Charles I Jones | The past and future of economic growth: a semi-endogenous perspective

Summary of the lecture (Atkinson Memorial Lecture 2021)

Charles I Jones (Stanford) presents a semi-endogenous growth framework to explain both the past stability of long-run economic growth and likely features of its future path, with special attention to how ideas (and research) drive growth.

1) Starting point: why long-run growth has been so steady

• The lecture begins with the empirical puzzle: U.S. GDP per person has followed an approximately steady growth trajectory for over a century, interrupted only briefly (e.g., the Great Depression).
• Jones explains this through the non-rivalry of ideas: once an idea is created, it can be used simultaneously by many people without depletion.
• This non-rivalry implies increasing returns at the macro level, motivating models where growth depends on the scale of innovation activity.

2) Core semi-endogenous growth model (ideas become harder to find)

Jones builds the simplest semi-endogenous model:

• Income per person depends on the aggregate stock of ideas.
• Ideas are produced using research effort, but research faces a key drag:
  • the “ideas get harder to find” mechanism, parameterized by β.
  • as ideas become harder, maintaining constant productivity growth requires ever-increasing research effort.

3) Evidence: research must grow fast to maintain TFP growth (“Red Queen”)

Using multiple empirical domains, Jones argues that data are consistent with the Red Queen implication:

• Aggregate U.S. evidence: TFP growth is roughly stable or slightly declining, while the number of researchers rises rapidly—suggesting ideas have become harder to generate/locate.
• Moore’s law / semiconductors: chip density growth continued for decades, but the implied research effort rises dramatically over time, consistent with increasing difficulty (even in an innovation-heavy sector).
• From these exercises, the implied difficulty parameter β is positive (with estimates ranging widely by context; the example around Moore’s law suggests a relatively small but nonzero β).

4) Transition dynamics are extremely slow (Atkinson’s “past affects the future” angle)

A major theme is the slow adjustment of growth rates toward steady states:

• With realistic β values, it can take many decades to centuries for growth rates to move meaningfully toward their long-run behavior.
• Jones links this to major past research and institution-building events (DARPA, NIH, NSF, space program, Manhattan Project, etc.), arguing their effects can persist for a very long time.

5) Historical growth accounting: much growth was not “long-run” growth

Jones extends the model to do a growth-accounting decomposition for the U.S. since about the 1950s:

• The long-run component of growth is small: roughly ~0.3 percentage points per year of the ~2% growth rate can be attributed to the long-run mechanism (population growth in the model).
• Large parts of historical growth came from “transitory” drivers such as:
  • rising educational attainment
  • increased labor force participation (notably women)
  • improvements in misallocation and talent allocation
• He quantifies misallocation using related research (back-of-the-envelope), and then attributes the remaining TFP growth to a mix of population growth and research intensity.

Key implication: because these transitory channels appear to have already leveled off, future growth should slow unless something changes.

6) Why future growth might be slower

(a) Research effort is already slowing (global diffusion issues)

• Jones discusses how to measure “U.S. growth” when research is global and ideas diffuse internationally.
• He reports that research growth has slowed since the early 2000s for the U.S. and OECD aggregates (with smaller slowdowns globally), which—under the model—implies slower future growth.

(b) Population growth is slowing—and may become negative

• Jones argues that globally falling fertility rates can lead to declining researcher populations.
• In the model, if researchers decline exponentially, living standards eventually stagnate:
  • the “empty planet result”: once the scale of idea production falls enough, growth slows toward stagnation even if ideas are still produced at some level.
• He emphasizes that the boundary between fertility slightly above and below replacement is macroeconomically sharp, even though it looks continuous at the individual/family level.

7) Why growth might not be as slow as that bleak scenario implies

Jones offers offsetting forces:

1. Finding “lost Einsteins” (expanding the researcher base)

   • More potential inventors globally as developing countries rise (e.g., China/India coming closer to the frontier).
   • Underutilization of women in invention/innovation: patenting rates by women were very low in the past and remain far from parity.
   • Evidence that talent is broadly distributed, but opportunity/access is not.
   • Combining these, he argues the global research effort could rise substantially (potentially by factors like 3 or more), boosting long-run incomes.

2. Automation and AI might accelerate research

   • Jones models automation as shifting research tasks from scarce labor/researchers toward machines/capital.
   • Under his mechanism, more automation can increase the growth rate of ideas because it reduces reliance on diminishing or constrained human research capacity.
   • He cautions: automation has been happening for over a century, yet aggregate growth hasn’t accelerated dramatically—so the question remains whether AI will “change the game” structurally.

8) Open research questions

Jones concludes by highlighting key uncertainties:

• How large are the model’s increasing returns to ideas (parameter γ)?
• What is the social rate of return to research, and are societies underinvesting in basic research?
• How much do long-term historical events and slow transition dynamics actually account for today’s growth?
• Why hasn’t automation historically produced faster growth—and will AI differ?

---

Presenters / contributors mentioned

• Charles I Jones (speaker; Atkinson lecturer)
• Margaret Stevens (Oxford Economics Department head; introducing speaker and discussing Atkinson)
• Hilary (name not fully captured in subtitles; appears to be part of the opening/moderation team)
• Russell (mentioned as the person calling on questions via Zoom chat)
• Professor Peter Neary (mentioned as recently deceased; honorary doctorate recipient)
• Tony Atkinson (late Sir Tony Atkinson; memorial focus; discussed as inequality and earlier growth economist)

Contributors cited in research/model context (as reflected in the transcription)

• Paul Romer, Aghion and Howitt, Grossman and Helpman, Krugman and Eaton and Kortum
• Eckis/Bernstein/Melitz-type firm dynamics authors (subtitle list partial names)
• Joan Robinson, Joseph Stiglitz, Robert Solow
• Nick Bloom, John Van Reenen, Michael Webb
• Bill/Adam? (Bills Cleno as transcribed; likely Hsieh & Klenow family name pair)
• Eric Hurst
• Claud? (Chang Shea as transcribed)
• Jean-Felix Bruyette, Bricker and Ibbitson (“Empty Planet” authors)
• Daron Acemoglu and others (automation literature referenced; multiple names in subtitles)
• Joseph Sierraa? (Sierra & Mogul as transcribed), Restrepo
• Alvin Roth? (not clearly; a phrase like “Bell chatty chair of Alvin renin” appears to refer to “lost Einsteins” literature—possibly related to Jones citing prior work by others as transcribed)
• Philippe Aghion and Ben Jones (automation paper authors as transcribed)