Reid Hoffman on AI’s environmental impact and the path to sustainability

RANA EL KALIOUBY: The world feels pretty chaotic right now – whether you’re looking at global politics, the state of the economy, or doom scrolling on social media. And sometimes it feels like we’re just standing on a beach, watching as the tsunami is about to crash. It can feel hopeless. But in these moments, Reid Hoffman has a different mindset than most, and it’s something I really admire about him.

REID HOFFMAN: I think that the entrepreneurial thing, the thing that you and I do and everyone else says, you always look at, okay, well in crises, how do you make opportunity? Because that’s the way that we make progress. And so I think part of the thing for thinking about this as citizens is say, okay, the tsunami is going to essentially crush our coastal town. What kinds of things should we be doing? How should we be preparing to rebuild? What kinds of technologies should we be using? How do I make something as good from this as possible? And I think that’s the right human and forward approach.

EL KALIOUBY: It’s a hopeful outlook. And one that I strongly agree with. Which is why I am so excited to have Reid on the podcast today. Yes, he’s co-founder of LinkedIn and one of the lead investors of our time, but he’s also a deep thinker who’s provided me with a lot of guidance over the years.

In April we celebrate Earth Day, and the whole month is about sustainable living. So in this moment of uncertainty, which includes looming climate disasters, I wanted to bring Reid on to give us a dose of pragmatic hope. We talk about the concerns people have about AI and the environment, the push for more sustainable energy sources, and why Reid thinks AI could be the platinum bullet to solving climate change.

I’m Rana el Kaliouby, and this is Pioneers of AI – a podcast taking you behind-the-scenes of the AI revolution.

[THEME MUSIC]

Hi Reid. Thanks so much for joining us on Pioneers of AI.

HOFFMAN: A pleasure. Great to be here.

Copy LinkWhy AI can expand human agency instead of diminishing it

EL KALIOUBY: Well, so it’s been a hot second since we last connected. And in the meantime, you’ve written a book called Super Agency. Congratulations. And your whole thesis or premise for the book is that AI can amplify, not reduce our agency. So tell us more.

HOFFMAN: So one of the things that I realized, there’s obviously a whole bunch of negative discourse around AI. There’s fears about privacy. There’s fears about jobs, there’s fears about democracy and all the rest. And I think all of this kind of resolves at its core to a fear of human agency.

And one of the things that was interesting when I looked at this is that this is the history of human societies confronting new technologies. It goes all the way back to the written word and the printing press, but also cars and electricity and mainframes and all the rest, which is this technology may be the destruction of human agency, human society, and so this is not new.

That we have this concern in terms of what we’re doing. And of course, each time you look back across the thousands of years of human society, the technology actually increases human agency, it transforms it. We lose some, but we gain lots more.

Copy LinkSeparating valid energy concerns from the bigger climate opportunity

EL KALIOUBY: So there are so many things we can talk about, especially related to AI’s impact on society in the future. But for today, I wanna focus our conversation on AI’s impact on the environment. You, like me, are a big believer in the positive potential of AI. But I guess I wanna begin our conversation by addressing some of the real concerns that people have around using AI and in particular, the amount of energy AI consumes. And I’ll just share a few data points. So according to the Department of Energy, data centers account for about 4.4% of the total electricity use in the US, but that’s expected to go up as demand and adoption of AI increases. So by 2028, that number could go up to even 6.7 to 12%. There are a lot of people who are concerned about this. Do you think these concerns are valid? And if so, how do we address them?

HOFFMAN: Well, look, the baseline concern is a valid one to say, hey, we’re gonna be creating this new world of intelligence. It’s gonna use a lot of electricity in training and in having computational agents everywhere. Energy, when it has non-green sources, contributes to creation of carbon and other kinds of things.

And is this — and you’re telling me it’s gonna be an exponential increase in energy — is that gonna be, even off this initial small percentage, an exponential increase in carbon creation, is that gonna have a climate impact? So that’s not an unreasonable thought. But then when you start actually looking at the details of it, you realize the story is quite the reverse.

First, all of the hyperscalers — Microsoft, Google, Amazon — are all making very strong commitments to make their new data centers green energy. And this isn’t just, okay, so when it moves from 4% data centers to 8% data centers, the next 4% will be completely carbon neutral.

That’s also great. But in fact, all of them are doing essentially venture capital on green energy, which is like geothermal, hydro, solar, et cetera. They’re all investing in new kinds of energy companies. The whole challenge with energy companies is they have to get to scale in order to be cost effective and competitive in the market against coal and against regular gas. And once they get there, then obviously that becomes energy that’s available for all of the other purposes. Like even if you say, all right, data centers are 4% and then 8%, you go, well, actually that leaves 96%, 92% for all the other things we do.

Can we make those green as well? And so this AI revolution, this data center revolution is funding the venture capital for making clean energy across all of this. That’s observation number one. Observation number two is the question around, well, once you apply intelligence to things, intelligence can actually make everything more efficient in what it’s applied to.

And so it’s like, well, let’s be applying AI and intelligence to climate, to energy. So now it ranges everything from, call it the mundane, which is, hey, can you be running your household much more effectively? Like when do you turn off the heat or the air conditioning? Similar across all the different electricity, lights and everything else in the household, to actually your grid. And we’ve already seen DeepMind applying it to some of the most efficient grids, Google’s computing centers.

So you can apply AI to all of these things to make the energy much more efficient. And then the very last piece of it, I think, is there’s some hope that as we get AI better and better, it can actually affect our progress of science, like whether it’s making new materials that might be much better for carbon capture or energy consumption or carbon prevention in the atmosphere, fusion power, all of these other things. And it doesn’t even have to be super-genius AI creates fusion. It could simply be AI helping the people who are building fusion power operate more efficiently and effectively. Just think of it like giving them a Google search engine for their work in order to improve. And so I’m hugely positive on AI being one of the many reasons for accelerating AI to actually help with climate change.

Copy LinkHow cleaner power sources could make AI more sustainable

EL KALIOUBY: So let’s deconstruct all of that. I first wanna start with, okay, how can we create more sustainable AI? So that’s kind of your reason number one. And I wanna actually deconstruct it and use the tech stack as a framework for how to think about this. So let’s start at the very beginning, which is kind of the lowest layer of the stack, which is what kind of energy is needed to power these data centers and this idea of cleaner energy. And in particular, I’m very curious about nuclear energy. I’m a tiny investor in a company called Deep Fission. They create these modular fission reactors that are a mile deep in the ground, so it’s supposedly very safe. How do you think about nuclear energy and other forms of clean energy? And also let’s talk about the regulatory process to bring these technologies to the world, to actually be used to power our data centers. How far are we from that reality?

HOFFMAN: Well, the history of nuclear technology is kind of the classic microcosm of people being overly fearful and negative about technology in its early stage versus iterating to more positive. So like for example, early in the US we went, oh my god, nuclear can create radiation, can do these problems, let’s shut it all down, let’s not do any of this. And you have Europe as the kind of case study, France especially, where they said, well, no, let’s continue to iterate on it, build on it. And then we’ve got our technology, which is coal and oil, and is massively contributing to climate change.

So all the environmentalists who were anti-nuclear actually directly contributed to essentially the warming of the planet and climate change by saying not nuclear. And then of course they wanted not oil either, but society always says I must have energy. And now we have the next generation of technology. I myself am also an investor in various fusion things — Pacific Fusion and Helion and so forth — which I do entirely out of a kind of a pro-human-society stance. I have no idea, like, you invest a dollar here and you make more dollars there. For me, it’s a philanthropic investment. And by the way, fusion is always kind of the holy grail because it doesn’t have the same radioactive components. But thorium and micro reactors and others are actually also really good.

And then of course, you also have things like the Bill Gates project with TerraPower, which is, hey, we know how to use nuclear byproducts to make them into the fuel that creates the electricity. And so it even gets rid of these things. I think the general lesson among technology critics is not that technology can’t be bad in some incarnations, but the question is always how do you shape it?

EL KALIOUBY: Yeah. There’s obviously been — you alluded to some of that — there have been some really public nuclear disasters. How do we change public perception around nuclear energy and get the average person to be more excited about these kinds of approaches?

HOFFMAN: Well, I think it’s one of the responsibilities of leadership because part of how human beings naturally operate is that we tend to be storytelling creatures. And so it’s the same reason why when you see a plane crash, you go, oh my God, planes are totally dangerous, even though in almost every metric today, driving is far more dangerous than getting on a plane. And yet people continue to be very nervous ’cause they see a plane crash and they go, plane crash, out of my control and dangerous. And so they respond the same way when you have a Fukushima, when you have some kind of nuclear disaster.

And by the way, it isn’t to say concern is actually bad. Old power plants are not great and you actually have to think about how do we build the new future power plants? The iterative development within technology is really, really key. And so, back to your question, part of the thing is to say, well, we’re on top of this. We’re navigating concerns. And by the way, if we go and create a whole bunch of nuclear power, both fission and fusion, that makes a substantive difference here. That will make your life a whole lot better. That will obviate climate change.

And that’s where I think leadership comes in. Leadership is helping people understand why it’s important to do certain things, why it may be important to invest in something years in advance of it happening. That’s part of the reason why leadership goes, we need a bridge here. We’re gonna build a bridge. It’s gonna take years to build this bridge, but at the other side of the bridge, it’ll be amazing. And that’s the kind of thing that we have to do collectively as society’s leaders.

EL KALIOUBY: In a minute, we keep moving up the tech stack, looking at the ways AI interacts with our environment. Could innovations on the chip and model front lead us to a greener future? Stay with us.

[AD BREAK]

Copy LinkWhy chips and compute infrastructure will shape the next AI era

EL KALIOUBY: So let’s look at the next layer up the tech stack, the chips. So there is a lot of demand for both training AI, but also on the inference side. I fundamentally believe the whole chip landscape is very ripe for disruption. We’ve actually had a couple of innovators in this space on the podcast. One company, Etched, they are doubling down on the thesis that transformer models are gonna power AI. So they’re building specialized chips for transformer models. And then of course, Jonathan, CEO and founder of Groq, who is really focused on the AI inference side. How do you look at the chip landscape and where do you think the opportunity is as it relates to more efficient hardware?

HOFFMAN: Well, I definitely think it’s once more the general principle of iteration of technology against a set of concerns that we as society and markets are asking for, which tends to get you some progress. And so I think on the chip side, this is part of the reason why we want this kind of technological innovation — because not only can you get energy efficiency in the production of compute and intelligence, you might get a lot more production of compute and intelligence. There’s a whole stack of things going on there. I think one of the things that we’re seeing in the global landscape right now is the creation of compute infrastructure as the modern cognitive industrial revolution, a kind of economic competitive race.

EL KALIOUBY: Can you elaborate on that?

HOFFMAN: Well, basically if you say the next generation — and I think with AI we’re in the cognitive industrial revolution — part of the reason I say cognitive industrial revolution is because it’s both the amazement of what our future economy and our future productivity will look like. And I think it’s as or more impactful than the Industrial Revolution. And we don’t have anything of our modern society without the Industrial Revolution. We don’t have a middle class, we don’t have broad spread education, we don’t have broad spread medicine — all of this stuff comes from the Industrial Revolution. Now, that’s the upside of what we’re moving towards. Part of the reason I also use Industrial Revolution is because the transition will be challenging.

We as human beings don’t do transitions in society well; it tends to amplify chaos and uncertainty. The Industrial Revolution was an example of that and we have to try to do it better, but it’ll still be difficult and painful as we’re making it happen.

But part of the reason some people say, well, we’ll just slow it down, we’ll just stop it — it’s like, well, there are 8 billion people on the human planet and other people aren’t gonna slow it down. So the story of, you adopt the industrial revolution and we won’t, plays very badly.

So you have to kind of play it through. And that’s the competitive race we’re in. And then that resolves a lot to compute fabric because when you get to the full stack of things that makes AI work, a bunch of it is major data centers with a lot of compute in them, both for the training and learning and also for the inference. And a bunch of that within those data centers is the chips that power it. And that’s part of the reason why, in the commercial competition, the CHIPS Act is saying, hey, let’s only provide chips to essentially our friends and partners in terms of how we’re operating.

Let’s make that happen. And by the way, that’s also of course important within the energy consumption and environment, but it’s also important in the raw production of compute, which is one of the things that powers the cognitive industrial revolution. And that cognitive industrial revolution will be almost all of the products and services of the future. So you need to be steering into that.

Copy LinkThe race to build more efficient AI models beyond transformers

EL KALIOUBY: Yeah. And of course that includes another important layer, which is the actual models. I’ll just give one example. There’s an MIT startup called Liquid AI and they’ve created this alternative architecture for models, which is liquid neural networks. These are much more energy efficient and they consume a lot less data to train. Yann LeCun, who’s the godfather of AI, has been pretty vocal that he doesn’t really think transformer models — he actually thinks that approach is gonna be obsolete in the next few years. To me, all that solidifies or underscores that we are gonna continue to need and see innovation on the models front. What’s your point of view?

HOFFMAN: Well, I think we are gonna see a ton of innovation on this. I think the most fundamental abstract way of looking at the AI revolution is how do we apply scale compute to learning systems? And the thought that we just happened to have the first kind of scale probabilistic model with Transformers and with large language models happens to be the one true holy grail — that seems a little improbable. It is obviously amazing and there are things that have of course been created in the last decade that previously were unimaginable among any of the experts.

And one of the things I find really interesting about the exploration of transformers is it kind of moves computer science into natural science because no one still has really good theories about why these transformers work at that kind of scale.

That’s part of the reason why there’s dispute amongst intelligent people like between LeCun and the people at OpenAI and Anthropic, et cetera — to say, actually, how far will this go? And that’s because it’s gone a lot farther than anyone would’ve really imagined 10 plus years ago. Now I am a believer that it will continue to go, that transformers will continue to do magical things. But believing that doesn’t mean that there won’t be other additional really amazing things. And I do think that there are other new techniques coming out in AI that will make a very big difference here.

But obviously when we get to acceleration within the software layer — the algorithms that put this together, what is the shape of the learning algorithm, everything else — that’s obviously one of the places where a huge amount of acceleration happens, because it’s the acceleration of the world through bits versus atoms, and bits can move, pun intended, at lightspeed compared to atoms. And that’s actually one of the things I think — not surprising as you worked your way through the stack — we get to one of the greatest accelerants and most plastic areas, where we will also see intense work, not just in the next decade, but in the next years.

Copy LinkWhat responsibility users have in steering AI toward better outcomes

EL KALIOUBY: Yeah. Amazing. What do you think is the consumer’s responsibility in all of this? Should we all be using AI less to be more cognizant of the environment, or should we be really aware of which tools from which companies we’re using that are greener or cleaner models or architectures? What do you think is the role of the average consumer?

HOFFMAN: Well, almost if you go to your current percentages that you started with, you say, look, data centers are 4% today and AI stuff is probably 10% or less of that in terms of the data centers. Your choice as a consumer about making any real impact — you’re not actually having any energetic impact today on this. As charted out earlier, I think all of the progress on this is likely to make not only your AI usage more efficient, but everything else in energy usage across the world and across major data energy uses much more efficient as well. So I don’t think you need to as a consumer try to steer your AI usage away from trying to be energy efficient.

What I do think is important for you to do as a consumer — there are at least two reasons. One of them is helping steer the technology to being more human. As you engage with it and see it and say, hey, I really like this, I can use this for various things, this is creating problems, I’m worried about this, et cetera — providing that feedback, whether it’s in things you post online or polls you vote on, I think that engagement is really important. The second part that’s really important is I think one of the under-described things about the cognitive industrial revolution is which societies, which industries, which companies are going to define the future of these products and services. Those are gonna be the ones that are actually engaged in using it.

And so I actually think part of the race is not just the AI development of the raw capabilities for building new products and services and making intelligence available everywhere, but also human beings using it, deploying it, engaging with it. Everybody engaging with it and using it in their own work, in their group’s work, in their company’s work, in their industry’s work, in their society’s work. And I think that engagement is also really important.

So part of what I tend to do when I’m talking to people is say, look, go start using it today. And I think AI is already good enough that if you haven’t found something where it’s important to your work today in using it, then you haven’t tried hard enough.

EL KALIOUBY: In a minute, why Reid thinks that AI will be the platinum bullet that solves climate change. We’ll get to that, and why talking to animals might be part of the solution. Stay with us.

[AD BREAK]

Copy LinkHow AI can optimize the grid and cut massive energy waste

EL KALIOUBY: You have said that AI could be the platinum bullet to solving climate change. And we’ve already kind of tackled some of these, but I wanna double click on a few of them. As you know, we’ve had Astro Teller at Masters of Scale Summit and on the podcast, and one of the big projects they’re working on is how to use AI to solve the electric grid problem. Let’s unpack that a bit. What is the problem and how can AI help solve it?

HOFFMAN: So part of the thing with the grid is the energy grid is what transforms energy and moves it all around, and there’s a lot of energy that gets essentially wasted in grid transport and in inefficiencies in the grid. And so one of the things we were talking about earlier is that intelligence applied can make lots of things more efficient and it can make grids more efficient. And you might look at your house as a little mini grid — do you need to heat all the rooms, air condition all the rooms, or can you predict usage and do it in just the right amount? Lights on and off. Efficiency of use of everything from charging your car and devices to use of the washing machine and all the rest. That’s like a mini grid where you could go, hey, if I can save, call it 30% of my power here, that’s a massive cost savings and everything else just by application of intelligence.

And then of course you apply that to the world’s grid — your city, your state, et cetera. Everything from your house to your entire society is one of the places where basic application of intelligence — this isn’t like, hey, we invented new fission or new fusion — can have massive energy savings, massive cost savings, massive carbon protection. And knowing Astro and his smart work and Google’s smart work, that’s the direction they’re gesturing at.

EL KALIOUBY: Yeah, absolutely. So you also invest in a project called the Earth Species Project, which I think is so cool. Can you talk a little bit about what they do and what’s their approach to using AI to combat climate change?

HOFFMAN: Earth Species Project is essentially like one of the things that comes out of large language models is translation. Now apply that to sentient animals, apply that to whales and dolphins, apply that to apes and chimpanzees, apply that to corvids and crows and say, hey, what can we get in translation for these species? And they definitely already have some pretty interesting results from the fact that these various creatures do engage in language more than your average human being thinks they do.

EL KALIOUBY: And it’s so fascinating because if you’re able to have this kind of connection and interconnection with other species in the world, I think the thesis is that that builds a lot of empathy and compassion for the entire universe, right? And hopefully that’s a very different and unique way to addressing climate change, having each of us have empathy towards —

HOFFMAN: That makes it look like, if you say, hey, we actually have empathy towards other species other than ourselves — which is easier to have when you can have a conversation — that may lead to a better caretaking stance on the entire globe, which would be obviously just amazing.

Copy LinkWhere AI could unlock breakthroughs from animal communication to medicine

EL KALIOUBY: Exactly. So then let’s tackle the last leg, which is AI as a technology that is accelerating and unlocking innovation in other industries, whether it’s biology, scientific discoveries, material science, robotics. Where are you most excited about AI’s opportunity to accelerate innovation?

HOFFMAN: Well, I’m excited across a number of different vectors. One of the things that I’ve done this year, as you know, is with Siddhartha Mukherjee. I co-founded a company called Manas AI, which is accelerating drug discovery, because you basically say, well, what’s one of the killers across all human groups, all ages, all races, all genders? It’s cancer. And part of it is there are so many different kinds of cancer that there is no one pill to cure all cancer. And this is one of the places where AI and intelligence amplification can be massively beneficial. So that’s obviously one. But obviously there are a bunch of others, whether it’s the creation of physical materials, or all kinds of things in robotics, in terms of everything from manufacturing. All of these things can be really amazingly done. And so it’s part of the reason why I’m so excited about what AI can mean for humanity and one of the reasons why I put a little bit of energy into writing Super Agency.

EL KALIOUBY: Yeah. And one of the kind of tenets of this book that I find really fascinating is that it is a collaboration and a partnership between humans and AI, right? It’s not competition, it’s really this partnership. But it does pose a question that I like to ask all my guests on the show, which is if AI can be so intelligent, so smart, so creative, so efficient, so empathetic even — because I’ve spent many, many years building emotion AI — then what does it mean to be human in this age of AI?

HOFFMAN: Well, I think that as we evolve technology, our concept of what it is to be human also evolves. It’s part of the, you know, like in several different books, I’ve kind of described us as more “homo techne” than homo sapiens. And it’s almost like the technology is what wraps into the thinking, but also everything else.

I mean it’s everything from glasses, the world we see, clothing, fire — all of these things evolve what it is to be human. Like for example, if you were in hunter-gatherer tribes before the village, before agriculture, you’d have a very different conception of what it is to be human than once you go to the Aristotle where citizens of the polis exist within cities and towns and villages. And I think that even on the cognitive side, with the printing press, people would describe it as, oh my gosh, this is gonna be the complete degradation of human capability, because no longer is this really absolute important ability called memory absolutely front and center and the most important part of intelligence. And yet, actually, now you know, part of the progress has been that memory is important, but not as important as it was a century and a millennia ago. And so part of what we’re going to be learning is to say, well, what’s the right way that we dance with this technology? Like for example, here’s the simplest thing — clearly for decades, computers play chess better than human beings.

We haven’t stopped playing chess, right? There’s still lots of exploration in it. There’s still lots of watching of it. There’s still lots of doing of it because we as human beings can still engage in all these things even as we use the technology to help us become even better. And I think that’s where it is across empathy. That’s where it is across cognition. That’s where it is across what it is to be human. And we need to be steering towards that as part of embracing our more human future.

EL KALIOUBY: Yeah, I love that. So my takeaway is, how can each and every one of us steer technology to become more human and amplify all of these human virtues, including empathy and of course, agency.

HOFFMAN: Yeah, exactly.

EL KALIOUBY: Reid, this was wonderful. Thank you so much for joining us on the show until the next.

HOFFMAN: Until the next time, and may it be soon. Always a pleasure.

EL KALIOUBY: When we talk about AI and the environment, I think we need to hold two truths: One, that AI does consume energy, leading to CO2 emissions as well as other environmental impacts, and two, that innovations in AI could lead to climate solutions. It’s too simplistic to position AI as a net positive or net negative when it comes to climate. We may be watching a tsunami approaching the beach, but we still have time to act.

If we invest in new, more sustainable energy sources, innovate across the AI tech stack to build AI that is greener, and continue developing AI technologies that reshape our relationship with our environment, I’m hopeful we’ll see change.