What one founder's past says about AI's future

0:00This message comes from AT &T. Staying connected matters. That's why AT &T has connectivity you can depend on. or they'll proactively make it right. That's the AT &T Guarantee. Terms and conditions apply. Visit att.com slash guarantee for details. This is the TED Radio Hour. Each week, groundbreaking TED Talks. Our job now is to dream big. Delivered at TED conferences. To bring about the future we want to see. Around the world. To understand who we are. From those talks, we bring you speakers and ideas that will surprise you. You just don't know what you're going to find. Challenge you. We truly have to ask ourselves, like, why is it noteworthy?

0:44And even change you. I literally feel like I'm a different person. Yes. Do you feel that way? Ideas worth spreading. From TED and NPR. I'm Manoush Zomorodi. It was about a year after 9-11. This is Mustafa Suleiman. In 2002, Mustafa was a student at Oxford University when a friend decided to start a helpline for young British Muslims. A good friend of mine at the time had just started the first prototype one evening session. I think it was a Thursday evening of offering counseling services on the phone. Mustafa was intrigued. He ended up joining the project to help grow it. And we became essentially co-founders.

1:38Quickly, their service was in hot demand. You know, it was a pretty remarkable time because, you know, young British Muslims were feeling judged by 9-11 as though they were responsible or somehow complicit. There was a lot of Islamophobia. But what they found these young people needed was really just someone to talk to. Most of the challenges that they were working through were, you know, the sort of conventional things that, you know, a young person might deal with, like relationships, bullying, self-harm, increasing homophobia in the community. and fundamentally I think people were just looking to find support in a language that made sense to them, like a London accent using sort of Islamic terminology and having a broader understanding of whether it was a Bangladeshi or Pakistani culture or an Arab culture combined with the religious foundation and just being able to reflect back that empathy went a long way.

2:4020 years later, the Muslim youth helpline is still going strong. And Mustafa Suleiman now lives in Silicon Valley, where he's been working on another kind of helper. I'm the CEO of Microsoft AI, and I work on building new AI systems and products. These days I'm working on an AI companion. With artificial intelligence and the backing of one of the biggest tech companies in the world, Mustafa Suleiman wants us each to have our own digital assistant. I can get philosophical, creative, or poetic if you like. Available 24-7, kind of like a personal helpline. I can be your sounding board as you work through challenges.

3:25I can write stories, brainstorm ideas. Think of me as your superpower in your pocket. How can it serve you? How can it support you? How can it save you time? How can it reduce your anxiety? How can it help make you feel smarter and more confident and more prepared to go about your day with no judgment, no pressure? Hey, Manoush. Nice to hear from you. What's on your mind today? Microsoft calls this assistant co-pilot, and it was released at the end of 2023. I've been playing around with it. Well, I need to interview Mustafa Suleiman, your creator. What do you think I should ask him? I go for topics like his vision for AI's role in society, the balance between innovation and ethics, and what he sees as the most exciting future possibilities for AI.

4:20Okay, these are good. Now, can you please write the intro for this episode? Sure thing. Here's a possible intro. Hey, it's Manoush Samarodi, and today's TED Radio Hour is dedicated to a deep dive with none other than Mustafa Suleiman. Co-founder of DeepMind and author of The Coming Wave, Mustafa has been a pivotal force in AI, driving both groundbreaking advancements and sparking intense debates. In this episode, we'll discuss the promises and perils of artificial intelligence and whether we're truly prepared for the wave of change it's bringing. Is AI evolving too fast for society to adapt? Are we doing enough to mitigate the risks?

5:07It's an hour of insight, reflection, and some tough questions. So, let's jump in. Okay, that was pretty good. As Co-Pilot noted, Mustafa Suleiman is a longtime pioneer in the field of AI. After he dropped out of Oxford University, he worked for London's mayor. He was even part of an NGO's delegation to the Copenhagen Climate Negotiations in 2009. He wasn't impressed. There were so many sort of difficult egos and entrenched sort of interests. And I just felt a sense of sort of frustration and I felt kind of demoralized after that experience. Mustafa wanted to change the world. Governments and nonprofits didn't seem terribly effective.

6:00But there was one company that was managing to change the behavior of millions of people. Facebook was exploding at that time. It had got to 100 million monthly active users in the course of a couple of years. And it was pretty clear to me that that was going to have a profound impact, more so than anything I could do in, you know, sort of the world of social activism. Around then, Mustafa started hanging out with the older brother of a friend, a software genius named Demis Hassabis, who had been designing computer games since he was a teen. Demis saw how to make Mustafa's vision possible. And he was coming at the world from a completely different perspective, which I found really interesting.

6:49His belief was that we would just simulate, you know, complexity in the world. This new suite of methods in AI self-learning systems were coming up strong and looking likely to work. And, you know, we really just clicked on that kind of technical and sociocultural view of sort of how to positively impact the world. Together, the two decided to start one of the first AI companies ever. They called it DeepMind. It was very clear from that, even those early stages, that if we got the technology right and it was going to be this decade that, you know, led to major, major breakthroughs, then the consequences for humanity were going to be significant.

7:36I've been lucky enough to be working on AI for almost 15 years now. Mustafa Suleyman picks up the story from the TED stage. Back when I started, to describe it as fringe would be an understatement. Researchers would say, no, no, we're only working on machine learning, because working on AI was seen as way too out there. In 2010, just the very mention of the phrase AGI, artificial general intelligence, would get you some seriously strange looks, and even a cold shoulder. You're actually building AGI, people would say. Isn't that something out of science fiction? People thought it was 50 years away, or 100 years away, if it was even possible at all.

8:19talk of AI was, I guess, kind of embarrassing. People generally thought we were weird. And I guess in some ways we kind of were. The ironic thing is that many people still don't think it's possible. Many people still think that we're crazy. And at the time, people really thought we were crazy. I mean, it was so far out there. It was really strange. And we were a strange group of people. I mean, you know, Demis, our third co-founder, Shane, who, you know, basically a mathematician, spent his entire career thinking about how to formalize a definition of intelligence that we could use to measure our progress in the company.

9:04We were we were misfits and kind of outsiders and there weren't very many people willing to back us. So naming the company DeepMind, what was it that you pictured in your mind that you hoped to achieve? Yeah, I mean, we wrote our business plan in the summer of 2010 and took it to Silicon Valley to shop it around a bunch of people. And the strapline for the business plan was, you know, building artificial general intelligence safely and ethically. And then that evolved into a two line mission, which was solve intelligence and use it to make the world a better place. and that dual frame was kind of the foundation of the company.

9:53Our belief that science and technology was the engine of progress and there are some downsides, but I certainly think this is the engine of creating civilization in a more healthy and sustainable way for the very long term. And if you think about it, everything that is of value in our world today is a product of us humans being able to take materials from our environment and recompose those into products and services and other compounds that are useful to us, from our lights to our carpets to everything that you see in front of you today. In 2014, DeepMind ended up being acquired by Google. That must have been huge for you in terms of money, resources.

10:44You were off to the races. Yeah, it was a huge event. I mean, it was the largest acquisition Google's ever made outside of the U.S. We became Google's primary AGI bet. And we were empowered with tremendous resources, both people and computation, to go and both do the hard science, but also work on really important applied practical product problems. And that's where I kind of really honed my craft, if you like. As a product maker, it was just the most amazing experience. And as early as 2015, I actually ran a hackathon project in my applied group at DeepMind. And the theme of our hackathon was to find high-impact ways of using our technologies for good.

11:38And so there were lots of application, there were sort of lots of prototype hackathon experiments in healthcare or in energy systems, both of which went on to become significant parts of DeepMind Applied. And my group published three papers in Nature showing human level performance on, for example, classifying eye diseases, the top 50 eye diseases from OCT scans, three dimensional eye scans, showing that we could perform as well as a panel of radiologists in identifying cancerous tissues in mammograms. showing that we could predict the onset of sepsis and acute kidney injury, as well as the top, you know, renal doctors using vast amounts of data.

12:33And this was way back in sort of 2016, 2017, 2018, and really helped to kind of lay a foundation for the application of large-scale machine learning to, you know, tough social problems. And that was very much my motivation.

12:53In a minute, more of the incredible breakthroughs that Mustafa and DeepMind had at Google that changed the way medicine works today and why Mustafa ended up leaving the company. The company was just, frankly, being too slow to get things into production. Today on the show, the CEO of Microsoft AI, Mustafa Suleiman and the future of artificial intelligence. I'm Manoush Zamarodi, and you're listening to the TED Radio Hour from NPR. PR. Stay with us.

13:32This message comes from the Michael J. Fox Foundation. Did you know that every six minutes someone in the U.S. is diagnosed with Parkinson's? It's the fastest growing brain disease in the world. Chances are you or someone you know is impacted by Parkinson's. The Michael J. Fox Foundation is on a mission to find a cure. Now is a critical time for Parkinson's research, and your support will help drive progress toward a cure. Get involved today. Visit michaeljfox.org slash get involved. This message comes from Charles Schwab with their original podcast, Choiceology. Choiceology is a show about the psychology and economics behind people's decisions.

14:12Download the latest episode and subscribe at schwab.com slash podcast. This message comes from Michelin. More than a tire company, Michelin is an innovation company. Now taking on space, developing an airless wheel for space exploration. Motion for life. More at michelinman.com slash y hyphen Michelin slash innovation. This message comes from Dell Technologies. Dell AI PCs are newly designed to help you do more faster. That's the power of Dell AI powered by Intel Core Ultra processors. Upgrade today by visiting Dell.com slash deals. Hey, it's Manoush. This past year, we have brought you stories about AI, relationships, climate change, neurotechnology, dinosaurs, privacy, human behavior, and even one about what it means to create thriving public spaces.

15:11All of these public spaces that we take for granted, you know, all of the social infrastructure. We work really hard to bring you all of these stories because that is our public service. Kind of like a park or a public library. And libraries are these beautiful hubs that can take on the shape of whatever community that uses them really needs. Public media is infrastructure that we all can use. It's free and it's for everyone. And our mission is to help create a more informed public. So this time of year, I want to thank you for listening and to say that one of the best ways you can support this public service is by signing up for NPR+.

15:54With one subscription, you are supporting all the work that NPR does and you get sponsor-free episodes and perks across NPR's catalog of 25 podcasts. It's a great time to join because throughout December and January, Plus listeners have access to even more insights from TED speakers to help you kick off 2025 right. From making big life decisions to being more hopeful to carving out time for what is important to you, we have got you covered. Just visit plus.npr.org. You can also find the link in our episode notes. And the other way you can give is to make a donation at donate.npr.org. Your gifts are tax deductible either way.

16:44Thank you so much for being here. And now let's get back to the show.

16:51It's the TED Radio Hour from NPR. I'm Manoush Zamorodi. Today on the show, we're spending the hour with Mustafa Suleiman, the CEO of Microsoft AI, who was the co-founder of one of the first AI tech companies ever, DeepMind. The company was acquired by Google in 2014. And Mustafa and his co-founder Demis Hassabis went on to have numerous scientific breakthroughs, including a project called AlphaFold. AlphaFold uses AI to figure out incredibly complex protein structures in molecules, the building blocks of every biological process in our bodies. It won Hassabis the 2024 Nobel Prize in Chemistry.

17:37Here he is on the TED stage. The rule of thumb is that it takes one PhD student, their whole PhD, so four or five years, to uncover one structure. But there are 200 million proteins known to nature. So you could just, you know, just take forever to do that. And so we managed to actually fold using AlphaFold in one year all those 200 million proteins known to science. So that's a billion years of PhD time. AlphaFold was just one of the ways that computers were solving biological problems, says Mustafa Suleiman. I mean, this really, AlphaFold, is really the first time we started hearing about, I guess, the term computational biology, this idea of using tech and science to rethink how biology works and then getting it out into the world, changing the way we treat diseases or maybe developing crops that are more resilient, all kinds of ways that what you figured out at DeepMind would potentially change the world, which was the goal.

18:41Absolutely. That was the core mission of the company from day one. How could we take that process of synthesis and prediction and turn that into a general purpose system that could use all the data and all the compute that we have in the world to try and solve these very hard problems, whether it's growing crops in drought resistant environments, whether it's more efficient ways to do water desalination, whether it's long term battery storage. But in 2022, Mustafa decided to leave DeepMind and Google. There were reports that he clashed with Google leadership over using technology for military projects, and that some employees had complaints about his management style.

19:29Mustafa says he was simply frustrated with the pace of innovation and was anxious to get an AI product out into the world that everyone could use. You know, I really wanted to scale these large language models and get them into production as quickly as possible. And the company was just, frankly, being too slow to get things into production. And I felt that it was an opportunity to sort of have all the resources that, you know, that I needed. So I went out and fundraised over a billion dollars and, you know, just work extremely fast with a very small team. And I think at the time, I was and I am still now very much a believer that these AI companions are going to become an everyday part of our lives.

20:27Mustafa's AI called Pi debuted in 2023. Chatbots have taken the world by storm. Several months later, rival company OpenAI put out its AI product, ChatGPT, for free. ChatGPT answers questions and writes essays. They read like the work of a human. An artificial intelligence tool has the potential to change the way we live. This was the point when AI went mainstream. And Mustafa's small, nimble startup just didn't have the resources of these bigger tech companies. The truth is the pace of big tech had really accelerated in 2023. Google, Microsoft, OpenAI, everyone was really going for it and essentially made these huge models available to everybody for free, which sort of changed our business model.

21:21At that very tough moment, Microsoft CEO Satya Nadella approached Mustafa and made him an offer. You know, the offer that he made was, well, look, we've got all the computation and all the data and all the distribution that you could dream of. Come and run consumer products here at Microsoft and really build the future of AI here. And that was a huge offer. So the product that we're building at Microsoft is called Copilot. And the reason why we've called it that is because Copilot is an aid. You know, it's a conciliary. It's an assistant. It's in your corner, aligned to your interests, on your team, backing you up.

22:02Yeah, I actually talked to Copilot about my interview with you. How do you feel about Mustafa Suleiman, considering that he is your creator? I'd say I'm intrigued by Mustafa. Just so you know, she said she's intrigued by you, just as she is with many innovators in the tech world. And his work has pushed the boundaries of what's possible. I will say it was very helpful in terms of suggesting topics to bring up with you, but it could not check the weather. And is that simply because of how the access or the data, the information that the agent or co-pilot has access to, that it's not immediate yet?

22:43Yeah, there are loads of weaknesses of our co-pilot today, but all of those things will come. I mean, we're working on all of those things. It will have permanent memory, session to session, infinitely across time. It'll be able to access your files and folders if you choose to make those available, your email, your calendar, and be able to browse the web for you in the future. and all of that will be integrated into these like seamless conversational, very friendly, polite experiences. I literally was talking to it last night about what to watch and we were going back and forth on whether I would enjoy Pan's Labyrinth and whether I've got the time to watch all of Dune because I haven't seen Dune 1 or Dune 2.

23:26I mean, it reminds me a little of the hotline for Muslim youth that you're describing. It's helpful, it's infinitely patient, It's supportive. Are we talking mostly about companionship and mental health resources that this can provide? Or how do you see it? No, I think that the cool thing about Copilot is that it doesn't judge you for asking a stupid question. Even if you have to ask that question three times over in five different ways, you know, even your best friend might be like, hey, come on, man. I mean, you're asking me this again? Seriously? whereas, you know, CoPilot's here for it. There's obviously some similarities to stuff I've done in the past, and I guess it's kind of inspired by nonviolent communication, if I'm honest with you.

24:14It's certainly not like a mental health app or, you know, anything like that. It's just got a little bit of kindness and empathy. It's got some emotional intelligence, and I think that's no bad thing. Gosh, is that where we've gotten to, that technology has to tell us how to communicate with each other better? non-violently? Well, it doesn't tell us, it just... Demonstrates. Yeah, exactly. It demonstrates. But that's what technology has always done. The choice architecture, the buttons, the colors, the language, that is shaping our behavior. Whether it's an infinite scrolling feed or whether it's an encouragement to go and film your lunch for Instagram or create a little video for TikTok, I mean, all of those inputs shape behavior.

25:07And so we have to be super thoughtful about what those inputs actually are, because technology shapes us in return, and we're in this constant cyclical sort of feedback loop and interaction. And that is kind of what's propelling us forward as a civilization, and it's very powerful. And so far, so good. It's actually been, you know, very, very productive over the last couple of centuries. Science has massively delivered for us. But we shouldn't just assume that that's going to happen naturally or inevitably. We have to really be deliberate and thoughtful about what the consequences are ahead of time.

25:48In the book that you wrote that came out in 2023, you really tried to put what's happening with AI in a historical context. So if the printing press let people own and share information and the personal computer let people search and disseminate information, tell me how you're thinking you can explain to people what AI will do for people now. Each new wave of technology is fundamentally a new interface. It's a new interlocutor, a translator, a way of sort of accessing and creating new information, new tools, new knowledge. So if the last wave of social media and web search help people to access information, this wave is going to help us to invent and create new ideas, be it in science or in culture and media and entertainment.

26:46And I think everybody is ultimately going to have an AI companion, just as we have a search engine or a smartphone. And just as we use a browser, you'll just ask your computer in natural language, you know, can you write that contract and check that it's OK? Can you create that new piece of software for me? And you're just going to describe what it is. Can you help me plan that trip, you know, for my parents that are coming into town? So, you know, that kind of breakthrough is a change in the interface, which changes itself what we can actually get done. And I think it's going to be pretty transformational.

27:26With the invention of computers, we quickly jumped from the first mainframes and transistors to today's smartphones and virtual reality headsets. Information, knowledge, communication, computation. In this revolution, creation has exploded like never before. And now a new wave is upon us, artificial intelligence. These waves of history are clearly speeding up as each one is amplified and accelerated by the last. And if you look back, it's clear that we are in the fastest and most consequential wave ever. The journeys of humanity and technology are now deeply intertwined. In just 18 months, over a billion people have used large language models.

28:14We've witnessed one landmark event after another. Just a few years ago, people said that AI would never be creative. And yet AI now feels like an endless river of creativity, making poetry and images and music and video that stretch the imagination. People said it would never be empathetic. And yet today, millions of people enjoy meaningful conversations with AIs, talking about their hopes and dreams and helping them work through difficult emotional challenges. AIs can now drive cars, manage energy grids, and even invent new molecules. Just a few years ago, each of these was impossible. And all of this is turbocharged by spiraling exponentials of data and computation.

29:03Last year, Inflection 2.5, our last model, used 5 billion times more computation than the deep-mined AI that beat the old-school Atari games just over 10 years ago. That's nine orders of magnitude more computation. 10x per year, every year, for almost a decade. Over the same time, the size of these models has grown, from first tens of millions of parameters to then billions of parameters, and very soon, tens of trillions of parameters. If someone did nothing but read 24 hours a day for their entire life, they'd consume 8 billion words. And of course, that's a lot of words. But today, the most advanced AIs consume more than 8 trillion words in a single month of training.

29:56And all of this is set to continue. The long arc of technological history is now in an extraordinary new phase.

30:08I think the way to think about this is that we're at the very earliest stages of the development of this technology. Today, when you need to go and sort of ask a question like that, you go to a search engine like Google, you type in a query, you get these 10 blue links. You have to then go to the web page. You have to read all of this sort of complicated stuff formatted in 25 different ways. And that takes time. And it means that you don't always want to invest two or three minutes to go and make sense of that. You don't always have the energy for it. Whereas you can always just like quickly send off a text or quickly make a phone call or leave a voice note.