Against Criti-Hype: Artificial Intelligence, Marxism, and the Politics of Deskilling

Debates about artificial intelligence are organized around an agreement that goes unnoticed because it sits under a loud disagreement. Firms selling the technology say it will replace human thinking, automate knowledge work, and remake civilization, and many of the sharpest critics accept that description in full and reverse the valuation, arguing that because AI will replace thinking it will destroy meaningful labor. Forecasts that close to half of American employment sat at high risk of computerization circulated for a decade as both sales material and indictment, and the same numbers did both jobs without modification (Frey and Osborne 2017; Brynjolfsson and McAfee 2014). Lee Vinsel named the pattern criti-hype, and the point of the term is that criticism borrowing its factual premises from marketing performs promotional labor for the thing it opposes.

A recent Behind the News segment featuring Hagen Blix shows where the reduction happens. The framing runs through Braverman and the computer numerical control machine, where knowledge once held by skilled machinists was transferred into the equipment, and then carries the case over to artificial intelligence as though the extension were obvious, with mental workers standing where metal workers stood (Behind the News 2026; Braverman 1974). The obviousness is the tell, since an analogy that need not be argued is usually one whose conclusion was available before the evidence.

What Blix and Glimmer argue in print is different and considerably better. Their case is that fears about artificial intelligence are displaced fears about capitalism imagined as an autonomous agent, and that cheapened skills, downward pressure on wages, and expanded insecure work are outcomes capital would prefer rather than outcomes the technology delivers by itself, which they say plainly is not inevitable (Blix and Glimmer 2025). The reduction happens in transmission. Detached from the argument that produced it and repeated on its own, a claim about what capital wants becomes a claim about what machines do.

Much of the underlying position is correct and I do not want to give any of it up. Firms pursue technologies that lower labor costs and tighten managerial control, and describing technology as politically neutral has always been a way of making those decisions vanish from view (Braverman 1974). The numerical control case shows it cleanly, since a record-playback alternative that preserved machinist knowledge existed, worked, and lost on grounds of control rather than efficiency (Noble 1984).

Trouble starts when the insight hardens into a law of motion, so that machines performing interpretive work implies the deskilling of whoever once did it. Held that way, opposition to a technology stops being a conclusion reached about a particular case and becomes a starting position, which is roughly where a great deal of Anglophone left commentary on artificial intelligence now sits, including commentary drawing on arguments considerably better than the one it ends up repeating. The inference was contested inside the sociology of work almost as soon as it appeared, and decades of case studies never resolved in either direction because occupations upgrade and degrade at once across sectors and firms (Attewell 1987; Form 1987). Skill is not a single measurable quantity in any case, and much of what gets counted as skill is a claim about status and bargaining position rather than a description of cognitive demand (Vallas 1990). Machines have in any case been interpreting for a century without producing the predicted collapse, since sensors read vibration and tolerance continuously, process control software flags anomalies no operator could see, and imaging systems read scans. What close observation of computerized plants found was not the disappearance of judgment but its relocation, since information technology automates a process and simultaneously renders it visible in ways that place new demands on whoever watches the screen (Zuboff 1988). Whether those demands become developed capacities depends on organization, and formally identical structures produce opposite results depending on whether they are built to enable or to coerce (Adler and Borys 1996).

This is the point of the paleo-Marxist critique of labor process theory, which holds that the field became insufficiently Marxist by attending only to valorization while neglecting Marx on the socialization of production, so that the long-run development of technical knowledge and cooperative complexity gets treated as an anomaly rather than a tendency (Adler 2007). The period that produced Braverman also produced arguments that automation had opened a real choice between mass production and skill-intensive flexible production, with the outcome turning on which coalitions won (Piore and Sabel 1984).

None of this is a claim that American work has been getting better, and the strong reading draws its persuasive force from the fact that it has not. Demand for cognitive skill reversed around the turn of the century, with college-educated workers pushed down the occupational ladder into jobs once held by high school graduates, who were pushed further down or out (Beaudry, Green, and Sand 2016). Polarization hollowed the middle rather than lifting the average, and growth concentrated in low-wage service work requiring little of the systems thinking the upgrading story anticipated (Autor and Dorn 2013).

The question is what did that, and the technological answer fails first. This was not the forecast. The dominant economic account held that technological change was skill biased, that demand for educated labor would keep rising, and that American inequality reflected an education supply falling behind a race it had been winning (Goldin and Katz 2008; Autor, Levy, and Murnane 2003). The prediction failed on its own terms, and the timing never fit either, since inequality accelerated in the 1980s in ways computer diffusion does not track and returns to schooling moved in patterns the model cannot accommodate without repeated adjustment (Card and DiNardo 2002; Mishel, Schmitt, and Shierholz 2013).

What happened instead was a divestment from human capital that no technology required. Employers who complain loudest about skill gaps have cut the training that would close them, and the evidence for a general shortage of skilled American workers is thin while the evidence for collapsing employer investment is not (Cappelli 2015). Firms dismantled the internal labor markets that once carried workers through long careers with employer-funded upgrading, transferring the cost of skill formation onto individuals told to think of themselves as their own enterprises (Cappelli 1999). Fissuring finished it, since a company that contracts out its work has no reason to develop people who are not its employees and every reason to demand they arrive pre-qualified (Weil 2014).

Shareholder primacy made this rational at the level of the firm, since money spent developing workers is money not returned to owners, and once the corporation was reconceived as a bundle of assets managed for share price rather than an organization that retains and reinvests, buybacks became the standard disposition for earnings that had financed training (Lazonick and O’Sullivan 2000; Lazonick 2014). Deunionization removed the countervailing force, taking with it the wage floor and the joint apprenticeship machinery unions had administered, and it accounts on its own for a substantial share of the growth in American wage inequality (Western and Rosenfeld 2011; Rosenfeld 2014).

That these were choices rather than necessities becomes obvious as soon as the comparison crosses a border. The same machines arrived in Germany, Austria, Denmark, and Japan and produced different skill regimes, because employers, unions, and states were bound into institutions that made training a shared obligation rather than a private cost to avoid (Thelen 2004; Busemeyer and Trampusch 2012). Skill formation is a collective good markets underproduce, since any firm that trains risks having the investment poached, and only institutions that penalize defection sustain a high-skill equilibrium (Streeck 1989; Becker 1964). The United States dismantled what little of that apparatus it had and called the result technology.

The comparison is running again right now, and artificial intelligence makes it unusually clean. Chinese labs have spent two years giving the technology away, shipping frontier-competitive models under MIT and Apache licenses at a cadence Western firms have not matched, while the state expanded the human capital pipeline rather than treating it as redundant. More than six hundred Chinese universities offered artificial intelligence undergraduate degrees by 2026, against thirty-five institutions approved in 2018, and at least ninety Double First-Class universities had established AI schools or colleges (China Daily 2026a). The Ministry of Education added thirty-eight undergraduate majors to the national catalogue for the 2026 cycle, concentrated in integrated circuits, embodied intelligence, energy, and materials, and leading universities expanded intake in those fields (China Daily 2026b).

The United States ran the opposite experiment in the same window, and on top of four decades of falling state appropriations per student that had already shifted the cost of higher education onto debt (Newfield 2016; Mettler 2014). Doctoral admissions at fifty-five major research universities fell fifteen percent for fall 2026, the second consecutive year of substantial reduction, driven by contraction and uncertainty in federal research funding, while the National Science Foundation closed a social science directorate and the following budget request sought deeper cuts still (AAU 2026; Brennan Center for Justice 2026). No technological development required any of this. If artificial intelligence had made trained scientists redundant, someone forgot to tell the country expanding its engineering faculties while giving the models away.

The direction of technological development is selected rather than given, and tax codes that subsidize capital over payroll tilt it toward displacement instead of augmentation before any engineering decision is made (Acemoglu and Johnson 2023). Automation only marginally better than the labor it replaces displaces workers without generating the productivity gains that would fund reabsorption elsewhere, a bad outcome firms still have reason to choose under current incentives (Acemoglu and Restrepo 2019). Cheap computation does not determine whether human judgment becomes more valuable or simply cheaper. Institutions determine that.

The political cost of criti-hype is fatalism. If deskilling follows from the machine, there is nothing to organize against and the available responses are mourning and refusal. If it follows from training budgets, contracting structures, bargaining coverage, tax treatment, and appropriations bills, it has an address and decision makers who can be named. The numerical control case is instructive for the reason its citers usually skip, which is that the outcome turned on a fight one side lost and could have won (Noble 1984).

The danger is not that machines are learning to think. It is that a story about machines learning to think is available to explain away forty years of decisions about whose capacities were worth developing, and that the industry and its critics both have reasons to keep telling it.

A companion piece takes up the stronger versions of this argument, in David Harvey and Aaron Benanav, and asks why the capital intensity that would have to hold for labor to become inconsequential is itself a political arrangement rather than a fact about machines.

Sources

• AAU (Association of American Universities). 2026. New PhD Admissions Data Show Threat to U.S. STEM Workforce, Breakthroughs, Innovations. AAU Data Exchange report.
• Brennan Center for Justice. 2026. The Cost of the Trump Administration’s Attacks on Research Funding. Brennan Center for Justice.
• China Daily. 2026a. “Universities Cater to AI Study Demand.” China Daily Hong Kong, June 21.
• China Daily. 2026b. “China’s Top Universities Expand Enrollment in AI, Strategic Fields.” China Daily Hong Kong, June 25.
• Acemoglu, Daron, and Simon Johnson. 2023. Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity. PublicAffairs.
• Acemoglu, Daron, and Pascual Restrepo. 2019. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives 33(2):3–30.
• Adler, Paul S. 2007. “The Future of Critical Management Studies: A Paleo-Marxist Critique of Labour Process Theory.” Organization Studies 28(9):1313–1345.
• Adler, Paul S., and Bryan Borys. 1996. “Two Types of Bureaucracy: Enabling and Coercive.” Administrative Science Quarterly 41(1):61–89.
• Attewell, Paul. 1987. “The Deskilling Controversy.” Work and Occupations 14(3):323–346.
• Autor, David H., and David Dorn. 2013. “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market.” American Economic Review 103(5):1553–1597.
• Autor, David H., Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118(4):1279–1333.
• Beaudry, Paul, David A. Green, and Benjamin M. Sand. 2016. “The Great Reversal in the Demand for Skill and Cognitive Tasks.” Journal of Labor Economics 34(S1):S199–S247.
• Becker, Gary S. 1964. Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education. National Bureau of Economic Research.
• Behind the News. 2026. “A Vision of Democratic Socialism, with Michaela Brangan” (segment with Hagen Blix). Jacobin Radio, July 27.
• Blix, Hagen, and Ingeborg Glimmer. 2025. Why We Fear AI: On the Interpretation of Nightmares. Common Notions.
• Braverman, Harry. 1974. Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. Monthly Review Press.
• Brynjolfsson, Erik, and Andrew McAfee. 2014. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. Norton.
• Busemeyer, Marius R., and Christine Trampusch, eds. 2012. The Political Economy of Collective Skill Formation. Oxford University Press.
• Cappelli, Peter. 1999. The New Deal at Work: Managing the Market-Driven Workforce. Harvard Business School Press.
• Cappelli, Peter. 2015. “Skill Gaps, Skill Shortages, and Skill Mismatches: Evidence and Arguments for the United States.” ILR Review 68(2):251–290.
• Card, David, and John E. DiNardo. 2002. “Skill-Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles.” Journal of Labor Economics 20(4):733–783.
• Frey, Carl Benedikt, and Michael A. Osborne. 2017. “The Future of Employment: How Susceptible Are Jobs to Computerisation?” Technological Forecasting and Social Change 114:254–280.
• Form, William. 1987. “On the Degradation of Skills.” Annual Review of Sociology 13:29–47.
• Goldin, Claudia, and Lawrence F. Katz. 2008. The Race between Education and Technology. Harvard University Press.
• Lazonick, William. 2014. “Profits Without Prosperity.” Harvard Business Review 92(9):46–55.
• Lazonick, William, and Mary O’Sullivan. 2000. “Maximizing Shareholder Value: A New Ideology for Corporate Governance.” Economy and Society 29(1):13–35.
• Mettler, Suzanne. 2014. Degrees of Inequality: How the Politics of Higher Education Sabotaged the American Dream. Basic Books.
• Mishel, Lawrence, John Schmitt, and Heidi Shierholz. 2013. Assessing the Job Polarization Explanation of Growing Wage Inequality. Economic Policy Institute Working Paper.
• Newfield, Christopher. 2016. The Great Mistake: How We Wrecked Public Universities and How We Can Fix Them. Johns Hopkins University Press.
• Noble, David F. 1984. Forces of Production: A Social History of Industrial Automation. Knopf.
• Piore, Michael J., and Charles F. Sabel. 1984. The Second Industrial Divide: Possibilities for Prosperity. Basic Books.
• Rosenfeld, Jake. 2014. What Unions No Longer Do. Harvard University Press.
• Streeck, Wolfgang. 1989. “Skills and the Limits of Neo-Liberalism: The Enterprise of the Future as a Place of Learning.” Work, Employment and Society 3(1):89–104.
• Thelen, Kathleen. 2004. How Institutions Evolve: The Political Economy of Skills in Germany, Britain, the United States, and Japan. Cambridge University Press.
• Vallas, Steven P. 1990. “The Concept of Skill: A Critical Review.” Work and Occupations 17(4):379–398.
• Weil, David. 2014. The Fissured Workplace: Why Work Became So Bad for So Many and What Can Be Done to Improve It. Harvard University Press.
• Western, Bruce, and Jake Rosenfeld. 2011. “Unions, Norms, and the Rise in U.S. Wage Inequality.” American Sociological Review 76(4):513–537.
• Zuboff, Shoshana. 1988. In the Age of the Smart Machine: The Future of Work and Power. Basic Books.
• Charles, Will. “Labor Will Not Become Inconsequential.” BadHabitus.
• Charles, Will. “Artificial Intelligence Needs Intelligent Societies.” BadHabitus.
• Charles, Will. “The AI Substitution Trap.” BadHabitus.