Architects of Intelligence

by Martin Ford

  When it gets to that point, it makes the military look more like police. Is that good in the long term? I don’t think anyone can guess. It’s less destructive than nukes—it can’t be more destructive than nukes!

  MARTIN FORD: Do you worry about a race with China in terms of advancing artificial intelligence? They have over a billion people, so they have got more data and along with that, fewer constraints on privacy. Is that going to give them an advantage in moving forward?

  YANN LECUN: I don’t think so. I think currently progress in the science is not conditioned on the wide availability of data. There may be more than 1 billion people in China, but the proportion of people who are actually involved in technology and research is actually relatively small.

  There’s no question that it will grow, China is really progressing in that direction. I think the style of government and the type of education they have may be stifling for creativity after a while. There is good work coming out of China, though, with some very smart people there, and they’re going to make contributions to this field.

  There was the same kind of fear of the West being overrun by Japanese technology in the 1980s, and it happened for a while and then it kind of saturated. Then it was the Koreans, and now it’s the Chinese. There are going to be big mutations in Chinese society that will have to happen over the next few decades that will probably change the situation completely.

  MARTIN FORD: Do you think that AI needs to be regulated at some level? Is there a place for government regulation for the kind of research you’re doing and the systems that you’re building?

  YANN LECUN: While I don’t think there is any point in regulating AI research at the moment, I do think there is certainly a need for regulating applications. Not because they use AI, but because of the domain of applications that they are.

  Take the use of AI in the context of drug design; you always want to regulate how drugs are being tested, how they are deployed, and how they are used. It’s already the case. Take self-driving cars: cars are regulated, and there are strict road safety regulations. Certainly, those are application areas where existing regulations might need to be tweaked because AI is going to become preponderant.

  However, I don’t see any need for the regulation of AI at the moment.

  MARTIN FORD: So, I assume you disagree quite strongly with the kind of rhetoric Elon Musk has been using?

  YANN LECUN: Oh, I completely and absolutely disagree with him. I’ve talked to him several times, but I don’t know where his views are coming from. He’s a very smart guy and I’m in awe of some of his projects, but I’m not sure what his motivation is. He wants to save humanity, so maybe he needs another existential threat for it. I think he is genuinely worried, but none of us have been able to convince him that Bostrom-style, hard take-off scenarios are not going to happen.

  MARTIN FORD: Are you an optimist overall? Do you believe that the benefits of AI are going to outweigh the downsides?

  YANN LECUN: Yes, I would agree with that.

  MARTIN FORD: In what areas do you think it will bring the most benefits?

  YANN LECUN: Well, I really hope that we figure out the way to get machines to learn like baby humans and animals. That’s my scientific program for the next few years. I also hope we’re going to make some convincing breakthrough before the people funding all this research get tired, because that’s what happened in previous decades.

  MARTIN FORD: You’ve warned that AI is being overhyped and that this might even lead to another “AI Winter.” Do you really think there’s a risk of that? Deep learning has become so central to the business models of Google, Facebook, Amazon, Tencent, and all these other incredibly wealthy corporations. So, it seems hard to imagine that investment in the technology would fall off dramatically.

  YANN LECUN: I don’t think we’re going to see an AI winter in the way we saw before because there is a big industry around it and there are real applications that are bringing real revenue to these companies.

  There’s still a huge amount of investment, with the hope that, for example, self-driving cars are going to be working in the next five years and that medical imaging is going to be radically revolutionized. Those are probably going to be the most visible effects over the next few years, medicine and health care, transportation, and information access.

  Virtual assistants are another case. They are only mildly useful today because they’re kind of scripted by hand. They don’t have any common sense, and they don’t really understand what you tell them at a deep level. The question is whether we need to solve the AGI problem before we get virtual assistants that are not frustrating, or whether we can make more continuous progress before that. Right now, I don’t know.

  When that becomes available, though, that’s going to change a lot of how people interact with each other and how people interact with the digital world. If everyone has a personal assistant that has human-level intelligence, that’s going to make a huge difference.

  I don’t know if you’ve seen the movie Her? It’s not a bad depiction in some ways of what might happen. Among all the sci-fi movies on AI, it’s probably one of the least ridiculous.

  I think a lot of AI-related technology is going to be widely available in the hands of people because of hardware progress. There’s a lot of effort now to develop low-power and cheap hardware that can fit in your smartphone or your vacuum cleaner that can run a convolutional network on 100 milliwatts of power, and the chip can be bought for 3 bucks. That’s going to change a lot of how the world around us works.

  Instead of going randomly around your room, your vacuum cleaner is now going to be able to see where it needs to go, and your lawnmower is going to be able to mow your lawn without running over your flowerbeds. It’s not just your car that will drive itself.

  It might also have interesting environmental consequences, like wildlife monitoring. AI is going to be in the hands of everyone because of progress in hardware technology that is specialized for deep learning, and that’s coming in the next 2 or 3 years.

  YANN LECUN is a Vice President and Chief AI Scientist at Facebook, as well as a professor of computer science at New York University. Along with Geoff Hinton and Yoshua Bengio, Yann is part of the so-called “Canadian Mafia”—the trio of researchers whose effort and persistence led directly to the current revolution in deep learning neural networks.

  Prior to joining Facebook, Yann worked at AT&T’s Bell Labs, where he is credited with developing convolutional neural networks—a machine learning architecture inspired by the brain’s visual cortex. Yann used convolutional neural nets to develop a handwriting recognition system that became widely used in ATMs and at banks to read the information on checks. In recent years, deep convolutional nets, powered by ever faster computer hardware, have revolutionized computer image recognition and analysis.

  Yann received an Electrical Engineer Diploma from Ecole Superieure d’Ingenieurs en Electrotechnique et Electronique (ESIEE) in Paris, and a PhD in Computer Science from Universite Pierre et Marie Curie in 1987. He later worked as a post-doctoral researcher in Geoff Hinton’s lab at the University of Toronto. He joined Facebook in 2013 to establish and run the Facebook AI Research (FAIR) organization, headquartered in New York City.

  Chapter 7. FEI-FEI LI

  If we look around, whether you’re looking at AI groups in companies, AI professors in academia, AI PhD students or AI presenters at top AI conferences, no matter where you cut it: we lack diversity. We lack women, and we lack under-represented minorities.

  PROFESSOR OF COMPUTER SCIENCE, STANFORD CHIEF SCIENTIST, GOOGLE CLOUD

  Fei-Fei Li is Professor of Computer Science at Stanford University, and Director of the Stanford Artificial Intelligence Lab (SAIL). Working in areas of computer vision and cognitive neuroscience, Fei-Fei builds smart algorithms that enable computers and robots to see and think, inspired by the way the human brain works in the real world. Fei-Fei is Chief Scientist, AI and Machine Learning at Google Cloud,
where she works to advance and democratize AI. Fei-Fei is a strong proponent of diversity and inclusion in artificial intelligence and co-founded AI4ALL, an organization to attract more women and people from underrepresented groups into the field.

  MARTIN FORD: Let’s talk about your career trajectory. How did you first become interested in AI, and how did that lead to your current position at Stanford?

  FEI-FEI LI: I’ve always been something of a STEM student, so the sciences have always appealed to me, and in particular I love physics. I went to Princeton University where I majored in Physics, and a by-product of studying physics is that I became fascinated by the fundamentals of the universe. Questions like, where does the universe come from? What does it mean to exist? Where is the universe going? The fundamental quest of human curiosity.

  In my research I noticed something really interesting: since the beginning of the 20th century, we’ve seen a great awakening of modern physics, due to the likes of Einstein and Schoenberg, who towards the end of their lives became fascinated not only by the physical matter of the universe but by life, and biology, and by the fundamental questions of being. I became very fascinated by these questions as well. When I started to study, I realized that my real interest in life is not to discover physical matters but to understand intelligence—which defines human life.

  MARTIN FORD: Was this when you were in China?

  FEI-FEI LI: I was in the US, at Princeton Physics, when my intellectual interest in AI and neuroscience began. When I applied for a PhD there I was very lucky, and to this day, it’s still a bit of a rare combination to do what I did—which was both neuroscience and AI.

  MARTIN FORD: Do you think then that it’s an important advantage to study both of those fields rather than to focus exclusively on a computer-science-driven approach?

  FEI-FEI LI: I think it gives me a unique angle because I consider myself a scientist, and so when I approach AI, what drives me is scientific hypotheses and the scientific quest. The field of AI is about thinking machines, making machines intelligent, and I like to work on problems at the core of conquering machine intelligence.

  Coming from a cognitive neuroscience background, I take an algorithmic point of view, and a detailed modeling point of view. So, I find the connection between the brain and machine learning fascinating. I also think a lot about human-inspired tasks that drive AI advances: the real-world tasks that our natural intelligence had to solve through evolution. My background has in this way given me a unique angle and approach to working with AI.

  MARTIN FORD: Your focus has really been on computer vision, and you’ve made the point that, in evolutionary terms, the development of the eye likely led to the development of the brain itself. The brain was providing the compute power to interpret images, and so maybe understanding vision is the gateway to intelligence. Am I correct in that line of thinking?

  FEI-FEI LI: Yes, you’re right. Language is a huge part of human intelligence, of course: along with speech, tactile awareness, decision-making, and reasoning. But visual intelligence is embedded in all of these things.

  If you look at the way nature designed our brain, half of the human brain is involved in human intelligence, and that human intelligence is intimately related to a motor system, to decision-making, to emotion, to intention, and to language. The human brain does not just happen to recognize isolated objects; these functions are an integral part of what deeply defines human intelligence.

  MARTIN FORD: Could you sketch out some of the work you’ve done in computer or machine vision?

  FEI-FEI LI: During the first decade of the 21st century, object recognition was the holy grail that the field of computer vision was working on. Object recognition is a building block for all vision. As humans, if we open our eyes and look around our environment, we recognize almost every object we look at. Recognition is critically important for us to be able to navigate the world, understand the world, communicate about the world, and do things in the world. Object recognition was a very lofty holy grail in computer vision, and we were using tools such as machine learning at that time.

  Then in the mid-2000s, as I transitioned from a PhD student to become a professor, it became obvious that computer vision as a field was stuck, and that the machine learning models were not making huge progress. Back then, the whole international community was benchmarking autorecognition tasks with around 20 different objects.

  So, along with my students and collaborators, we started thinking deeply about how we might make a quantum leap forward. We began to see that it was just not going to be sufficient for us to work with such a small-scale problem involving 20 objects to reach the lofty goal of object recognition. I was very much inspired by human cognition at this point, and the developmental story of any child, where the first few years of development involves a huge amount of data. Children engage in a huge amount of experimenting with their world, seeing the world, and just taking it in. Coincidentally, at was just at this time that the internet had boomed into a global phenomenon that provided a lot of big data.

  I wanted to do a pretty crazy project that would take all the pictures we could find on the internet, organize them into concepts that mattered to humans, and label those images. As it turned out, this crazy idea turned into the project called ImageNet, with 15 million images organized into 22,000 labels.

  We immediately open-sourced ImageNet for the world, because to this day I believe in the democratization of technology. We released the entire 15 million images to the world and started to run international competitions for researchers to work on the ImageNet problems: not on the tiny small-scale problems but on the problems that mattered to humans and applications.

  Fast-forward to 2012, and I think we see the turning point in object recognition for a lot of people. The winner of the 2012 ImageNet competition created a convergence of ImageNet, GPU computing power, and convolutional neural networks as an algorithm. Geoffrey Hinton wrote a seminal paper that, for me, was Phase One in achieving the holy grail of object recognition.

  MARTIN FORD: Did you continue this project?

  FEI-FEI LI: For the next two years, I worked on taking object recognition a step further. If we again look at human development, babies start by babbling, a few words, and then they start making sentences. I have a two-year-old daughter and a six-year-old son. The two-year-old is making a lot of sentences, which is huge developmental progress, something that humans do as intelligent agents and animals. Inspired by this human development, I started working on the problem of how to enable computers to speak sentences when they see pictures, rather than just labeling a chair or a cat.

  We were working on this problem using deep learning models for a few years. In 2015, I talked about the project at the TED2015 conference. The title of my talk was How we’re teaching computers to understand pictures, and I discussed enabling computers to be able to understand the content of an image and summarize it in a human, natural-language sentence which could then be communicated.

  MARTIN FORD: The way algorithms are trained is quite different from what happens with a human baby or young child. Children for the most part are not getting labeled data—they just figure things out. And even when you point to a cat and say, “look there’s a cat,” you certainly don’t have to do that a hundred thousand times. Once or twice is probably enough. There’s a pretty remarkable difference in terms of how a human being can learn from the unstructured, real-time data we meet in the world, versus the supervised learning that’s done with AI now.

  FEI-FEI LI: You totally nailed it, and this is why as an AI scientist I wake up so excited every day because there’s so much to work with. Some part of the work has inspiration from humans, but a large part of the work does not resemble humans at all. As you say, the success today of neural networks and deep learning mostly involve supervised pattern recognition, which means that it’s a very narrow sliver of capabilities compared to general human intelligence.

  I gave a talk at Google’s I/O conference
this year, where I was again using the example of my two-year-old daughter. A couple of months ago, I watched her on a baby monitor escape from her crib by learning the cracks in the system, a potential path to escape from the crib. I saw her open her sleeping bag, which I had particularly modified in order to prevent her from opening and get herself out. That kind of coordinated intelligence to a visual motor, planning, reasoning, emotion, intention, and persistence, is really nowhere to be seen in our current AI. We’ve got a lot of work to do, and it’s really important to recognize that.

  MARTIN FORD: Do you think there will likely be breakthroughs that allow computers to learn more like children? Are people actively working on how to solve this problem?

  FEI-FEI LI: There are absolutely people working on that, especially within the research community. A lot of us are working on the next horizon problem. In my own lab at Stanford, we are working on robotic learning problems where the AI is learning by imitation, which is much more natural than learning by supervised labels.

  As kids, we watch how other humans do things and then we do it; so, the field is now starting to get into inverse reinforcement learning algorithms, and neuro-programming algorithms. There is a lot of new exploration, and DeepMind is doing that. Google Brain is doing that; Stanford is doing that; and MIT is doing that. I’m very hopeful that in our lifetime we’ll be seeing a lot more AI breakthroughs, given the incredible amount of global investment in this area. We also see a lot of effort in the research community to look at algorithms beyond supervised learning.

  Dating when a breakthrough will come, is much harder to predict. I learned, as a scientist, not to predict scientific breakthroughs, because they come serendipitously, and they come when a lot of ingredients in history converge. But I’m very hopeful that in our lifetime we’ll be seeing a lot more AI breakthroughs given the incredible amount of global investment in this area.