The Big Nine: The Future of AI

So when we think about artificial intelligence, we tend to think that a lot of that research is being done in venerated universities like this one. In actuality, the entire future of AI and as a result, the future of our human existence is being built by just nine companies. Three are in China, six are in the United States. So let's begin with a couple of questions that are on my mind and should be on your mind too. If the fastest growing power to change our future was concentrated in the hands of only nine decision makers, would you worry if those decision makers were driven by market forces or politics instead of what's good for you? Is that something you would care enough about to do something? So when we think about artificial intelligence, we tend to think that a lot of that research is being done in venerated universities like this one. In actuality, the entire future of AI and as a result, the future of our human existence is being built by just nine companies. Three are in China, six are in the United States. The six that are in and these are it. So it's Baidu, Alibaba, Tencent, Google, Microsoft, Amazon, Facebook, IBM and Apple. The ones that are in the United States, I like to call the G Mafia. And they serve two masters. So they serve the whims of Wall Street. Because these are publicly traded companies and they also serve Silicon Valley, there is a transactional relationship at best with Washington D.C. so, so oftentimes these companies, and by the way, I don't think that they are intentionally doing evil, although we keep seeing that in our headlines over and over again. I believe that these companies were founded with our best interests in heart. I know engineers working at these companies who are trying to do good for the future of humanity. But unfortunately there's this expectation that if they don't constantly deliver brand new shiny products and services on some kind of schedule, or as though research and development can be scheduled in advance to coordinate with annual meetings, that somehow they're failing. So there's this rush to productize and commercialize their work. And that is not the case in China. So in China, the bat, that's Baidu, Alibaba and Tencent, these companies are very much under the thumb of Beijing, even though they are technically public independent companies. And what they're doing right now fits into the Belt and Road initiative. So this is an infrastructure initiative that actually has a ton of digital components. So this is not just about building roads and bridges all throughout the old Silk Road route route. This is also about laying 5G and fiber and giving great connectivity to all of the people who are there. That's happening at the same time that China has a President Xi Jinping who is brilliant, and there have been some changes in how the government functions, which means he's effectively president for life. And while in the United States we're rushing to develop commercial products for AI, in China, it's a different story. China sees AI as the connective tissue between it and a new world order. A new world order in which Chinese ideals, which butt up against Western democracy and certainly call into question what we think about free speech, are at the forefront, not just in China, but in many different countries at the moment, 58 countries that are part of the BRI pilot. So there is no single event horizon. Everybody looks out into the future for some day when AI will magically appear and when all of you, because I know all of you have sit through your meetings and look at all of the news, content and everything else that's happening, there's a tremendous amount of misplaced optimism and fear. Most people have a fundamental misunderstanding of what AI is, what it isn't, and what it's capable of. So here's what I'm going to do. In 14 minutes and 42 seconds, I'm going to demystify AI so that everybody who leaves here can go have a conversation with their staff. I'm going to give you some context in history. You're going to feel really bad, and then I'm going to tell you how we can fix our futures. So AI is not brand new. It's been in some form of development for literally hundreds of years, going back to this person. Shout out her name if you know her. I guess it's on the screen, right? Shout it out anyways. Ada Lovelace. Yes, Ada Lovelace. And Charles Babbage, who she worked alongside, built something called the Analytical Engine. And in the footnotes of academic papers she was translating, she went off on a tangent and described the first algorithm. She talked about how machines might someday do other things besides simple tabulation. That in fact, they could compose music, they could make autonomous decisions, and someday we would have computers that could think on its own. So that was the word that she used to describe it. Fast forward a while to the 1939 World's Fair, and Westinghouse debuted a computer that looked pretty similar to humans. This is it.

And so, ladies and gentlemen, with a great deal of pride and pleasure, I present to you Electro, the Westinghouse Moto Man. Electro, come here. And here he comes, ladies and gentlemen, walking up to greet you under his I am a smart fellow, as I have a very fine brain of 48 electrical relays. It works.

So basically, what's inside of Electro is a telephone system. And so they're pre recorded, but he was able to respond in real time. And the announcer had a little telephone and that he was using to signal prompts. And at some point, Electro talks about his brain, and the announcer says, well, not everything is always right because he's just learning. And that takes us to Alan Turing, who I know you've heard of before, and of course, the Turing Test. This is when you have a computer and a person and an interrogator, all separate, and the interrogator asks questions to both the computer and the other human. And. And the point at which the interrogator can no longer tell the difference between human answers and computer answers is when that test is passed. And when that test is passed, we think that the computer is thinking, all right. What's really interesting is that up until this point, computers were actually pretty dumb. So they had compute power, but they couldn't do much beyond simple call and response. And then things changed. So there was this big gathering at Dartmouth University in 1956, where Marvin Minsky and John McCarthy, who were two of the founders of modern AI, got a bunch of people together. A bunch of people. Well, they got a certain type of people together because, as it turns out, they invited this massive list. Neuroscientists, computer scientists, people working on radar, people working on all different aspects of what would become the system that we use today. And though they reached out to all of these people, there was only one woman invited. It was Marvin Minsky's wife. And I think she was there to take care of the dog for the summer. So there were no people of color, there were no women. And this group of people are the ones who laid the groundwork for modern AI. In fact, it was at this meeting where the term artificial intelligence was coined. So for 60 years, we've been trying and failing over and over again to teach machines to think. To think just like we do. And then this happened. This is Lee Sotol losing a game of Go to a computer to Google's DeepMind algorithm. And you don't have to play Go to understand the significance of this. Computers playing chess is something they can accomplish through brute force and strategy. Go is a lot different. Go is not just about capturing a king. Go is about capturing space and force. And because of the different ways that the pieces can move, it's kind of mathematically impossible to get through that game, using brute force alone. This is about a new kind of strategy, in fact, teaching machines how to think. As it turns out, DeepMind, their AlphaGo system, is an AI computer program that can't just think. And it was capable of outthinking a human. And as it turns out, you've been using AI every single day for years. You just don't recognize it. What you've been using is something called artificial narrow intelligence or weak AI. And this is when a system can perform a single narrow, insignificant task at the same level as us or better. So your car, like the fuel injection, the anti lock brakes, that's ani, your, your phone, the maps, the Gmail that you use, where it will automatically complete an email for you, your email spam filter, your travel experience, everybody who works in risk and compliance in your organizations, they're using ani and everybody who built these systems, these were relatively small group of researchers. And the way that all this works is they started with a data set, they selected a framework, they built the algorithms, they trained those algorithms, and then they either corrected the result or optimized it in some way. But guess what? The people who are building all of this work in small offices in just a few places, and they don't represent all of us. So this is ani. And at this point you might be thinking, wtf, right? Because when we think about AI, we anthropomorphize it. So we think about Skynet and we think about Westworld, the greatest television show in American history. And we think about Zuck, who tells us that everything is going to be wonderful because AI is going to save us all. Or we listen to Elon Musk who says that the robots are going to kill us tomorrow. The reality though, is far more terrifying than Skynet. So if we can get away from these layers of abstraction and get down to the details to talk about what's really happening, that's scary. There's a lot of uncertainty. But that also means that we've got some hope for the future. Now, those working at the G Mafia and at the Bat form AI tribes, they are brilliant, they are insular, and they are pretty much homogenous. The problem is that they all tend to study at the same few universities, so that's not unlike many other fields. I actually went to school right across the walkway there at the Columbia J School, so a lot of journalists who go to graduate school tend to go to the same few places. So this is true of every field. However, when it comes to computer science and AI, Related fields, there is zero room for extracurriculars, which means that at most, if you're at a good university, there's a single ethics class which everybody has to take and everybody kind of hates. And that information doesn't percolate through the rest of the curricula. There's no cultural anthropology, there's no world religion. There's no opportunity to take comparative lit. And one of the key studies that people learn as they go through these programs is something called the trolley problem, which you're probably familiar with too. It comes from Western moral philosopher Immanuel Kant, and it goes something like this. You've got a car or a trolley, and what should the trolley do if it starts going out of control? Should the trolley veer into one direction and kill just one person, or is it better the driver, or is it better to veer into a crowd and kill all of those people? You can't just do this using quantification. You have to think through the moral implications. And so that's what everybody is learning how to do, and that's what they're training our machines. But it's a binary conversation. Killing a human is bad, keeping a human is good. This is a very, very rigid interpretation. And it works if you only have two choices. If the choices are crash into a tree and kill one or none, crash into a crowd and kill many. But that's not how the real world works. The real world isn't that rigid. And also humans are capricious, which means there are untold variables that we are constantly having to deal with at any given time. So the issue is that the Big Nine promise that they are making AI for the good of humanity. But how do we define good? I mean, if you and I, after this, went and got a coffee somewhere, would we align? Is your definition of good under every single circumstance the exact same definition that I have of good? What does good mean to you? What does good mean to your company? What does good mean not just to your company, but how it relates to other people. These are challenging questions that require flexibility. The challenge is that Silicon Valley and China are operating under totally conflicting assumptions right now. We have no baseline. There's no international alignment, not just on what is good, but but under what circumstances should we assume risk? What do ethics look like? How important is inclusion and diversity? How important is it to sandbox our AI systems which make automatic decisions before we can guarantee their safety? There is no baseline. There are no guardrails, which is a way of saying not only is there no alignment but basically there's no safety standards anywhere in place. There are no norms, there are no standards. And very, very few organizations and certainly very few departments are doing any kind of real risk assessment. And so here's what happens. So this is some research that some researchers did, some academics did. They decided to train AI to see if it could have gadar. They used a VGG deepface learning algorithm, a bunch of a corpus with a bunch of photos, and they successfully did it. I mean, so this is the. These are composites. So these aren't real people. But according to AI, this is the straight version of that person and this is the gay version of that person. I don't know about you, but I'm not comfortable with this. All right, here's Google search. I typed in CEO to images. By the way, this learns from us in real time using machine learning algorithms. Looks like a pretty white bunch of dude CEOs, right? So there is a woman behind this black box. Does anybody know the first woman CEO who pops up in this search? Shout her out if you know her. Yeah, no, it's not. It's CEO Barbie. It's funny until it's not, right? Software engineer last week used some deep learning algorithms and Nvidia to generate a whole bunch of people who don't exist. So these people, all, a few of them, look janky. These are all totally realistic looking photos. And by the way, we can also do similar things now with text and with voice and with video. So if you think we've had a problem with fake news, actual fake news, up until this point, you have absolutely no idea what's coming. We talk a lot about. I told you, we talk a lot about bias. We talk a lot about these future problems as though they are on some event horizon far in the distance. But they're not. We are doing a lot of talking, but we are not doing any fixing. There is no mechanism right now, and certainly no economic incentive to prioritize safety over speed. Not in the United States, not in China. And by the way, it's not just our two countries who are using these systems. So there's an entire ecosystem of startups. Some of your companies are probably partnered with these companies. And none of you are incentivized to prioritize safety over speed. And having lived in China and witnessed what happens when you take shortcuts, because not only are you not incentivized to prioritize safety over speed, but in fact the economic incentives are to do more faster. I don't know. And in the past Year. This is not a made up number. In the past few years, China has had 500,000 food health scandals due to quarter cutting. We're talking about training autonomous decisions to make systems to make decisions for us. I don't want this way of working to impact me in the future. ANI is just the beginning. So here's where we go. AI intersects. It's the third era of computing. It's not a tech trend. Okay, so. And it's not a buzzword. It intersects with all other fundamental technologies. Autonomous driving, genomics, CRISPR, blockchain robotics, 5G, every other thing that you can think of. And the problem is that AI has always been measured using deception and replication. So I'm going to end on this, which is what happens. When the DeepMind researchers took that system that played Go, they beat everybody in the world. There were no other grandmasters. So they retired their system and then brought it back online to essentially play itself. It didn't take long for the system to get back up to the same level of playing GO as it was before. And then something interesting happened. What happened was that it just took a few days to go from playing at the same level as grandmasters to playing at a superhuman level in a way that was not explainable. And nobody understood. The system had made a choice to get rid of human strategy in favor of a strategy that it created without any humans in the loop. Google's AI now creates better machine learning code than the researchers who created it. And here's the problem. The problem is that AI is making choices rather than decisions. There is now a third version of this, which is Alpha 0. And instead of the system going from nothing and no humans to playing a game, it's using hierarchical reinforcement learning and multitask learning to learn multiple things at once in ways that we don't entirely understand. And that, my friends, is AGI. You are hearing about. You hear people all the time pontificate, when is artificial general intelligence coming? Because we keep thinking about these anthropomorphized robots. I'm telling you, it's here. We see already plenty of movement all over the place. We see examples of this all the time. So my question is, what happens when AI is deployed to monitor and make decisions about us? China's social credit score. I know you know what it is, is currently monitoring people using this technology and doing things like preventing them from buying airline tickets. It's proved. This is all data from China. Five million people couldn't buy a train ticket. Last year. About 300,000 people were really, really good at their jobs and qualified for a promotion. But their social credit score, that was built by and decided by an AI decided that they were no longer eligible. And why should you care if you don't live in China? Because China is in the process of deploying this technology to not just those 58 countries like the Philippines and Vietnam, but all of their partners too. Which means that if we don't have social credit scores in a decade, we could get locked out from the global economy. I know that sounds crazy, but we're already seeing data and signals in the present. So here's where I want to end. I don't actually think that the Big Nine are doing this intentionally. I don't think that the Big Nine are actually the villains of this story. In fact, I think that they are our best hope for the future. There is another way forward, and it involves all of us. So for the Big nine, slowing down development, prioritizing inclusivity, making their systems interoperable are great places to start. Developing new processes to evaluate ethical implications of their work is something that can be done. Making that work transparent is something that can be done as well as instituting new codes of conduct. But it's not just them. Wall street has to make some changes too. They've got to give these folks some breathing room and allow some of the investment and some of the R and D spending to cover things like risk and safety. Washington, D.C. has to stop turning its head against and away from what's happening in the Valley. We don't have any national strategy on, not just AI, but on many other technologies. We are not doing any serious funding into these areas. So the private sector is left to pick up the tab. They need to build relationships and not just focus on transactions. And we need to start looking at China as a militaristic pacing threat, which we don't do. And to acknowledge that while it's interesting to have giant planes and ships, our future wars are going to be fought in. I'm sorry? Our future wars are going to be fought in code, not in conduct. And you have a stake in this, too. We need courageous leadership under extremely challenging circumstances. And so I'm going to end with one question. I started with a few. And here's where I'm going to end. Will you join me in making these changes? I hope you will and I hope you would join me right afterwards. I'll be out there signing books and I can answer any questions that you might have. Thank you.