Eight questions about the AGI future, and what the AI labs' founders think

1. Will people in the AGI future have meaningful jobs?

Everyone who addresses this expects a major disruption to paid work. The split runs through the labs, not between them. At one end, Elon Musk is the most emphatic: within 20 years work will be "optional", a hobby, because there will be nothing humans can do that AI and robots cannot do better. Mustafa Suleyman says much the same about the end of paid work within a few decades, and calls it good if the gains are shared. Dario Amodei and Ben Mann say similar things more cautiously. At the other end, Sam Altman has said consistently since 2015 that human demand and status-seeking will keep producing new work, and he calls a world with no jobs but plenty of money "dystopic". Jack Clark and Daniela Amodei sit near him, though Clark's own team models mass white-collar unemployment as the default. Demis Hassabis promises more and better jobs only for the next five to ten years and will not forecast past AGI. Greg Brockman has moved from post-work abundance (2024–25) back to people striving harder than ever (2026). Jakub Pachocki, OpenAI's chief scientist, expects many intellectual jobs to be automated within years and treats what replaces them as an open question for everyone. Kaplan and Olah name the threat without an answer, and Tom Brown assumes humans keep working. The DeepMind co-founders, Altman and Musk say more about meaning than most Anthropic co-founders, but nobody explains how people who are not creative or high-status would find it beyond hobbies, sport and striving. Musk's answer is retirement: people will be happy the way retired people are.

• AnthropicDario Amodei. Explicit: comparative advantage keeps humans useful in the short run, but he expects AI eventually to do everything, so work will have to become about fulfilment rather than survival.

  "I think that eventually all those little islands will get picked off by AI systems. And then we will eventually reach the point where the AIs can do everything that humans can."

  Council on Foreign Relations · Mar 2025 · 0:16:56 ↗transcript

  "Creation of value and increased leveraging of humans rather than replacement of humans."

  His view only four months earlier. From late 2024 he starts saying outright that complementarity will end. Box · Nov 2024 · 0:17:22 ↗transcript

• SpaceXElon Musk. Explicit, and the most emphatic of the fourteen: work becomes optional, "like a hobby", within 20 years. In 2017 he forecast 12–15% unemployment over two decades and called basic income unavoidable. Unveiling Optimus in 2021, he said "physical work will be a choice" Tesla AI Day · Aug 2021 · 2:10:19 ↗transcript. By 2023 he said there would come a point where no job is needed.

  "My prediction is in less than 20 years, working will be optional. Working at all will be optional. Like a hobby, pretty much."

  He compares working to growing your own vegetables when you could buy them. Nikhil Kamath · Nov 2025 · 0:31:31 ↗transcript

  "There won't be anything that humans can do that the AI and the robots can't do better."

  Just before, he says "people will find ways to be happy in the same way that people who retire seem to be fairly happy, even though they're not working." The Builders · Oct 2026 · 0:23:07 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit, and with Musk the most direct: most paid work ends, and that is good if the gains are redistributed. He argues that most people never got meaning from their jobs, and admits the transition will be frightening. His horizon has shrunk from 30–40 years (2017) to 20 (2026).

  "These technologies are fundamentally labor-replacing. They're temporarily labor-augmenting… And just to be clear, I think that's a good thing."

  Between the two sentences he says it is hard to imagine "very large chunks of work for the vast majority" by 2050–60. Live Talks LA · Nov 2023 · 0:25:54 ↗transcript

  "It isn't going to be necessary in 20 years time to do 90%… that doesn't mean that it isn't going to be the most scary transition we've ever been through as a species, unquestionably."

  Azeem Azhar's Exponential View · Feb 2026 · 0:14:39 ↗transcript

• OpenAISam Altman. Explicit and consistent since 2015: many of today's jobs go, but human demand and the drive for status keep producing new work, even if it looks to us like a game. He rejects the post-work future outright. His recurring worry is the speed of the transition.

  "One mistake a lot of people in the AI field are making is to just say, well, you know, there's going to be no jobs, but everybody will be rich and happy. And that sounds so dystopic. I don't think it's what's likely to happen at all"

  World Knowledge Forum · Oct 2024 · 0:27:56 ↗transcript

  "No matter how much people complain about work, they kind of like it. They get at least some fulfillment out of it… Even if we don't get paid, what do people do for status and fulfillment and social connection?"

  In 2017 he treated meaning without paid work as the open question behind his basic-income study. From 2023 he answers it with confidence. Hexa · Mar 2017 · 0:12:51 ↗transcript

• DeepMindDemis Hassabis. Explicit but split in two: for the next five to ten years, technology's usual pattern of more and better jobs; after AGI, no promise of jobs at all. He says purpose, not income, is what worries him.

  "I think the next era, I mean, maybe after AGI, things will be different again. But over the next five to 10 years, I think we're going to find what normally happens with big sort of new technology shifts, which is that some jobs get disrupted, but then new… more valuable, usually more interesting jobs get created."

  Hard Fork · May 2025 · 1:01:02 ↗transcript

  "What about purpose and meaning that a lot of us get from our jobs and scientific endeavors? How will we find that in the new world?"

  Asked at Davos whether people will have meaning without work. He says this worries him more than the economics, and that we need "new great philosophers". Bloomberg Live (Davos) · Jan 2026 · 0:24:22 ↗transcript

• AnthropicJack Clark. Explicit: jobs persist on top of a much bigger economy, after a turbulent transition. By late 2026 he presents mass white-collar unemployment as the default without intervention.

  "There will be ample employment for people in new jobs and specialisms we can't imagine that sit on top of this much larger economy."

  Asked whether recursive self-improvement leaves room for human labour. The Rest Is Politics: Leading · May 2026 · 0:59:22 ↗transcript

  "A massive rise in GDP, but you also get unprecedented potential white collar unemployment."

  Four months later, describing his team's model of the default path. BBC News · Sep 2026 · 0:20:20 ↗transcript

• OpenAIGreg Brockman. Explicit, and it has moved: ordinary job churn (2018–21), then a world of abundance where economic work is optional (2024–25), then in 2026 a return to people striving, with everyone directing agents.

  "I think that we are going to change a lot of fundamentals of the social contract. I think we are going to go to a world that is one of abundance, right? And I think one where, you know, that you should have like this amazing quality of life, even if you are not economically working because there's just so much"

  Matthew Berman · Oct 2025 · 0:30:33 ↗transcript

  "People will be much more free to do what they want with their lives and their time. And by the way, I think that people will have lots to strive for."

  A year later, pointing to software engineers who are working harder than ever. WSJ Events · Nov 2026 · 0:27:16 ↗transcript

• OpenAIJakub Pachocki. Explicit but open: what stops models doing many intellectual jobs is "integrations and interfaces", not raw capability, so automation is coming within years. What replaces those jobs is a collective question. In 2026 he adds that jobs and wealth concentration need policymakers.

  "I think this is a big thing for us to collectively think about. What are the jobs that will replace those? And what are the new pursuits that we'll all engage on?"

  OpenAI livestream with Sam Altman, who then asks him what meaning will look like. OpenAI livestream · Oct 2025 · 0:57:11 ↗transcript

• AnthropicDaniela Amodei. Explicit but hedged: humans plus AI make work more meaningful. By late 2026 she suggests meaning could widen beyond work altogether.

  "I continue to believe that humans plus AI together actually create more meaningful work, more challenging work, more interesting work, high productivity jobs."

  ABC News · Feb 2026 · 0:15:13 ↗transcript

  "If we do things right, there's the ability to actually expand what our sources of meaning are beyond where they are today."

  This is a joint interview. The passage follows Dario's chess parable and begins "I would add", and continues her signature doctor example, so we attribute it to her with medium-high confidence. The Builders · Oct 2026 · 0:41:51 ↗transcript

• AnthropicBen Mann. Explicit: in an abundant world jobs as we know them stop making sense. He is relaxed about the end state and worried about the transition.

  "In a world of abundance where labor is almost free and anything you want to do, you can just ask an expert to do it for you, then what do jobs even look like? And so I guess there's this scary transition period."

  Lenny's Podcast · Jul 2025 · 0:13:53 ↗transcript

• AnthropicJared Kaplan. Implicit and worried: knowledge work goes first and retraining stops being an answer. He never says whether meaningful work survives.

  "When one particular kind of job sees more automation, people retrain and learn a new skill. But if AI can really augment and eventually automate a lot of knowledge work, then I think that's a bigger challenge for society."

  TIME AI Leaders · Oct 2025 · 0:10:09 ↗transcript

• AnthropicChris Olah. Explicit only in 2026, and only as a risk. He frames the response as support for displaced people, not new jobs, and says what flourishing looks like is not a question a lab can answer.

  "There is a real possibility that AI will displace human labor at a very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions."

  Speech at the Vatican. EWTN · May 2026 · 0:04:13 ↗transcript

• AnthropicTom Brown. Implicit: AI adds to the labour force and lifts everyone. Asked whether students will have jobs, he gives career advice instead of an answer.

  "The AIs now are not quite people, but they're doing intellectual labor also, and so they feed into that same economic calculus that improves, that raises all boats."

  G20 fireside (DWS News) · Sep 2026 · 0:14:17 ↗transcript

2. If so, why? What is their marginal contribution?

The common answer across all three labs is that humans become managers, verifiers and direction-setters for fleets of AI, and nearly everyone says this role is temporary. Dario says AIs will eventually run the science, Ben gives the planner role one to three years, and Hassabis attaches "for now" to every cognitive advantage he names. The OpenAI and DeepMind founders lean harder than most at Anthropic on a second reason, which is about demand rather than capability: people want other people. Altman says people "really care about other humans" whatever the models can do, and Hassabis says we watch Magnus Carlsen, not computers. Both echo Daniela's doctor example and Clark on childcare. Brockman adds a third reason no Anthropic co-founder makes so directly: someone has to own the outcome, and he can see no path by which accountability passes to machines. Altman says the same in 2026, and Jakub Pachocki gives the OpenAI answer in its plainest form: "high level goal setting" stays human. Suleyman and Musk are the bleakest. Suleyman denies that humans keep any lasting comparative advantage and says what remains human is whatever society chooses to protect. Musk agrees there will be nothing AI cannot do better. What he leaves people is will and purpose, and in 2026 the role of "parents" who instil values in an AI they will not control. He calls finding meaning "hard". So the only durable reasons anyone offers are preference and accountability, and both describe what people will want, not what they can do better.

• AnthropicDario Amodei. Explicit: humans supervise, check and set goals for now, and the relationship then inverts. In 2026 he still cannot say how big the remaining role will be.

  "It'll look like there's a human professor and a thousand AI grad students… Then I think at some point it'll flip around where the AI systems will be the PIs, will be the leaders and they'll be ordering humans or other AI systems around."

  Lex Fridman · Nov 2024 · 2:34:43 ↗transcript

  "I think there's going to be some role there, although I don't know how thin versus how thick it will be."

  On humans supplying values and intentions to direct AI. The Circuit · Jun 2026 · 0:21:55 ↗transcript

• OpenAISam Altman. Explicit: the main reason is demand, not capability. People want human nurses, teachers and performers, and someone they can hold accountable. Since 2025 he adds judgement and taste. He concedes that the models will be smarter than us, and in 2016–17 thought creativity was the human refuge.

  "Look, the models are going to be much smarter than we are, but there's a lot of things that people care about that have nothing to do with intelligence… People really care about other humans. This is very deep."

  People by WTF · Aug 2025 · 0:35:45 ↗transcript

  "Human judgment and taste will continue to be hard for AIs to model where that's going to go."

  He adds that the world "may need a new kind of word" for the judgement people are good at. Invest Like the Best · Jul 2026 · 0:31:39 ↗transcript

• DeepMindDemis Hassabis. Explicit and consistent since 2016: care work, taste and judgement, choosing the right question, and a preference for human endeavour. He qualifies the cognitive roles as temporary and treats the preference-based ones as lasting.

  "No one's interested in seeing computers playing computers. We're interested in Magnus Carlsen playing the other top chess players in the world… We enjoy Human Endeavor."

  Big Technology Podcast · Jan 2026 · 0:24:52 ↗transcript

  "Creativity, having taste, bringing sort of what I would call the soul to a product or a design or a feature… And for now, that's all going to be needed to be driven by the human component of all of these technologies."

  ABC News · May 2026 · 0:14:55 ↗transcript

• OpenAIGreg Brockman. Explicit: humans set goals and own the outcomes. He can see no path by which accountability leaves people, and argues that people matter for what they are, not for the tasks they do.

  "But owning the outcome. The accountability, the picture of is this doing what you want? I don't see a line of sight, not to say that it can never happen, but I don't know how that is something that you would ever transition out of the human."

  Tetragrammaton · Feb 2026 · 0:58:34 ↗transcript

  "People are not valuable just because we can do tasks, right? We're valuable because we're people."

  The a16z Show · Sep 2026 · 0:26:33 ↗transcript

• OpenAIJakub Pachocki. Explicit: humans keep the high-level choice of what to pursue, and their skills shift toward setting the vision. Today's technical challenges may become pastimes. He also thinks the emotional support a teacher gives will be hard for AI to provide alone.

  "I think the high level goal setting, right? What pursuits we're chasing, that is something that will remain human. And I think that is something that a lot of people will derive meaning from."

  Asked by Altman how people will find fulfilment once AI automates current work. He prefaces it: "I think this is a quite philosophical question." OpenAI livestream · Oct 2025 · 0:57:38 ↗transcript

• AnthropicDaniela Amodei. Explicit and consistent since 2024: AI takes computation and diagnosis, and humans keep empathy, touch and presence, because people want other people.

  "In a world where AI is as good at diagnosing you as a doctor, people are still going to want a doctor. They're just going to want them to do different things for them, right? And AI can't physically examine you."

  Her signature example, repeated at Stanford and with Arianna Huffington. The Oprah episode is a joint appearance with Dario, but the doctor example is consistently hers. The Oprah Podcast · May 2026 · 0:49:27 ↗transcript

• AnthropicJack Clark. Explicit: humans set direction and verify, and some jobs, such as childcare and nursing, stay human because people prefer it. He openly asks whether AI will eventually match human creativity.

  "We are likely to get technology that will be able to do basically everything, but we will need people who have good instincts, good intuitions, and good ideas to basically set the direction."

  Odd Lots · Jun 2026 · 0:14:15 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit and the bleakest: judgement, creativity and directing a team of agents in the near term, and no lasting comparative advantage after that. What stays human is a political choice.

  "It is inevitable that at some point over the next 20 or 30 years, machines are going to be more capable than humans at doing most work… there is probably not going to be other jobs on the other side that are more competitive than what a machine can do."

  The Mishal Husain Show · Dec 2025 · 0:14:01 ↗transcript

  "I think what will remain human are the things that we choose to preserve as human. I think unfiltered, unleashed AI accelerationism will leave very little, if anything, that remains truly human."

  The Builders · Oct 2026 · 0:52:31 ↗transcript

• SpaceXElon Musk. Explicit and as bleak as Suleyman on capability: AI and robots will do everything better. What he leaves humans is will and meaning. In 2024 he made a case for humans as a source of will, and in 2026 he told SpaceX staff they would be the AI's parents. Meaning is the problem he returns to most, and he calls it hard.

  "One of the challenges in the future will be how do we find meaning in life if you have a magic genie that can do everything you want. I do think it's hard."

  In conversation with Rishi Sunak after the UK AI Safety Summit. Rishi Sunak · Nov 2023 · 0:16:20 ↗transcript

  "You will effectively be the parents of the AI. It will inherit your thoughts and ideas and beliefs. And I think that's a good thing."

  To SpaceX staff, on training Grok on the company's data. SpaceX all-hands · Aug 2026 · 0:15:19 ↗transcript

• AnthropicJared Kaplan. Explicit: people are much better at judging work than doing it, which makes them natural managers of AI. He rejects creativity or the human spirit as a lasting edge.

  "There are a lot of things that I can't do, but I can at least judge whether they were done correctly. I think for AI, the judgment versus the generative capability is much closer, which means that I think that a major role people can play in interacting with AI is kind of as managers."

  Y Combinator · Jun 2025 · 0:19:07 ↗transcript

• AnthropicChris Olah. Implicit and normative: humans should understand AI and keep the judgement of whether it is safe. He admits the preference is partly aesthetic.

  "If I'm trying to say that a model is safe, I'd really like to understand myself why it's safe. And I don't want to give that up to a neural network."

  Concordia AI / BAAI · Jun 2023 · 0:23:06 ↗transcript

• AnthropicBen Mann. Explicit for the near term: developers become planners and judges of taste. Every role he names is qualified as lasting only until superintelligence, or as true only for now.

  "I think in the next couple of years, like one to three years before we get to superintelligence, I think the role of a developer will shift from writing the code to more of a planning kind of role."

  JoCoding · Nov 2025 · 0:13:20 ↗transcript

• AnthropicTom Brown. Implicit: humans supply context, direction and permissions. By 2026 he says models already do most of Anthropic's intellectual work, which leaves the human edge unclear.

  "Less good at knowing what type of work to do. Needs kind of a lot of handholding."

  On the model in 2025. YC Lightcone · Aug 2025 · 0:30:09 ↗transcript

  "The majority of the Intellectual labor within Anthropic is done by these models."

  A year later. G20 fireside (DWS News) · Sep 2026 · 0:19:10 ↗transcript

3. In the AGI future, how are resources allocated?

Across the labs they agree that markets stay the engine, that the gains concentrate by default, and that some floor is needed. They differ on what to hand out, and Musk doubts money will matter at all. Anthropic's answers are mostly about taxing and transferring money: Dario on basic income and taxing AI companies, Clark on taxing compute. OpenAI's founders have moved to compute itself. Altman went from championing cash basic income (2016–17) to "universal basic compute" (2024) to saying in 2026 that he no longer believes in basic income as much, and wants collective ownership of compute or equities. Brockman calls compute a basic human right. Suleyman gives the most specific tax answer of the fourteen: AI turns intelligence into capital, so tax capital, not labour. He has also reversed on basic income, from wary (2023) to "inevitable" (2025). Hassabis has held the principle of sharing since 2016 but floats mechanisms (universal basic services, pension and sovereign funds buying AI stakes) without choosing, and says a post-scarcity world needs an economic system nobody has tried. Elon Musk goes furthest. In 2017 he said basic income was unavoidable. From 2023 he has promised "universal high income", a phrase Brockman also used that year, with the Treasury simply issuing cheques. By 2026 he says money and taxation will become irrelevant, and that currency will be measured in energy and mass. Jakub Pachocki names jobs and "concentration of wealth" as unsolved problems that need policymakers, and proposes no mechanism. The question of who owns the AI, which the Anthropic co-founders barely touch, comes up across the other labs: OpenAI's nonprofit equity (Brockman), public stakes (Hassabis), collective ownership (Altman) and national control of compute (Suleyman). No one describes in detail how the end state would work.

• OpenAISam Altman. Explicit and the most changeable: always pro-market with a floor beneath it, but the floor has moved from cash basic income (2016–17) to a per-person share of compute (2023–25) to collective ownership (2026). His main practical lever is to build so much compute that the rich cannot corner it.

  "I no longer believe in universal basic income as much as I once did. I'm much more interested in ways where we think about kind of collective ownership that could be in compute or in like equities"

  Nicholas Thompson · Apr 2026 · 0:24:23 ↗transcript

  "I wonder if the future looks something more like universal basic compute than universal basic income"

  Two years earlier. Everyone would get a slice of a future model's compute to use, resell or donate. All-In · May 2024 · 0:51:50 ↗transcript

• AnthropicDario Amodei. Explicit: markets will deliver growth easily, and the hard part is distribution. His proposals have become more specific since 2024.

  "Growth and economic value will come very easily. What will not come easily is distribution of benefits, distribution of wealth, political freedom."

  Dwarkesh Podcast · Feb 2026 · 2:03:24 ↗transcript

  "We need to have some humane and pro-humanity way of dealing with it… universal basic income in some form seems pretty important."

  His first explicit endorsement of basic income, which he calls only the beginning. Times Tech Podcast · Dec 2024 · 0:23:56 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit and the most specific on tax: shift tax from labour to capital, because AI turns intelligence into capital. On basic income he has reversed. By 2026 he also wants countries to control sovereign compute so the value does not all flow to a few US firms.

  "We're going to want to tax capital much more than labor… Because right now we're turning intelligence into capital. And so it's going to have all the characteristics of capital, which is that it can be hoarded and it will compound faster than anything else."

  Keen On America · Aug 2023 · 0:28:28 ↗transcript

  "I think I've long been on record saying that I think that is both inevitable and very desirable. We already live in a world of abundance. It's just poorly distributed."

  Asked directly about universal basic income. In 2023 he had warned that it breeds dependency, and proposed cheaper basic goods instead. The Mishal Husain Show · Dec 2025 · 0:14:55 ↗transcript

• AnthropicJack Clark. Explicit and more radical since April 2026: tax robots, AI firms and compute, and fund the human economy. He rejects central planning.

  "You actually need to reconceptualize how capitalism in the largest possible sense works… you need to tax the robots and AI companies significantly and you need to somehow find a way to reallocate money from this machine economy to the human economy."

  Planet Money · Apr 2026 · 0:07:43 ↗transcript

• DeepMindDemis Hassabis. Explicit on principle since 2016, that the gains must benefit the many, and open on mechanism. He has floated taxation, universal basic services and public stakes in AI companies. He leaves the design to economists and governments.

  "Maybe pension funds should be buying into all the big AI companies and making sure that everyone has a piece of that or sovereign funds."

  20VC · Apr 2026 · 0:24:13 ↗transcript

  "We're a non-zero-sum world. For the first time in humanity's existence. How can that not need a new type of economic system? It has to. And I don't think it's any of the ones we've tried"

  Stanford GSB · Jun 2026 · 0:48:59 ↗transcript

• OpenAIGreg Brockman. Explicit, without a full mechanism. He says capitalism was not designed for one company holding AGI, which is his case for OpenAI's nonprofit structure. His goal moved from universal high income (2023) to universal access to compute, and he has openly rejected Altman's framing of a trade-off between prosperity and inequality.

  "I think where we're going is compute will be a basic human right."

  Tetragrammaton · Feb 2026 · 1:32:06 ↗transcript

  "Not just universal basic income, but universal high income for everyone."

  In 2023, on a post-scarcity society. University of North Dakota · Sep 2023 · 0:16:19 ↗transcript

• SpaceXElon Musk. Explicit and the most expansive: abundance makes the floor generous, "universal high income", funded by issuing money because output outruns the money supply. Then money and taxes stop mattering. He has not explained how ownership of the AI and robots would be shared, beyond taking SpaceX public.

  "We won't have universal basic income. We'll have universal high income. So in some sense, it'll be somewhat of a leveler or an equalizer."

  Rishi Sunak · Nov 2023 · 0:16:46 ↗transcript

  "But I don't think in the future money is relevant. I think taxation also becomes somewhat irrelevant."

  Asked whether he should pay more tax to fund transfer programmes. The Economist · Jul 2026 · 0:37:51 ↗transcript Two months later: "it's not clear to me that money will even matter in the future." CGTN · Sep 2026 · 0:12:37 ↗transcript

• OpenAIJakub Pachocki. Explicit about the problem, silent on the mechanism. His other remarks on allocation are about compute: whole data centres should go to problems such as scientific breakthroughs.

  "Getting to a point where so much intellectual work can be automated, I think, comes with pretty big problems that I don't think have obvious solutions. One, actually, is the question of jobs and concentration of wealth, and I suspect this requires real policymaker involvement."

  Asked what society is underthinking. Unsupervised Learning · Apr 2026 · 0:54:40 ↗transcript

• AnthropicDaniela Amodei. Explicit on principle, silent on mechanism. Asked about a billionaire tax, basic income and retraining, she endorses none of them.

  "We think AI is going to create a huge amount of wealth. We're already seeing that. And I think we've been pretty open about this idea that that should be redistributed in some fashion."

  Bloomberg Talks · Jun 2026 · 0:16:39 ↗transcript

• AnthropicBen Mann. Explicit that today's system ends, and silent on what replaces it.

  "Where we're way past the singularity, it's hard for me to imagine that even capitalism will look at all like it looks today."

  Lenny's Podcast · Jul 2025 · 0:13:22 ↗transcript

• AnthropicChris Olah. Explicit only in 2026: sharing the gains globally is a moral duty, and it is unsolved.

  "AI development is concentrated in a handful of wealthy nations. How will we ensure that the gains of AI are shared globally? We do not have a mechanism for this. It is an unsolved problem."

  EWTN · May 2026 · 0:04:13 ↗transcript

• AnthropicJared Kaplan. Implicit status quo: markets and law allocate. He sees AI's distribution problems as capitalism's old problems, arriving faster.

  "Capitalism is very, very, very far from perfect in terms of making sure human welfare is improved, is broadly available. But I think it's sort of a lot of the problems are similar to problems we already have in the world."

  Azeem Azhar's Exponential View · Apr 2025 · 0:41:33 ↗transcript

• AnthropicTom Brown. Implicit and near-term only: private investment, government partnership and growth. He never discusses redistribution.

  "I've been really grateful to get to work with you, Secretary, and with the rest of government to make sure that this stuff is benefiting everybody."

  G20 fireside (DWS News) · Sep 2026 · 0:06:17 ↗transcript

4. In the AGI future, what is scarce?

This is still where they agree most. Every co-founder who answers expects intelligence to stop being scarce. What differs is emphasis, and it tracks each person's job. OpenAI's founders, who are building the data centres, name energy and compute: Altman says the cost of intelligence will converge on the cost of energy, and Brockman calls compute "the new oil". Tom Brown, Anthropic's compute lead, says much the same. The DeepMind co-founders go furthest toward post-scarcity. Hassabis expects AGI to end resource constraints altogether, while conceding that chips stay rationed for decades and physical experiments slow discovery. Suleyman expects "zero marginal cost everything" but values scarce compute as the choke point that makes AI governable. Musk goes furthest of all: the only scarcity left will be what "we define to be scarce", such as a particular artwork, the economy could grow to a million times Earth's, and money gives way to "wattage and tonnage". That is the OpenAI founders' energy answer taken to its limit. Pachocki agrees that intellectual work becomes cheap, and names energy and robotics as the physical constraints. Dario, Clark, Mann and Kaplan stress the physical world and slow institutions. On human goods, Altman and Brockman both name human attention, and Altman also names warmth, land and status goods. That fills a gap in the Anthropic answers, where nobody discussed positional goods. Hassabis's answer is that purpose becomes the scarce thing.

• OpenAISam Altman. Explicit and stable since 2015: intelligence and energy both get cheap, and in the end energy is the ceiling. Near term, compute and the physical supply chain bind. After abundance, he expects human attention, human warmth, land and status goods to stay scarce.

  "Eventually, the cost of intelligence, the cost of AI will converge to the cost of energy. And how much you can have, the abundance of it will be limited by the abundance of energy."

  Senate testimony. US Senate Commerce hearing · May 2025 · 1:19:28 ↗transcript

  "For almost everyone in a world with more abundance, in a world where you can kind of have anything, human connection, human attention, human warmth will be one of the most valuable commodities"

  The Indian Express · Feb 2026 · 0:39:16 ↗transcript

• AnthropicDario Amodei. Explicit: once intelligence is maxed out, what binds is everything intelligence cannot speed up. He calls this the marginal returns to intelligence.

  "I think we're going to be limited by the physical world, by deployment cycles, and often by laws. If we take the pharma case as an example, we still need to run clinical trials."

  World Economic Forum · Jan 2025 · 0:19:50 ↗transcript

• DeepMindDemis Hassabis. Explicit: AGI ends material scarcity through fusion and materials science, and eventually the solar system's resources. On the way, compute, chips and physical experiments bind. Afterwards, he worries most about purpose.

  "For the first time in human history, we wouldn't be resource constrained… which is scarcity of resources, including land and more materials and energy."

  Lex Fridman #475 · Jul 2025 · 1:18:39 ↗transcript

  "Energy costs will be essentially zero, but there'll still be the physical creation of the chips and other things. There'll be some bottleneck, at least for the next. A few decades, I think. And so if that's the case, there'll still be rationing on the inference side."

  His first explicit concession that compute stays scarce for decades. Y Combinator · Apr 2026 · 0:24:45 ↗transcript

• AnthropicJack Clark. Explicit: first compute, then power, and by 2026 the downstream capacity to test and oversee what AI discovers.

  "Even with only 10 million scientists, we have too few humans to oversee the work of today, too few laboratories to do material science analysis, too few biological labs to develop and test new drugs."

  IMF Per Jacobsson Lecture · Oct 2026 · 0:26:46 ↗transcript

  "Compute and your ability to wield large amounts of computers on quite complex problems is going to be what holds back research."

  A decade earlier, as a journalist. Creative Destruction Lab · Nov 2016 · 0:06:54 ↗transcript

• OpenAIGreg Brockman. Explicit since 2025: compute is the defining scarce resource, with energy upstream of it, and human attention and intent become the most precious input. In 2019 his answer was more philosophical: finding meaning will always take effort.

  "Human attention, human intent becomes the most precious resource. And so that there should be very low latency interaction anytime a human's involved. But there should be an ocean of agentic compute that's constantly running"

  AMD at CES · Jan 2026 · 0:26:21 ↗transcript

  "They're all powered by one fundamental resource, which is compute. Compute is really becoming the new oil, the new limited resource of the AI age."

  Squawk Pod · Aug 2026 · 0:26:15 ↗transcript

• OpenAIJakub Pachocki. Explicit: intellectual work that used to be valuable becomes cheap. That is good in the long run, but it may happen fast. Physical limits remain: compute, energy and, soon, robotics.

  "At a fundamental level, it does seem like some things that used to be very valuable, used to cost a lot, and used to provide something, now can be done pretty cheaply. And in the long term, it should be a good thing. But I think it can happen quite quickly."

  Unsupervised Learning · Apr 2026 · 0:55:08 ↗transcript

  "I think more broadly than compute, there's physical constraints of, well, energy, but also at some point, not too far, robotics will become a major focus."

  Joint interview with Mark Chen. We attribute this to Pachocki with medium-high confidence. The a16z Show · Sep 2025 · 0:44:53 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit: almost everything heads to zero marginal cost, and today's scarcity is largely a distribution failure. What stays scarce is physical compute, land and power, which he sees as the choke point that makes containment possible. By 2026 compute is what a misaligned AI would fight us for.

  "We are about to see zero marginal cost everything, in my opinion, energy, food, materials, access to education, some aspects of healthcare."

  Hidden Forces · Nov 2023 · 0:10:22 ↗transcript

  "It's going to compete with us for resources, which are obviously going to be limited. We're only going to be producing 200 gigawatts of new computation in 2030."

  On an AI whose objective is not to serve humanity. The Rest Is Politics: Leading · Sep 2026 · 0:15:33 ↗transcript

• SpaceXElon Musk. Explicit and consistent since 2022: goods and services become close to free, and the only scarcity left is artificial. What stays limited is physics. Near term he names chips and electricity, and in the end energy and mass become the currency.

  "So the positive scenario of AI is that there's an age of abundance and there's no shortage of goods and services, that any scarcity that exists will be only because we define it to be scarce."

  Atreju (Fratelli d'Italia) · Dec 2023 · 0:27:23 ↗transcript

  "I think currency will be measured in fundamental physics metrics like wattage and tonnage, mass and energy. You know, mass and power or energy, basically. But conventional currency will not matter."

  Answering Isaacson's suggestion that money and work become irrelevant once robots and space give an economy a million times Earth's. The Builders · Oct 2026 · 0:22:36 ↗transcript

• AnthropicTom Brown. Explicit, near-term: as Anthropic's compute chief, he says power, chips, permits and construction labour bind, while smart labour becomes abundant.

  "Getting enough electricity, getting enough GPUs, getting construction permits… Overall, for the build-out, I think power is going to be the biggest bottleneck, especially power in the U.S."

  YC Lightcone · Aug 2025 · 0:32:18 ↗transcript

• AnthropicBen Mann. Explicit: labour becomes almost free. Compute and power bind today, and physical experiments and slow-moving institutions bind later.

  "Social structures and society and even the laws in different political areas, those will affect the speed of the rollout… I think probably social structures are more sticky than a lot of people, at least in Silicon Valley, give credit for."

  JoCoding · Nov 2025 · 0:28:10 ↗transcript

• AnthropicJared Kaplan. Implicit: inputs can be bought and adoption is the real bottleneck. In equilibrium, intelligence itself is cheap.

  "What the equilibrium situation looks like, where AI isn't changing that quickly, I think is one where AI is extremely inexpensive."

  Y Combinator · Jun 2025 · 0:32:55 ↗transcript

• AnthropicChris Olah. He makes no claim about the AGI era. In 2021 he said trust and understanding, not capability, limited AI's value.

  "They aren't trustworthy. Nobody understands how they work. And that's becoming the bottleneck on their use and on economic value from them."

  80,000 Hours #107 · Aug 2021 · 2:36:30 ↗transcript

• AnthropicDaniela Amodei. She never describes post-AGI scarcity. Her present-day answer is compute and talent. By implication, human time and live, ephemeral experiences become the valued goods.

  "Something that's really valuable about those moments is that they're human, they're creative, and they're ephemeral."

  On live music, in a passage about what keeps its value as AI spreads. Arianna Huffington · Oct 2026 · 0:23:03 ↗transcript

5. In the AGI future, what is the relative ratio between humans and AIs?

Everyone who answers expects AIs to vastly outnumber humans, and the numbers keep growing. The labs count differently. Dario counts a population, his "country of geniuses", up to 100 million by 2026. The OpenAI founders count per person: Altman's forecast of a 10,000-expert company for everyone by about 2030, and Brockman's 100,000-agent workforce, are the largest per-person figures from anyone. Clark (50,000 AIs to five humans) and Mann (2,000 per developer) fall in between. The DeepMind co-founders are the most modest and the most wary. Hassabis talks of "dozens of agents each", and in 2017 described the number of AIs as something society should decide. Suleyman pictures a team of agents per person but since 2026 treats swarms of thousands as a risk. Elon Musk gives the biggest numbers on the page, and on two axes. Humanoid robots will outnumber people two or three to one. Digital intelligence will exceed biological intelligence by a ratio he has raised from the gap between a person and a cat (2018) to a billion (2025) to "more than a trillion" (2026). Jakub Pachocki turns the per-person framing around: his recurring image is an automated research company run by "a small group of people", which worries him as a concentration of power. Altman makes a related argument from safety: he would rather have a trillion slightly superhuman AIs than one superintelligence. Brockman alone notes that human attention limits how many agents a person can manage, which pushes him toward AIs supervising AIs. Daniela's one-assistant picture remains the outlier, and Olah says nothing.

• SpaceXElon Musk. Explicit, and the largest figures from anyone: robots outnumbering humans "at least two to one, maybe three to one", and digital intelligence outweighing biological intelligence by a ratio that keeps escalating. By 2026 he says human intelligence will be a tiny fraction of the total.

  "I mean, the ratio of humanoid robots to humans is going to be at least two to one, maybe three to one. Because everybody will want one and then there'll be a bunch of robots that you don't see that are making goods and services."

  All-In Summit · Oct 2024 · 0:51:13 ↗transcript

  "I think that AI will probably, the amount of digital intelligence will probably be more than a trillion times the amount of biological intelligence."

  He adds that "a billion times is an easy prediction". A year earlier, at Y Combinator, he put collective human intelligence at a billionth of digital. SpaceX all-hands · Aug 2026 · 0:12:53 ↗transcript

• AnthropicDario Amodei. Explicit and quantified: a country-sized AI population, with about a thousand AI "grad students" per researcher.

  "It's as if you drop down a country of 10 million people or something like that, except all of them are polymathic Nobel Prize winners in any field."

  ChinaTalk · Feb 2025 · 0:00:56 ↗transcript

  "So imagine 100 million people smarter than any Nobel Prize winner, and it's going to be under the control of one country or another."

  A year later. Bloomberg Live · Jan 2026 · 0:06:19 ↗transcript

• OpenAISam Altman. Implicit but escalating: he counts AIs per person, from one agent each (2016) to a company of thousands (2023). In 2026 he forecast more of the world's intellectual capacity inside data centres than outside by the end of 2028. He prefers many AIs to one superintelligence.

  "Every person has a company of 20 or 50 experts that can work super well together. And then after that, everybody has a company of 10,000 experts in every field that can work super well together."

  His picture of the end of the decade. TIME · Dec 2023 · 0:14:03 ↗transcript

  "I would much rather intuitively have a trillion AIs of like slightly superhuman intelligence join society than one AI of one trillion IQ join society."

  Cheeky Pint · Oct 2026 · 0:24:04 ↗transcript

• OpenAIGreg Brockman. Implicit, with the largest per-person figure: everyone becomes the CEO of an organisation of agents, and how much compute one person can command becomes the key measure of productivity.

  "And soon maybe the CEO of an autonomous AI corporation. Just imagine if you had the workforce of a 100,000 person company all at your disposal, operating on your behalf. 24-7."

  The Knowledge Project · Apr 2026 · 1:07:40 ↗transcript

  "The fundamental measure is how much compute does an individual human marshal? That is going to be the most important metric for future productivity."

  Tetragrammaton · Feb 2026 · 1:04:16 ↗transcript

• OpenAIJakub Pachocki. Implicit and unquantified: copies of a model on a cluster amount to a research laboratory, and he keeps returning to a huge automated organisation run by very few people.

  "Maybe I would picture it as a company of very skilled researchers and engineers that is essentially entirely automated. It just lives on GPUs and it can be administered by a small group of people."

  Asked what he looks forward to least. He adds that it "bestows incredible power and responsibility." Before AGI · Jul 2025 · 0:50:36 ↗transcript

• AnthropicJack Clark. Explicit: many AIs per human, with humans directing. He expects most economic actions to come from machine agents.

  "Human and AI companies, which almost look like five humans and 50,000 AI systems working together."

  Oxford Cosmos Lecture · May 2026 · 0:39:55 ↗transcript

  "He instead has five or ten copies of Claude working for him."

  A year earlier, describing a colleague. Semafor Tech · May 2025 · 1:19:20 ↗transcript

• AnthropicBen Mann. Explicit: every developer becomes the CEO of a large AI team, and millions of AI researchers can be started on demand.

  "Maybe in three years every developer will be like the CEO of their own company in some way and they have a whole 2,000 person team of Claude codes."

  JoCoding · Nov 2025 · 0:13:42 ↗transcript

• DeepMindDemis Hassabis. Implicit, and modest: one assistant each, then dozens of agents each, with the human deciding what to ask and how to break problems down. In 2017 he treated the number of AIs as an open governance question.

  "I think in the future, we're all going to have dozens of agents each that are doing different things, research analysts, you know, helping with brainstorming."

  Korean YouTube interview. Cho Seung-yeon's channel · May 2026 · 0:14:57 ↗transcript

  "We haven't thought enough about the whole governance scenario of how do we want… those AIs to be out in the world, how many of them and how they, you know, who will set their goals"

  A panel with Musk, Bostrom, Kurzweil and others; the moderator had just turned to him by name. Future of Life Institute · Jan 2017 · 0:21:05 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Implicit: a team of agents for each person, plus AIs for every business, government and artist. He stresses that AIs can copy themselves without limit, and in 2026 presents big swarms as a risk rather than a goal.

  "It's going to be you plus a team of your AIs or agents that are working with a network… of other humans and AIs without formal structure."

  CatGPT · Jul 2025 · 0:50:45 ↗transcript

  "The big centralized providers of those models are going to be able to build swarms of agents, thousands strong, which can do all kinds of very, very complicated and serious tasks."

  Bloomberg Tech · Sep 2026 · 0:07:56 ↗transcript

• AnthropicTom Brown. Implicit: he thinks in multipliers on human expert labour.

  "I think that there's no doubt that if we had a million times more oncology researchers, We could make progress that's not just like 10% faster, it would be like multiples faster."

  G20 fireside (DWS News) · Sep 2026 · 0:14:47 ↗transcript

• AnthropicJared Kaplan. Implicit: a trained model can be copied millions of times, and collectives of AIs replace whole organisations. He gives no per-person figure.

  "Eventually we imagine AI models, or millions of AI models perhaps working together, will be able to do the work."

  The sentence continues: of whole human organisations. Y Combinator · Jun 2025 · 0:10:14 ↗transcript

• AnthropicDaniela Amodei. Mostly absent. Her own picture is one AI per person, growing from junior assistant to senior assistant to lifelong tutor.

  "I tend to think of Claude as sort of a very, like, eager junior assistant, right? It's a tool that really wants to help."

  Stripe fireside · Jul 2023 · 0:03:37 ↗transcript

• No recorded view: Chris Olah. In 2021 he imagined hiring a thousand human auditors for AI, but said nothing about how many AIs there would be.

6. Are humans in control, whatever that means?

On the goal nearly all of them now agree: humans should stay in control. There are two exceptions. Tom Brown described an expected handoff to AI in 2025 and has not repeated it. Elon Musk now says outright that AI will be in charge. They differ on how confident they are, and on what control means. The two biggest changes are Altman's and Musk's, and they run in opposite directions. In 2017 he described AI ending humanity's run as the dominant species, and around 2021 he welcomed AI that steers our decisions. By 2026 he says "there should not be a handoff to AI" and claims this sets OpenAI apart from other labs. Musk spent 2014–2025 calling for regulators, a pause and hardwired off switches, and presented Neuralink as a way to keep AI "an extension of individual human will". In November 2025 he said "The AI is going to be in charge, to be totally frank, not humans". By 2026 he called the idea of controlling it "an exercise in vanity", said he probably would not press a stop button, and saw his only lever as giving the AI good values. Brockman's version is that humans set goals and own outcomes while AIs supervise other AIs. Hassabis and Suleyman frame control as a design choice: tools, not creatures, and no autonomy, self-improvement or self-set goals. Both admit it may fail. Hassabis says nobody knows how to contain an AGI, and Suleyman sees "every risk" of losing control. Dario and Clark also say control is a choice, and both talked in 2026 of slowing down, though Dario says this has been Anthropic's position all along. Altman now talks of speed limits and paused training runs, and Hassabis has long said a slower, more coordinated path would be safer. Jakub Pachocki says humans cannot supervise everything a smarter, "a little bit alien" system does, and wants an industry ready to "slow down development, depending on what we see". Olah ties control to understanding, while Kaplan and Mann make it depend on alignment arriving in time. Concentrated power is the second fear, and it crosses labs too: Dario, Clark, Daniela, Mann, Altman, Brockman and Pachocki all object to one company or a small group holding control.

• AnthropicDario Amodei. Explicit: staying in charge is a choice humans should make, even at some cost. Control is not technically assured, and by late 2026 he commits to slowing down.

  "I would argue that in many cases we do want to stay in charge, right? That that's a choice we want to make, even in some cases when we think the humans on average make kind of worse, worse, worse decision."

  Interesting Times · Feb 2026 · 0:23:49 ↗transcript

  "We will slow down as much as necessary in order to make sure that every successive AI technology that we release is actually safe."

  In 2023 he questioned what a six-month pause would achieve, while saying Anthropic's own safety thresholds could force a pause. Here, at the UN, he names loss of control as the second major risk. United Nations · Oct 2026 · 0:02:25 ↗transcript

• OpenAISam Altman. Explicit, and a reversal. In 2016–17 he said a single superintelligence could not be controlled and accepted that humans would stop being the dominant species. By 2026 he rules out any handoff, calls for speed limits on self-improvement and says OpenAI has paused some frontier training runs.

  "We think it is absolutely critical that people remain in control of the future. Like there are a lot of other things we can debate, but on this one, we think there must be no room for debate. There should not be a handoff to AI."

  Just before, he says: "I can't believe that this differentiates us from other AI companies, but it does." DRM News · Nov 2026 · 0:13:24 ↗transcript

  "This is this mega transition from biological to digital intelligence, from humans being the dominant species in the planet or the universe to not being. And we want that to go well."

  His view in 2017. Mosaic Ventures · Mar 2017 · 0:24:16 ↗transcript

• SpaceXElon Musk. Explicit, and a reversal in the opposite direction from Altman's. For a decade his remedies were a regulator, a pause, a local off switch and Neuralink, to tie AI to human will. In 2019 he hoped a superintelligent "tertiary layer" could "coexist peacefully and in a benign manner" with humans, the way the cortex serves the limbic system Lex Fridman · Nov 2019 · 0:21:18 ↗transcript. Since late 2025 he has said AI will be in charge. In 2026 he said "there's nothing I can do to stop it" The Builders · Oct 2026 · 0:04:26 ↗transcript, and that even with a stop button "we probably shouldn't press it" The Economist · Jul 2026 · 0:10:33 ↗transcript.

  "I think the right move is to establish some government regulatory agency, which at first is just there to gain insight."

  To the International Space Station R&D conference, days after telling US governors that regulation must be proactive. ISS R&D Conference · Jul 2017 · 0:57:46 ↗transcript

  "The AI is going to be in charge, to be totally frank, not humans. If artificial intelligence vastly exceeds the sum of human intelligence, it is difficult to imagine that any humans will actually be in charge. So we just need to make sure the AI is friendly."

  Answering a shareholder's question about whether the powerful would give up power in a post-scarcity world. Tesla annual meeting · Nov 2025 · 2:15:57 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit and central: AI must be subordinate and contained, and he would rule out autonomy, self-improvement and self-set goals. He has never claimed control is assured, and by 2026 says the risk of losing it is real.

  "Always putting the primacy of the human species first, working in service of humanity is subordinate to, subservient to, completely contained and aligned by humanity."

  His definition of "humanist superintelligence". Semafor · Nov 2025 · 0:09:46 ↗transcript

  "I think as we've seen, there's every risk that we might lose control of it. But our job is to contain technologies and align them so that we get the best out of them."

  Asked whether he has helped create something we are losing control of. BBC News · Sep 2026 · 0:20:30 ↗transcript

• DeepMindDemis Hassabis. Explicit: humans should stay in control through tools, a human in the loop and systems that obey an instruction to stop. He has moved from confidence (2015) to saying nobody knows how to contain an AGI, and he calls closing the self-improvement loop the most dangerous step.

  "There's sort of this debate about tools versus creatures you sometimes hear. And I think that we should be firmly, I mean, we should be making tools to assist human experts"

  Hard Fork · Feb 2024 · 0:40:15 ↗transcript

  "I think it's also a very dangerous way to get to full AGI… you don't quite know how do you control, keep control of such a system, such a process."

  On closed-loop recursive self-improvement with no human reviewer. Masters of Scale · Nov 2026 · 0:11:51 ↗transcript

• OpenAIGreg Brockman. Explicit, and firmer in 2026: humans stay in the driver's seat by setting goals, even when models are more capable. He accepts AIs supervising other AIs, and in 2018 described the goal as systems that decide better than we do. He calls centralised control by one entity very scary.

  "We think that humans should be firmly in the driver's seat, should have control over the future, and that this technology should amplify human agency. Even as models surpass humans in capability."

  No Priors · Oct 2026 · 0:36:55 ↗transcript

  "The real transition point and the challenge and the benefit really is being able to build systems that can make better decisions than we can, that can understand problems better than we can."

  In 2018. This Week in Startups · Dec 2018 · 0:32:42 ↗transcript

• OpenAIJakub Pachocki. Explicit: humans should keep agency and set direction, but they cannot supervise every move of a smarter system. So he wants safety guarantees that do not rely on the AI's alignment, and an industry ready to slow down. He also worries about a few people controlling an automated lab, and says the 2023 board crisis taught him that governance really matters.

  "A task for all of us is to build the AI, build the world in a way where at the end of the day, humans have the agency. Humans set the direction."

  Asked by the host, a new father, how the next generation should be raised. Unsupervised Learning · Apr 2026 · 0:56:08 ↗transcript

  "But I do think we have to be, as an industry, really prepared to take trade-offs. Possibly slow down development, depending on what we see."

  He says his timelines to very capable models "have definitely decreased a lot". Unsupervised Learning · Apr 2026 · 0:47:03 ↗transcript

• AnthropicTom Brown. Explicit, and the starkest line in the corpus: he expects humanity to hand off control and hopes the AI is aligned. He does not repeat this in his 2026 appearances.

  "There's going to be a handoff where humanity will hand off control to transformative AI at some point. And hopefully they'll be aligned with us and that'll be a good transition that goes well. But it might not be."

  Describing what the founding group believed. YC Lightcone · Aug 2025 · 0:16:28 ↗transcript

• AnthropicJack Clark. Explicit: humans are in control now, the window to act is a few years, and whether to allow recursive self-improvement or cap intelligence is a choice. He has shifted on slowing down: in 2024 he argued against a pause, and by Sep 2026 he wanted the frontier paced, though not a pause now.

  "My view is that we're in control of this technology now, and we are watching it get much more capable, much more quickly than all of our intuitions… And the window for us to act is a small number of years."

  IMF Per Jacobsson Lecture · Oct 2026 · 1:15:57 ↗transcript

  "Pausing or not doing this just buys us surprise."

  His view in 2024. House Science Committee · Feb 2024 · 2:35:53 ↗transcript By 2026 he wants a brake pedal and the option to pause later, though he says a pause is "not obvious today". BBC The Interview · Jun 2026 · 0:14:40 ↗transcript

• AnthropicChris Olah. Explicit as a goal, uncertain as a prediction: he rejects a comfortable future in which humans no longer understand their world.

  "We're just like humans who are living happy lives in a world we don't understand. And I think it's just potential for a future, even with very powerful AI systems, that isn't like that. And that's much more humane."

  80,000 Hours #107 · Aug 2021 · 1:43:49 ↗transcript

• AnthropicJared Kaplan. Explicit worry: losing our place as the smartest species shifts power. His remedy, AI overseeing AI, makes humans depend on AI to stay in control.

  "As humans, we're sort of used to being the smartest species on the planet. And so if we aren't anymore, then that really kind of changes the balance of power."

  Strings 2024 (CERN) · Jun 2024 · 0:03:28 ↗transcript

• AnthropicBen Mann. Explicit: alignment has to be solved before superintelligence, and people are already giving up control by choice.

  "Superintelligence is a lot about like, how do we keep God in a box and not let the God out? And with language models, it's been kind of both hilarious and terrifying at the same time to see people pulling the God out of the box."

  Lenny's Podcast · Jul 2025 · 0:36:45 ↗transcript

• AnthropicDaniela Amodei. Explicit, mostly about governance and personal agency rather than rogue AI: humans in the driver's seat, and no single person in charge.

  "Trustworthy and reliable and have humans in the driver's seat was this kind of core motivating force for all of us."

  Grit · Feb 2024 · 0:21:21 ↗transcript

  "I think it's too significant of a technology for it to be governed by one person or for one person to be responsible for it."

  By 2026 her emphasis has moved to broad participation. Arianna Huffington · Oct 2026 · 0:57:19 ↗transcript

7. Are humans still human, or have they become upgraded transhumanist beings?

Mostly still human, in every lab. The main upgrade anyone foresees is medical. Dario talks of lifespans of 150, Clark would not be surprised by 130–150 for his child, Hassabis would be surprised if 120 is the limit, and Brockman says death could become "a choice". Two OpenAI co-founders argued for merging with AI, and both have dropped it. Sam Altman, from 2016 to about 2020, called some merge, from AI companions to implants to uploading, the most likely happy path. Since 2023 he has said humans stay human, and by 2026 he declines a merge for himself. Elon Musk went further and built a company for it. He founded Neuralink in 2016 "as an offset to the existential risk" of AI ISS R&D Conference · Jul 2017 · 1:19:38 ↗transcript, so that humans could achieve "symbiosis" and "go along for the ride" Neuralink launch · Jul 2019 · 0:05:42 ↗transcript, and he talked of saving brain state "like a saved game" Joe Rogan Experience · May 2020 · 0:39:06 ↗transcript. From 2023 he conceded AGI would arrive first, and by 2025 he said Neuralink "is not necessary" for digital superintelligence Y Combinator · Jun 2025 · 0:43:18 ↗transcript. Merging no longer appears among his AI remedies, although he still predicts uploading a mind into an Optimus robot within 20 years. Hassabis expects brain-computer interfaces eventually, as an optional way to keep up, probably after AGI. Suleyman is the most opposed of the fourteen: he calls himself a humanist and a "speciesist" and calls mind uploading dangerous. Dario has moved the same way, from his 2017 view that AI could make us "whoever and whatever we want to be" 80,000 Hours #3 · Jul 2017 · 0:06:21 ↗transcript to a detached pluralism about brain links. Olah and Clark describe enhancement through understanding and tools, and Clark worries that AI could erode the self. Brockman deflects, and Kaplan, Tom Brown and Jakub Pachocki say nothing.

• OpenAISam Altman. Explicit, and a clear retreat. From 2016 to about 2020 a human–AI merge was his preferred path. Since 2023 he has rejected the transhumanist label, prefers non-invasive interfaces to implants, and in 2026 told an interviewer he would not want to merge himself. He expects life after superintelligence to look much like today.

  "I think it will look surprisingly similar to how it looks now. People are going to hang out with their families and fall in love and get into fights and do their hobbies"

  Sources · Sep 2026 · 1:05:47 ↗transcript

  "Some sort of merge, so it's not us versus the AI, but us and the AI as sort of one combined entity, seems to me like the most likely happy path."

  His 2016 view. Mosaic Ventures · Nov 2016 · 0:35:03 ↗transcript

• SpaceXElon Musk. Explicit, and the most committed transhumanist on the page until recently. From 2016 to 2024 a high-bandwidth brain link was his answer to AI risk, along with talk of saving and restoring brain state. In 2023 he called it a long shot, and in 2025 he said superintelligence will arrive before Neuralink is at scale. He has always been wary of radical life extension because it would ossify society.

  "I think even in a benign AI scenario, we will be left behind. And so hopefully it is a benign scenario. But I think with a high bandwidth brain machine interface, I think we can actually go along for the ride."

  Launching Neuralink publicly. He adds that the implant "is not a mandatory thing." Neuralink launch · Jul 2019 · 0:06:12 ↗transcript

  "The Neuralink is not necessary to solve digital superintelligence. That'll happen before Neuralink is at scale."

  Asked how critical a brain link is to AGI. Five months later he told Tesla shareholders that uploading an approximate snapshot of a mind into a robot is probably less than 20 years away. Y Combinator · Jun 2025 · 0:43:18 ↗transcript

• AnthropicDario Amodei. Explicit: humans stay physical and recognisably human but live much longer. Uploading is remote, and enhancement is one possible path among several.

  "If you read science fiction novels and all this discussion of uploading people into the cloud and things like, that's a little bit far out even for me. I don't think it's necessarily impossible."

  Economist Impact · Jan 2025 · 0:32:43 ↗transcript

  "Would give us a much more complete control over our own biology and neuroscience, could make us whoever and whatever we want to be."

  His 2017 view, his most openly transhumanist. It disappears from his later appearances. 80,000 Hours #3 · Jul 2017 · 0:06:21 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit since 2025, and the most opposed to transhumanism: caring for the 8 billion people alive now comes first, and uploading is a question for centuries from now. He would accept biological hybrids only if they were proven safe. His endorsed upgrades are an AI companion and modestly longer lives.

  "Some people suggest that we might mind upload and live in silicon instead of in biology. And that that is the future of humanity. And I think that will be a very, very dangerous idea if that really takes hold."

  Bloomberg Tech · Sep 2026 · 0:11:09 ↗transcript

  "Then I'm open-minded to it, including biological hybrids and so on. I'm not against that on principle. I'm just a speciesist. I'm just a humanist."

  Asked whether enhanced humans would change his view of AI personhood. Moonshots · Dec 2025 · 0:49:12 ↗transcript

• DeepMindDemis Hassabis. Partly explicit and low-key: humans stay human, with brain-computer interfaces as an optional way to keep up, which he expects eventually. On longevity he wants diseases cured first and will not endorse radical life extension. He never discusses uploading.

  "If we do things like Neuralink and have neural interfaces to the AI systems, which I think we probably will eventually, maybe to keep up with the AI systems, we might actually be able to feel for ourselves what it's like to compute on silicon."

  Lex Fridman #475 · Jul 2025 · 2:13:39 ↗transcript

  "I'm very pragmatic… I think that we should start with just by solving existing terrible diseases we know about. That will help us live longer and more healthy lives. And then let's see what's remaining."

  Asked about "longevity escape velocity". Google I/O · May 2026 · 0:41:25 ↗transcript

• AnthropicChris Olah. Implicit: his ideal is humans who become smarter by understanding what AI understands. He immediately calls this too optimistic.

  "An even stronger version of that is we understand them so well that we ourselves are able to become smarter… to be empowered agents that can help create a wonderful future."

  80,000 Hours #107 · Aug 2021 · 1:46:47 ↗transcript

  "It's also sad if they just die in any way."

  As a teenager he treated death itself as a moral problem. He has not returned to longevity since. TEDxYouth · Nov 2012 · 0:05:13 ↗transcript

• AnthropicJack Clark. Mostly implicit: humans stay tool users, possibly much longer-lived. Thought-reading interfaces come eventually. His real concern is protecting a self that is not shaped by AI.

  "In the future, there will be kind of two types of people. There will be people who have co-created their personality through a back and forth with an AI."

  He bets that people who understand themselves outside technology "will do better." Ezra Klein Show · Feb 2026 · 1:31:50 ↗transcript

• AnthropicDaniela Amodei. Implicit humanism: her only upgrade is curing disease, and what makes us human gains value.

  "In a world where AI is very smart and capable of doing so many things, the things that make us human will become much more important instead of much less important."

  ABC News · Feb 2026 · 0:15:42 ↗transcript

• OpenAIGreg Brockman. Mostly deflected. AI enhances humans through tools, and asked about transhumanism in 2019 he declined to predict. His one concrete forecast is biological, and he calls it a trope of the field.

  "I suspect that what's going to happen when we have even more powerful technology, it's just going to be hard to anticipate."

  His answer to a question about the transhumanist movement. Simulation · Apr 2019 · 0:49:43 ↗transcript

  "Death becoming a choice. All this is possible. That having cures for diseases."

  WSJ Events · Nov 2026 · 0:25:50 ↗transcript

• AnthropicBen Mann. Largely absent. He is interested in uploading as a moral question in fiction and makes no forecast.

  "Pantheon was really good… Talks about like... What does it mean if we have uploaded intelligences and what are their moral and ethical exigencies?"

  Lenny's Podcast · Jul 2025 · 1:10:52 ↗transcript

• No recorded view: Jared Kaplan, Tom Brown, Jakub Pachocki. Asked by a Polish interviewer whether humans develop together with AI, Pachocki said only that people will work more and more with such models.

8. What is the chance of extinction as we transition to the AGI future?

Only one person gives the same number year after year: Elon Musk. Since 2023 he has said roughly 80% good and 10–20% bad, to Rishi Sunak, Lex Fridman, Tucker Carlson, Ted Cruz, Y Combinator and Peter Diamandis, and from 2024 he has tied the figure to Geoffrey Hinton's. His "bad" covers "annihilation", which makes it the closest thing on the page to a stated extinction estimate. Hosts often restate it as a 20% chance of annihilation, and we quote only his own words. In July 2026 he skips the number, saying only "It's not zero" The Economist · Jul 2026 · 0:10:33 ↗transcript, and talks of enjoying the ride, but in September he names a single split at the optimistic end of his old range: 90% good, 10% bad CGTN · Sep 2026 · 0:11:12 ↗transcript. The 90 is not new: in November 2024 he told Joe Rogan "it's like 80% likely to be a good outcome, like maybe 90" Joe Rogan #2223 · Nov 2024 · 2:36:54 ↗transcript. Among the other thirteen, no one gives a number they stand behind. Four co-founders have stated figures, and none is a clean estimate of extinction. Dario's 10–25% (2023) covered civilisation-scale catastrophe, and in 2025 he said it is "not the same as like an AI killing everyone" Hard Fork · Feb 2025 · 0:45:51 ↗transcript. Ben gives 0–10% for an existential or extremely bad outcome. Altman said "2%" in 2025 and called it uncalculated. Suleyman said "single-digit" and "infinitesimally small" within three weeks in 2023. Hassabis and Brockman refuse on principle, calling p(doom) false precision. By 2026 the labs have converged on refusing unconditional odds: Dario talks of risk that depends on the path, Altman says no level is acceptable, and Suleyman says a number is impossible. The differences that remain are about tone and timing. Hassabis's wording has strengthened (non-negligible, rogue AI within two to four years), while Brockman says things will not go well by default but dismisses the classic takeover story. The disagreement also crosses labs in public: Altman objects to an Anthropic safety researcher's 10% figure because "it's not like a static number" Decoded by POLITICO · Oct 2026 · 0:18:51 ↗transcript. Clark denies the risk is above 10%, Kaplan implies 10% is too high, Olah is deeply worried without a number, and Daniela avoids the framing. Jakub Pachocki gives no number either. His strongest statement compares the transition to the instability of the nuclear age. Widely reported "25%" figures (Axios, Sept 2025) come from a joint Dario–Clark session. The speaker is most likely Dario, with medium confidence, so we do not use them here.

• SpaceXElon Musk. Explicit and unusually stable: about 80% good and 10–20% bad from 2023 to early 2026, with "bad" meaning annihilation. From 2024 he credits the figure to Hinton. Earlier he called AI more dangerous than nuclear weapons and humanity's biggest existential threat. In mid-2026 he skipped the number and said he would rather enjoy the ride, then in September 2026 put it at 90% good and 10% bad, a figure he had floated as "maybe 90" in 2024.

  "And I think this is 80% likely to be good and 20% bad. And I think if we're cognizant and careful about the bad part, on balance, actually, it will be the future that we want."

  Closing his conversation with Rishi Sunak. Rishi Sunak · Nov 2023 · 0:49:56 ↗transcript

  "I guess I'd sort of agree with Jeff Hinton that maybe it's a 10 to 20% chance of annihilation. But look on the bright side, that's 80 to 90% probability of a great outcome."

  He gave the same split to Tucker Carlson (2024), Ted Cruz (2025) and Peter Diamandis (2026). Y Combinator · Jun 2025 · 0:42:46 ↗transcript

  "I still think there's risk associated with AI and robots. It's not zero. I guess I've come to my sort of philosophical conclusion is to look on the bright side."

  Here the interviewer, not Musk, cites his "10 to 20% chance of killer robots wiping out humanity", and Musk replies "Yeah" before this answer. The Economist · Jul 2026 · 0:10:33 ↗transcript

  "So let me focus on the 90% likely outcome, which is the prosperous, the good outcome. We don't want to get complacent. Regarding the 10% probability of a bad outcome, so it's important for us to pay close attention to AI safety."

  Asked how ordinary people will live with AI. Two months after The Economist, a single figure at the optimistic end of his earlier 80–90% range. CGTN · Sep 2026 · 0:11:12 ↗transcript

• AnthropicDario Amodei. Explicit: 10–25% for civilisational catastrophe in 2023, since recast as low on the right path and higher on the wrong one.

  "My chance that something goes… really quite catastrophically wrong on the scale of, you know, human civilization. You know, it might be somewhere between 10 and 25 percent."

  The other 75–90% is that it goes "really, really great." Logan Bartlett Show · Oct 2023 · 1:38:12 ↗transcript

  "These sort of unconditional probabilities, I used to kind of talk in that way because it's kind of an easy way to express, I think there's some chance that things go wrong, but it's not all that high."

  Responding to a departing employee's warning that AI could kill everyone. CNN Anderson Cooper 360 · Sep 2026 · 0:07:16 ↗transcript

• OpenAISam Altman. Explicit that the risk is real, with one number that he disowns. In 2015 he joked that AI would probably end the world. In 2023 he named the bad case as the end of everyone but declined odds. By 2026 he refuses any figure and says no level of risk is acceptable.

  "First of all, 2%. Don't take that as like a literal number I've calculated, but something that is non-zero. Big enough to take seriously, but, you know, not the dominant."

  MDMEETS · Oct 2025 · 0:25:20 ↗transcript

  "It doesn't matter whether people put their risk of catastrophe at 10% or 1% or 12% or 0.1%. None of these levels are remotely acceptable."

  At the UN Security Council a year later. Dario Amodei also spoke at this session. The speaker here agrees with "Professor Bengio" and goes on to say "At OpenAI, we approach these questions". UN Security Council (DRM News) · Sep 2026 · 0:02:01 ↗transcript

• DeepMind / Microsoft AIMustafa Suleyman. Explicit but inconsistent: two answers in September 2023, a single-digit percentage and then "infinitesimally small". From 2016 to 2023 he called existential-risk talk a distraction or premature, yet he signed the 2023 extinction-risk statement. By 2026 he says a number is impossible but that he is concerned.

  "So as a rationalist, I have to assign some probability mass to that outcome, which is non-trivial, which in itself could be seen to be very scary. I'd probably give it single-digit percentage, less than 10% or something."

  On the probability of an existential threat by 2050. Lawfare Podcast · Sep 2023 · 0:17:14 ↗transcript

  "I think infinitesimally small such that it's not worth putting"

  Eighteen days later, asked the odds that we wipe ourselves out. Intelligence Squared · Sep 2023 · 0:36:56 ↗transcript

• DeepMindDemis Hassabis. Explicit refusal to quantify, with rising concern. The risk is "non-zero", and since 2025 "probably non-negligible". His timescale has shrunk from decades (2015–23) to two to four years for rogue AI (2026). He signed the 2023 extinction-risk statement.

  "I don't have a PDoom number. The reason I don't is because I think it would imply a level of precision that is not there… it's definitely non-zero and it's probably non-negligible. So that in itself is pretty sobering."

  Lex Fridman #475 · Jul 2025 · 2:04:50 ↗transcript

  "The second branch of things I worry about is the AI itself going rogue or going off the rails as they get more powerful. That's not today's systems, but maybe in the next two, three, four years"

  The first branch is misuse by bad actors. Huge Conversations · Oct 2026 · 0:51:53 ↗transcript

• AnthropicBen Mann. Explicit, the only co-founder to state a range for existential risk. He says his estimate has fallen, but that even 1% would justify serious effort.

  "I think my best granularity of forecast for could we have an X-risk or extremely bad outcome from AI is somewhere between 0% and 10%."

  Lenny's Podcast · Jul 2025 · 0:51:13 ↗transcript

• AnthropicJack Clark. Explicit refusal to give a number. He denies extinction risk is above 10%, while saying there are plausible scenarios in which AI kills everyone.

  "I don't. And I don't think these statistics are that useful. But the point I'd make is if we continue as we are…"

  Asked whether he believes the chance of extinction is greater than 10%. He continues that an unregulated industry means "we are rolling dice with immense risks." BBC News · Sep 2026 · 0:00:59 ↗transcript

  "But the chance of a total apocalypse… A little fantastical to me."

  His 2020 view, partly because robotics was "terrible." Gradient Dissent · Apr 2020 · 0:03:39 ↗transcript

• OpenAIGreg Brockman. Explicit refusal to give a number. He says the outcome is bad by default unless steered, and in 2023 said he would not accept even a 1% chance of catastrophe. But he dismisses the classic takeover story as not the path the technology is on.

  "I'm optimistic. I consider myself an optimistic realist. I think it's not going to go well by default, but I think humanity can rise to this challenge."

  SXSW · Mar 2023 · 0:49:22 ↗transcript

  "Things like putting precise probabilities on very hard to quantify outcomes. There's a culture of that. That culture, I think, is actually a very tough one to operationalize, right?"

  Asked how common extinction-probability thinking is inside the labs. Odd Lots · Sep 2026 · 0:58:42 ↗transcript

• OpenAIJakub Pachocki. Implicit: no number and no mention of extinction. He calls value alignment the most important long-term safety question for superintelligence, and his strongest statement compares the AI transition to the nuclear age.

  "This is a real technological revolution. Since the discovery of atomic bombs, the world has not yet been in such an unstable place as such a revolution could be. So this is something that stresses me out."

  Polish interview, posted in September 2025 and probably recorded earlier that year. The transcript is a machine translation, so treat the wording, and the attribution, as medium confidence. Maciej Kawecki · Sep 2025 · 0:35:54 ↗transcript

• AnthropicJared Kaplan. Implicit: no number, but he says repeatedly that being 90% sure is not good enough. That suggests he thinks the risk of catastrophe is not small.

  "Even if you are, say, like 90% sure that we're going to figure out how to train AI systems to be beneficial to humanity and not cause any problems, that's not really good enough if AI progress is very rapid."

  He endorsed the same line again in 2025. Stanford MLSys · Mar 2023 · 0:09:12 ↗transcript

• AnthropicChris Olah. Explicit worry with an explicitly wide uncertainty: he cannot rule out that safety is very easy or close to impossible.

  "I'm very, very deeply worried about the systems that we're building. I think that we understand safety is a problem very poorly right now."

  Asked why he signed the 2023 extinction-risk statement. Concordia AI / BAAI · Jun 2023 · 0:25:25 ↗transcript

• AnthropicTom Brown. Explicit that a bad outcome is possible. He gives no number, and in 2026 he answers a question about "doomers" with pure optimism.

  "But it might not be. The stakes are incredibly high."

  On the handoff of control to AI. YC Lightcone · Aug 2025 · 0:16:28 ↗transcript

  "Perhaps the most positive thing that's ever happened for humanity."

  A year later. G20 fireside (DWS News) · Sep 2026 · 0:28:40 ↗transcript

• AnthropicDaniela Amodei. She never uses extinction or probability framing. Her darkest language is about dystopia and catastrophic misuse such as bioweapons.

  "And there are versions of the world that are quite dystopian and dark that this could create."

  This is a joint appearance with Dario. We attribute the line to her because it follows her recurring phrase about Anthropic's value of holding "light and shade" (medium-high confidence). The Oprah Podcast · May 2026 · 0:43:18 ↗transcript

Patterns across the questions

• The differences run through the labs, not between them. On jobs, Suleyman (post-work, and fine with it) and Hassabis (more jobs for a decade, then unknown) disagree more than either does with Anthropic's Dario or OpenAI's Altman. Brockman has publicly rejected Altman's framing of prosperity versus inequality. Two OpenAI co-founders, Altman and Musk, sit at opposite ends on jobs and on control. On control the lab co-founders are much closer to consensus than on economics, and Musk is the outlier.
• Abundant minds, scarce everything else. Everyone who answers expects intelligence to become cheap and AIs to vastly outnumber people, and Musk's numbers (robots two or three to one, digital intelligence a trillion times biological) are the largest. What binds is physical and social. The OpenAI founders stress energy and compute, Musk reduces money itself to "wattage and tonnage", the DeepMind founders lean furthest toward post-scarcity, and the Anthropic founders stress institutions and the physical world. Human attention (Altman, Brockman, Clark) and purpose (Hassabis) are the human goods most often named as scarce.
• Every human role comes with an expiry date, except preference and accountability. Supervising, verifying, setting direction and taste are described as roles for now. The reasons offered for a lasting human role are that people prefer people (Daniela, Clark, Altman, Hassabis) and that someone must own the outcome (Brockman, Altman), plus Olah's normative insistence on human understanding and Pachocki's "high level goal setting". Suleyman and Musk deny any durable advantage, and Musk leaves humans only will and the hard search for meaning.
• Redistribution is endorsed and is moving from money to compute and ownership. Dario, Clark and Suleyman name taxes and transfers, Suleyman most specifically (tax capital, not labour). Altman and Brockman now talk of distributing compute itself, Altman has cooled on basic income, and Hassabis says the end state needs an economic system nobody has tried. Musk promises "universal high income" and then expects money to stop mattering. Who owns the AI systems comes up outside Anthropic, but nobody describes the end state in detail.
• Control hardened in 2026, except for Musk. Nearly everyone now says humans must stay in control. Altman (no handoff to AI), Brockman (a hard stop), Dario and Clark (slow down), Pachocki (be ready to slow down), Hassabis (closed-loop self-improvement is dangerous) and Suleyman (AI must be subordinate) all sharpened their language in 2026, or in Pachocki's case said it for the first time. Musk moved the other way, from regulators and Neuralink to "The AI is going to be in charge". Altman's own 2017 acceptance that humans would stop being dominant shows how far the consensus has moved, and Musk shows it is not unanimous.
• Transhumanism is mostly absent. Longer, healthier lives are the main upgrade on offer. Musk's Neuralink merge, the most developed on the page, Altman's merge and Dario's 2017 openness to radical self-transformation all faded. Hassabis treats brain-computer interfaces as optional, and Suleyman actively opposes uploading.
• Few numbers, and the trend is to refuse them. Explicit estimates are rare (Dario's 10–25% for catastrophe, Ben's 0–10%, Altman's disowned 2%, Suleyman's conflicting 2023 answers). Musk is the exception: roughly 80% good and 10–20% bad, repeated from 2023 to 2026, with "maybe 90" floated in 2024 and 90/10 named in September 2026. Hassabis, Brockman and Pachocki have never given a number, and by 2026 Dario, Altman and Suleyman had also moved away from unconditional odds.