## Machine Intelligence and Human Judgment

## Source details

**Canonical URL:** [Machine Intelligence and Human Judgment](https://www.imf.org/en/publications/fandd/issues/2025/06/machine-intelligence-and-human-judgement-ajay-agrawal)

## Other formats

- [Markdown version](/en/publications/fandd/issues/2025/06/machine-intelligence-and-human-judgement-ajay-agrawal/index.md)
- [Structured JSON version](/en/publications/fandd/issues/2025/06/machine-intelligence-and-human-judgement-ajay-agrawal/index.json)
- [Bundle manifest](/en/publications/fandd/issues/2025/06/machine-intelligence-and-human-judgement-ajay-agrawal/bundle-manifest.json)

## Bibliographic details
- Authors: AJAY AGRAWAL, JOSHUA GANS, AVI GOLDFARB
- Published: June 2, 2025

---

### Overview
- AI could reverse the widening inequality driven by technology, or aggravate it.
- The article explores how AI’s effects on prediction and judgment will shape markets, wages, and the distribution of power.
- Publication: F&D Magazine, June 2025.
- Authors: AJAY AGRAWAL, JOSHUA GANS, AVI GOLDFARB.

### Scenarios for prosperity vs instability
- Prosperity scenario:
  - Productivity and economic growth could soar.
  - Industries from health care to education to technology could be revolutionized.
  - Office tasks handled with flawless efficiency could free people to pursue more meaningful endeavors and lower service costs, raising living standards.
- Instability scenario:
  - Knowledge workers and professionals could face mass unemployment as geniuses perform tasks at a fraction of the cost.
  - Eroded wages and job security could collapse the middle class and deepen inequality.
  - Concentration of geniuses under a few corporations or nations could monopolize wealth and power, stifling innovation and increasing geopolitical tensions.
  - Loss of social value for human creativity could produce widespread unrest and existential questions of purpose.

### Evidence from recent studies (key statistics preserved exactly)
- Roldán Monés (Esade Business School): generative AI in a university debate competition
  - Higher-ability debaters were 12 percent more likely to win debates when using generative AI.
  - Little change to the outcome for lower-ability debaters.
  - Interpretation: AI amplified advantages for high-ability debaters by aiding credibility, rhetoric, and rebuttal (judgment-related gains).
- Brynjolfsson et al. (Stanford) on call center employees:
  - AI tools increased productivity (measured by number of problems resolved per hour) by 14 percent on average.
  - Novice and low-skilled workers saw a 34 percent improvement.
  - Minimal impact on productivity of experienced and highly skilled workers.
  - Interpretation: AI substituted for human prediction, disproportionately aiding lower-skilled workers and reducing productivity gaps.

### Role of judgment versus prediction
- Decision theory framing: prediction assigns probabilities to outcomes; judgment assigns values to consequences.
- Where differences between workers are prediction-based:
  - AI prediction substitutes for human prediction.
  - Lower-skilled workers benefit disproportionately; income disparity in that industry may decrease.
  - Example consequence: back-office and call center wages may increase in India relative to the US.
- Where differences are judgment-based:
  - AI augments higher-skilled workers who better identify promising suggestions.
  - AI amplifies rewards for judgment, widening productivity differences and income disparity.
  - Potential geographic concentration of high-value work in regions that supply judgment-intensive talent (e.g., US innovation hubs).

### Geographic and distributional implications
- Regions with more skilled workers, stronger research institutions, and advanced technological infrastructure will likely capture disproportionate economic benefits.
- If AI’s amplification of judgment-intensive, high-value tasks outweighs its equalizing effect on prediction-intensive tasks, global economic inequality will deepen.
- Potential long-term consequences:
  - Greater concentration of wealth and influence in select cities or countries.
  - Growing disparity in technological leadership, research funding, and geopolitical influence.
  - Redefinition of which forms of judgment remain scarce, dependent on regional adaptation of the workforce.

### Policy recommendations (three principal levers)
- Expand access to high-quality education and training that emphasizes complex decision-making skills to sharpen judgment across regions.
- Promote global talent mobility and knowledge exchange to distribute judgment necessary for effective AI use more broadly.
- Create incentives to spread the ability to generate valuable AI predictions beyond traditional power centers through funding, infrastructure, and AI adoption incentives.

### Transition dynamics and research imperative
- AI is advancing rapidly, but complementary factors (management practices, infrastructure, education, regulations, customer demand) change slowly, limiting short-term impact.
- Long-term global economic impact will be significant; economic stability depends on managing the transition.
- Computer scientists raced ahead developing the technology; economists and policymakers must catch up with research to guide policy toward global stability and prosperity.

### Editorial update and authorship
- Editor’s Note (June 16, 2025): Article updated to remove references to “Artificial Intelligence, Scientific Discovery, and Product Innovation” by Aiden Toner-Rogers, a paper that MIT has said should be withdrawn from public discourse because it has no confidence in the research.
- The authors and affiliations:
  - AJAY AGRAWAL — Geoffrey Taber Chair in Entrepreneurship and Innovation, University of Toronto’s Rotman School of Management.
  - JOSHUA GANS — Jeffrey S. Skoll Chair in Technical Innovation and Entrepreneurship, University of Toronto’s Rotman School of Management.
  - AVI GOLDFARB — Rotman Chair in Artificial Intelligence and Healthcare, University of Toronto’s Rotman School of Management.
- Citation highlights in the article: Agrawal, Gans, and Goldfarb 2018; Brynjolfsson, Li, and Raymond 2023; Roldán Monés 2024.

*Source: Machine Intelligence and Human Judgment — F&D Magazine, June 2025*

---


_Source: https://www.imf.org/en/publications/fandd/issues/2025/06/machine-intelligence-and-human-judgement-ajay-agrawal_