Transcription
We will become a hybrid species. We'll still be human, but we will be enhanced by AI. We will no longer be limited by our biology. We'll be free to live life Without Limits. We're going to find solutions to diseases and aging.
Don't you think that superhuman AI is a marketing invention?
In 2024, I was honored to be on Time's list of the 100 most influential people in AI. I've been working on AI ever since—for 61 years, longer than anyone else alive—and predicted computers would reach human-level intelligence by 2029. And some say it will happen even sooner.
Let me start by explaining how we're going to connect our brains to the cloud. The host of the conversation is Dr. Matt Keki, a science popularizer and a digital ambassador of the European Union.
"It's a great pleasure having you here in New York."
"Thank you, Ma, for inviting me to participate in this interview."
Everyone talks about AI today. What did people say about AI 60 years ago? Did people believe that AI would be able to change the world?
Almost no one had heard of AI or knew what it was when I first started in the field. The term artificial intelligence was coined by Marvin Minsky and John McCarthy at a conference at Dartmouth College in 1956. Four years later, when I was 12 years old, I worked on a computer for the first time, programming statistical tests for a nonprofit to help them evaluate their learning programs for children. I was lucky enough to have access to one of 12 computers in all of New York City. I worked during the overnight shift and noticed that when the computer considered the data I entered, the lights on the front panel would dim and glow, almost like a dance. It looked like it was deep in thought. That's when I first became interested in the connection between computers and human intelligence. I've been working on AI ever since—for 61 years, longer than anyone else alive.
When I was 16, I met with Dr. Frank Rosenblatt at Cornell University, who developed a famous neural net called the perceptron, which was getting a lot of attention. It could recognize images, so I brought printed letters from the alphabet and fed them into the computer to see if it would recognize them, but it didn't work very well. Well, at that time, there was very little confidence in the power of neural networks. Marvin Minsky, who became a mentor for 50 years and invented the neural net in the 1950s, but had become pessimistic about its potential. He wrote a book with his partner, Seymour Papert, called Perceptrons that basically proved a perceptron could not recognize disconnected letters. So, for example, it couldn't recognize that a short line with a separate dot on the top was the letter I. The book was influential and killed all funding from neural nets for about 25 years. Now I know the perceptron failed because it was a single-layer neural net. The key word there is single. Today, multiple-layered neural nets are the backbone of machine intelligence and are driving the recent incredible breakthroughs in large language models and so forth in AI.
Shortly before Minsky died in 2016, he told me that his opinion had changed and he believed neural nets were the future of AI. Skepticism about AI has continued all the way up until the past few years when large language models like ChatGPT and Gemini caught everyone's attention.
I have read thousands of your predictions that have become true. You predicted several decades ago that a computer worth a thousand dollars would be able to perform trillions of operations per second. You indicated that payment would be possible with a digital card. What kind of person are you?
When I was a kid, my parents taught me the power of ideas can change the world. I used to write around my neighborhood in a go-kart that I made, collecting mechanisms from old broken bicycles and cameras and other things that I found, because I believed I could solve any problem if I figured out how to put them back together in just the right way. I knew then, when I was 5 years old, that I wanted to become an inventor.
One essential lesson that I've learned over my career as an inventor is that timing is critical—for everything from inventions to investments to romance. Even the best invention will fail if it's released at the wrong time. So in 1983, I began tracking the rate of technological progress to help develop and time my inventions, and that's when I discovered information technology was advancing at an exponential rate before anyone knew it was happening. This trend continues today, and I've used it to make all of my predictions. So take a look at this chart that I created. It shows the amount of computing power you can buy for one constant dollar, and it has a logarithmic scale. So on a logarithmic scale, it shows a smooth and steady line indicating consistent exponential growth since the first programmable computer was released in 1939. Despite major world events like World War II, economic depressions, and regardless of the hardware measured, the exponential trend in computational power spans relays, vacuum tubes, transistors, integrated circuits, and you can't see any indication of where one type of hardware ends and another begins. This is because each new information technology makes the next innovation easier. For example, we use one chip to design a better chip. This is why you can buy a smartphone that's twice as good as a smartphone two years ago for half the price. In 1939, the first programmable computer performed 0.007 calculations per second per constant dollar. Today, the latest Nvidia H100 chip performs 500 billion calculations per second per constant dollar. That's a 75 quadrillion-fold increase. Let me say it again: that's a 75 quadrillion-fold increase in the amount of computation you get for the same amount of money in the past 85 years.
Did you believe several decades ago that humanity would reach the point where it is today?
In 1999, I wrote a book called *The Age of Spiritual Machines* and predicted computers would reach human-level intelligence by 2029. I was able to see it coming by extrapolating the historical trend shown in my chart that we just looked at. This stirred up a lot of concern and skepticism. Stanford University organized a conference of AI experts from around the world—several hundred of them—to discuss my prediction, and 80% of them thought it would take 100 years. Today, consensus among AI experts has fallen in line with my original prediction, and some say it will happen even sooner. So I've technically gone from being an optimist to a pessimist without changing my prediction at all. I'm confident we'll achieve human-level AI by 2029 because of three converging trends that are happening now. First, as I've said, the price-performance of computation per constant dollar is growing exponentially. Second, AI researchers are constantly developing new algorithms that enable AI to learn and reason more efficiently. And third, the availability of massive data sets to train neural nets is skyrocketing.
You say that humanity will create AGI by 2029. Don't you think that superhuman AI is a marketing invention?
When I talk to one of the creators of OpenAI, Ilya Sutskever, he pointed out that humanity is developing together with AI. Today, we are the reference point of assessing whether something is or is not AI. The problem is that if we develop together with AI, something that we would consider superhuman AI today will not be able to be considered such in a few years. It's true that humans and AI are developing together as we blend our capabilities and knowledge into one. However, we will not fully merge with AI until the nanotechnology in the 2030s. So, for now, AI's superhuman capabilities are extremely apparent in many ways. Take large language models as an example. You can ask them any question in any subject, and they will immediately provide a well-written, thoughtful, and clear answer. And if you ask it again, it'll give you a different answer. But it's also true. What human can do that with any question? AlphaFold is another compelling example. One of the biggest scientific challenges that existed until recently was the protein-folding problem. Proteins are the building blocks of life, and their shape determines their function. If we can understand how they fold into their unique shapes, we can create medicines to reprogram our biology away from disease. And that's happening. For 50 years, scientists were unable to find the shapes of most proteins. Then, in 2022, DeepMind's AlphaFold 2 AI model found the shapes of 200 million proteins—nearly all known proteins—not just in humans but in all of nature. These were released publicly for free, which marked a profound step forward in medicine because it unlocked information that scientists need to synthetically produce proteins with the desired function, like simulating our immune system to fight cancer or Alzheimer's. And just a few months ago, DeepMind announced AlphaProto, which can create new proteins that can manipulate the interaction of molecules. AI breakthroughs like this are not limited to medicine and language; they're happening in all industries. They're transforming energy, manufacturing, farming—basically everything we care about. Once AI learns all skill sets, it will blend them together like humans. An early example is Google's DeepMind's Gato, a single generalist neural network that can perform over 600 different tasks, including text chatting, controlling a robot arm, playing video games, and so on. It's far from perfect, but it lays the groundwork for building a unified brain-like system that doesn't have to farm out different tasks to separate neural networks.
Are You Afraid Of Death?
I think most people fear death, as I do. It's an incredibly lonely concept, and it's tragic when loved ones die. Even 20, 30, 40 years after they have died, we still want them back. Death is not something to celebrate, but we've lived in a world where people just accept it. You hear it all the time: "Oh, life's short; you have to make the most of it." Some people say they don't want to live past 100, but I'd like to see them say that when they are 100. We've convinced ourselves that death is a good thing and gives life meaning, but when we get a life-threatening illness, we become desperate to find a cure. When faced with a choice, people will choose to live. It's only those who are in unbearable physical, mental, or spiritual pain that choose death. We are the species that transcends biology. We have; not done that; our life expectancy would still be 20, which it was a thousand years ago. So overcoming the limitations of biology is not a new story. It's for why I want to live indefinitely. It again comes down to living one day at a time. I want to live to see tomorrow because I want to see my loved ones, and I want to continue working on my creative projects. I don't see a time when I would not feel that way.
I ask because you spend a lot of time pointing out that humanity will one day achieve immortality. You pointed out that our brains will go to the cloud; they will be able to live there forever. I wonder if we can then talk about immortality. Shouldn't we ask a different question? For example, can such a state be called life?
As AI merges with medicine, we're going to find solutions to diseases and aging. At the beginning of the COVID pandemic, Moderna used a range of AI tools to help design and optimize mRNA sequences and discovered the sequence of its successful vaccine in just two days. They also used AI to speed up the manufacturing and testing process, which saved many lives. Biosimulations like these aren't going to be limited to viruses like COVID. With the exponential growth of computation, we'll soon have the ability to rapidly test billions of possible molecular sequences to find cures ultimately for all diseases. By around 2032, people who are diligent with their health are going to reach what we call longevity escape velocity. This is when scientific breakthroughs will add more time to our remaining life expectancy than is going by right now. As you live through a year, you get back about four months of life from scientific progress, so you're only losing about eight months of your life expectancy for each year that you live. But as I've said, medicine is advancing exponentially, so by around 2032, we'll be getting back a whole year of life as we live through a year, and after that, we'll get back more than a year back for each year that we live, so we'll be going backwards in time as far as our health is concerned.
As we merge with AI in this way, we will become a hybrid species. We'll still be human, but we'll be enhanced by AI.
Do you consider people who have cochlear implants or pacemakers or prosthetic limbs to be less alive?
Of course not. These are early examples of merging with machines. Our machines are going to continue to shrink in size and gain in power to the point where they are invisible and inside our bodies. As we merge with technology, we will no longer be limited by our biology. We'll be free to live life Without Limits.
People will have their own cerebral cortex and an artificial cerebral cortex. Thanks to this artificial cortex, will they be able to search for information just like we do today using Google?
Well, let me start by explaining how we're going to connect our brains to the cloud. Nanotechnology is a field that's manipulating materials measured in nanometers. One nanometer is one billionth of a meter. For perspective, the head of a pin is about a million nanometers wide. Scientists are in the early days of figuring out how to build nanorobots the size of cells that will function like today's robots—sensing data, processing information, taking action, communicating with each other—all on a molecular level. This idea might sound futuristic, but my research shows a steady trend leading to a nanotechnology revolution in the next 15 years. The amount of computation that once took up an entire floor of a building now fits on a smartphone in your pocket, and soon what now fits in your pocket will fit inside a blood cell and will be far more powerful. In the 2030s and 2040s, nanobots will swim in our bloodstream; they'll perform medical tasks with precision, deliver drugs straight to the source, drill through clogged arteries. Ultimately, they will go into our brains non-invasively through our capillaries, provide wireless communication between our neocortex, which is the top layer of our brains, and additional digital neurons hosted in the cloud. Think of it like having your phone, but in your brain. If you ask a question, your brain will be able to go out to the cloud, similar to the way you do on your phone now, only it will be instant; there won't be any input or output issues, and you won't realize it has been done. The answer just will appear in your brain like it is part of you. Some people say they don't want nanobots in their body, but lots of people didn't want to use early cell phones either, yet today they take them everywhere and never leave home without them. Once we connect our brains to the cloud, our intelligence won't be limited by the small size of our skulls, and ultimately, it will expand a millionfold.
Have you ever wondered whether AI will be able to feel pain?
The lines between AI and biology are going to blur as we merge together, so it will be difficult to determine whether our pain is felt biologically or digitally. This is similar to the question of consciousness: Will AI be conscious? There's no definitive scientific test to confirm or deny consciousness. It's a philosophical question. We assume that each of us is conscious if we seem conscious, but that shared human assumption breaks down as we start to move away from humanity. Some people share the idea that animals are conscious, but not everybody does, and that's where the whole issue of animal rights comes from. Are insects conscious? There's really no way to prove it. In the future, AI will act as though it is conscious, and therefore we will treat it as though it is.
You currently work at Google. We don't know much about your work during a public sphere. Can you tell us about the last important project you are working on at Google?
I've been with Google for over 12 years. I was first involved with natural language processing. If you've ever used Gmail, you'll know that your phone will give you three auto-reply options to choose from when responding to a message. That was created by my group. It's part of a larger trend to make technology more intuitive and accessible, and it's an early example of breaking down barriers between computers and humans. A more recent example is using large language models to write code based on our natural language descriptions. In the past, we've had to invest hundreds of hours learning how to code to create new programs. The same is true in the arts. My father was a composer, and when he wanted to hear his compositions, he had to hire an entire orchestra to play them, which was very expensive and time-consuming. If he were alive today, he'd simply play them with a software program in seconds at no cost. Old barriers like this have made it difficult to bring creative ideas to fruition, but with AI, composers will be able to hum a tune and then work with the digital Beethoven to orchestrate it into a symphony.
In 2011, you were ranked high in the Time weekly ranking of the most influential people in the world. Why do you think that is?
Actually, in 2024, I was honored to be on Time's list of the 100 most influential people in AI. They broke the list down into a handful of categories, and I was number one in the thinkers category. There was a feature article in the magazine that said they chose me because of my accurate predictions over the past few decades, including my 1999 prediction that AI would reach human-level intelligence by 2029. I've been thinking and writing about the future of technology for decades, and I've created a framework called the Law of Accelerating Returns to understand the exponential nature of information technology. This has spawned a worldwide movement of movies, plays, songs, books about the merging of human and machine. I'm often asked what my advice is for the next generation, and my answer is somewhat old-fashioned: My advice is to pursue your passions. And despite pushback from skeptics, you need to be most concerned with what turns you on, what drives you forward, and what excites you, because you'll have more ability to do that with the support of AI. As we merge with AI, our creativity and innovation will grow. We'll be able to create things that weren't possible before.
You are the author of the book about the singularity. What is the singularity, and how will it change our lives?
The Singularity is a point when humans fully merge with artificial intelligence by connecting our brains to the cloud and expanding our intelligence a millionfold. I believe it will happen by 2045. We borrow the term Singularity from physics because it represents a transformation so profound and rapid that it's difficult to comprehend with our current biological minds. It's like asking a mouse what it's like to be human. The mouse doesn't understand the question, let alone have an answer. Once we connect our brains to the cloud, we'll be freed from biological limitations. We'll choose and change our physical appearance, customize natural and virtual environments, shaped with human and AI friends. We'll tap into creativity and skills that are unimaginable today, like visualizing shapes in 11 dimensions or speaking every language, which AI can already do. We'll have the capability to recreate ourselves and our world. The pace of technological change will be so rapid and its impact so deep that human life will be changed forever.
"It was a great honor. Thank you for your time."
"Thank you, Ma. It's been a pleasure talking with you."