Simon Johnson on Banking, Technology, and Prosperity

Simon Johnson — MIT economist, former IMF chief economist, and co-author with Daron Acemoglu of Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity — talks with Tyler Cowen ten days after the collapse of Silicon Valley Bank, so the banking material is a live, unsettled reaction to an ongoing crisis, not a retrospective. The conversation then turns to the book’s argument: technology’s gains are never automatic, and whether automation creates new tasks or simply displaces workers is a matter of institutional choice, not technological necessity.

Key ideas

1. The Silicon Valley Bank response was live improvisation, not settled policy. Recorded ten days after SVB’s failure, Johnson describes an unresolved regulatory scramble: a 2018 rule change had raised the systemic-oversight threshold from $50 billion to $250 billion in total assets, letting SVB slip below the line that would have triggered closer supervision. He endorses Sheila Bair’s proposed alternative — haircut SVB’s uninsured depositors while insuring uninsured deposits at every other bank — as a way to punish the specific failure without triggering system-wide contagion.
2. ‘So-so technology’ is Johnson’s diagnostic for when automation fails workers. Coined by Daron Acemoglu and Pascual Restrepo, the term names technology adopted to displace labour even when it does not raise productivity — the self-checkout kiosk being the clearest case: work shifts onto the customer, and no wage increase follows. This is the mechanism at the centre of Power and Progress.
3. Whether technology’s gains are shared is a matter of choice, not inevitability. Johnson rejects the ‘techno-optimist’ view that productivity gains flow automatically to ordinary people. The Industrial Revolution eventually raised living standards broadly, but only after a century in which — by Johnson’s account of Manchester’s factories and the coal-mine child-labour inquiries of the 1840s — conditions for workers got worse, not better.
4. AI risks repeating the pattern of the past fifty years of ‘so-so’ digital technology. Johnson expects AI, absent deliberate task creation, to displace workers from well-paying jobs without lifting their productivity or wages — while paying algorithm designers more. He is agnostic on where AI’s commercial value will ultimately concentrate, noting OpenAI’s then-$30 billion valuation was strikingly modest next to the scale of ChatGPT’s adoption.
5. Institutions predict long-run resilience, not short-run growth. Pressed on why institution-based theories failed to predict South Korea’s takeoff or Ethiopia and Ghana’s more recent stalls, Johnson concedes institutions offer little forecasting power over one, two, or five years — their value is a ‘hysteresis effect’ that makes sustained growth harder to reverse once institutions strengthen, not a short-term growth signal.
6. He disputes Joel Mokyr’s reading of how fast the Industrial Revolution paid off. Where Mokyr’s own research finds wages rose quickly in industrialising English regions, Johnson insists — citing Engels on Manchester and the 1840s coal-mine inquiries — that ordinary living standards did not visibly improve until after 1850, roughly a century after the Revolution began.
7. Writing the book changed his view of the Middle Ages. Johnson had assumed a ‘dark age’ of stagnant innovation between Rome and the Renaissance; researching the book showed him substantial medieval innovation in agriculture, commerce, and early industry — innovation that enriched the church and state (cathedral-building funded by squeezed peasant labour) without reaching ordinary people, the same failure-to-share pattern he applies to the Industrial Revolution and, potentially, to AI.

Content

The Silicon Valley Bank weekend, ten days on

Cowen opens on banking, not the book, because the interview was taped on 21 March 2023, days after SVB’s collapse. Johnson situates the failure in a specific regulatory history: he had testified to the Senate in 2015 that banks above $50 billion needed systemic scrutiny, but ‘the view coming out of the reforms or changes of 2018 was that $250 billion total assets was where systemic attention should start to be paid. Silicon Valley Bank was smaller than that.’ The 2018 rollback of the Dodd-Frank threshold, on his account, is why SVB escaped the closer supervision a larger bank would have faced.

On the remedy, Johnson credits Sheila Bair’s proposal from that weekend: SVB’s uninsured depositors — mostly holding high-quality but rate-depressed long-term government bonds — should have taken a haircut of roughly ten cents on the dollar, while every other bank’s uninsured deposits should have been guaranteed to stop the run from spreading. He is careful to distinguish this from insuring all deposits everywhere, which he calls a last resort: ‘I think insuring all deposits is something you only do when you’re absolutely desperate.’ His preferred narrower fix is a dedicated insurance category for small-business transaction accounts — payroll and working capital — so a nine-person startup is not forced into treasury management it has no capacity for.

Cowen presses him on the credibility problem: once regulators demonstrate they will protect uninsured deposits at one bank, the signal is ‘infinitely large’ regardless of the failing bank’s size, because roughly half of US bank deposits are uninsured. Johnson does not resolve this tension — he calls it ‘a very big problem, Tyler, honestly’ — and instead points to the practical stakes: a flight of deposits from regional and mid-size banks to the largest banks would disrupt regional lending, since large banks are not well suited to replace what community and regional banks do for non-financial borrowers.

Credit Suisse, AT1 bonds, and the limits of ‘too big to fail’ for small states

A parallel thread covers Credit Suisse’s forced merger into UBS the same week. Johnson notes the equity holders received $3 billion for a bank worth $8 billion days earlier, while the bank’s AT1 (‘CoCo’) convertible bonds were written to zero — a reversal of the normal capital-structure hierarchy that he expects to reprice the entire CoCo market. Switzerland’s capacity to backstop UBS turns, in his framing, on the relative size of the Swiss National Bank’s roughly $1 trillion in foreign-exchange reserves against Credit Suisse’s roughly $500 billion balance sheet — a comparison he contrasts with Iceland in 2008, where the state could not stand behind its outsized banks at all. He remains uneasy that the rescue leaves Switzerland with one now-larger bank rather than several smaller ones, an outcome that runs against the diversification he favours.

The Power and Progress thesis: technology and choice

Turning to the book, Johnson frames its argument as a challenge to ‘some modern techno optimism view… that technology just kind of happens. It raises productivity. It improves how people live… and everyone benefits eventually.’ The historical record, he argues, is far messier: technology sometimes pays off broadly, but only when institutions push its gains toward ordinary workers rather than away from them.

The book’s central mechanism is ‘so-so technology’ — Acemoglu and Restrepo’s term for automation adopted because it displaces labour, even when it fails to raise productivity. Johnson’s illustration is the supermarket self-checkout kiosk: ‘you shift the work onto the consumers. You’re not making the workers more productive. You don’t see increases in the wages in supermarkets where they adopt self-checkout kiosks.’ He contrasts this with the earlier Henry Ford-era pattern, in which automation created new tasks that raised unskilled workers’ productivity, and that productivity was shared through higher wages via unions and other bargaining channels — ‘that whole mechanism has broken down more recently.’

Pushed by Cowen on whether this squares with the data — income inequality has risen since the 1970s during a period of comparatively low productivity growth, not high — Johnson agrees the coincidence is real but reframes it: digital technology since the 1970s was often ‘quite disappointing in terms of productivity effects, but it was nevertheless deployed because management thought it would be helpful to displace workers.’ Technology change failing to become productivity growth, and inequality widening regardless, are for him the same phenomenon, not a contradiction.

AI and who captures the gains

Johnson extends the so-so technology diagnosis directly to AI: ‘the piece that we really focus on and are concerned about is that more workers may be displaced from previously well-paying jobs, and without creating new tasks. Therefore, worker productivity doesn’t go up. You may pay some people — maybe the designers of AI algorithms — more money, but most people may well be paid less in real terms.’ He preserves real uncertainty in this claim — displacement ‘may’ happen, wages ‘may well’ fall — rather than asserting it as settled.

On where AI’s commercial value will land, Johnson professes genuine agnosticism. Cowen notes OpenAI’s roughly $30 billion valuation looked modest relative to ChatGPT’s unprecedented adoption speed; Johnson’s response invokes the standard cliché that technology’s near-term impact is overestimated and its long-term impact underestimated, without predicting whether gains will concentrate in foundational models, downstream applications, or diffuse to users and workers. He treats social media as a cautionary precedent — 1990s hopes that the internet would distribute power away from large corporations gave way to concentration and, in his view, real social harm — while stopping short of predicting AI repeats that path.

Institutions, prediction, and the middle-income trap

On what protects a country from the middle-income trap, Johnson offers South Korea and Ireland as his favoured cases: broad-based education systems, entrepreneurship, and deep integration with the world economy. But Cowen presses the harder question — institutions are supposed to be the master variable, yet in 1960 few predicted South Korea’s takeoff or the Philippines’ and Sri Lanka’s relative stagnation, despite similarly-rated institutions and English-language advantages at the time.

Johnson largely concedes the point on forecasting: ‘I think that if institutions have some predictive value, it’s fairly long term… I don’t think it helps you that much. Think about one year, two years, even five years.’ What institutions buy, in his account, is a ‘hysteresis effect’ — once a country grows and strengthens its institutions, relapse becomes harder, even though the initial takeoff was not foreseeable from institutional quality alone. Pressed further on his own recent misses — he had expected promising futures for Ethiopia and Ghana that did not materialise — Johnson does not defend the predictive power of the institutions framework beyond this long-run, non-linear claim.

Rereading the Industrial Revolution and the Middle Ages

Cowen puts a direct challenge from Joel Mokyr’s own research: in English regions with an Industrial Revolution, wages rose ‘a fair amount pretty rapidly.’ Johnson, while crediting Mokyr as an important influence on the book, holds his ground on timing: citing Engels’s account of 1830s Manchester and the parliamentary inquiries into 1840s coal-mine child labour, he maintains ‘I don’t think people really saw much by way of gains until after the 1850s’ — roughly a century after the Revolution’s start. He grants Mokyr’s point about early regional wage gains but frames the broader shared-prosperity story as a much slower, later payoff.

Asked what most changed his mind while writing the book, Johnson names the Middle Ages without hesitation. He had absorbed the standard ‘dark age’ narrative of stagnant innovation between Rome and the Renaissance; closer research showed him substantial medieval advances in agriculture, commerce, and early industry — but gains that enriched the church and state rather than ordinary people, funnelled into cathedral-building financed by coerced peasant labour. ‘Next time anyone sees a cathedral, think about them as symbols of medieval despair and increasing productivity, but the failure to create shared prosperity.’ He accepts Cowen’s pushback that the ‘dark age’ framing was roughly accurate for several centuries (perhaps 800–1000 AD), even if wrong as a description of the full millennium.

Science policy and the case for wider geography

On his earlier book Jump-Starting America (with Jonathan Gruber), Johnson reports growing more, not less, optimistic about place-based science investment, despite mRNA vaccines and generative AI both emerging from existing coastal hubs (Boston-Cambridge, San Francisco) rather than the book’s proposed revival cities. His argument is depth of talent outside expensive coastal markets, and spillovers from public basic-science investment that a purely private funder would not make — his standing example is the Human Genome Project, turned down by venture capital in the 1980s for lack of a monetisation path, which he credits with roughly 300,000 subsequent jobs.

See also

• Simon Johnson — speaker
• Tyler Cowen — host
• Daron Acemoglu — co-author of Power and Progress; the episode’s technology-and-choice argument is his and Johnson’s joint thesis
• Daron Acemoglu on the Struggle Between State and Society — companion Conversations with Tyler episode with Johnson’s co-author
• Joel Mokyr — Johnson disputes Mokyr’s reading of how quickly the Industrial Revolution raised wages
• Joel Mokyr on Clans, Corporations, and a Culture of Growth — companion episode; Mokyr’s own account of the Industrial Revolution’s slow payoff for ordinary people
• Fortress Balance Sheet — Johnson’s SVB analysis locates the failure in regulatory threshold design and deposit-insurance architecture, a different axis from this concept’s management-conservatism framing
• Jamie Dimon on Building JP Morgan Chase, the Fortress Balance Sheet, and Not Blowing Up — companion episode where the Fortress Balance Sheet concept originates
• What Makes Economies Grow — Johnson’s institutions-and-prediction discussion sits inside this theme’s institutions-camp debate
• Neal Stephenson — Johnson names Snow Crash his favourite novel and discusses its predictive track record