Benjamin Todd is my favorite AI expert. He ran the first nonprofit to be part of Y Combinator, became a key thinker in the Effective Altruism movement (and did long, serious self-examination in the wake of the FTX scandal), and was years ahead of the curve in assessing the risks posed by AI.
Now, he has a book out entitled 80,000 Hours: How to Have a Fulfilling Career That Does Good, based on the work of his eponymous organization. The premise of 80K is simple: human beings have roughly 80,000 hours to devote to their careers, and, if you want to do good, it’s best to think carefully and rationally about how to allocate them. But its answers are counter-intuitive. Don’t just “follow your passion” — figure out which problems are the most pressing ones to solve, and how your talents and interests can be put into the service of solving them.
When I first met Ben a few years ago (we first intersected in the meditation world), I remember thinking that including AI as one of those “most pressing” concerns (alongside climate disaster, runaway nukes and global pandemics) seemed like science fiction. This is why he is the futurist, and I am not.
So, I was really happy that Ben made the time for this conversation, where we dig into some of the questions I’ve been most obsessed with lately: how to prepare for the massive economic changes AI will probably bring, what parents can do to help their kids navigate them, and where AI/AGI is on the spectrum from Meh to Apocalypse.
Importantly, Ben isn’t just riffing here; he’s one of the few broad-audience experts on the subject who actually does the research and demystifies the hype on both sides. In many ways, he’s a quant.
So I wasn’t surprised to learn as much as I did. Ben helped me understand why computer chips will become like a country’s population. He persuaded me that a “strategic pause of AI development” might actually happen, with China signing on. He made me realize how AI states might become like petrostates.
But I was surprised that Ben thinks we’ve got the risks wrong — and by “we” I now include Pope Leo, who released a massive Encyclical on AI earlier this week. For Pope Leo, the greatest AI risks have to do with massive unemployment, economic chaos, and the loss of our distinctive sense of humanity’s value. As serious as those risks are, Ben is more worried about the Really Bad Stuff. Like AGI spinning out of human control. Or AI-run bioweapons. Or a 24/7 AI-powered surveillance state.
Yet somehow, Ben doesn’t end in despair. He thinks we may be at a pivotal moment in human history — and thus a time in which committed people really can make a difference. He thinks China is more amenable to collaboration and agreement than AI-Boosters insist. And he thinks contemplative practices like meditation have an important role to play.
I hope you enjoy our conversation.
Is it the end of the world as we know it?
Jay: Ben, I’m really stoked to have this conversation with you. The metaphor that immediately came to mind when I was preparing for it was Walter Benjamin’s Angel of History. His metaphor is that the angel of history is facing backward: we can see what’s happened in the past, but we’re being blown backward into the future and have no idea what’s coming. So, I think it would be really useful to disaggregate the fearful hype about AI from what I at least see as very realistic, in terms of maybe a two-to-five-year time horizon, and then we can expand it further out. So — is AI taking all of our jobs and changing our society, or what?
Benjamin: Thanks for having me. I think the near-term job loss stuff I would actually argue is a bit overrated. But I do think there’s at least a one-in-three chance that there is far more powerful AI in the next two to five years. There are a few different ways to see this, and there’s so much talking past each other in these debates because people are often talking about different time horizons. But I think at this point you can really look and just think: if we extrapolate the trends that have happened for the last three or four years forward another three or four years, it’s looking pretty crazy.
Jay: Even without some of the more optimistic exponential growth models you see in the AI world. I have friends who’ve lost jobs already — even just in the humanities. I have a well-known, well-placed editor who’s a friend of mine and is having trouble finding work. And obviously in education as well. I guess — how would you prepare? When 80,000 Hours came out, the putative audience were people with a lot of opportunities in front of them who wanted to make a difference. And now I’ve been teaching at law school, and my nephew is 23. It feels like a completely different world.
Benjamin: I agree the vibes are pretty different, though I would actually say if you want to make a difference, now is arguably the best time ever, in that we’re facing such an important moment in history that any role you can play in that potentially has this huge long-term impact. Those types of opportunities just weren’t around in the past.
Jay: What do you mean by that? What’s the moment in history? You can pick your mega-crisis — is it the threats to democracy, the transformation of AI, something else? And my instinctual reaction is one more of futility than opportunity, so I’d love to hear more.
Benjamin: I was talking about AI in particular. I think if AI kind of stops now, then it’s a normal technology — about as big a deal as the internet, or maybe a little bigger. But if you could actually have AIs that do the job of a human remote worker, that puts us into a totally different economic situation — more on the scale of the industrial revolution, or more. And there’s at least a chance that happens in the next two to five years. That would make these years and their immediate aftermath one of the most pivotal moments in history. We’re talking about transformation of every aspect of society, not just narrowly about AI itself. And can you actually do anything to help? It seems to me there’s a lot we could do to increase our chances, at least a little bit, of navigating this transition. That’s what we’ve been trying to encourage people to work on these last ten years at 80,000 Hours.
Jay: I’d be really interested to hear more. I’m familiar with the 80,000 Hours discourse from about five years ago, so I’m behind. What are some of the ways? And I assume you’re not including “bomb the data centers” as one of the career paths.
Benjamin: No, definitely don’t endorse any violence. But there is a question of what the biggest risks are. We do think that if we have fully autonomous AI, loss of control of that AI is still one of our top-ranked risks. But we’ve added some extras to the list. One being AI-enabled bioweapons, which is hopefully a relatively widely agreed-upon concern. And then we also have concentration of power. There are several different ways in which AI is very centralizing — for instance, it makes possible literally 24/7 surveillance of every person, and you can actually get AI to follow everyone around and synthesize it into something usable, which just isn’t possible currently because there isn’t enough staff to process all the data. So there are a lot of new pathways to concentration of power that weren’t possible in the past. And then we have a bunch of other, more niche things on the website. Gradual disempowerment is also interesting — even if we don’t lose control of AI, don’t have concentration of power, and avoid a bioweapon, you could still see society evolving in troubling ways.
Jay: I feel like thirty percent unemployment is still problematic enough.
Benjamin: Well, our focus is on risks that could be truly existential — permanent loss of civilization’s potential. But if unemployment goes far enough, there’s an analogy to oil states, which tend to treat their citizens pretty badly because they don’t actually need them economically — they just get the money from the oil. In the world of the future, all states kind of become this, except what matters is how many computer chips and robots you own — that’s where all your economic power comes from, and you care much less about labor. In a sense, it’s Marx ultimately being right: capital becomes all that matters. And that seems pretty bad for democracy.
Jay: Is the theory that the amount of compute remains the metric even in a five-year time horizon — he who has the most compute wins? Maybe we should define what that means for listeners.
Benjamin: However many computer chips you have is however many AI workers you can run. That essentially determines the population of your future AI economy. If you have ten times more computer chips than someone else, you essentially have ten times the workforce, so you can produce ten times as many things — including military drones and whatever else you want to produce.
A strategic pause of AI Development
Jay: On a scale of zero to ten — ten being Sauron the Dark Lord — where on the scale of Palantir concern are you when it comes to global surveillance technologies enabled by AI?
Benjamin: I don’t know what Palantir specifically is doing, but I think we should be very worried. AI makes possible basically perfect surveillance, and that would be a complete gift to any would-be dictators — it would make it much more possible to stifle dissent than it’s ever been in history.
Jay: It’s interesting that you say that, because I think a lot of my liberal-minded Palantir critics don’t actually listen — they only get the clips of Alex Karp being weird, but they don’t really listen to the theory, which is actually similar to what you just said. If we agree that this technology is happening, our hands aren’t great, but they’re a little better than the bad hands of the really bad actors, whether it’s China or others. So it’s actually really important to strengthen Western democracy to the extent it is a democracy. Because the other guys are racing toward this technology also. Is that consonant with what you actually think? And is there a response that allows for a little more prudence, or are the AI boosters right that we’re in a foot race with even worse actors?
Benjamin: This gets us to one of the things we could do to reduce some of the risks, which is agreeing on a strategic pause to AI development. That’s not going to happen in the next couple of years with the current administration. But you could imagine a few years in the future, when it’s much more obvious to everyone that AI is an even bigger deal than it is now — maybe after some kind of dramatic event in the news where an AI got out of control or did something that killed someone. And we know that if AI starts being able to do AI research itself — which the people in the labs say they think they’re pretty close to — you could get an actual acceleration of the rates of progress we’re seeing now. At that point, it would be really good if everyone could say, let’s not speed through a massive leap in AI performance in three months, but spread it out over a year or two to give us more time to do the safety research, prepare policies, and understand the implications better. Even the difference between three months and twenty-four months is significant.
I also think the Chinese would have a lot of incentives to join in with a slowdown, because they’re still behind the US. The benchmarks say they’re about seven months behind, but the Chinese models seem to be gaming the benchmarks — when you move to more qualitative assessment, they’re more behind than that suggests. And the US still has about ten times more computing power than China. So if the US said to China, “We’re going to slow down for a year as long as you join us” — and the Chinese government has seemed genuinely concerned about AI risks, with high-level leadership conferences discussing them — there’s a good chance they would actually participate in that deal.
Jay: What I hear in your slowdown theory is that the current Silicon Valley discourse — China is this gigantic machine, they’ll cheat on any agreement, it’s a winner-take-all race — maybe that’s a little exaggerated. Maybe these people are amenable to reason, and might actually be incentivized to cooperate because they’re behind. Does that sound right?
Benjamin: Yeah, it seems like it’s giving up a bit too early. And some of these risks — like bioweapons and loss of control of autonomous AI — are just terrible for everyone. So everyone has a big incentive to prevent them.
What are we all supposed to do with ourselves?
Jay: By way of transition — both to 80,000 Hours and to the contemplative side — I’m working on this essay. I don’t know if you’re old enough to remember the early meme about “fully automated luxury gay space communism.” The idea was, once the robots can do everything, what are humans going to do? All the resources will be shared, everything will be nice because stuff will be taken care of, and we can just drink martinis on the space station. I loved that when it was satire twenty-five years ago. But the thing I still haven’t heard a good answer to is: what are people supposed to do? The fifty percent of humans who are now basically NPCs living on UBI. It feels to me that work and career and livelihood comprise a big part of people’s meaning-making, and as that erodes because of robotics and AI, I have some concerns.
Benjamin: From the perspective of our own lives, that would be a big transition. But from the big picture, that’s what happens if we succeed — in a way it’s a relatively small problem, because we’ve survived and we’re living in a post-scarcity society. It would be a big thing to adapt to, but in some ways it’s not that bad a problem to have on the scale of “we might be killed by robots.”
The more serious response is: everyone gets retired eventually, and people actually become happier in retirement on average. I feel like people can adjust to that — especially since we’d all be doing it at the same time. It won’t be that you’re mooching off everyone else and feeling low-status for it; it would be happening collectively. The other analogy is the aristocrats in history, who may have had some ennui but seem to have had a pretty good time. I think most people in the past would have preferred to be an aristocrat to someone working, if they’d had the choice.
Jay: I don’t know if that’s even true. I think the aristocracy depended on status — it wasn’t just the cocktail parties and the good food, it was knowing they were at the top of the heap.
Benjamin: That’s fair, that is a difference.
Jay: It’s hard for me to imagine the political will to have UBI [Universal Basic Income] at a level that allows for happiness, rather than basic subsistence.
Benjamin: Remember, we’re talking about full automation of work, in which case the economy is going to be at least probably a hundred times bigger in terms of output than it is now. Currently, social welfare broadly — various types of transfers — is more than ten percent of GDP, probably more like twenty. If GDP grows by a hundredfold, unless transfers drop to something like 0.1% of GDP, they’re going to go up a lot. And this doesn’t even require actual transfers of money, because a lot of what’s happening is just everything becoming extremely cheap. All services basically become free and much better, and all physical goods become way cheaper because robotic labor is really cheap. So the same income today can buy you ten times more than in the past. And there would be massive political pressure for some amount of redistribution. That percentage would have to shrink so much for us to not be significantly better off, given that much increase in overall wealth.
Is the Economic Chaos Overhyped?
Jay: This parallels something you and I were talking about a couple of years ago — the difficulty of weighing existential risks versus horrible things that aren’t existential risks. Some of the non-FTX-related criticism of effective altruism has been along those lines: there’s this very long time horizon, which is a feature not a bug, intentionally. And so there are things which may happen in the distant future — robots kill us, bioweapons — but then there are all of these other things happening in the interim that are pretty lousy for a lot of people.
Benjamin: I actually think the risks we’re talking about come before the point at which we have mass automation of work, because that requires very comprehensive AI that has fully working robotics, where it’s trusted and we’re happy to take all humans out of the loop on every job. Many jobs are these social things. I think we face the bioweapons risk long before we face full automation and no one having a job anymore.
Jay: Sure, but we face the white-collar job loss and tech job loss now. That feels like it’s happening in 2026 already. I have an eight-year-old daughter, and it seems clear that her education — and even elite institutions — have no idea how to prepare young people for an AI future. How would you prepare your child for this future?
Benjamin: On the white-collar unemployment point, I do agree that the pace of change in the next few years could be so fast that there might be a lot of transitional unemployment, and that has relevance to these other risks because it makes a lot of people really angry — in particular people with a lot of political influence, like the upper-middle classes who control the media and so on. So I think it could be a pretty wild time.
But I would also say it’s actually still quite hard to find this in the statistics. There was a widely reported paper about how hiring of software engineers aged 22 to 25 was down 20%. But what people didn’t report was that hiring for all other ages was actually up, such that net-net software engineering hiring is actually up, and wages are up too. So even in the profession that’s been most automated so far, wages and total employment are still slightly up. I feel very unsure whether we will have near-term large-scale white-collar unemployment. The type of people who become software engineers will also have other options. In theory, stuff that AI can do well will go down in value, but there’s all the other stuff AI can’t do, which will still command wages. And if we’re getting wealthier because stuff is being automated, those other things should actually see their wages go up.
Jay: Maybe I’m also biased for being old. My partner does development research for a large nonprofit — something AI does really well — and it hasn’t come for his job yet, but we’re glad we’re in our fifties because we feel sure it will come for his job in just the next few years.
Benjamin: I talk about this in the book. With most jobs, you get this pattern where as you get partial automation, wages actually go up and employment is steady, because it makes you more productive. So each grant-maker who can now write ten times as many grant applications is actually probably generating more revenue for the nonprofit — they’d potentially even hire more grant-makers or pay you higher wages. But then once you get to really thorough automation — ninety percent or more — that tends to bring employment down. This has happened a lot in history. The example I use is ATMs and bank tellers.
When ATMs arrived, everyone thought bank tellers would lose their jobs. And the number of tellers per branch did drop from about ten to five — you needed way fewer tellers to run each branch. But because that made each branch so much cheaper to run, banks opened so many more branches that teller employment actually increased for twenty years after the introduction of ATMs. But then smartphones meant people just don’t go to bank branches anymore — they use online banking — and now bank teller employment is crashing. We’ll likely see this pattern in many, many jobs. Software engineers even now seem to still be at the point where they’re just getting way more done, and the amount of software people want is basically unlimited — so they’re just producing more. But maybe in a few more years you could see that start to curve down. It’s very hard to predict with each job at what point you get to that crossover.
What Should Young People Do?
Jay: Let’s turn to the book. If you were eighteen, what would you advise, and what does the book say to advise, in this high-uncertainty moment?
Benjamin: Is this in terms of the impact people have, or more from a personally-looking-after-yourself perspective?
Jay: Both. It’s the ikigai diagram, right? I want to have an impact, but I also want to put food on the table. And even if someone just wants to get rich, it’s hard to know how to advise that person either.
Benjamin: The opportunities for getting rich seem bigger than ever in some ways — look at how much people in AI are making. But from an impact point of view, I like to think in terms of three scenarios for AI.
One is the scenario the people at the labs are basically telling us they think is going to happen: in a couple of years they automate AI R&D, apply millions of AIs to improving AI itself, and we have human-level AI somewhere around 2028 to 2030. That’s an incredibly transformative moment happening in the near future.
They could also be too optimistic, and we might not reach that point until the early 2030s. And even when we reach it, it might not be possible to automate AI R&D. There’s what people call fast takeoffs and slow takeoffs. A fast takeoff is when AI can improve itself purely in terms of software, and you get an acceleration of progress. If that’s not possible, then you just have to wait for more and more computer chips to be built, which takes longer — a slower takeoff that might take five or ten years, pushing things into the late 2030s.
The third scenario is that we’re way further away from any of these things, AI hits a plateau relatively soon, and it might be decades before we get to anything like AGI.
What you want to do is different in each scenario. If you’re 18, one way to reason is that in the really short-term scenarios, you just can’t really do anything because you’ll only be graduating in 2029, which might be too late. More people we advise these days are dropping out of college, so I think it’s worth — while you’re in college — doing a bunch of internships and side projects, and if any of them take off, considering dropping out early. There’s a stronger case for that than in the past.
In the middle scenario, if you’re 18 now, I think you should be mostly focusing on that medium-term picture where AGI arrives in the early 2030s with a slower takeoff — in which case there’s a fifteen-year pivotal period ahead. You’d be thinking: what’s the most I could contribute in that time horizon? And there’s actually a lot of time. A lot of the most influential people in AI now are very young, because younger people can learn new technologies faster and tend to get to the forefront of a new field more quickly. A lot of people in policy and government are also really young — it’s quite possible to end up in staffer positions making really crucial decisions in your twenties.
In the third scenario, it’s more like normal career planning. And it’s also possible AI does peter out, in which case it’s just normal planning as in the past.
The Role of Meditation
Jay: You and I have hung out a fair amount, and I know you devoted — especially during your sabbatical — a lot of energy to things that might look like personal fulfillment and self-actualization. Meditation is one of those things. Is that orthogonal to what we’ve been talking about for the last half hour, or do you see a nexus between our various ways of self-actualizing and these massive changes taking place?
Benjamin: One quick thing: if you feel like you can’t help with any of these challenges — though I’d suggest most people actually could help a bit, even just part-time — then the question becomes more like, how do I prepare for a post-automation world? Maybe this is also what makes me more sanguine about the “we all lose our jobs” risk, because I just feel like it’s so easy to fill up a day with hobbies and cooking and taking a walk and exercise and playing games. And meditation would be something I feel like I could really spend many more years devoted to if I had the time — though I might just be a bit weird on this dimension.
The way it does more overlap, for me at least: this is a very scary time if you really think about what we’re facing and how above our pay grade it all feels. And I think meditation has helped me, most of the time, get to a place where it’s more like I just feel I’m doing my part, doing the work, and I’m able to get on with it and not be trapped in constant worrying.
Jay: I’m working on a book about consolation during the polycrisis and what that looks like. That part I feel very good about — if the world is going to hell in a bucket, I do know how to better enjoy the ride, as that rock and roll song goes. I do know how to build resilience and be available for people who are suffering, and to tend to my own suffering when it arises.
I remember hearing from a well-known Buddhist teacher that he felt the role of the Buddha-Dharma was to provide hospice care for humanity. I didn’t love that framing — it doesn’t feel just. And it also didn’t feel like it actually reflected the reality on the ground, whether it’s climate, democracy, or AI — my three major fears. It felt like the monastic himself was being a little controlled by fear, as opposed to something like equanimity.
It sounds like you oscillate a little, like any human would, between those poles — some days feeling anxious about Skynet becoming self-aware, other days more like you’re contributing what you can and finding satisfaction in the effort since we can’t control the results.
Benjamin: It’s possible I’m just still not truly facing what is happening. What tends to happen to me is that every time a new breakthrough occurs, it feels more and more real, and I’ll often go through some days of feeling more anxious about it. But I think in significant part because of meditation, it tends to move through pretty fast when I get back to work. In the past I would have been thrown off for longer.
On the hospice care metaphor — I try to think instead about the bodhisattva path. The goal of practice is your own personal reduction of suffering and well-being, but beyond that the really crucial part is also being able to help the well-being of all beings. I want my practice to support me in being a better constructive actor who’s making the world better — not just being off in a cave in the jhanas feeling great.
Jay: That’s a great place to close, because the book is some of that. There’s an aspect of service in 80,000 Hours that feels like it’s manifesting that in its own existence — it’s walking the walk. This is an organization, and now a book, that is trying to help people make a difference in how they make a difference. It’s the ripple effect of the bodhisattva vow in paperback. Or hardcover.
Benjamin: Yeah, I spent my career trying to help other people figure out what they should do with theirs — and this hopefully has a multiplier effect.
Jay: You’re the guidance counselor to the stars. Thanks, man. I really appreciate you taking the time, and I’m enthusiastic to see the book get a nice reception.
Benjamin: Great to see you. Thanks so much.
I hope you enjoyed this conversation — I certainly did. Here’s more from 80000 Hours about the skills AI makes most valuable.
Some other things I’ve enjoyed this week:
I feel very seen my this McSweeney’s piece, “A Camping Trip with Young Kids as Imagined By Me Before Having Kids.” 100%
The UnPopulist ran great coverage of Trump’s insane $1.8 billion slush fund and the horrifyingly anti-democratic ‘Election Security Czar.”
And finally, this lovely piece in the Times about allowing yourself to get weirder as you get older. Also 100%.
See you next week.
