◆Book Ontologytheorem-class
One book, taken apart and mapped: its claims, its lineage, its hinges, its afterlife.

Technology and the Rise of Great Powers

How Diffusion Shapes Economic Competition
by Jeffrey Ding · Princeton University Press, 2024

Great-power transitions are driven by capacity to diffuse general-purpose technologies across the whole economy, not by monopolizing innovation in leading sectors; the binding constraint is skill infrastructure.

ontology compiled 2026 · rubric v1.4
IPayload — what the book says

The theory: general-purpose technologies deliver power transitions only through decades-long, economy-wide diffusion, and diffusion breadth is gated by skill infrastructure, the institutions that widen the ordinary engineering skill base. States that diffuse win; states that merely innovate first do not. Three historical cases plus a statistical test and an AI application carry it:

Full table
CaseChapterGPT trajectoryDiffusion verdict
Britain's rise (First IR)3mechanization, ironbroad diffusion capacity, not sector monopoly, drove the rise
America's ascent (Second IR)4machine tools, interchangeable partsmechanical engineering education and standards widened adoption; the US out-diffused Britain
Japan's challenge (Third IR)5computers, electronicsinnovation leadership without matching diffusion breadth; the challenge stalled
Statistical test6computerizationsoftware engineering skill infrastructure predicts computerization across countries
US-China in AI (Fourth IR)7AI as GPTthe decisive variable is diffusion and skill infrastructure, not benchmark leadership
Claims register
  1. GPT diffusion capacity, not leading-sector monopoly, explains which great powers rose in each industrial revolution. (p. chs. 1-2)
  2. GPTs impact growth through decades-long diffusion phases, not at introduction. (p. ch. 2)
  3. Skill infrastructure — engineering education, standards bodies — is the institutional gate on diffusion breadth. (p. ch. 2)
  4. Britain's First-IR rise ran through broad mechanization diffusion. (p. ch. 3)
  5. America's Second-IR ascent ran through mechanical engineering skill infrastructure and interchangeable-parts diffusion. (p. ch. 4)
  6. Japan's Third-IR challenge stalled for want of diffusion breadth despite innovation leadership. (p. ch. 5)
  7. Software engineering skill infrastructure statistically predicts computerization across countries. (p. ch. 6)
  8. AI is a GPT; US-China outcomes will turn on diffusion capacity rather than innovation races. (p. ch. 7)
  9. Policy should fund broad technical education and standards, not moonshot races. (p. ch. 8)
  10. The GPT classification method is stated explicitly (appendix tables). (p. appendix)
IIContext network — the book in its field

One graph, two directions: an ego network centered on the book (merged cards will make it a true knowledge graph). Everything left of the book node flows in (inside out: the citation anatomy, section V). The lower-right flows out (outside in: reception and citing literature). Node size = mentions in the book; hollow dashed nodes were never engaged. Hover a node for its role and evidence; click legend entries to highlight strands (multi-select).

1920 1940 1960 1980 2000 2020 GPT ECONOMICS AND INNOVATION STUDIES IR AND POWER-TRANSITION THEORY WAVE SCHOOL AND ECONOMIC HISTORY RECEPTION AND CHALLENGES DING 2024 Bresnahan-Trajtenberg 1995 Lipsey-Carlaw-Bekar 2005 P. David 1990 Rosenberg 1963/1982 von Tunzelmann 1995 Edgerton 2007 Cohen-Levinthal 1990 Thompson 1990 Modelski-Thompson 1996 Rostow 1960 Kennedy 1987 Gilpin 1981 Beckley 2018 Horowitz 2010 Schumpeter 1939 Kondratiev 1935 Freeman-Louca 2001 Perez 2002 Smil 2005 Fogel 1964 Mokyr 1990 Donaldson-Hornbeck 2016 Lepgold Prize 2025 NBR roundtable 2024 Hounshell 1984 Moe 2007 M.Z. Taylor 2016 Comin-Hobijn 2010 Goldin-Katz 2008 Abramovitz 1986 Mansfield 1968 Veblen 1915 Lundvall 1992

The author

Jeffrey Ding (b. ). An AI-governance researcher writing backward into economic history: the cases exist to discipline a forecast about AI. That explains the four-count periodization taken from the policy discourse, the staged fight with an IR literature rather than with economics, and the book's real audience, which is Washington rather than the historians it mines.

Career line
  • DPhil, University of Oxford; researcher at the Centre for the Governance of AI (GovAI)
  • Assistant Professor of Political Science, George Washington University
  • Author of the ChinAI newsletter, translating and analyzing Chinese AI discourse
  • This book: 2025 Lepgold Book Prize (best book in international relations); Chicago Council Read of the Year; Choice Outstanding Academic Title; ISA STAIR Book Award

Formation

  • advisor: Duncan Snidal (PhD supervisor, Oxford DPIR) — rational-institutionalist IR theory; visible in the book's mid-range, operationalized-mechanism style
  • nest: Allan Dafoe's Centre for the Governance of AI ('especially Allan Dafoe'), then Stanford CISAC (Sagan, Trinkunas), then GWU (book workshop: Beckley, Brooks, Finnemore, Newman)
  • coauthors: Allan Dafoe (strategic-assets work); ISQ and RIPE articles preceded the book
  • admired influences: Edgerton credited by name for the diffusion-centrism critique; the GPT economists credited openly
  • formation note: Rhodes Scholar; the acknowledgments contain the striking fact that William Thompson and Dan Drezner — the book's named adversaries — commented on the manuscript: the staged fight was staged with the opponent's cooperation, which strengthens the fairness ledger and softens the scarecrow charge

Reception

2025 Lepgold Book Prize for the best book in international relations; Chicago Council on Global Affairs Read of the Year; Choice Outstanding Academic Title; ISA Science, Technology and Art in International Relations Book Award; NBR roundtable. Rapid consecration inside IR (Lepgold and three further awards within a year) and heavy uptake in the AI-policy world; the economic-history and wave-school literatures it mines have not pushed back in print, which mirrors the book's own one-way citation practice toward them. Our review-in-progress (4IR project) supplies the missing cross-examination: right mechanism, wrong drama.

Reviews: NBR book roundtable · Anthony Canevello (Inter Populum) · GovAI book talk

Sources: jeffreyjding.github.io · politicalscience.columbian.gwu.edu · www.nbr.org · interpopulum.org · www.governance.ai

IIICore and reverse outline

GPTs deliver growth only through decades-long, economy-wide diffusion; diffusion breadth is gated by institutions that widen the ordinary engineering skill base (education systems, standards bodies); states with wider skill infrastructure convert the same technologies into productivity leadership and power transitions.

GPT emerges (innovation phase)→Decades-long diffusion phase begins→Skill infrastructure gates adoption breadth→Broad diffusion yields productivity leadership→Productivity leadership yields power transition

Scope: Great powers during industrial revolutions; GPT trajectories only (not all technologies); economic foundations of power, with military competition downstream
Rival, per the author: Leading-sector theory: power transitions from monopoly profits in a few new industries (Rostow, Thompson, Modelski-Thompson, Gilpin lineage)
Actual nearest rival: The wave school he mines but never argues with (Perez's finance mechanism, Smil's indivisible saltation, Freeman-Louca's wave clock) and the in-domain ghost, Horowitz's adoption-capacity theory of military diffusion

LOAD-BEARING 1 · Introduction Stakes the puzzle (why do technological revolutions reorder power?) and stages the leading-sector conventional wisdom the book will overturn.
  • The staged rival: leading-sector monopoly theory (Rostow, Thompson, Modelski-Thompson) (p. ch. 1)
LOAD-BEARING 2 · GPT Diffusion Theory The mechanism: GPT impact arrives through extended diffusion, gated by skill infrastructure; built openly on Bresnahan-Trajtenberg, Lipsey-Carlaw-Bekar, David, von Tunzelmann, Edgerton and absorptive capacity.
  • Diffusion breadth over innovation leadership; skill infrastructure as the gate (p. ch. 2)
APPLICATION 3 · The First Industrial Revolution and Britain's Rise Case one: Britain's rise as diffusion story.
LOAD-BEARING 4 · The Second Industrial Revolution and America's Ascent The strongest case: US mechanical engineering education and standards out-diffused Britain; historical judo on the leading-sector school's own sectors.
  • US skill infrastructure, not sector monopoly, drove the ascent (p. ch. 4)
APPLICATION 5 · Japan's Challenge in the Third Industrial Revolution The negative case: innovation leadership without diffusion breadth stalls (against Freeman-Clark-Soete's Japan-will-lead forecast, used as foil).
LOAD-BEARING 6 · A Statistical Analysis of Software Engineering Skill Infrastructure and Computerization The large-n test: skill infrastructure predicts computerization.
  • Cross-country evidence for the mechanism (p. ch. 6)
APPLICATION 7 · US-China Competition in AI and the Fourth Industrial Revolution The transfer: AI as GPT; diffusion capacity as the decisive variable in US-China competition.
SUPPORTING 8 · Conclusion The policy program: diffusion capacity over innovation-race monomania.

Argumentative movement. A clean theorem architecture: one theory chapter, three historical applications arranged as most-likely and least-likely cases, one statistical corroboration, one forward transfer, one policy landing. The drama is front-loaded: chapters 1-2 stage a conventional wisdom whose defeat the cases then perform; read without the staging, the same chapters are a well-operationalized contribution to a diffusion consensus that economics reached decades ago. The book's real fight (with the innovation-race policy discourse) is won convincingly; its staged fight (with leading-sector theory) was over before it began.

IVHinges — where it stands or falls

Each hinge opens to its dependencies, evidence, vulnerability and concessions.

STAGE Leading-sector theory is the reigning conventional wisdom that must be overturned (p. chs. 1-2) weakened
If wrong, falls: The book's novelty claim and its drama; not the mechanism itself, which stands on its own evidence
Evidence offered: Leading-sector language throughout ('leading sector' 163 times); Thompson and Modelski-Thompson as named holders; the innovation-race policy discourse as the live foil
Vulnerability: Measured citation clusters run roughly 6:1 for GPT-and-diffusion over leading sectors; Fogel refuted leading-sector indispensability in 1964; the two camps mostly never argued with each other, so the contest is retrospective, assembled by the book
What remains: A real if smaller target: Thompson and Modelski-Thompson genuinely held the leading-sector view, and the policy world's innovation-centrism is no scarecrow
Author concedes: his own conventional-wisdom section, read closely, claims the IR literature has a GAP (Kennedy's chain with no tech-growth mechanism) rather than a reigning rival theory
EMP GPT diffusion capacity, not innovation leadership, drove the three historical transitions (p. chs. 3-5) intact
If wrong, falls: The theory's evidentiary base and the historical judo of testing leading-sector scholars' own sector choices
Evidence offered: Britain's rise (ch. 3), America's ascent via mechanical engineering skill infrastructure (ch. 4), Japan's stalled challenge in the third revolution (ch. 5), plus the ch. 6 statistical analysis of software engineering skill infrastructure and computerization
Vulnerability: Case separability: Smil's Age of Synergy treats 1867-1914 as one indivisible saltation, which attacks the IR-2/IR-3 split the cases rely on; and Donaldson-Hornbeck 2016 more than doubles Fogel's railroad effects, weakening the anti-indispensability ammunition he cites (while its market-access logic arguably supports diffusion thinking)
What remains: The mechanism survives the D-H revision better than the citation does; the case architecture needs the periodization defended, not assumed
MECH Skill infrastructure (engineering education, standards bodies) is the binding constraint on diffusion (p. chs. 2, 4, 6) intact
If wrong, falls: The operationalization that makes the theory testable, and the entire policy payoff (educate and standardize rather than race)
Evidence offered: US mechanical engineering education in the second revolution; ch. 6 large-n analysis of software engineering skill infrastructure and computerization
Vulnerability: Finance is entirely absent: Perez's installation-frenzy-crash account of how every historical GPT era actually unfolded is never addressed, though it is the largest live alternative account of the same episodes; a mania-and-crash variable may confound skill-infrastructure effects
What remains: Skill infrastructure as A binding constraint survives; AS THE binding constraint is unproven against the unexamined financial channel
DEF The four-revolution periodization and GPT classifications are sound working units (p. chs. 1, 7; appendix) intact
If wrong, falls: The case boundaries; though less than it appears — the mechanism may not need a revolution count at all, which is a strength
Evidence offered: Appendix GPT classification tables; the four-revolution frame organizing chs. 3-7
Vulnerability: The scholarly lineage counts five (Freeman-Louca, Perez); the four-count is Schwab's popularization; Smil's singularity and Gordon's one big wave are the live objections to any count
What remains: Nearly everything: the diffusion mechanism is count-independent
Author concedes: the appendix method makes the GPT classification criteria explicit
SCOPE The framework transfers to AI: diffusion capacity will decide US-China competition in the fourth revolution (p. ch. 7) weakened
If wrong, falls: The book's policy relevance and its public reception; the historical argument stands without it
Evidence offered: AI-as-GPT argument; US-China comparison of AI skill infrastructure and adoption capacity (ch. 7)
Vulnerability: AI may diffuse unlike prior GPTs (API-mediated adoption lowers the skill threshold); the complexity counterargument to easy diffusion (Gilli and Gilli 2019) is not engaged; and if Perez is right, the fourth revolution's frenzy-crash dynamics will confound diffusion measures in real time
What remains: A disciplined forecast frame either way: the variable to watch is diffusion infrastructure rather than benchmark leadership
Author concedes: presented as implication and forecast, not as tested result
VCitation anatomy — inside out
Thompson96combat
David, Paul A.75constitutive
Kennedy58adversarial
Rosenberg58constitutive
Gilpin51adversarial
Modelski44adversarial
Mokyr43evidentiary
Rostow36foil
von Tunzelmann20constitutive
ghostsHorowitz, Kondratiev, Mansfield, Veblen, Lundvall — zero mentions
Full citation table
NameWorkNEngagement
Thompsonleading-sector world-leadership corpus (1990; with Modelski 1996)96combat
David, Paul A.dynamo, GPT diffusion lags (1990)75constitutive
KennedyThe Rise and Fall of the Great Powers 198758adversarial
Rosenbergmachine-tool convergence 1963; Technology and American Economic Growth 1972; economics of technical change58constitutive
GilpinWar and Change 1981; US Power 197551adversarial
Modelskilong cycles; with Thompson 199644adversarial
MokyrLever of Riches; British IR scholarship43evidentiary
Rostowleading sectors and stages of growth36foil
von TunzelmannTechnology and Industrial Progress 199520constitutive
BresnahanGPTs, with Trajtenberg 199520constitutive
LipseyEconomic Transformations (with Carlaw, Bekar) 200520constitutive
BeckleyUnrivaled 2018; net power20evidentiary
Freemanwith Louca, As Time Goes By 2001; 1982 foil15extractive
Schumpeter193911courtesy
SmilCreating the Twentieth Century 2005; 20107extractive
FogelRailroads and American Economic Growth 19647extractive
Edgertoninnovation-centrism critique (2007-2011)5constitutive
Cohen and Levinthalabsorptive capacity 19905constitutive
PerezTechnological Revolutions and Financial Capital 2002; 19834courtesy
HorowitzThe Diffusion of Military Power 2010—ghost
Kondratievlong waves—ghost
HounshellFrom the American System to Mass Production 198411evidentiary
MoeGovernance, Growth and Global Leadership 2007; 200940adversarial
Taylor, M. Z.The Politics of Innovation 2016; 201210adversarial
Comin (and Hobijn)cross-country technology diffusion measurement (CHAT)17evidentiary counts-only
Jovanovic and RousseauGPTs, Handbook of Economic Growth 200510constitutive counts-only
Goldin and KatzThe Race Between Education and Technology 20086evidentiary counts-only
Organski (and Kugler)power transition theory5evidentiary counts-only
Abramovitz'Catching Up, Forging Ahead, and Falling Behind' 1986 (social capability)2courtesy
Mansfieldthe economics of technology diffusion (firm/industry studies)—ghost
VeblenImperial Germany and the Industrial Revolution 1915 (the original latecomer-borrowing thesis)—ghost
Lundvallnational innovation systems—ghost

Positionality (acknowledgments): Oxford DPhil lineage (Centre for the Governance of AI), IR and security-studies community

VIDimension vector — experimental
coherence10computed
extractiveness1computed
combativeness3computed
empirical density9computed
synthesis share6computed
theoretical ambition7judgment
falsifiability discipline7judgment
scope breadth6judgment
prescriptiveness7judgment
performativity2judgment

Computed axes reproduce from raw fields (validate_card.py); judgment axes are rubric-anchored calls, not measurements. Cross-book comparison waits on a corpus.

Axis justifications and caveats
VIIVerdict

Steel-man: the best operationalization yet of diffusion-centered technology competition, with the policy payoff that education and standards beat innovation races. Weakest: the staged leading-sector orthodoxy (weakened hinge), the consistently extractive citation practice toward the wave school whose frameworks are never confronted, and the unengaged finance channel that is the largest live alternative account of the same GPT eras.

Method and verification

Extraction via pdftotext of the Princeton 2024 PDF; page cites are by chapter (the extraction has no page map, and page-level quotes remain flagged for re-verification, per the source dossier's own queue). Mention counts are regex counts verified against the 4IR project's DING_REVIEW_NOTES.md ledger (2026-07-20), which carries footnote-level verification for the Perez, Smil, Fogel and Horowitz claims and the OpenAlex citation-cluster measurement. Engagement types follow that dossier's passage-verified ledger. Reception verified by web search 2026-07-21 (Lepgold and other awards). Census block records gaps. Computed vector axes reproduce via validate_card.py; data_exhibits is a flagged estimate. Working dossier: cards/ding-2024.md, pointing to DING_REVIEW_NOTES.md in the 4IR project as the primary record. Compiled 2026-07-21, rubric v1.4.