The Singularity at the IMF Lectern: To Whom, and on Whose Behalf, Did Jack Clark Speak?
The Machine Is New, the Chain Is Old: What a Monopoly Executive Who Says "The Singularity Has Begun" Said, and What He Left Unsaid
Dear Young Comrades,
On 17 October 2026, during the IMF–World Bank Annual Meetings in Bangkok, Jack Clark, one of Anthropic's co-founders and its head of policy, delivered the Per Jacobsson Lecture, one of the IMF's most prestigious platforms. Two days later he published the text of the lecture in his own newsletter under the title "In which I tell the IMF the Singularity has begun".
The lecture fits into five sentences:
- The singularity has begun. Artificial intelligence no longer just writes software; it is accelerating science itself, in mathematics, biology and laboratory automation.
- The evidence: jumps on benchmark tests from 2% to as much as 94% in two years, and Anthropic researchers now producing about 10 times as much code as in the company's first four years.
- The promise: "a billion virtual scientists" could be added alongside the world's roughly 10 million scientists. US GDP could reach 44 trillion dollars by 2030.
- The request to states: build electricity, data centres and supply chains. Create laboratory and robotics capacity. Move discoveries quickly into the clinic and the factory.
- The brake: let there be strict human oversight in biology. Let rules be placed on AI companies. Let the technology be developed "slowly and deliberately".
Our reading is five sentences too:
- The leap in the productive forces is real; denying it is not our job. But the singularity is happening in the productive forces, not in the relations of production. Property, wages and profit remain in place, the very old models that Vinge says "must be discarded".
- The 44 trillion dollars Clark cited is the most extreme scenario in Anthropic's own model. In the same scenario, labour's share of national income falls from 60% to 45%, and unemployment in knowledge work rises to 18%. That half of the story was not told in the IMF hall.
- "A billion virtual scientists" is capital distilled from the collective labour of real scientists and owned by a handful of companies. The role left to the scientific worker is that of a foreman supervising thousands of copies.
- The infrastructure demanded of states means the socialisation of costs and the privatisation of profits. Hearing this from the lectern of an institution that has prescribed austerity to peripheral countries for half a century is especially instructive.
- The call to "go slowly and set rules" is the language of a cartel order whose tempo is set by the monopolies. The real question is still the same: In whose hands? Under whose control? For whose benefit?
This piece builds on "On Artificial Intelligence and the Technological Singularity", which we published in August. In that piece we criticised the singularity as "a modern opium", as a myth of the ruling class. Two months later, the co-founder of an AI monopoly has formally proclaimed this myth to the world's finance ministers and central bankers.
Now let us go into the details.
The Lectern, the Speaker, the Audience
What is the Per Jacobsson Lecture?
Per Jacobsson was a Swedish economist who served as Managing Director of the IMF from 1956 to 1963. Since 1964, the foundation bearing his name has organised a lecture on monetary and financial policy at the IMF–World Bank annual meetings. Those who take the lectern are mostly central bank governors, former IMF officials and leading names in mainstream economics.
So this is not a tech fair stage. The audience is the cadre that makes the world's fiscal and monetary policy. A technology company executive speaking about AI from this lectern is a sign that capital now sees artificial intelligence not as a "sector" but as a question of macroeconomic regime.
Who is the speaker?
Jack Clark comes from journalism; as a technology reporter he wrote about data centres. In 2016 he became OpenAI's policy director; in 2021 he was in the team that founded Anthropic, and he is the company's head of policy. For years he has published the "Import AI" newsletter.
In his lecture he positions himself as both narrator and actor: he says he is "one of the principal actors" in what is happening in AI and that he tries "to be candid about his own part". Let us take this declaration of honesty seriously; but let us also note one thing: a confession of personal responsibility is no substitute for an analysis of corporate interest. Nowhere in the lecture are Anthropic's commercial interests, its valuation or its investor relations discussed. We saw the same pattern in "The Class of a Resignation", our piece on Jacob Coxon, who resigned from Anthropic: the problem lies not in individual conscience but in the law of competition.
Who is the audience?
Perhaps the young generation of this country knows the IMF only from textbooks. Let us briefly recall: the decisions of 24 January 1980, the programmes after the 2001 crisis, "structural adjustment", privatisation, cuts in public spending, the suppression of wages... For half a century, the IMF has been the institution that prescribes austerity to peripheral countries.
Now, in the same hall, these words are spoken: spend on energy, data centres, laboratories and robotics. According to Reuters, Clark told policymakers to prepare for this level of growth and to be ready to spend. So the question is not "should the state spend?" but for whom should the state spend? Spend on schools, hospitals and pensioners, and "fiscal discipline" is broken; spend on data centres, and it becomes "investment in the future".
What Did Clark Say? The Course of the Lecture
First, let us be fair and report the lecture as it was.
The claim of a singularity
Clark draws on the 1993 definition by the mathematician and science fiction writer Vernor Vinge: the singularity is the point "where our models must be discarded and a new reality rules". According to Clark, this point has arrived, because AI is now accelerating scientific progress itself.
The evidence
| Field | Clark's example |
|---|---|
| Software | Anthropic researchers produce about 10 times as much code as in the company's first four years; one person does work that teams used to do |
| Cybersecurity | The same capability strengthens both defence and attack, giving rise to an international race |
| Mathematics | AI-proposed solutions for Navier–Stokes, the Jacobian Conjecture and some open Erdős problems |
| Biology | Viruses designed from scratch, drug candidates identified with AI, AI-designed drugs in trials |
| Robotics | Remotely operated laboratory equipment; acceleration in battery and fuel cell research |
| Benchmarks | Humanity's Last Exam: 8% (January 2025) → 67.7%. FrontierMath: 2% (November 2024) → 94%. TheoPhysEval: 98% for GPT-6-Astra once experts corrected errors in their own answer key |
The preview reports published before the lecture mention one more figure: according to Crypto Briefing, as of August 2026 Claude carries out 26% of Anthropic's R&D activity; in February the figure was below 1%.
The promise
Clark recalls the economists' observation that "ideas are getting harder to find": the unit cost of research is rising. Training a researcher takes decades. Yet hundreds or thousands of copies of an AI model can be run instantly. According to UNESCO there are about 10 million scientists in the world; Clark asks: what happens if a billion virtual scientists are added alongside them?
And he gives the figure: according to Anthropic's modelling, in a scenario where AI capable of carrying out broad tasks autonomously spreads rapidly, US GDP could reach 44 trillion dollars in 2030. Clark says this is a scenario, not a forecast.
Recommendations to states
- Infrastructure: electricity capacity, data centres, AI supply chains. Clark says he has been saying this for ten years, but few politicians have listened. Data centre capacity is today concentrated in the US, China and, increasingly, the Middle East.
- Physical science capacity: laboratories, robotics, testing facilities.
- Bottlenecks: "Infrastructure alone is not enough." The transfer of discoveries into the clinic, the factory and the field must be accelerated.
- Developing countries: countries that "supply the basic inputs" can leapfrog by building testing facilities and growing their scientific communities.
Risks and the brake
- "The strictest human oversight and transparency" in biological research.
- How much oversight to apply to AI conducting AI research is a hard decision for labs like Anthropic; rules are needed.
- The question of how far we need to understand equipment and science that AI designs for its own way of working rather than for humans.
- Developing the technology "slowly and deliberately"; distinguishing between what can be done and what should be done.
In closing, Clark recalls the fears of climate catastrophe and scarcity from his youth, refers to Carl Sagan's "Pale Blue Dot", and says humanity has not centuries but a few years to find wisdom and restraint.
What Was Left Unsaid: The Lecture's Negative
The first step in a class reading of a text is to look at what is not said.
| What Clark said | What he did not say | Our question |
|---|---|---|
| The singularity has begun | Property relations have not changed | A singularity for whom, and in whose ownership? |
| GDP could be 44 trillion dollars | In the same scenario, labour's share falls to 45% | If the pie grows, who gets the slice? |
| A billion virtual scientists | The data that trains them is the labour of real scientists | Whose property is the general intellect? |
| One person does a team's work | What happens to the rest of that team? | Where are those who were laid off? |
| Let states build infrastructure | Who will use the infrastructure, and who will get the profit? | The cost is society's; whose is the profit? |
| Developing countries can leapfrog | Computing power stays at the centre | A leap, or a new dependency? |
| Let us go slowly and deliberately | Who sets the speed? | Whose foot is on the brake pedal? |
| Let rules be placed on AI companies | Who will write the rules? | Regulation, or a barrier to entry? |
The words unemployment, wages, union, strike, labour share, inequality do not appear in the lecture. Labour comes up only once: in the sentence saying that AI "augments, extends and automates" some aspects of human labour. It is said that scientists will become "more important and more powerful"; but which scientists?
Where Is the Singularity Happening? Productive Forces and Relations of Production
First, let us turn to ourselves
In "On Artificial Intelligence and the Technological Singularity", we criticised the singularity narrative as an ideological tool of the ruling class and said that the real issue is not machines gaining consciousness but the ownership of technology. We stand by this thesis.
But let us be honest: the benchmark jumps Clark lists are not invented. A mathematics score rising from 2% to 94% in two years, a model carrying out a quarter of a laboratory's R&D, a system that finds errors in the experts' answer key... These point to a real qualitative leap in the productive forces. Marxism does not deny the development of the productive forces; on the contrary, it regards it as the material basis of historical change. We said the same thing in "OpenAI's 722 Mathematics Papers and the Closed Laboratory of the General Intellect": the problem is not the existence of the capability, but its accumulation as private property in a closed laboratory.
In September, meanwhile, we explicitly abandoned the "techno-feudalism" framework of our August piece (see "Reading Anthropic's 2030 Model Through a Class Lens"). What we face is not a feudal order of rent but monopoly capitalism operating through the wage relation. This correction matters for reading Clark's lecture too: Clark proclaims a new age, but everything he describes consists of capitalism's oldest mechanisms, such as accumulation, competition and state support.
Reading Vinge's sentence the other way round
The Vinge definition Clark relies on defines the singularity as the point where "our models must be discarded". So which models are being discarded?
- Private property is not being discarded. The models, the weights and the computing power are the property of a handful of companies.
- Wage labour is not being discarded. From the data labeller to the software engineer, labour power is still being sold.
- The profit motive is not being discarded. The 44-trillion-dollar figure is itself the language of valuation.
- Competition between nation states is not being discarded. Clark himself speaks of an "international race" in cybersecurity.
So if there is a singularity, it exists only in the productive forces. The relations of production remain stubbornly the same. This is exactly the contradiction Marx describes in the preface to A Contribution to the Critique of Political Economy: at a certain stage of development, the productive forces come into conflict with the property relations within which they have been operating; from forms of development of the productive forces, these relations turn into their fetters.
Let us recall our slogan once more: The machine is new, the chain is old.
The function of singularity talk
"Singularity" is a term from physics; it describes the point at the centre of a black hole where the known laws break down. Using this word for a social process does three ideological jobs:
- Inevitability: the process is presented like a natural phenomenon. One does not organise against a natural phenomenon; one "adapts" to it.
- Ahistoricity: saying "the old models are invalid" implies that class analysis, the theory of value and the calculation of labour's share are invalid too.
- Urgency: "We have not centuries but a few years." Urgency is the oldest way of sidelining democratic debate. The decision must be made quickly; therefore it must be made by "those who know", that is, the monopolies and the state cadres they consult.
The Untold Half of the 44 Trillion
The 44 trillion dollars Clark cited at the IMF comes from the economic scenarios published by the Anthropic Institute in September. We examined these scenarios in detail in "Reading Anthropic's 2030 Model Through a Class Lens". Let us recall:
| Scenario | US national income in 2030 | Labour's share | Unemployment in knowledge work |
|---|---|---|---|
| Today's structure (baseline path) | about $33.5 trillion | 60% | – |
| Moderate | $34.1 trillion | 59.4% | – |
| Significant | $36.3 trillion | 56.1% | – |
| Extreme (the one Clark cited) | $44.4 trillion | 45.2% | 17.9% (overall unemployment 11.9%) |
Read the table once more. The figure Clark cited in the IMF hall is the figure of the most extreme scenario. In the same scenario:
- The economy grows by about a third.
- Labour's share falls by 15 points; total labour income barely moves.
- Knowledge workers' wages fall by about 11.5%.
- One in six people in knowledge work is left unemployed.
In other words, almost all of the growth goes to capital. In his lecture Clark mentioned growth, but not distribution. Yet those in the hall know better than anyone what distribution means: a falling labour share means shrinking aggregate demand, a shifting tax base and shaken social security systems.
There are also those who view the figures with scepticism. According to Startup Fortune, Gary Marcus describes Anthropic's framework of tens of trillions of dollars as "fantasy"; the same report points out that the bonds of the data centre company CoreWeave have lost value. We are not betting on whether the figure will hold. For us, both possibilities are instructive:
- If the figure holds: capital will take most of the growth; labour's share will collapse.
- If the figure does not hold: this discourse is the language of fictitious capital (valuations inflated by expectations about the future). When the bubble bursts, the bill will again be paid by workers, pensioners and taxpayers.
Either way, singularity talk is also capital market talk. When an executive of an AI company says "the singularity has begun", he is also telling investors and the state: "put your money here".
"A Billion Virtual Scientists": The Second Expropriation of the General Intellect
Who trained the virtual scientist?
Clark's most striking image is "a billion virtual scientists". But these virtual scientists were not born out of nothing. The data that trained them is centuries of scientific publications, open access archives, textbooks, code repositories, questions answered on forums, laboratory notebooks... In other words, the collective labour of real scientists and of all humanity.
In the Grundrisse, Marx gives this accumulation the name "general intellect": social knowledge becoming a direct force of production. What we have been saying since "The Expropriation of the General Intellect and the AGI Illusion" is this: artificial intelligence is the general intellect turned into private property. Clark's "billion virtual scientists" is in fact an item of fixed capital, distilled from the labour of ten million real scientists (and their predecessors) and sitting on the balance sheets of a few companies.
October is the month of International Open Access Week. Librarians, researchers and activists who have fought for decades for scientific knowledge to be open to all are watching the open knowledge they produced become raw material for closed models (see "What Is International Open Access Week, and Why Does It Matter?"). Open input, closed output: that is enclosure.
The role left to the scientific worker: Foreman
According to Clark, a single researcher will be able to manage hundreds or thousands of AI copies. He presents this as "scientists will become more important and more powerful".
What is actually being described is Taylorism brought into science:
- The research process is broken into pieces.
- Creative and interpretive work is given to the machine; supervision and approval to the human.
- A handful of "senior" researchers become foremen; the rest, the assistants, postdoctoral researchers and technicians, become redundant.
Most of the world's 10 million scientists are already precarious: an academic precariat working on short-term projects and project-based contracts, for low pay. The other side of the sentence "one person does a team's work" is that the rest of the team is left unemployed. This is exactly what we call mental Taylorism.
Whose property will the discovery be?
Clark wants discoveries to be "moved quickly into the clinic and the factory". So who will own the patent on a drug candidate proposed by an AI? The owners of the data that trained the model, the company that runs the model, or the public that funds the laboratory? This question is absent from the lecture. Yet it is exactly this question that determines the price of the drug, that is, who lives and who dies.
The Call for Infrastructure: The Cost Is Society's, the Profit Is the Monopoly's
Clark's most concrete recommendation to states is infrastructure: electricity, data centres, supply chains, laboratories.
At first glance this recommendation looks like "public investment". But let us ask these questions:
- Energy: Data centres consume enormous amounts of electricity and water. When the cost of expanding capacity is passed on to households through their electricity bills, will we call it "public investment" or a "transfer from households to the monopolies"?
- Land and incentives: The tax exemptions, cheap land and zoning concessions given to data centres come out of whose budget?
- Use: Whose models will run on the infrastructure the state builds? The public's, or the monopoly's?
This is a familiar pattern of capitalism: the cost of infrastructure is socialised, the return is privatised. That is how it went with the railways, with power plants and with the backbone of the internet. We asked the same question in "Against Whom Is 'Physical AI' Arming?", our piece on physical AI and laboratory automation.
For Turkey, too, this debate is not abstract. We saw it in what we wrote on "Turkey's Artificial Intelligence Action Plan (2026–2030)" and on "the Directorate General of Public Artificial Intelligence": the state treats AI as a matter of "infrastructure and competition"; labour's share, data ownership and democratic oversight are not on the agenda. Clark's lecture at the IMF can be read as an international endorsement of this view.
The Promise of a "Leap" and the International Division of Labour
Clark offers developing countries a hope: a country that "supplies the basic inputs" can leap forward by building testing facilities and growing its scientific community.
Note the phrase: "supplies the basic inputs". These inputs are cobalt, lithium, rare earth elements, copper, water, cheap energy and cheap labour. The data labelling workers are in Kenya, the Philippines and Venezuela; the miners in Congo; the chip factories in Taiwan; the computing power in the US, China and the Gulf.
This picture is not new. It is the current form of the international division of labour that Lenin described in Imperialism:
- The centre: capital, technology, patents, computing power.
- The periphery: raw materials, cheap labour, energy, "testing facilities".
History has shown how realistic the promise of a "leap" is. The "information society" promise of the 1990s and the "bridging the digital divide" programmes of the 2000s did not bring the periphery closer to the centre; they reproduced dependency in new forms. Today, too, a country's "leap" is impossible without touching the ownership of models, weights and computing power. We discussed this in "The Illusion of 'Indigenous and National' Artificial Intelligence" as well: a nationalist "our own model" move is no solution either, as long as class relations stay the same.
One more detail: the lecture was given in Bangkok. The host, Thailand, with its electronics assembly and the foreign investment it attracts through incentives, is itself a peripheral economy of the kind Clark describes.
"Slowly and Deliberately": Whose Foot Is on the Brake Pedal?
At the end of his lecture, Clark says the technology should be developed "slowly and deliberately", that rules should be placed on AI companies, and that international rules must be "unusual and serious". At first glance this looks like a call for responsibility.
But we have heard this call before. We examined Dario Amodei's call about tempo in "Whose Foot Is on the Brake?". We said this there, and we repeat it here:
- Who sets the speed? The one saying "let us go slowly" is the one running at the front. If, when he slows down, his rivals will have to slow down too, that is not responsibility but a cartel agreement.
- Who writes the rules? Complex, expensive compliance rules hurt small players, open-source communities and public research most. The big monopolies, meanwhile, can easily bear the cost of compliance. Regulation thus turns into a barrier to market entry.
- Who does the oversight? Clark says that how much oversight to apply to AI conducting AI research is "a hard decision for labs like Anthropic". The lab makes the decision. The overseen does the overseeing. The question posed in our pieces "Can Companies Audit Themselves?" and "Whose Hand Is on the Shutdown Button?" is still unanswered.
Our idea of a "brake" is different: the tempo of technology should be set by workers and society, who use it, are affected by it and own the data that trains it. AI use in the workplace becoming a subject of collective agreements, data centres' energy and water consumption being subject to the consent of local people, public models for public research... The brake should be not under the monopoly's foot but in the hands of organised labour.
Reactions: Who Said What?
In the mainstream press, the lecture was reported largely through its growth figure. CNBC, Reuters and the Asian press (e.g. Seoul Economic Daily, Manila Times) ran the headline "AI could vastly boost US GDP". The crypto and start-up press read the lecture as an investment opportunity and a warning of "sudden change".
| Who | What they said | Why they said it | Do they question ownership? |
|---|---|---|---|
| Mainstream business press | "GDP could explode" | Its investor readership is interested in the growth figure | No |
| Start-up/crypto press | "Sudden change is coming, get ready" | It feeds the valuation and investment cycle | No |
| Gary Marcus and the sceptics | "The figures are fantasy" | They criticise technical hype | No; they only question whether the figure is accurate |
| The financial market (CoreWeave bonds) | Silent but meaningful: a loss of value | Doubt about the return on infrastructure debt | No; but it is pricing in bubble risk |
| Our reading | "The singularity is in the productive forces; the chain is in place" | We ask who owns labour and the general intellect | Yes |
Note: the enthusiasts and the sceptics are arguing on the same ground: Will there be growth or not? No one asks who will the growth go to? The class perspective exists precisely to fill this gap.
Concrete Tasks for Young Comrades
The lesson to draw from this lecture is not "AI is bad". The lesson is this: we have to tie the leap in the productive forces to the struggle to change the relations of production. Concretely:
- Bring the question of distribution into every debate. When someone says "AI will increase GDP by this much", ask: What will happen to labour's share? To wages? To working hours?
- Let scientific workers organise. Research assistants, postdoctoral researchers and laboratory technicians must put the use of AI in academia, workloads and employment on the union agenda. Eğitim Sen (the Education and Science Workers' Union) and university workers need to speak up on this.
- Make AI in the workplace a subject of collective agreements. IT workers should demand that agreements set out how AI tools will be used in the workplace, how productivity gains will be shared and who cannot be dismissed.
- Monitor infrastructure investment. Follow data centre incentives, energy deals and water use in Turkey. The professional chambers can lead public oversight in this field (see "Reading and Developing the TMMOB's Artificial Intelligence Report").
- Defend open science and open models. Against publicly funded knowledge becoming raw material for closed models, demand models trained with public funds and under public oversight. That is why the theme of ÖYD's (the Free Software Association's) Odak 2026 conference matters: "The People's Data, the People's Model."
- Do your own "singularity" reading. The next time you see a "the singularity is here" story, ask four questions: What is being sold? From whom was it taken? To whom is it being sold? Who pays the price?
Not a Singularity, but Class Struggle
Dear Young Comrades,
At the IMF lectern, Jack Clark invoked Carl Sagan and called on humanity for wisdom and restraint. It is hard not to agree with this call. But wisdom is not multiplying knowledge without asking to whom it belongs. Nor is restraint the monopolies deciding who will brake in a race whose tempo the monopolies set.
Clark said "the singularity has begun". Perhaps, as far as the productive forces are concerned, he is right. Machines are doing things that could not even be imagined two years ago. But these machines are humanity's collective intelligence of thousands of years turned into the property of a few companies. And that property relation remains in place as the oldest of the old models that Vinge says "must be discarded".
The real singularity, the moment when "the old models are discarded" for real, will come on the day this general intellect is returned to those who produced it. That will be the day when history is written not by machines but by organised labour.
The machine is new, the chain is old. And what will break the chain is not the machine; it is us.
Our Related Pieces
- On Artificial Intelligence and the Technological Singularity (9 August 2026): the foundational text this piece builds on.
- Reading Anthropic's 2030 Model Through a Class Lens: a detailed analysis of the 44-trillion scenario and a self-criticism of techno-feudalism.
- Whose Foot Is on the Brake?: Amodei's call about tempo.
- The Expropriation of the General Intellect and the AGI Illusion
- OpenAI's 722 Mathematics Papers and the Closed Laboratory of the General Intellect
- Reading GPT-6 Astra Through a Class Lens
- "We Stopped Them All": A Monopoly's Grammar of Self-Absolution
- The Class of a Resignation
- Whose Hand Is on the Shutdown Button?
- Can Companies Audit Themselves?
- Against Whom Is "Physical AI" Arming?
- What Are the Artificial Intelligence Summits Selling?
- The Week's Class Developments in Technology — 13–19 October 2026
Sources
- Jack Clark, "Import AI 477: In which I tell the IMF the Singularity has begun", 19 October 2026 (full text of the lecture).
- Jack Clark, post on X announcing the livestream of the lecture, 17 October 2026.
- CNBC, "AI rapidly improving, could vastly boost U.S. GDP, Anthropic's Clark tells policymakers", 17 October 2026.
- News Tribune (Reuters), "AI rapidly improving, could vastly boost US GDP, Anthropic's Clark tells policymakers", 19 October 2026.
- Crypto Briefing, "Anthropic's Jack Clark takes the AI 'sudden change' warning to the IMF" (preview before the lecture).
- Startup Fortune, "Anthropic's Jack Clark Tells the IMF AI Could Push US GDP to $44 Trillion".
- Seoul Economic Daily, "Anthropic Co-Founder Sees U.S. GDP Rising 50% in Four Years", 18 October 2026.
- Manila Times, "Anthropic's Clark sees AI boosting US GDP", 19 October 2026.
- TokenPost, "Anthropic Co-founder Jack Clark to Discuss AI's Economic Impact at IMF Meetings".
- Karl Marx, Grundrisse ("Fragment on Machines"); A Contribution to the Critique of Political Economy, Preface (1859).
- V. I. Lenin, Imperialism, the Highest Stage of Capitalism (1916).
- Vernor Vinge, "The Coming Technological Singularity" (1993).
This piece is one of Knowledge Commons' living documents; it may be updated as things develop. It is shared under a CC BY-SA 4.0 licence: you are free to share it, citing the source and under the same licence.
