AI Reskilling and the Panic Cycle
AI is a big deal for knowledge work. Probably not as big as writing, the printing press or the pocket calculator. But the deskilling panic is every bit as loud. If history holds, here’s what comes next: a massive shift in skills, education rebuilt around it, some real casualties, a few things we’ll miss, and a net gain for society.
Open any panic to read the critic’s own words and what the evidence later showed. Entries marked with a speaker are read aloud as they open.
The panics, on a log scale of years before now
Quotations checked word for word against the original texts; translations from Latin are mine. Spoken quotations are performed by AI-designed voices, not recordings of the authors, and only texts out of copyright are voiced. Every figure carries its sample and window.
01
The losses were real
The panics named real losses. None of them predicted the net.
Socrates, as Plato tells it, warned that writing would make people forgetful. He was right: memory moved onto the page, and we remember less for ourselves. He was also wrong about what mattered, and we only have his warning because someone wrote it down. The same shape repeats down the timeline. Trithemius praised the scribe in a book he had printed. Sousa feared the amateur musician would vanish, yet 12% of American adults still played an instrument in 2012 (NEA, n = 37,266).
Not every panic was wrong. Heavy satnav use does seem to wear away spatial memory (Dahmani and Bohbot, n = 50, with 13 followed for three years), and Wikipedia did carry more minor errors than Britannica (3.9 against 2.9 per entry in Nature’s 2005 comparison). I have kept those on the timeline, marked as they are, because a case made only from the wins is a sales pitch.
02
Thinking has always included its tools
A notebook is part of the mind that uses it.
Clark and Chalmers argued in 1998 that a notebook doing the work of memory is part of the thinking, not a crutch outside it. Hutchins showed that a cockpit, not a pilot, remembers the landing speeds: the crew and the markers on the airspeed dial do it together. Risko and Gilbert’s review finds that offloading generally improves performance, across perception, memory, arithmetic and spatial tasks.
So the useful question was never whether we lean on tools. We always have. The question is what we do with the capacity they free up.
03
The new work arrives
Most of today’s work is in jobs that did not exist in 1940.
About 60% of US employment in 2018 was in job titles that did not exist in 1940 (Autor and colleagues, from some 35,000 Census job titles). Between 1982 and 2014, as spreadsheets spread, US bookkeeping clerks fell by about 740,000 while accountants and auditors grew by about 530,000 (BLS Current Population Survey).
To an extent, the pessimists have a point. Acemoglu and Restrepo find that from 1987 to 2017 automation displaced work faster than new tasks replaced it. The net is not guaranteed: it depends on how quickly new work is invented, and that is partly a choice.
04
What’s new is the speed
The worry hasn’t changed. The clock has.
Writing to print took nearly two thousand years. ChatGPT reached an estimated 100 million monthly users two months after launch, where TikTok took about nine months and Instagram two and a half years (UBS and Sensor Tower, reported by Reuters, February 2023). Education adapted to the pocket calculator within a decade: in 1980 the US National Council of Teachers of Mathematics asked that problem solving be the focus of school mathematics, and that programmes “take full advantage of the power of calculators and computers at all grade levels”. Hembree and Dessart’s meta-analysis of 79 studies found calculators improved pencil-and-paper skills, as long as they were introduced appropriately.
AI asks for the same move, faster. Training people to hold on to the old skills of knowledge work is playing the wrong game. The new game is working with AI: defining outcomes, questioning the output the way Socrates questioned his students, and owning what ships.
05
Society nets out. People don’t have decades.
The guards are personal.
The early evidence for AI looks like the calculator’s. In a trial with 453 professionals, ChatGPT cut writing time by 40% and raised quality by 18% (Noy and Zhang, 2023). In a call centre of 5,172 agents, the least experienced gained most, 34%, reaching in two months what had taken six. Yet on a task beyond GPT-4’s competence, consultants using it were right 60 to 71% of the time against 84.5% without it (Dell’Acqua and colleagues, n = 758).
Society had generations to absorb each tool on the timeline. A professional may not. In four Polish clinics, experienced endoscopists who had worked with AI for a few months then found pre-cancerous growths in 22.4% of colonoscopies done without it, against 28.4% before (Budzyń and colleagues, 2025: 19 endoscopists, 1,443 colonoscopies). It is an observational study and the clinics got busier at the same time, so it is a warning rather than a verdict. So I argue for three personal guards. Context: know what the AI was given, what it assumed and what it left out. Expertise: you can’t challenge an assumption you can’t recognise, so keep enough of the domain to judge the output, and grow it. Executive function: keep the planning, the questioning and the deciding in human hands. Managing not-quite-humans is the new core skill, and it stays yours.
Times change. Humans adapt.
(The sooner, the better.)
The everyday practice
DRIVE: five steps for working with AI
The guards say what to keep. DRIVE is how you keep it, every working day: Decide the outcome before you open it, Request with the context it lacks, Iterate while you still hold the thread, Validate with the expertise you kept, and Evolve what you learned into how the next piece of work begins.
Work on AI with your team
I run AI effectiveness programmes for teams and organisations, grounded in how our minds really work: a whole-team picture of where people actually sit, DRIVE in everyday practice, and a promotion path so that what one person learns becomes what everyone can use. If that would be useful, get in touch and we can talk about what your team needs.
Or find out where you stand with AI today
The Cognition Survey is the instrument at the start of the programme. It is free, anonymous, takes about eight minutes, and it will not flatter you. You get a detailed report on where you land, and where the opportunities are.
Related: The Trust Trap, the loop inside one person; AI Workslop and the Slop Seep, where the checking goes; AI Improves Itself. So Can Your Team., the two loops: the labs’ and yours
References
The warnings
Plato. Phaedrus, 274e–275b. Trans. B. Jowett.
Trithemius, J. (1494). De laude scriptorum. Mainz: Peter von Friedberg, ch. VII, fol. 10a.
Gesner, C. (1545). Bibliotheca universalis. Zurich: Froschauer, sig. *3v.
Thoreau, H. D. (1854). Walden, “Economy”.
Sousa, J. P. (1906). The Menace of Mechanical Music. Appleton’s Magazine, 8, 278–284.
Wertham, F. (1954). Seduction of the Innocent. New York: Rinehart.
Carr, N. (2008, July/August). Is Google Making Us Stupid? The Atlantic.
Gentzkow, M., & Shapiro, J. M. (2008), introduction, on 1950s concerns about television.
Gorman, M. (2007, June 11). Web 2.0: The Sleep of Reason, Part I. Britannica Blog.
Christian Science Monitor (1986, May 9). The great calculator debate.
Orben, A. (2020). The Sisyphean Cycle of Technology Panics. Perspectives on Psychological Science, 15(5), 1143–1157.
What followed
Blair, A. (2010). Too Much to Know: Managing Scholarly Information before the Modern Age. Yale University Press.
Dittmar, J. E. (2011). Information Technology and Economic Change: The Impact of the Printing Press. Quarterly Journal of Economics, 126(3), 1133–1172.
Steinwender, C. (2018). Real Effects of Information Frictions: When the States and the Kingdom Became United. American Economic Review, 108(3), 657–696.
National Endowment for the Arts (2013). Highlights from the 2012 Survey of Public Participation in the Arts.
Tilley, C. L. (2012). Seducing the Innocent: Fredric Wertham and the Falsifications that Helped Condemn Comics. Information & Culture, 47(4), 383–413.
Gentzkow, M., & Shapiro, J. M. (2008). Preschool Television Viewing and Adolescent Test Scores. Quarterly Journal of Economics, 123(1), 279–323.
Hembree, R., & Dessart, D. J. (1986). Effects of Hand-Held Calculators in Precollege Mathematics Education: A Meta-Analysis. Journal for Research in Mathematics Education, 17(2), 83–99.
National Council of Teachers of Mathematics (1980). An Agenda for Action: Recommendations for School Mathematics of the 1980s.
Giles, J. (2005). Internet encyclopaedias go head to head. Nature, 438, 900–901.
Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google Effects on Memory. Science, 333, 776–778.
Camerer, C. F., et al. (2018). Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015. Nature Human Behaviour, 2, 637–644.
Dahmani, L., & Bohbot, V. D. (2020). Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports, 10, 6310.
Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). Brain Drain. Journal of the Association for Consumer Research, 2(2), 140–154.
Ruiz Pardo, A. C., & Minda, J. P. (2022). Reexamining the “brain drain” effect. Acta Psychologica, 230, 103717.
Hartmann, M., Martarelli, C. S., Reber, T. P., & Rothen, N. (2020). Does a smartphone on the desk drain our brain? No evidence of cognitive costs due to smartphone presence in a short-term and prospective memory task. Consciousness and Cognition, 86.
Thinking with tools, and the new work
Clark, A., & Chalmers, D. (1998). The Extended Mind. Analysis, 58(1), 7–19.
Hutchins, E. (1995). How a Cockpit Remembers Its Speeds. Cognitive Science, 19(3), 265–288.
Risko, E. F., & Gilbert, S. J. (2016). Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676–688.
Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2024). New Frontiers: The Origins and Content of New Work, 1940–2018. Quarterly Journal of Economics, 139(3).
Acemoglu, D., & Restrepo, P. (2019). Automation and New Tasks. Journal of Economic Perspectives, 33(2), 3–30.
US Bureau of Labor Statistics. Employment and Earnings, January 1983, Table 23; Current Population Survey 2014, Table 11.
AI at work
Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192.
Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at Work. Quarterly Journal of Economics, 140(2), 889–942.
Dell’Acqua, F., et al. (2026). Navigating the Jagged Technological Frontier. Organization Science, 37(2), 403–423.
Budzyń, K., et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. The Lancet Gastroenterology & Hepatology. doi:10.1016/S2468-1253(25)00133-5.
Reuters (2023, February 2). ChatGPT sets record for fastest-growing user base, analyst note (UBS; Sensor Tower).