-up windows keep coming up, people get sick and tired and just click yes. A very tiny percentage of people will click ‘no’ a lot and miss out on the convenience while a large number of people will just get annoyed with the pop-up windows.  It's a pretty poor user experience.

Is there a risk of a race to the bottom, where countries compete to have the least effective data privacy?

More data collection involves the highest risk of damaging individual privacy, which will then damage corporations and that will force countries to have stricter laws. It’s like capital punishment: you could argue that countries with the strictest laws have the least crime, but there are side effects of that. I’m not an expert in policy and governance, so I can only point out the possible benefits and downsides.

Do you think attitudes to IP rights vary in different parts of the world?

Overall AI is not very heavily patented because the originators of the discovery of deep learning didn’t patent it. It’s unclear whether each permutation warrants a patent. There are some patents held by AI giants. Judging by how much the algorithms are proliferating, I would guess there are not many extremely strong patents that could be enforced at this time, such as the voicemail patents or some of the encryption security patents.

We encourage our portfolio companies to file patents but only when they have something valuable. Theoretically if a company has a paramount advantage in IP, that is absolutely something we would consider as part of the investment and business strategy process – but it’s all theoretical. We have never invested in a company purely based on the strength of a patent. If you have something defensible and get sued, you can use it to defend yourself. These are nice to have. In no case have patents been the number one consideration. Again, we look at founders' and companies' critical abilities to implement technologies into products and business, beyond inventing them.

Are there any other areas in the world that you think will emerge in AI in coming years?

In terms of research, I think in Canada there is extraordinary talent and there are several other fairly strong countries. Hong Kong and Singapore are decent. But in terms of implementation, none of these countries have an ecosystem to turn the expertise into economic advantage. You need a strong venture capital ecosystem to drive the technologies to the right application areas and relentlessly focus on user needs and use that to push the scientists to improve the technologies. Outside the U.S. and China there are no countries with such ecosystems. Israel to some extent has them but I don’t think many countries are aware of that. They are focused on technologies targeting their own industries and that is where they go wrong.

Will companies be constrained by their own markets then? Will a Chinese company be able to expand to the US, where the environment is very different?

Regulatory is part of it but there’s more. The entire ecosystem is different – users, language, expectations, how you build and advertise a product and acquire users. Even assuming no regulatory issues, it’s hard to succeed. The US by default exports its products to developed countries so those products are standardized. But China has a good opportunity to get into the developing countries and regions which have similar demographics – such as south-east Asia, the Middle East, Africa and probably India and potentially South America. I think Chinese AI and mobile technologies will make some inroads internationally but probably not in developed countries. These countries probably account for two-thirds of the population of the world but only a tiny percentage of GDP – so short-term not worth a lot but long-term a lot.  I am predicting that down the road, Chinese built technologies have a good chance to penetrate half of the world. 

What technologies will be key to the commercial development of AI and scaling them up?

The cloud enables a company to not have to buy a lot of machines but to rely on a cloud-based solution. That helps start-ups get going and the US has a substantial lead. On top of that I think there is something we can call AI platforms – allowing non-AI experts to do AI. Many companies aspire to do that but no one has a complete solution yet. What made Mac, Android and Windows accessible to many users? They made development toolkits available so you can make applications without knowing the deep kernels. Clearly Google is in the lead and Google Cloud and TensorFlow is likely to become the default. Facebook is doing Facebook PyTorch that lowers the barrier to development. Amazon, Tencent, Alibaba and Baidu are all doing similar things to let non-AI experts and regular engineers inject AI into their regular applications. Whoever cracks that will have potentially a Windows or an Android.

What about the social and cultural impacts of AI?

There are many risks. One is if AI takes away all the entry-level jobs, how do people get to the higher jobs? How do we promote people in the future? There are other risks too, such as AI security – hacking into a phone and everything it controls.

AI is also a bunch of numbers that are undecipherable multiplied together in ways inexplicable to humans – if someone hacked in and changed a thousand numbers, how would people know? What would it cause? These are all questions that will need new ways of security. There are dystopians who worry that if you give AI a goal it will pursue that. I don’t really buy that as a threat to humans, as in most scenarios and known issues, human controls are behind the technologies. Most of AI’s impact is yet undiscovered and we will discover and hopefully solve most of the impacts, and be troubled by the new ones that keep coming up.