Charles Jones and Christopher Tonetti on AI and automation:

Summary by Claude.

Let me break down what Jones and Tonetti are arguing.

The Core Question

How much of past economic growth came from automation, and what does that tell us about AI’s future impact?

The Key Insight: “Weak Links”

The paper’s central concept is that economic production involves complementary tasks — you need all of them to produce output. Think of it like a chain: your output is limited by your weakest link, not your strongest.

This has a counterintuitive implication: even if you automate 95% of tasks with infinite productivity, you don’t get infinite output. You’re still constrained by that remaining 5%.

What They Found Historically

1. Machines improve faster than humans
The growth rate of capital productivity (how good machines are at tasks) exceeds labor productivity by at least 5 percentage points per year across all sectors they studied.

2. Automation’s real benefit
The gain from automation isn’t the act of switching itself — when you automate a task, you do it at the point where capital and labor costs are equal, so there’s no immediate productivity jump.
The real gain is that once automated, that task now benefits from rapidly-improving machines rather than slowly-improving humans.

3. Automation explains most historical growth
If you “froze” automation at 1950 levels — keeping all subsequent improvements in machine and labor productivity but not automating any new tasks — you’d lose essentially all TFP growth in the private business sector.

What This Means for AI

Here’s where it gets provocative:
Growth will accelerate — their model predicts growth rates rising to 5%+ annually over the next 75 years.
But the acceleration is remarkably slow:

By 2040: GDP only 4% higher than trend
By 2060: GDP only 19% higher than trend

Why so slow?

Weak links again. Even as AI automates more cognitive tasks, the economy remains constrained by whatever tasks humans still perform.
The software example is striking: Software is ~2% of GDP. Automating all software tasks with infinite productivity would raise GDP by… about 2%. Weak links mean you can’t escape the constraint of everything else.

Three Possible Futures

They model three scenarios based on whether all tasks eventually get automated:

Full automation (capital share → 100%): Explosive growth, infinite income in finite time

Permanent weak links (some tasks never automated): Growth eventually slows to the rate humans improve (~0.5%/year)

Asymptotic automation (labor share → 0 but never reaches it): Growth still explodes, but the capital share stays stable

The remarkable finding: all three paths look nearly identical for the next 75 years. The differences only emerge in the distant future.
Important Caveats
The authors are refreshingly honest about limitations:

“The key ‘weak link’ in our results so far is the use of ChatGPT to estimate the automation rate”

They literally used an LLM to estimate what fraction of tasks have been automated in each sector over time — and acknowledge this needs refinement.
The Big Picture

This paper pushes back against both AI hype and AI dismissiveness:

Against hype: Even with rapid AI progress, weak links mean growth acceleration is slow. No singularity next decade.

Against dismissiveness: Automation has been the primary driver of historical growth. AI continues this trend.

The framework is elegant: growth happens because we keep shifting tasks from slowly-improving humans to rapidly-improving machines. AI is just the latest chapter in a centuries-old story.

What aspects would you like me to dig deeper into? The math behind weak links, their empirical methodology, or the future simulations?