Bounty and Spread
Overview
Erik Brynjolfsson and Andrew McAfee’s framework for understanding the economic impact of digital technologies in The Second Machine Age (2014):
- Bounty: The total amount of goods, services, and value produced grows enormously — abundance replaces scarcity
- Spread: The distribution of that bounty becomes increasingly unequal — winner-take-all dynamics intensify
Both forces are real and simultaneous. The challenge is that bounty is easy to celebrate while spread creates real suffering.
The Mechanisms
Bounty
- Exponential improvement in computing (Moore’s Law, “second half of the chessboard”)
- Digitization creates near-zero marginal cost for information goods
- Recombinant innovation: new combinations of existing technologies create exponential possibilities
- GDP understates true bounty because it misses free digital goods (Wikipedia, GPS, search)
Spread
- Stars and superstars: Digital scale allows the best performers to serve entire markets, displacing average performers
- Skill-biased technical change: Technology complements high-skill workers, substitutes for routine workers
- Capital vs. labor: As machines do more, returns shift from wages to capital ownership
- Industrial Revolution parallel: it took 50+ years for benefits to trickle down (Frey documents this in detail)
The Historical Parallel
Both the first and second machine ages follow the same pattern:
- Transformative technology arrives
- Total output surges (bounty)
- Distribution worsens dramatically (spread)
- Eventually, institutions adapt and benefits broaden — but “eventually” can mean generations
AI as Potential Spread-Reverser
Autor argues that AI might break the pattern. Computerization widened spread by concentrating decision-making power among elite experts — the 40% of adults with college degrees — while automating the mass-expertise jobs that built the middle class. AI could reverse this by extending expert-level decision-making to workers with foundational training but not elite credentials (see Expertise Democratization).
The empirical evidence supports this hypothesis. Across three major studies, AI’s productivity gains are concentrated among the lowest performers:
| Study | Domain | Average gain | Bottom-performer gain |
|---|---|---|---|
| Noy & Zhang (2023) | Professional writing | 40% faster | Least effective writers reached median quality |
| Brynjolfsson et al. (2023) | Customer service | +14% productivity | Novices reached experienced-agent level in 3 months vs 10 |
| Dell’Acqua et al. (2023) | Management consulting | +40% quality | Bottom performers gained most; top performers gained least |
If this pattern holds at scale, AI compresses the skill distribution rather than stretching it — the opposite of computerization’s effect. Bounty still grows, but spread could narrow rather than widen. Autor frames this as “not a forecast but a claim about what is attainable” — the outcome depends on institutional choices, not technological inevitability.
Mollick documents the same equalizer pattern and adds a warning: the workers who gain most are also most vulnerable to cognitive surrender, because they have the least independent expertise to deploy when the AI fails (see The Jagged Frontier).