“Artificial Intelligence is very powerful, but also very limited. With good data, AI can do far better than people. But it will never have the capability to think, to be self-aware or to have the creativity humans have.”
Dr. Kai-Fu Lee is the Chairman & Chief Executive Officer of Sinovation Ventures, a leading venture capital firm. Before this, Dr. Lee held various leadership roles at Microsoft, SGI, Apple and most recently Google, where he was the president of their China business.
Artificial Intelligence has tremendous power to help humans do their jobs better. Yes, some “routine” jobs can and may be replaced by AI, but most will be supplemented by its presence. AI can handle System 1 tasks (repetitive, routine), whereas humans beings will handle System 2 tasks (thinking, analysis).
The next big opportunity for AI is in health care. Here in the United States, it can synthesize health records and analyze bodily function far more accurately than physicians. In third world countries, AI can bring doctors to places that may never have had them before.
LISTEN AND SUBSCRIBE
SPEAKER
Kai-Fu Lee
Chairman & CEO
Sinovation Ventures
MODERATOR
Anthony Scaramucci
Founder & Managing Partner
SkyBridge
EPISODE TRANSCRIPT
John Darsie (00:08):
Welcome back to SALT Talks. My name is John Darsie. I'm the managing director of SALT, which is a global thought leadership forum that encompasses finance, technology, and geopolitics. With the SALT Talks, we try to do what we do at our in person SALT conferences, which is to provide a platform for big exciting ideas, as well as to provide our audience a window into the minds of subject matter experts. Today, we are pleased to welcome Dr. Kai-Fu Lee to SALT Talks. Dr. Lee spoke at SALT 2019 about artificial intelligence and machine learning.
John Darsie (00:47):
It was a fascinating talk that was led actually by Anthony's son, AJ. Kai Fu is the chairman and chief executive officer of Sinovation Ventures, which is a leading venture capital firm with about two billion in AUM, and it focuses on the development of the next generation of China's high-tech companies. Prior to founding Sinovation in 2009, Kai-Fu was the president of Google China, as well as a senior executive at Microsoft, SGI, and Apple.
John Darsie (01:18):
He's the co-chair of the Artificial Intelligence Council for the World Economic Forum and is the author of a New York Times bestselling book, which I highly recommend you read. It's called AI Superpowers. It was published in the fall of 2008, and Dr. Lee tells us that his next book is coming out in about a year and a half. And we look forward to reading that one as well. Kai Fu will be interviewed today by Anthony Scaramucci, the founder and managing founder of SkyBridge Capital, as well as the chairman of SALT.
John Darsie (01:46):
And I'll turn it over to Anthony and Kai-Fu now to begin the interview.
Anthony Scaramucci (01:51):
John, thank you very much. I think Kai-Fu's book was 2018, the fall of 2018, not 2008. So it was a way more current book, but it was a fascinating discussion about artificial intelligence. And I want to start with your origin, Dr. Lee, if you don't mind, because I think it's always fascinating for people. You mentioned at the SALT Conference that you sort of got interested in AI in your sophomore year in college. And so just take us through the steps of what you were thinking about then and how it led to where you are now.
Dr. Kai-Fu Lee (02:26):
Sure. Anthony, thank you, and John, for giving me the chance to talk to this great audience. Yeah, I got fascinated back in about 1980, my sophomore year, while I went to Columbia. And I was given an introduction to artificial intelligence, and I thought this would be the last technology for humanity. That we will figure out our brain and then we would build these amazing robots and life would be wonderful. Didn't quite work out that way. I think there are people who have those either dystopian or utopian beliefs about AI.
Dr. Kai-Fu Lee (03:06):
But what really happened in the last five years is all these AI-based machine learning technology started to work, and they started doing amazing job one task at a time. But these are amazing tools for us to use, but we are actually nowhere close to building that artificial general intelligence, something that's equal to our brain.
Anthony Scaramucci (03:35):
Obviously I've sat it on things at Singularity University and listened to Elon Musk. We hear about artificial intelligence, and you and I are old enough to remember the computer from 2001 Space Odyssey. What is realistic for us? What is realistic for people living on the planet today in terms of where artificial intelligence can go? And what will our grandchildren and our great-grandchildren see from the world of artificial intelligence in terms of the exponential development?
Dr. Kai-Fu Lee (04:11):
Right. AI is actually very powerful, but also very limited. It's powerful in the sense that if you have one single domain in which you have a lot of data and you throw that data at AI and tell it to learn something, something objective, something meaningful, and it will do that task better than people. The first wave of such applications was the internet. That's why Google, Facebook, Amazon are so good at targeting us individually based on what we've done in our history about things that we might want to read or we want to buy.
Dr. Kai-Fu Lee (04:49):
Then came the financial institutions. So automated loans and investment, quantitative investment, and insurance, they're coming up next. And then data in businesses. We're using Zoom now. We're creating data. That data can be mined. Smart things can be built from that for education, work, business, and pretty much any tradition industry. This is all happening right now. This is not for our grandchildren. This is happening now. Then AI will start to see and hear, but not just with eyes and ears.
Dr. Kai-Fu Lee (05:30):
AI can already recognize objects at higher accuracy than people and understand and recognize speech at higher accuracy than people. But what's more is now there are all these new sensors being plugged to AI, so AI can make decisions by aggregating all these sensors. The sensors can, for example, see things that people cannot see. They can sense temperature, humidity. They can do 3D reconstruction. So pretty much AI perception will definitely outdo human perception. Then finally, AI will move.
Dr. Kai-Fu Lee (06:08):
Of course, we already have our beloved Roomba, but AI will do much better than that. Beyond that, AI will be in the factory, warehouse, buses, highway, and then eventually cars, autonomous vehicles, so that it can move with a similar capability as people. You have these four waves adding up together building a lot of very valuable tools and applications to help us. But none of it has the capability of thinking, self-awareness, true understanding, compassion, creativity. All of that is missing.
Dr. Kai-Fu Lee (06:49):
It is simply taking one task, lots of data, with human telling it what to optimize, and then optimizing it so well that for that task it beats people.
Anthony Scaramucci (07:00):
Let's talk about that other wave though. Is it possible, Dr. Lee? I've obviously read your book, but I want you to explain it to others about the consciousness, about the empathy, the ability to change conversation, picking up emotional cues or body language from somebody. Is that something in the future for AI?
Dr. Kai-Fu Lee (07:22):
Okay. So we need to segment that into a few things. The true feeling, the way we feel, the consciousness, the self-awareness that we have, we currently have no idea how to build that for AI. So that could be 20 years, 30 years or longer away. We don't know. But it's probably not soon because we have no idea how to build that at all. However, can AI guess our emotion? Probably, because we give a lot of cues and AI can notice these small cues better than people.
Dr. Kai-Fu Lee (07:58):
So if you want to build a tool that has you and me talking to each other and have AI guess at any given period of time whether I was anxious, happy, sad, angry, suspicious, it can probably make a more accurate guess than people. Then can AI can pretend to be angry and happy? Well, of course, it can. If you have AI create a digital human that looks like us... You've all seen Deepfake. That's AI. We can build a Deepfake on Anthony and make that Deepfake speak like Anthony and appear angry or appear happy. So that is also possible.
Anthony Scaramucci (08:45):
I want you to make me happy, Dr. Lee. Okay? We'll focus on the whole Buddhist element of that.
Dr. Kai-Fu Lee (08:52):
Right. Right. We'll make sure we do that. I think this is really amazing that... The surprising thing about AI is for something that has absolutely no self-awareness, no human brain, and no feeling, it can exhibit feeling and perceive feeling. So that's the strangeness of AI. Similarly, you think AI machine translation works so well better than us. However, it doesn't really understand a word you say. It is merely mapping words to other words, having been trained on trillions of other words.
Dr. Kai-Fu Lee (09:32):
This is all data driven mechanism that exhibit somewhat intelligent behavior, but has no real understanding. It is just matching symbols and giving you other symbols.
Anthony Scaramucci (09:46):
But over the last 40 years while you've been studying and working on AI, you've obviously learned a lot about the human brain. Is the brain a computer, Dr. Lee? How would you describe the brain to somebody? If an alien landed here and you would say, "Okay, the human brain is," what based on your observation?
Dr. Kai-Fu Lee (10:10):
Well, it's still really unfathomable from a computer standpoint. We don't know how to simulate the brain. People are working on it. But if we think about what it is that makes us humans valuable, meaningful entities, what makes our lives full, it certainly isn't doing what AI already does very well. Daniel Kahneman wrote a very famous book, Thinking, Fast and Slow, in which he talked about System 1 and System 2 thinking. And in some sense, AI is doing the System 1 thinking, which is I see, I recognize, I heard, I heard something, and I recognize this word.
Dr. Kai-Fu Lee (10:58):
It's almost reflexive and almost perhaps muscle reflex. It's thing that we do without perception and without deep thinking. But what is interesting about the brain, as Dr. Kahneman said, is that we're able to think deeply, think strategically, think holistically, plan things in a very large space of possibilities, but we just know that if we do A, they'll respond by B, then we do C. So there's this very clear focus and awareness in making our decisions, and also with that, the ability to be creative, and also, of course, emotions and compassion.
Dr. Kai-Fu Lee (11:49):
So to answer your question, I think it's the System 2 stuff that makes us really unique. And that's why people can be brilliant like Einstein or Steve Jobs and that's why people can be compassionate like Mother Teresa. And these are special people and these are the special qualities that we have and that AI cannot do and possibly can never do.
Anthony Scaramucci (12:20):
That's my question. Could it ever be replicated based on your observations?
Dr. Kai-Fu Lee (12:26):
Well, there are many, many views on that because no one knows the answer. I think it maybe impossible to do, because we currently don't know how to do it. And also, I'd like to think that these technologies happen for a positive constructive reason, not because we want to build machines to replicate us. There's got to be something innate about us that makes us human, that makes this life meaningful. So I think we have to hold onto that belief that AI can't...
Anthony Scaramucci (13:01):
And the positive stuff about AI, our mutual friend, Peter Diamandis, has written a lot about the future and what he calls the abundance, and that there is a world ahead of us where through machine learning and AI and lots of other things that are going on in the world that we can end things like poverty, we can end sort of the income divide. So talk a little bit about how AI could be a part of that over the next generation. What do you envision?
Dr. Kai-Fu Lee (13:35):
Well, on the constructive side, clearly AI can make better decisions within limited tasks. AI can take over routine tasks that we have to do. If you think about all the System 1 stuff, those are more the routine tasks, right? If you think about the job of a receptionist, some of that job maybe very interesting, the human element, the warmth, the breathing, the compassion, the branding image on your customers, but a lot of that work is very boring. Show me your face. Show me your ID. Print you an ID. Who are you seeing?
Dr. Kai-Fu Lee (14:12):
Call the person. Well, the boring part can be done by AI, and you can extrapolate that to the job of an accountant, a lawyer, even a doctor. And these jobs, interestingly, AI will take care of the repetitive, routine, and quantitative. Things that we're not very good at. And then we get to focus on what we're good at, which is the System 2 thinking, the analytical, the creative, the compassionate, the human to human connection.
Dr. Kai-Fu Lee (14:48):
I think Peter and I share this belief that AI is here to take away the routine work so we can be liberated from it, and we can spend our time, all of it, on things that makes us uniquely human. That would be the most positive direction.
Anthony Scaramucci (15:09):
Can you talk a little bit about healthcare because I know that you have a belief that AI is going to certainly help us in diagnostic healthcare, research data? Enlighten us about where you think that's going using artificial intelligence.
Dr. Kai-Fu Lee (15:23):
Yeah. So healthcare is an area where AI really hasn't yet made a huge dent yet, but it is so perfectly designed for AI, because AI would work well in domains where you have large amount of data and very clear outcomes and labels and longitudinal data over years and decades. And that's exactly what the healthcare records have. And also, AI can basically deliver very targeted personalized determination. The reason we really get addicted to Facebook is it personalizes and shows us what it knows we want to see.
Dr. Kai-Fu Lee (16:12):
The reason we buy so much on Amazon is because Amazon shows us things that it knows that we as individuals want to see. Yet if you think about medicine, for each disease, we're largely all treated using a single prescription, or maybe for complex things like cancer, there maybe multiple types depending on each person's various background. But each person is unique and human doctors and human teaching of medicine just cannot possibly teach each doctor to treat each person uniquely according to that person's background and the DNA and genome sequencing and family history and so on.
Dr. Kai-Fu Lee (17:07):
But yet, when we have all the data from the patients from one country, that can be trained so that it can specifically target each individual with a treatment that is just right for that person. So that personalized medicine and training and diagnosis is something we can look forward to. Of course, it will have to overcome privacy laws, maybe anonymize the data, maybe use some technology to protect people losing their privacy, but I think that can be done.
Dr. Kai-Fu Lee (17:42):
And once that is done, what will happen to the future of treatment and healthcare is that for people who can afford it, which is basically most Americans today, you will get a human doctor aided by an AI doctor. The AI doctor will suggest to human doctor, ask few questions, take the answers, look it up, suggest treatments. And the human doctor will tease out all about your background and condition and also care about you, show compassion, connect to you, visit you at home, giving you a higher chance of recovery or survival.
Dr. Kai-Fu Lee (18:21):
That's the symbiotic combination that uses people for what people do best and machines for what machines do best. But finally, what's interesting is in poor areas, in under developed countries that cannot afford this expensive doctor who has to go through medical school and charges a lot of money because of the high salary, one could imagine a pure AI doctor that essentially draws no salary, runs on nothing except electricity, give decent treatment, significantly bring up the fatality rate, improving the treatment even for the poor reaches of the world.
Dr. Kai-Fu Lee (19:01):
So I see a lot of opportunities there. Of course, there are also things like robots and improve the intuitive surgical using robots to do surgery, AI for drug discovery, and also connecting AI to insurance and healthcare. Once it knows about you and your family history and your finances, it can design a perfect insurance policy for you that's much more economical than what you can buy from insurance companies. So I think it's endless when you connect all that data together.
Anthony Scaramucci (19:38):
So that brings up the question of further automation. And as we both know, the pandemic, unfortunately, has raised unemployment in the US to 14.7% and that's closing in on depression-like levels. Certainly we hope this is a temporary thing, but do you think it's accelerating the trends? Will it accelerate the use of AI? And will people that had traditional jobs, like the ones you're describing, will they lose out AI, or is it too soon for that?
Dr. Kai-Fu Lee (20:11):
Okay. First, on the AI impact on jobs before and then we get into the pandemic. While I believe in the symbiotic nature for AI in many human jobs as I described earlier, AI will take away many jobs as well, because if it can do 30, 50, 70% of different types of jobs, jobs of receptionists, a security guard, and entry level accountant, assistants, paralegal, and factory workers, drivers. So you list all of this. In a small number of cases, the whole job goes away because AI takes it over.
Dr. Kai-Fu Lee (20:53):
But in most cases, AI takes over 30, 50, 70%, but that still leads to a reduction of employment, because in a pool of workers, AI will take some jobs that it can do, leaving the rest for a fewer number of humans to do. Undoubtedly, there will be significantly fewer people working on today's white collar routine job and blue collar routine job. There will be other jobs created, but we don't quite know what they are yet, and they will tend to be more complex in nature, more creative in nature, or more human to human connection in nature.
Dr. Kai-Fu Lee (21:32):
Because if AI can do the routine jobs, then the jobs available for people would have to be elevated. There is a training gap. So while I believe there will be many more jobs created and the problem of taking the people whose jobs are displaced and retraining them for the jobs that are being created is an upleveling problem, is a training problem that somehow people have to understand what jobs are safe and get trained for it. So that's before the pandemic. Now, the pandemic will do some problematic things and also some constructive things.
Dr. Kai-Fu Lee (22:16):
The constructive thing that pandemic will do for our four habits is that it pushes us to much more online and digitized behavior. I mean, the fact that we're having this session here on Zoom and the fact that people are able to work from home and the billing kids are taking classes at home are signs that we are increasingly going online and increasingly getting comfortable with a digitized style of working. The opportunity is once digitized, you've got data. Once you got data, AI can work. That's the great thing about creating value and improving efficiency.
Dr. Kai-Fu Lee (23:08):
The problem of that is once AI can work and also outsourcing can work, jobs will be challenged. Imagine in the past, if you had a job that required you to go to the office, meet people and talk to people, then it seems hard for AI and robot to take it over. But now you're doing that job online and remotely and by video conferencing and then it will become obvious to the managers of the company that an AI could do your job too. The decision process maybe relatively simple. It can be learned. There's a technology called RPA, robotic process automation.
Dr. Kai-Fu Lee (23:57):
It's rapidly taking away these various types of white collar routine jobs. I believe the pandemic will lead to more digitization, online, and outsourcing and also automation as one unfortunate outcome. The other unfortunate outcome is that companies will have tighter budgets. They'll have to do cost cutting. And before, they might not think about, well, let me spend $2 million to replace $2 million of salary, $2 million of software to replace $2 million of salary or for some period of time.
Dr. Kai-Fu Lee (24:38):
People might not do it or maybe the company is making money, they feel if they did that, it would look bad. But now, everybody's scrambling. Everybody's tight. Everyone's cost cutting. So companies are going to be more willing to look at cost cutting...
Anthony Scaramucci (24:54):
It makes sense. Before I turn it over to questions from our audience, I want to talk a little bit about the relationship between Chinese government and the American government and the competition with AI. There are people in the United States that feel China is ahead of the United States. Perhaps they are. I don't know. Are they? Secondary question is, you and I, of course, want there to be a very healthy and strong bilateral relationship between the Chinese government and the US government.
Anthony Scaramucci (25:31):
But I'd like you to talk about those tensions if you don't mind, how they relate to AI, and where do we stand vis-a-vis the progress being made in AI, China versus United States.
Dr. Kai-Fu Lee (25:43):
Sure. AI turns out to be a technology that is not such a rocket science. There are probably a few dozen important discoveries. If you study them, if you get the technology, the code, you can probably implement AI after months of training, not even years of training. That is an advantage for China. While the US has more of the brilliant researchers who write up the papers, China has a larger army of engineers who are building solutions in the industry. And China's other big benefit is that China is a large country.
Dr. Kai-Fu Lee (26:28):
There is a lot of data. AI works better with more data. So that China has more engineers, more data, fewer brilliant scientists. So in some sense, US and China can and perhaps should in an ideal world be partners in this, where US is doing more the deeper research, the more complex technologies like autonomous vehicles, where China can do more the low-hanging fruit, the implementation, the things that requires a lot of data.
Dr. Kai-Fu Lee (27:01):
And then on domains like healthcare where Americans are extremely concerned about privacy and there are laws like HIPAA preventing aggregation of data in the US, perhaps Chinese companies can build models using advanced American medical technologies and AI technologies. But on Chinese data, it's anonymized, but there's no equivalent of HIPAA in China, so that aggregation can happen. And then the outcome can be shared by both countries. So in an idealistic and maybe at this point naïve viewpoint, the two countries are highly complimentary.
Dr. Kai-Fu Lee (27:44):
There's not an AI war going on. China can build all the things without great dependencies on American products, and US can, of course, build things on its own. But the two countries have such different talents they ought to work together. But that maybe pretty hard now.
Anthony Scaramucci (28:04):
Yeah, no, I get the tension. John, let's kick it over to some of our guests that are inside our chat room here. Ask Dr. Lee a question for us.
John Darsie (28:16):
Yeah. The first question is about GANs, generative adversarial networks. You talked about Deepfakes and things like that. What are the real benefits of GANs in terms of creating positive change to society? And what potential do they have to create general AI, and also what are some potential dangers of advanced AI becoming prevalent in society?
Dr. Kai-Fu Lee (28:44):
Generative adversarial networks are very cool technology. Basically you're building two networks, one to do what you want done, the other to be a critic. And then the critic will tell it, "Hey, this is not right," then it fixes itself, and it continuously improves itself. There are many, many applications of GANs. The one that's probably most infamous, notorious is the Deepfakes. It is using that technology that it manages to turn a video from some other third party into you or a voice and be converted that way.
Dr. Kai-Fu Lee (29:26):
When applied constructively to building entertainment and games and movies with full licensing of the properties, it's amazingly fun. But when you take a famous politician or movie star and put their faces on doing acts that they don't want to be seen doing, then it's a problem. It's the kind of technology where the technology is used by the bad people to do something, then the good guys catch up and catch them, then the bad guys improve again.
Dr. Kai-Fu Lee (30:03):
And unfortunately, because of the nature of the technology, that you have a good guy and a bad guy, basically the two networks, the good guy network and the bad guy network. They continue to iterate. And the moment you think you got a way to catch the bad guys, they take it into their training as well. It's very hard to say whether if we purely competed on the good guy/bad guys. The good guy continue to try to catch the bad guys doing the Deepfakes.
Dr. Kai-Fu Lee (30:35):
The bad guys continue to come up with yet another way to do a Deepfake. It's not clear whether this will lead to a good outcome or a bad outcome. My belief is often we have to resort to other technologies that will guarantee the worst case scenario doesn't happen. With respect to Deepfake, probably we'll need to move to some sort of a future blockchain assisted capture device which guarantees that this photo, this video is authentic. And it can catch anything that's been made on editing it.
Dr. Kai-Fu Lee (31:13):
Some technology like that is probably needed to absolutely guarantee the problem with Deepfakes. Otherwise, I would warn the people watching that we should expect there to be more Deepfakes happening in the social networks. We've got fake news. Now we've got fake video and fake voice. It's very, very hard to catch, and it's going to be a while before we eliminate it. People have to be advised not to believe everything you see even if it looks real.
Anthony Scaramucci (31:50):
Any other questions, John?
John Darsie (31:52):
Yeah. There are several more questions. I'm going to combine two questions into one. We talked about US and China in terms of where they are in the AI race, if you will. We have two questions about emerging market economies, ex. China, as well as Europe. How well are those economies doing in terms of advancing with AI and machine learning? And what potential does AI have to sort of bring emerging economies into a higher quality of life and into a more modern era?
Dr. Kai-Fu Lee (32:23):
Okay. In the current status, I think US and China are ahead of the other countries in terms of AI in an aggregate score, that is research plus implementation plus monetization. Right. Europe I think is very strong in research, but the entrepreneurial ecosystem is currently nowhere close to US or China. And unless that gets fixed, Europe is likely to be considerably behind in AI technology. India is another possible country that could do very well because it has also a large number of people and data. We have not quite yet seen that, but I think the potential is there.
Dr. Kai-Fu Lee (33:11):
And then there's obviously Russia, which is very good in math. There is Southeast Asia which is a large group of people, but not one culture, one language, and then it goes down from there in terms of likelihood of being very strong in AI. But what can AI do for these countries? First, the problem is that AI will create these hundred billion dollar companies, and they're currently pretty much all American and Chinese. The wealth is going to these companies. And AI will decimate a lot of the jobs most of which are routine jobs.
Dr. Kai-Fu Lee (33:57):
AI as a wealth creation is giving that wealth to US and China. In terms of replacing jobs, it will take more jobs away from developing countries because developing countries have more routine jobs. That is the seriously problematic part of AI for the rest of the world. There are some good news about AI in the developing worlds. It will dramatically reduce the cost of education because there will be virtual teachers, which can do a pretty decent job of teaching certain subjects, especially entry level ones.
Dr. Kai-Fu Lee (34:40):
There will be reasonable quality virtual AI doctors that will also provide better healthcare. So some services I think will help the people who are in the most extremely serious extent of destitute. But as a whole economically, it is a problem and I think all the countries have to pay attention about the impact of AI and find a path that makes sense for the country.
John Darsie (35:14):
We have a couple more questions. In science fiction, a very popular topic in movies like Blade Runner and others is the idea of consciousness and whether AI and technology will create immorality for humans in a way. Going into the science fiction aspect of that, do you think that AI will eventually be able, as Anthony was talking about earlier, replicate some aspects of consciousness and provide immortality for humans?
Dr. Kai-Fu Lee (35:45):
There a lot of different opinions on this subject. There are people who think it's imminent. It's within a decade. There are people who think it's another two or three decades, and there are people who think it might be never. I think it's hard to say which thinking is right. But I would like to think that first, we have no idea how to build consciousness. Secondly, we don't really understand what consciousness is. And thirdly, we people must be believe that we have a reason to be on this earth.
Dr. Kai-Fu Lee (36:24):
So I think it makes sense to believe that consciousness is the thing that makes humans unique and that it may not be buildable by machines. I think that will give us the confidence to go on. And I think it also is a plausible outcome and we should let people work on it. But until we see significant breakthroughs, there's no reason to believe that the age of the robots are coming. I think we're still quite a ways from that.
John Darsie (37:01):
Another question relates to the ethics of AI and where AI needs to make decisions in real time, some of which could involve law enforcement or conflict or war type scenarios. How do you program ethics into artificial intelligence?
Dr. Kai-Fu Lee (37:20):
This is a very important aspect and I think we're in a very early phase right now. Right now most AI programmers are not even taught ethics nor do they think they play a role in ethics. And that's important for the AI tools to change. And I think the AI education, some schools like Stanford and MIT, are starting to make sure that AI students are aware that their profession can impact good and bad, right and wrong in society. Just like doctors have to make an oath that they will do no harm. I think AI engineers will increasingly need to do that.
Dr. Kai-Fu Lee (38:10):
It's important also to note that when we read all of these AI disasters in the newspaper, not all of which are a result of AI not understanding ethics. There's usually a different explanation. For example, there are cases where people talk about certain company trained their HR system on AI. They didn't have a lot of women, so it became prejudice. It interviewed more men and fewer and fewer women. It became a downward spiral.
Dr. Kai-Fu Lee (38:41):
That story is true, but that could have been avoided if the programmer or the person who runs the AI over them recognized that their training set, their training data was not fairly balanced between men and women. And if engineers don't notice it, our tools ought to notice it. These kind of ethical issues many of which maybe solvable. The other that's talked a lot about is the autonomous vehicle. Trolley problem. Certainly it's an issue when the car is faced with different outcomes.
Dr. Kai-Fu Lee (39:28):
People talk about if you have two choices, one is 100% going to kill one person, the other is 52% and they kill two people. Which do you do? It is, in fact, a hard choice. But in reality, we have to remind ourselves that there are very few cases that you really have two people killing decisions in an autonomous vehicle. Secondly, we have to remind ourselves humans don't even have this program in. If you talk to all of the people who have been in accidents, who have caused accidents, got in trouble as drivers, they can usually hardly explain why did what they did.
Dr. Kai-Fu Lee (40:12):
I believe the glass half full would tell us is that if we program ethics in some reasonable way for the decision-making and with the powerful sensors that AI can see and the deliberate decisions as opposed to people just getting drunk or tired or sleepy and make a mistake, AI won't do that. At the end of the day, AI will really save a lot of lives. And while we do need to focus on training the engineers, building the tools, I think at the end of the day, AI will save so many lives.
Dr. Kai-Fu Lee (40:56):
That yes, there will be ethical issues and decisions and mistakes made, but in the grand scheme of things, AI doctor will save so many lives more than the few ethical mistakes it may make, and AI autonomous driver will save so many lives more than the few ethical decisions that it will make a mistake on. We have to look at in the grand scheme of things, not just focus on the one case where it appears to be not working.
John Darsie (41:28):
Well, Kai-Fu, we really want to thank you for joining us today. We're going to wrap it up there. I know you're in Beijing right now beginning your quarantine. The Chinese government, as well as other governments in Asia have done a great job of stamping out the virus. I want to thank you for taking the time to join us. Anthony, I don't know if you have any closing remarks for Kai-Fu.
Anthony Scaramucci (41:53):
We're grateful to you. We hope that we can get you back to the SALT Conference physically, Kai-Fu. Otherwise, we're going to have to create an artificially intelligent Kai-Fu to entertain our guests and educate our guests. But in the meantime, we wish you great health and great personal safety, and we look forward to seeing you at one of our next events. Thank you again for joining us today.
Dr. Kai-Fu Lee (42:15):
Yeah, see you in SALT. Bye, bye.
Anthony Scaramucci (42:17):
Okay.