This past week, as part of the Aspen Ideas Festival, I spoke with Kai-Fu Lee, the president and chair of Sinovation Ventures and a pioneer in artificial intelligence. We discussed his recent argument that AI has been of limited use in the response to the coronavirus crisis. And then we talked about the future of work and why he thinks that Covid-19 is going to accelerate trends toward automation. Because of the virus, and because of the way we all work now, we’re going to have many more robots and other machines in our factories, restaurants, and kitchens. A lightly edited transcript is below. You can watch the original video here.
Nicholas Thompson: You're a pioneer in artificial intelligence. You wrote my favorite book on artificial intelligence. You've taught us all a lot about artificial intelligence. And now we have perhaps the greatest crisis of our lifetime and you've given AI a B-minus in helping to resolve it! Why is that? Why such a low grade?
Kai-Fu Lee: Well, B-minus is a lot better than passing. It's not ideal. The reason is, AI works by accumulating a lot of data and seeing recurrence of similar events in order to make accurate predictions. And a pandemic is a once-a-century activity. There isn't a lot of experience building models and there isn't a lot of data. Despite that, there are many places where AI has added value. So therefore the B-minus.
Can you walk us through the places where artificial intelligence has been helpful in combating the coronavirus and the places where it hasn't done that much?
Two personal examples: I live in China, and contact tracing is working quite well. I get a red, yellow, or green signal on my phone telling me whether I may have been in contact with someone who has the virus and therefore need to do a checkup. That is a way of informing people about their status. The second thing is I was in quarantine when I returned to Beijing, and all the things I ordered by ecommerce—including takeout food—were delivered by robot. So I was really in zero contact with people because robotics are now working well enough within structured environments like apartment buildings, hospitals, stores, and office buildings.
Also, AI has made some contributions in helping drug discovery, discovering new antivirus vaccines, and AI has been used in warehouses to handle the massive number of packages that are sent by ecommerce. There are also prediction engines. Early on, there was a company called BlueDot that actually predicted that there may be a serious pandemic coming. Of course, it wasn't released to the world. Now that we've seen the damage of a pandemic, I think if there is a next pandemic or a second wave, the predictions will be much more accurate.
One area where people may have expected AI to be very useful where it hasn't been is in vaccine development. There are a few examples. For example, Moderna, when they were trying to figure out which protein to build for a potential vaccine, did use AI to model the options. But we did not have what many people hoped we'd have, which is the ability to simulate a human being and to identify potential vaccines. How far away are we from that?
I think that's impossible to predict right now. We're in a very urgent situation, so all the people working on vaccine development are using the low-hanging fruits; they are not really going all-out for that. One company we invested in called Insilico Medicine used generative chemistry aided by AI to predict the types of compounds that may block the spread of the virus. So there are similar techniques to that, basically retargeting existing AI technologies in places that are low-hanging fruits. I think what you're talking about, a brand new way of doing vaccine development, hopefully the pandemic will give us a running start and we'll make progress towards that but it would be, I think, too aggressive to predict that there will be a huge difference made by AI this time.
How high is the fruit? In the next pandemic, will we be able to very quickly develop a vaccine because of AI? Will it take 10 years until we're there?
Well I'm going to be an optimist and say yes, but I really don't have the solid grounding to prove that.
Let's talk about the industries that are going to change. The coronavirus is going to turn our economy upside-down. It's changing all kinds of industries. It's massively changing my industry, media. It's changing online education in crazy ways. It's changing telemedicine in fascinating ways. What are the industries that you think will change and will be most affected by artificial intelligence?
Clearly health care is one. AI has so much to offer in terms of personalized, targeted diagnosis, more accurate due to genome sequencing, new technologies like Crispr coming out, potentially combined with AI; also, there are a lot of inefficiencies in health care. Insurance was not designed with all the health care data. So I think all of these will compound and make health care plus AI the biggest potential. There is one issue with health care, which is whether the data can become accessible. In countries where there are strong protections, such as HIPAA, even with anonymized data, it may be hard to aggregate the data for training AI. And AI really runs on data.
Another is education, that you talked about. People are changing their habits about going to school. A billion kids across the world are learning online. And suddenly we see all these ways of using online AI technologies, whether it's AI teachers, AI to help you fix your pronunciation, AI to figure out what areas you're having trouble, in math or English, that can all be added to make the human-to-human interaction more about learning the methods, helping to motivate learning, individualize, but using AI for the routine part of education.
Lastly, I think work as a general category is shifting online. We're conducting all of our meetings, we're making investments potentially without ever meeting the entrepreneur. People are making deals online. This change of habit, of being willing to have meetings and make decisions, and helping to change the work process into a digitized process, this digitalization turns everything into data. Once you have data, you have AI potentially coming in to improve the margins, improve the efficiency. A huge potential challenge is you have AI potentially coming in to say, well, everything's digitized, why don't we use AI to do this workload instead of people? So that will accelerate automation and potentially cause a faster churn in terms of AI replacing people.
This is fascinating. Right now you and I are talking across screens. You're on the other side of the world, I'm in an attic in upstate New York. We should be in Aspen talking to each other, but here we are. So the fact that this conversation is all digital, and all the other work conversations I've had today and will have tomorrow are all digital, you think will lead to some kind of unknown advance in AI that will make human work disappear more quickly?
Yes. To give you an example, in the pre-pandemic world there are lots of companies that require people-to-people interaction. So people go to work, they have meetings, people take notes, they write on paper, they have records, and they call each other. But now that all the work is essentially operated and run online, everything from meetings to decisions to workflow becomes digital. And once it's digital, the company's management will see, oh, there's that part of expense report decision-making that could be done by AI. There's that part of customer service, we could simply have an AI agent rather than a human agent. Oh, the sales process, all this telesales could be done by AI with either automated speech generation or even synthesis of digital humans. So pieces of corporations and their workflow will become automated faster because it's already online and digital.
In your book, AI Superpowers—and let me just pause to say it's superb—you talk about the future of work, and you have two charts. One is about job and workforce, and you have a y-axis which is human interaction and an x-axis which is creativity. So a job that requires a lot of creativity and a lot of human interaction is not likely to be replaced by machines, like a CEO job. And a job that has not a lot of creativity and not a lot of human interaction, like a telemarketer, will be replaced by machines. How does your chart change in the post-coronavirus era?
I think the chart basically remains the same. The replacement curve, if you will, taking over the jobs, that's going to go faster. I also think the pandemic will potentially cause certain types of jobs to be more accelerated in their replacement. Because we would think health care, hospital jobs, we would want the human touch, empathy, etc. But human touch in the period of the pandemic increases the likelihood of spreading the virus. So if there were a smart robot that could move medical supplies and help the patients with testing their blood pressure, etc., we as humanity would be more willing to embrace that. Similarly, waiters and waitresses in restaurants. We would think there's value in the human interaction. However, those are dangerous jobs. And certainly at the fast food and lower-cost restaurants, more will be replaced by automation and AI. And actually in China, in many of the lower-cost restaurants, you see people taking orders on mobile phones. And the waiters or waitresses are just delivering the food and you'll also pay on your phones. So that already reduces the cost. And there are also many restaurants that have AI robots delivering the food because the restaurant, like the apartment building and the hospital, are structured environments. Robots’ work moving around in a structured environment is much easier. These are not robots that have feet and hands. Think of them as just carts that basically move to your table, make some sound to let you know to take your own order. So those kinds of jobs that we would think require some human interaction are potentially going to turn into automated jobs faster because of the pandemic.
Do you think using similar logic that we're soon going to have self-driving taxis or that the move toward self-driving vehicles will be accelerated because we don't want to be near human drivers?
Not as much, because in order to get to L5, which is the highest of the five levels of autonomous vehicles, there are still a lot of technical challenges to be overcome. So I think the taxis and Uber will continue to be needed. We will see AI taking over, automating human jobs in the warehouse, some of the manufacturing, driving on highways and buses, probably in that order. Taxis and Uber will be the last to be fully automated.
I just want to make sure I fully understand your argument, which is, we've all known that AI and robotics would be replacing human jobs. But there are two factors here that are accelerating it. One, we don't want as much human interaction, we don't want to be near the waiter. And two, so much more of our life is digitized and AI runs off data. Is that correct? Is there a third factor, or just those two?
No, I think those are the two main factors from the pandemic.
Let me ask you about something you mentioned a few minutes before. You were talking about health care and data and HIPAA, and you mentioned that countries that have strong health care regulations, like the United States, won't have as much data and therefore may not have as many advances in artificial intelligence. My view is that we're going to care a lot less about privacy in the future than we do now. So perhaps countries like the United States will start loosening up access to data, for better or for worse. Is that fair?
I think that would be a plausible outcome because I think people are starting to realize privacy is not a binary issue. It's also not an issue that trumps everything else. It needs to be considered in the context of public health, greater social good, and personal security. So while we want everyone to have their privacy safe from companies, as much as possible, when it provides a solution to the public health or greater security for each individual and perhaps some incredible convenience for people, then we should really consider it in the context of how much benefit it is providing and provide each person with some degree of choice. Because there will always be people who feel privacy is the most important. So to the extent that each country develops a [balanced] set of regulations, then the appropriate amount of data collection, anonymized data, can be aggregated and AI can be trained.
China is quite a few months ahead of the United States in its response to the coronavirus. The countries are clearly on different paths. Tell us what we have to expect for when we eventually get on the China trajectory. What do we have to expect in terms of AI and technology that will come along the way?
There are many things that are fundamentally cultural and very difficult. I think China has gone through SARS, as have many Asian countries. And also, culturally speaking, people are more willing to be disciplined for their own safety and for the safety of society. That's going to be hard to change until a country has experienced some of the challenges. But some of the possibly doable things include, I think, certainly contact tracing based on an understanding that privacy is important but public health is equally or maybe even more important. Second, when there is a significant spread, actually wearing masks is important, but it's really most important when everyone wears it. So when it's purely voluntary it's not going to be as effective. So when the pandemic was in a serious situation in China and other Asian countries, you'd go on the street and everyone is wearing a mask, because the purpose of a mask is mostly not to protect you, but to protect other people in case you have it. Therefore everyone has to wear it. These, I think, are the two biggest things, in case there is a continuing challenge or a second wave. If people would do that and get it close to zero, then actually the new normal as we see in China today is not that far from normal.
Actually, of the things that you've described in this conversation, I'm fine with wearing a mask, I'm fine with contact tracing, but I do really look forward to the robot delivering my dinner.
Well that actually works incredibly well. It's a little bit related to how people live. Most people like myself in China live in urban condos or apartment buildings. That makes the robot delivery much easier. You've got the delivery from the restaurant to the apartment building, then the apartment building manager handles delivering the last mile from the first floor to your apartment. Because in the US most people live in houses, it's a little more complex, because then the robot has to figure out how to deal with the last mile.
Thank you Kai-Fu. It’s always a pleasure to talk with you.
- We can protect the economy from pandemics. Why didn't we?
- Vaccine makers turn to microchip tech to beat glass shortages
- 15 face masks we actually like to wear
- It’s ridiculous to treat schools like Covid hot zones
- After the virus: How we'll learn, age, move, listen, and create
- Read all of our coronavirus coverage here
