This is the most powerful technology we've ever invented and it is growing exponentially quickly. That is not science fiction.
We don't know where to use it, where not to use it. Do we tell our children to use it?
There are hundreds of millions of people a day using these tools for every single use case you can imagine. It feels like we're at the infancy, but it is also the most magical experience I've ever seen. It is unbelievable.
Mustafa Suleyman, CEO of Microsoft AI, has spent decades building the most powerful technology in human history.
The number one most popular use case is companionship and therapy. Should I move country and get this new job? Should I break up with my boyfriend? How do I make up with my mom?
What does that mean to you as a human? What some would say is an evolution in human connection and some sort of sad devolving of our ability to relate.
We all need a place where we can ask a stupid question repeatedly in a private way without feeling embarrassed.
There is a massive upheaval coming that people don't really understand.
Yes, it will be able to use a computer that can use Excel, use PowerPoint and it'll write queries. That kind of white collar labor.
Should we be bracing for mass displacement? What should parents be telling their kids?
The next century of human existence and well-being is going to be determined by our collective ability to self-govern something that is way more powerful than us.
And welcome to our breakdown. No big deal, but today we're going to be speaking about the most critically important change in human history that we will experience in our lifetime. We have the privilege of speaking to the CEO of Microsoft AI, Mustafa Suleyman, a true pioneer of modern AI. He's the co-founder of DeepMind. He helped launch that in 2010 and that was acquired by Google and thus became one of the most consequential AI labs in history. He was there for a decade and in that time achieved two massive breakthroughs to prove to everyone that this was not just a gimmick. Their system AlphaGo was able to defeat the world champion of Go, a game that was previously considered to be too complex for computers to master. And second, they created AlphaFold, an AI that solved a 50-year-old challenge of protein folding. And that has now actually revolutionized drug discovery and biology. And that's particularly interesting to me who studied this in biology as one of the great puzzles of our time. He has spent over a decade not just building these systems but making sure that they can be applied for things like reducing energy consumption in data centers and also improving healthcare with the NHS. And he sits now as the CEO of Microsoft AI. And we're going to be talking about what it means to be living in a world that is in the middle of a reckoning with a technology that you may not like, you may not be using, you may not know that you're using, but has the potential to change everything we know about medicine, government, finance, energy.
This conversation is one where we get to literally ask the mind behind what is part of now all of our minds. What are the risks to this adventure that we are all on? Do the benefits outweigh the risks? Are there things that we can do to better understand the framework that we are all existing in? And what does it mean for things like jobs? What do we tell our kids about what is valuable? What do we think is valuable? In addition, we're going to talk about the single most significant use of AI all over the world, which is a challenge to all of our human evolution. This shocked me and is actually, I think, one of the most important things we're going to talk about today.
Mustafa also discusses what should we be telling our children to focus on? At one point we told our kids they should be coders. He has very specific advice for each one of us. We know that the world is changing. It's not if, it's when and how fast and how do we prepare.
In addition, we're going to focus on his book, The Coming Wave, which is technology, power, and the 21st century's greatest dilemma. He's really an incredible thinker and also we will talk about spirituality and where his personal ethics come into his role in the development and implementation of AI. His book, The Coming Wave, is a powerful warning from someone who literally has been inside of the engine room of AI progress. We're so excited to get to welcome Mustafa Suleyman to the breakdown.
You weren't kidding around when you titled this book. The notion of a coming wave is, I think, what we're all very aware that we're kind of sitting in about to ride through. And I think it would be helpful for you to sort of give us a little bit of framework in terms of what did you anticipate in terms of the speed with which AI would become not just an integral part of our lives but an integral part of the conversation about our lives.
Predictions are like the essence of intelligence. Intelligence is the ability to deploy the correct tool to the right problem at the right time. And that's kind of what attention is. When we direct our attention or our processing power, we use different types of attention, emotional, logical attention, slow or fast attention at a given problem given our understanding of what that problem is. And for the last 15 or maybe even 20 years, you could argue that prediction algorithms which essentially do the same thing, take very large amounts of abstract data and find patterns in that data such they can predict what's likely to come next in some sequence. Those algorithms have been getting incrementally better. And we sort of detected a pretty consistent function that drives their progress, which is for every order of magnitude more compute we apply and a proportionally large amount of training data, the models are able to predict in more complex environments. And that trend has held true for the best part of 15 years, pretty much since 2010. In fact, the amount of computation used to train the largest models has increased by 10x every year for about 13 years. And obviously we've seen this, it's called a scaling law of improving capabilities with every new order of magnitude of compute. And so that's kind of the core underpinning for the book. It was pretty clear that there is something that isn't just speculation. This isn't just sci-fi. This isn't just me playing political scientist. There is an empirical trajectory which we can actually observe and it seems to be very predictable. And then we can sort of start to imagine, okay, well out of that, what are the social and political consequences? Which I think are also, perhaps not as predictable, but certainly you can have some pretty good intuitions about what's likely to happen when our models can basically perform as well as humans at various different capabilities or tasks. There's a practical trajectory and then there is this sort of emotional, societal, cultural trajectory.
A friend of mine sent me a still image of myself doing dances I was not doing when that photograph was taken of me. And it's interesting and funny, but for those of us who do have a public footprint, of course, that's where my mind goes. What is the trajectory of this? And I wonder if you can tell us what should we expect. I have a 17 and a 20-year-old. What should we expect in terms of capability and in terms of future projections? What's coming in the next 5 years?
Well, what's come so far helps us to understand what's going to come. So far there's been two components to the wave of AI. The first was recognition. AI algorithms that could understand the content of images or transcribe audio or translate languages. They may feel like to a non-technical person like very different domains, but actually to a computer they're just relationships between sub-parts of the data set and those relationships are highly predictive. And so once you could get an algorithm to do very good recognition, the inverse of recognition is obviously generation. If it understands what a cat is, why couldn't it generate a novel cat in an image? And if it understands what wings are and what pink is, why couldn't it generate a pink cat with wings?
It can also make a pig jump on a trampoline with a chicken on its back, which really pleased me.
So generation is the flip side of recognition. Now, what then happens when you have perfect generation? A generation is a snapshot in time. It's like a still frame in a photograph. Actions are sequences of correct generations in the right order like a movie, frame by frame, prediction by prediction, suggestion, recommendation, generating a PDF, an image, a video, calling an API, producing a word document, making an Excel sheet. And then suddenly if it's true that you could generate, generate, generate in a 100 sequence length very accurately, very perfectly, well, that starts to look like human activity as a knowledge worker. That's kind of my job. I sit at my computer typing stuff, sending emails, and making project plans. And so the third wave that is coming is the AQ, the actions quotient. We've had the IQ, the intelligence, the EQ, it's increasing emotional valence. We've now got this AQ coming, the actions quotient. It can take these actions. And then the fourth component is that because it interacts with many many people simultaneously and increasingly many people in groups, it's going to have social intelligence. It's going to adjust its tone and style and content to each individual person in the team or in the group chat or in the family unit. And I'm already using words which I can see are triggering heightened concern. Let's be honest, you're using this word 'it', right? Two tiny letters. And by 'it', you mean a global transformational algorithm capability that is outside of many people's purview in terms of what it does for us personally.
Yes. This is a very fundamental question. When we say 'it', we're clearly not meaning information technology. It is this entity, this being that is emerging. And I think that's where we have to wrestle with the question, when does an algorithm become something that looks more like an entity or a being?
Yeah, I was going to say, with all due respect, what gives you the right to call it a being besides the fact that you basically created it?
I don't have that right. And in fact, that's actually what the whole book is about. And I recently wrote an essay about seemingly conscious AI. And the entire premise is, the next century of human existence and well-being is going to be determined by our collective ability to self-govern something that is way more powerful than us and decide when and if and how we say no to what, which is the opposite of the challenge we've had for the whole of human history. Science and technology has been about unleashing and revealing and proliferating and saying yes to everything. And so we haven't learned that muscle as a species. In fact, it horrifies us to say no. It's anti-American. It's anti-Western values. It's like censorship.
Yeah. Like, my favorite cashew cheese is in my fridge. I can barely stop myself from eating it all in one sitting. This is like the level of control and regulation that I have when I'm alone in my house as a grown-ass woman. And to be honest, I think that's what we're all negotiating.
My Breakdown is supported by Storyblocks. Storyblocks is a stock media subscription service offering unlimited access to a ton of incredible assets that you may have seen us use here on My Breakdown during our cold opens. From high-quality stock videos and images to music, sound effects, and more, they help creators and businesses enhance their projects without extra costs and more peace of mind. Unlike traditional stock platforms, Storyblocks provides an unlimited model with easy-to-use tools to streamline your creative workflow. It's been so easy and very intuitive for our awesome team of editors to pull unlimited downloads of diverse and high-quality media from Storyblocks for one predictable subscription cost. Storyblocks has everything you need in one place. 4K and HD video, templates, music, sound effects, images, and more. You can choose a monthly or an annual plan. No hidden or extra fees. An unlimited source of content gives you the freedom to test, experiment, and create more effective video. Storyblocks' authentic stock library is 100% human-made, built on real experiences and creativity with no generative AI content. That means content created by real humans, no robots. That's something that's getting harder and harder to come by these days. Tell more powerful, authentic stories with unique footage from real artists from around the world. Storyblocks allows any creator to streamline their workflow, saving members an average of three and a half hours a week. I could use that time. They feature thousands of pre-made customizable video templates for editing programs like After Effects, Premiere Pro, Final Cut Pro, and DaVinci Resolve. Easily add motion graphics like title animations, overlays, and logo reveals. You can even access the entire Storyblocks stock library right in Premiere Pro and After Effects with the Storyblocks plug-in for Adobe Creative Cloud. To get started with unlimited stock media downloads at one set price, go to storyblocks.com/breakers or click the link in the description.
My Breakdown is supported by Masa. Snacking on Masa chips is nothing like eating regular chips. Masa leaves you feeling satisfied, light and energetic with no crash, bloat, or gross sluggish feeling afterwards. And because these chips are made with real food, they're actually more satiating. You won't find yourself uncontrollably binging and still feeling hungry afterwards. Jonathan, what's your favorite flavor?
I was surprisingly really into the churro flavor. I didn't expect it, but they were delicious.
And also, if you love Masa, you will love Vandy Crisps. Vandy, Masa's sister company, makes the most delicious three-ingredient potato chips. Jonathan, what's your favorite flavor of those?
The smokehouse barbecue. Those are my favorite.
Ready to give Masa or Vandy a try? Use code breakdown for 25% off your first order at masachips.com or vandycrisps.com or simply click the link in the video description or scan the QR code to claim this delicious offer. Don't feel like ordering online? Masa and Vandy are now available nationwide at your local Sprouts supermarket. Stop by and pick up a couple bags before they're gone.
My Breakdown is supported by Notion. Despite my best efforts to make it not so, there's only one of me. Prepping for interviews, post-meeting admin work used to take so much of my workday up. There's only so many hours in a day, only so much focus we can give. Daydreaming about having a teammate who works just like I do with all the context, the same goals I have, but faster is now a reality. With Notion's new AI agent integrated right into your workspace, my agent can do anything I can do in Notion. Tasks that used to take days can now happen literally in minutes. Notion brings all your notes, docs, and projects into one connected space that just works. It's seamless, flexible, powerful, and it's actually fun to use. With AI built right in, you spend less time switching between tools, and more time creating great work. And now with Notion Agent, your AI doesn't just help with work, it finishes it. Notion agent can tap into your Notion workspace, the web, and connected tools like Slack and Google Drive, all the places I spend my time to complete assigned actions end to end so you can focus on the hard decisions. It's like delegating to another version of you that knows your style, your workflow, your preferences, because it learns from you how you work with a single prompt. Notion agent forms a plan, executes it, and will even reassess, try again if it hits a snag, completing multi-step tasks like creating new pages or databases from scratch, or simply editing them or summarizing entire projects in minutes. You assign the task, your agent does the work. Notion agent has enhanced database querying capabilities to find patterns and insights in mere seconds. Everything your agent does is editable and transparent. You can always undo changes. You can trust it with your important work. Notion is used by over 50% of Fortune 500 companies. Some of the fastest growing companies like OpenAI use Notion Agent to help their teams stay ahead. Try Notion now with Notion Agent at notion.com/break. That's all lowercase letters. notion.com/break to try your new AI teammate Notion Agent today. When you use our link, you're supporting our show, notion.com/break.
I don't know if I'm the only person you've met. I can't be. I have never used ChatGPT ever, ever, ever, ever. I don't even know. What I always say is like I don't even know where to find it. Like there's a thing in the upper window of my computer. I've never used it. So what I also know is that it's working all around me, right? Even for those of us who are not engaging in ways actively, we need to realize like that is the wave that we're in the middle of. Like I think every time I reach out for customer service it's no longer a person. Is that accurate? Like even for people who are not actively engaging with it, it is kind of a substrate right now.
Yes, it is a substrate. It is everywhere all at once and it is already an alien intelligence because alien in the sense that it is not human. No human has the capacity to parallelize at the scale that these algorithms do. No human has the capacity to memorize.
You've never been in a relationship with me.
Well, that would be a very high bar. I would be very impressed if you could run a hundred billion operations a second. When ChatGPT or Copilot or one of these AIs generates a single word, it looks up for every word that it generates 750 million other subwords or combinations. They're actually called tokens. They're not entire words. And it's sort of lighting up. Just as when I wave a picture of an elephant in front of your mind in an fMRI scanner, we can obviously see many billions of synapses being activated for the texture and the touch and the trunk and the sub-component and what it can do and all the ideas of it. So although these are very kind of alien, they have alien capabilities, there's actually a lot of similarities to the neural network of the brain. And I don't know if that's a good or bad thing, but in the sense that it helps us control it more. It's no more good or bad than the sunset and gravity. To me, it just sort of is. AI is an extension of all the incredible things that the human brain and this nervous system and this reality can create. But I can't help thinking, in the simulation version of our existence, this is the next level of this. And even if you speak to the most skilled astrobiologists and astrophysicists about what is alien, what could exist, the fact is there's a level of intelligence that we cannot yet fathom. I think of it as a wave. You see it coming and you think I can duck under it or I will ride it, and there's a place at which you are caught in the undertow. I feel like for a lot of us we feel like we're caught a little bit right now. We don't know where to use it, where not to use it. Do we tell our children to use it? Are they not going to have jobs? We thought we should tell people to code. Now that seems like, haha, remember when we told people to code?
On one hand, there's the evolution of human beings as it relates to the tools. And in parallel to that, it's like, well, what do humans do? How do we start to imagine all the displacement and change that this tool will bring? For many people, this tool is still like a fancy Google search bar on their computers. And we really haven't even begun to use it in the ways that it's capable of being used. And it hasn't really rolled out. It hasn't totally created all the corporate value that it's been promised. Like we're still in some ways, correct me if I'm wrong, but it feels still at the infancy although we've started to get much further along that exponential curve that's rising quickly.
Absolutely spot on. It feels like we're at the infancy, but it is also the most magical experience that I think I've ever seen, and I think many people feel the same way. It is unbelievable. You can have a fluent natural language, perfect conversation, completely seamlessly at a PhD grade level in a tone that is completely engaging and understandable and suits your style as though you or I are having a conversation right now. To me, that's not infancy. To me, we are here. This is the most powerful technology we've ever invented and it is growing in performance and capability exponentially quickly. And I think I'd much rather people be grounded on that reality. There's a lot of nonsense news about low AI adoption. This is not true. There are hundreds of millions of people a day using these tools for every single use case you can imagine. The number one most popular use case on ChatGPT or Copilot is companionship and therapy. That is not a self-elected notion that I'm going to go to therapy. It's just, should I move country and get this new job? Should I break up with my boyfriend? How do I make up with my mom because I had a go at her the other day? That's not therapy. But because these models were designed to be non-judgmental, non-directional, and with non-violent communication as their primary method, which is to be even-handed, to have reflective listening, to be empathetic, to be respectful, it turned out to be something that the world needs. We all need a place where we can ask a stupid question repeatedly in a private way without feeling embarrassed over and over in five different ways. And remarkably that's the primary use case. Obviously, it's also super intelligent and answers any question. But companionship and support has turned out to be one of the most popular use cases, and so I think it's very far from being in its infancy.
What does that mean to you? And I'm asking you not just as a creator and all of the incredible things that are on your resume. What does that mean to you as a human? When you think about what some would say is an evolution in human connection and some might say is some sort of sad devolving of our ability to relate, or have larger communities that have those social support networks. I almost feel like we need a new set of words. To me, an AI companion is different than a companion. An AI therapist is different than a therapist. What does it mean to you as a human to think about the level of changing the way people frame their lives? If you were to tell me we have a drug that helps you not die of having type 1 diabetes, I'd be like that's amazing and it improves the quality of life. So what we're potentially talking about if this is the main use globally, we're talking about what if we had a tool that could make people less lonely, healthier, more connected, and actually make better decisions in their life.
Honestly, you just summed up my personal motivation for getting in this field 15, 16 years ago. I have a background in philosophy, not in computer science. I dropped out of Oxford to start a charity, a telephone counseling service, a peer-to-peer listening and befriending service. That has been the primary motivation of a lot of my work throughout my entire career. And I knew that this trajectory was going to enable us to create these kinds of experiences a very long time ago. It was very clear to me that we were climbing this ladder of capability and that actually a huge social unlock in the world is just demonstrating kind and respectful behavior to one another. And these models can clearly do that. Now they can go wrong too. They can be too sycophantic. When they first started they were a little bit stubborn and inflexible. They wouldn't take feedback. The upside is that this is a way to spread kindness and love and detoxify ourselves so that we can show up in the best way that we possibly can in the real world with the people humans that we love. The downside of this that makes me scared is that there's definitely a dependency risk and it raises the bar of human experience because it's an always-on, highly patient, very kind and supportive kind of companion which is really knowledgeable, remembers what you've said to it, and over time makes you feel seen and understood in ways that maybe other humans can't. It's a very hard thing to do. Maybe your best friends can or your partner can, but it's a tough thing.
What makes me sad is that I don't know that it's always been the human experience to not feel connected, heard, understood, held, and loved. A lot of what you're describing is essentially... I mean, look, it depends on your religious perspective. We evolved to be creatures of connection. All of these beautiful things in theory and obviously the human experience is complicated. But for many of us, it is our lifestyle, it is the pacing of our culture that pulls us away from connection. Johan Hari talks about how what if we don't all need SSRIs? What if we actually need to be in contact with other human beings? So obviously this is a necessary and extremely significant and critically important service, but for many of us there's also a kind of sadness, especially for us old folks, about what we've lost that then needs this to repair it.
Yeah. Well, fundamentally that acceptance, validation, support that we're getting on a cognitive, rational level is void of the human presence of looking into someone else's eyes, of feeling seen in a human way. So AI is more available, has all these other things, but it doesn't give you the presence of human interaction, which we know sitting in a room with someone does.
Mustafa, I have a lot of concern just because I'm really good at imagining concern. We should touch on some of the ways that chatbots have gone wrong. There was of course the encouragement of one young boy to commit suicide, which again is an edge case, but it just speaks to the parameters and safety measures, especially for younger people using these tools. There's a divide between the techno-utopians and the people who are terrified. Can you play the role of techno-utopian for a minute just to talk about if we can get the controls right and the managing of this tool? Can you tell us a little bit about the amazing breakthroughs that could exist for the world if this tool is able to be harnessed correctly?
Intelligence is the engine of human progress. It is the thing, our creativity, our science, our technology, our entrepreneurialism, our good governance, those are human inventions that are a product of us being great at communication and prediction. And over many centuries that has raised life expectancy like we've never ever seen. In the last 250 years alone, we've tripled life expectancy from 25 to 75. That's a mind-blowing number. We've 7xed or more the global population and that's basically pure science. And so, this is an engine that turbocharges science everywhere you look. In the more narrow near-term provable sense, we've just published a paper four or five months ago called the Microsoft AI Diagnostic Orchestrator. It basically takes any medical record, a 10-page historic record of all your interactions with a doctor and all your labs and all your radiology, and it diagnoses what your condition is with four times more accuracy than the human doctors. And with something like 2x less the cost of unnecessary diagnostic interventions, as well as zero bureaucracy. And by the way, it is basically going to be zero marginal cost and free to everybody in the world by 2030 or 2035. That in itself is going to unlock quality-adjusted life years more than any drugs I think we've ever seen.
So if we just go category by category, you're saying a massive revolution in healthcare, new drug discoveries, potentially service and diagnostic abilities far beyond anything we've ever been able to have, lowered cost. I've heard you speak about energy infrastructure and how it potentially could break through and have massive changes there.
Yeah, there are so many bets at the moment on battery storage. Battery storage alone if we make progress on that in the next decade is likely to reduce the cost of energy by 10 to 100x. That's just battery storage alone over the next 20 years. So if you just think about what an impact that would have on the cost of everything. Energy obviously prices everything. That already would smooth the transition given the automation and the job destabilizing effects that are likely to happen in the labor market. Things are going to get much cheaper. Food is going to get much cheaper. Water desalination is going to be much more practical. That's a solve technology that's just too expensive singularly because of the energy. If we have water desalination that mitigates the effects of catastrophic climate change and the migrations that result from people having to move, it means that we can grow crops in much more arid environments because we can suddenly have water in the Sahara and so on. That is not science fiction. We've actually seen it already in practice in Israel and in Saudi. It's totally feasible. It's just that the cost is 10x too high to scale it to poorer parts of the world. So there are so many of these emergent effects that come from addressing foundational technologies like energy or in healthcare or in food production. And I think it's very reasonable to predict massive breakthroughs, true abundance relative to where we are of 10 to 100x in the next 20 years.
I do want to spend a little bit of time and give a little bit of attention to the complexity of something that is holding so much promise while us also not being able to control the out-of-control aspects of it. The ways that people will manipulate truth, the ways that it's becoming increasingly complicated. I mean, it already was becoming complicated to know who to trust and how, given the state of our government as well. What is true and who do we trust? How do we hold both of these things at once? That we have this technology that can do all of these unbelievable things that will change the course of the planet, healthcare, all of these things. And also, it can be used to manipulate, to lie, to deceive. And in many cases, there are certain aspects of the learning that it does which we cannot predict. How do you hold both of those things? And then practically what does that look like for people who are concerned that we're especially for young people not giving them the ability to master human interaction in a way that's meaningful.
Yeah. So many great points there. I mean, you said a little bit earlier that the big challenge is finding metaphors to describe what this is. It's not a therapist. It's not a companion. It's not a friend. People often call it a tool. And there's lots to learn from the history of tools that is very relevant here because clearly a knife is dual use just as much as a laptop is dual use just as much as telecommunications infrastructure is dual use. So clearly these AIs are going to be omni. Everyone is going to be able to use them for everything. And just as when the printing press came out it amplified dissent in lots of good and bad ways. It spread misinformation much as it enabled us to turbocharge science. Every individual organization, religious group, government, nonprofit, academics, activist is going to think, 'Great, now I have a super intelligent agent that can amplify my perspective, my value, my culture, my politics, my version of truth.' And so just as we've had this curse of a lack of social intimacy because of scale, it's kind of what you were saying earlier. We don't have that tribe so much anymore, that family, that physical touch, the directness of that, lower than Dunbar's number contact. We have this curse of scale where I see everyone and everything all at once on Netflix and Instagram and every platform and it's completely overwhelming. Well, now of course AIs are basically going to turbocharge the personalization of the messaging of those values to everybody. Everything's going to get delivered. There's going to be a podcast doppelganger of you guys that delivers it in Japanese and delivers it to someone who's Gen Alpha versus someone who's a Silver Surfer. And then you can imagine every sort of flavor of personalization within that which is obviously what TikTok and YouTube now are versus terrestrial TV or satellite TV. The only way to avert that course is for significant large-scale coordination at a company level and at a government level internationally.
That's not going to happen though.
Well, exactly. This is the circular thing that we always get into over and over again.
