Education and Its Limits

Education and Its Limits: What Daniel Susskind Gets Right — and Where I'd Push Back

Notes from Daniel Susskind's Gresham College lecture, "Education and Its Limits" (16 May 2026)

Daniel Susskind closed his Gresham College lecture series on the future of work with a talk that should land squarely in the inbox of every school leader, curriculum designer, and policy maker currently drafting an "AI strategy" (Susskind, 2026, 00:10:33). His core claim: more education remains our best response to technological disruption, but what we mean by "more education" has to change — and even a reformed version of education won't be enough on its own (Susskind, 2026, 00:16:29).

Below is my summary of the argument, followed by where I think it holds up and where I'd want to see it stress-tested.

The failure of "future-proofing"

Susskind opens with a cautionary tale from his own career. In 2014, the UK became the first country to mandate that all children learn to code from age five, sold by then-Education Secretary Michael Gove as preparation to "succeed in the 21st century" (Susskind, 2026, 00:18:33–00:20:14). Within a decade, generative AI systems were writing the bulk of their own code — Susskind cites Anthropic's own reporting that a large share of the code behind Claude Code was itself written by Claude Code (Susskind, 2026, 00:24:25–00:24:58). A skill sold as future-proof barely survived a single curriculum cycle (Susskind, 2026, 00:24:58–00:25:33).

His diagnosis isn't that policymakers picked the wrong skill. It's that the entire premise — that we can identify in advance which skills will stay valuable — is close to unfalsifiable optimism dressed up as strategy (Susskind, 2026, 00:25:33–00:26:08). He traces the same pattern across math: Timothy Gowers's tweets, months apart, going from joking that AI solving a famous graph theory problem was an April Fools' gag to reporting, straight-faced, that a system had solved a genuinely hard problem in combinatorial geometry (Susskind, 2026, 00:29:06–00:30:46).

My take: this is the strongest part of the lecture, and it's an uncomfortable one for anyone building an "AI-ready curriculum" right now — myself included. The honest response isn't to find a better list of future-proof skills; it's to admit the list-making exercise itself is the wrong frame.

If not future-proofing, then what?

Susskind reframes the goal as preparing people to navigate uncertainty rather than resolve it, via three levers (Susskind, 2026, 00:32:28–00:32:56):

What we teach. Two priorities: (1) a "back to basics" push on literacy, numeracy, and critical thinking, framed as a no-regret strategy — valuable whatever the future turns out to require (Susskind, 2026, 00:40:04–00:40:30) — and (2) explicit instruction in critical, skeptical use of AI systems, given documented failure modes like the roughly 1,500 legal cases Susskind references in which lawyers cited hallucinated AI-generated case law (Susskind, 2026, 00:41:34–00:42:10).

How we teach. He leans on Bloom's "two sigma" finding — that one-to-one tutoring pushes an average student to outperform 98% of peers in a traditional classroom (Susskind, 2026, 00:43:48–00:44:49) — and argues AI tutoring (his example: Khan Academy's Khanmigo) finally makes that model scalable (Susskind, 2026, 00:44:49–00:45:53). He also proposes borrowing simulated-practice models from aviation, nuclear operations, and Formula 1 engineering: junior professionals rehearsing high-stakes scenarios against AI-generated counterparts before facing the real thing (Susskind, 2026, 00:49:10–00:50:38).

When we teach. Education spending is radically front-loaded — tens of thousands of dollars a year in the US through age 18, and then, in his framing, next to nothing for adult reskilling (Susskind, 2026, 00:50:38–00:51:44). Singapore's SkillsFuture credits are the most ambitious example globally and still amount to roughly $3,000 for workers over 40 (Susskind, 2026, 00:51:44–00:52:18). He calls the imbalance incompatible with a world defined by uncertainty (Susskind, 2026, 00:52:18–00:52:49).

Where education stops being the answer

The most important move in the lecture, and the one most policy conversations skip, is Susskind's insistence that education only addresses part of the problem (Susskind, 2026, 00:54:28–00:54:54). He distinguishes:

• Structural unemployment — there simply aren't enough jobs, full stop.
• Frictional unemployment — jobs exist, but people can't take them, for three reasons: skills mismatch, geographic mismatch, and identity ("this isn't the kind of work people like me do") (Susskind, 2026, 00:54:54–00:55:22).

Education addresses only the first cause of frictional unemployment (Susskind, 2026, 00:55:22–00:55:48). It has nothing to say about geography or identity, and nothing at all to say about a structural shortage of work (Susskind, 2026, 00:56:18–00:56:46). For that scenario, Susskind argues we need a larger distributive state, not a return to 20th-century command-and-control production (Susskind, 2026, 00:57:14–00:58:11); new answers to the "contribution" question — how people feel they're pulling their weight without paid work (Susskind, 2026, 00:58:11–00:59:04); scrutiny of large AI firms' political rather than merely economic power (Susskind, 2026, 00:59:04–01:00:34); and new thinking about meaning and purpose outside employment (Susskind, 2026, 01:00:34–01:01:26).

A critical read

High confidence: the future-proofing critique is well-evidenced and directly actionable for curriculum design — it's a strong argument against locking a school's AI strategy around any specific tool or technical skill set, coding included.

Moderate uncertainty: the claim that declining PISA/PIAAC scores since 2009 are linked to social media's "addictive" design features is presented by Susskind as other commentators' hypothesis, not his own demonstrated finding (Susskind, 2026, 00:35:34–00:36:24) — it's correlational, and the lecture doesn't address confounds (curriculum changes, assessment methodology, economic shocks). I'd treat this as suggestive, not settled.

Where I'd push harder than the lecture does: Susskind's "basic skills = no-regret strategy" argument is intuitively appealing but somewhat circular — basics are defined as "whatever underlies future skills," which is close to unfalsifiable. An audience member (from Accenture) raised exactly this in the Q&A (Susskind, 2026, 01:03:42–01:04:46), and Susskind's answer — essentially "we know it when we see it" (literacy, numeracy, critical thinking) (Susskind, 2026, 01:05:51–01:07:07) — doesn't fully resolve the tension. It's worth asking, in an IB or DP context, whether "critical thinking" is actually a coherent, teachable, assessable basic in the way numeracy is, or whether it's a category that quietly expands to include whatever we currently value.

Also worth flagging: his empathy/soft-skills argument (that AI is encroaching even here, using accounting and therapy as examples) is a genuinely provocative claim but rests on anecdote rather than data in this lecture (Susskind, 2026, 01:08:50–01:10:40). It's a plausible hypothesis, not a demonstrated trend — I'd want harder evidence before building curriculum policy on it.

Why this matters for practice

For those of us working in DP/IGCSE/MYP settings, the lecture's most useful reframe isn't "teach AI literacy" — everyone already agrees on that. It's the structural point that adult, mid-career reskilling is where the system is thinnest (Susskind, 2026, 00:50:38–00:52:49), and that simulated/personalized learning models built for AI tutoring genuinely could compress the tutoring-effectiveness gap Bloom identified in 1984 (Susskind, 2026, 00:43:48–00:45:53). Both are underexplored territory relative to how much attention K–12 "AI-proofing" curricula currently absorb.

One more Q&A point worth carrying into practice: pressed on the digital divide by an audience member working in social mobility (Susskind, 2026, 01:13:50–01:14:22), Susskind conceded that these technologies likely widen the gap for people already outside education, while offering the partial silver lining that effective AI use is arguably better learned outside a traditional classroom than in one (Susskind, 2026, 01:15:52–01:16:26). That's a claim worth testing rather than assuming, particularly for students without home access or independent study skills.

Next step: I'd want to see whether any of the DP subject guides currently in circulation build in explicit "critical use of AI" as an assessed skill rather than a policy footnote — that's the gap Susskind's lecture points at most directly, and it's one worth testing against actual IB assessment criteria rather than leaving as an abstract goal.

---

Bibliography

Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6), 4–16.

Gresham College. (2026). Education and its limits [Public lecture]. Daniel Susskind, Barnard's Inn Hall, London.

Susskind, D. (2015). The future of the professions: How technology will transform the work of human experts. Oxford University Press. (Co-authored with R. Susskind)

Susskind, D. (2020). A world without work: Technology, automation, and how we should respond. Metropolitan Books.