## TECHNOLOGY’S BIFURCATED BITE

## Source details

**Canonical URL:** [TECHNOLOGY’S BIFURCATED BITE](https://www.imf.org/-/media/files/publications/fandd/article/2023/december/34-37-berg-final.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/fandd/article/2023/december/34-37-berg-final.pdf.md)
- [Structured JSON version](/-/media/files/publications/fandd/article/2023/december/34-37-berg-final.pdf.json)

---

### Historical context and observed labor-market outcomes
- Technological revolutions (electricity, internal combustion engines, telephones, modern medicine) have increased standards of living and life expectancy since 1900.
- Post-1960s U.S. experience: millions of net new jobs were created after Senator John F. Kennedy’s automation warning; contemporary unemployment rate cited as about "3.5 percent" and a "multi-decade-high ratio of employment to population."
- Despite job growth, wage inequality worsened beginning a few years after Kennedy’s speech, and "the share of income going to workers fell."
- Historical transitions have been economically and politically wrenching; for some vulnerable groups, losses lasted multiple generations.

### Four key channels through which technological change affects inequality
- Technological change that improves the productivity of skilled more than unskilled workers.
- Reductions in the cost of capital that complement chiefly skilled labor.
- Increased ability of machines to replace workers entirely for particular tasks.
- Increased concentration of market power in a few firms as a result of technology.

Findings and prior literature summarized in the source:
- Katz and Murphy (1992): evolution of relative wages explained as a race between demand and supply of skilled workers; increases in supply reduced the skill premium, while persistent demand increases raised it.
- Berg, Buffie, and Zanna (2018): treat AI and robots as a new type of capital that substitutes for some worker groups and complements others; over the past 30 years, substitutability between ICT and unskilled workers seems to have increased.
- Acemoglu and Restrepo (2020): automation of routine tasks explains most changes in relative wages across labor groups, with less role for skill-based technological change or foreign trade.
- Market-power channel: leading AI developers (e.g., Alphabet and Microsoft cited as examples in the source) can capture large shares of income from AI capital, rent technologies to other firms, and further depress labor share.

Implications drawn:
- Higher substitutability of workers with machines and AI increases wage inequality and the share of income accruing to owners of capital.
- During a possibly decades-long transition, many workers could see real wage declines.
- Firms large enough can "purchase and bury possible competitors," potentially stifling competition, limiting innovation, and worsening inequality.
- Automation can displace workers without producing large productivity gains if innovation is directed toward displacement rather than societal welfare (Acemoglu and Restrepo 2022).

### Generative AI, GPT, and shifting patterns of impact
- Rapid advances: early 2023 emergence of ChatGPT-4 marked a significant acceleration.
- Expert expectations revised: McKinsey-surveyed AI experts in 2019 expected computers to write at the level of the top "25 percent" of humans by "2050" and perform human-level creative tasks by "2055"; these estimates were revised to "2024" and "2028", respectively.
- Estimated workplace impact: one estimate suggests that once GPT is introduced into the work environment, about "20 percent" of workers could see at least half of their tasks affected.
- Empirical observations: GPT appears to increase productivity in creative tasks (writing, legal analysis, programming); in some studies the least-skilled participants benefit most and GPT-augmented input can be more creative.
- These patterns contrast with prior technology waves that emphasized automation of routine tasks and substitution of unskilled labor; GPT may reshape which worker groups gain or lose.

### Scenarios, risks, and open questions
- Possible outcomes are diverse:
  - Some wage inequality could be reversed if lower-skilled workers benefit disproportionately from generative AI.
  - Major corporations with superior data, computing resources, and talent could gain more economic and political power, exacerbating concentration and inequality.
  - The hypothetical prospect of artificial general intelligence (AGI) introduces further uncertainty; AGI would presumably be capable of any human intellectual effort.
- Under any scenario, substantial economic, social, and political upheaval is likely during the transition.

### Global implications
- Previous research suggested AI substitution for unskilled labor could widen global income disparities, disadvantaging lower-income countries (Alonso and others 2022).
- Generative AI introduces new uncertainty for country-level outcomes:
  - Potential upside: developing economies could benefit from AI as a "tireless universal tutor and expert programming assistant" that strengthens workforces.
  - Potential downside: limited access to data, expertise, and technological infrastructure could widen divergence.

### Policy-relevant considerations highlighted by the authors
- The distributional implications of AI depend on both the evolution of the technology and policy and societal responses.
- Key policy concerns implied by the analysis:
  - Address the distribution of benefits from AI technologies—who owns AI and how returns to AI capital are shared.
  - Monitor and, where appropriate, limit anti-competitive behavior by dominant firms that could stifle innovation and concentrate income.
  - Prepare for potentially long transitions with persistent effects on wages and labor shares, including policies to support displaced or vulnerable workers.
  - Assess global access to AI data, expertise, and infrastructure to avoid widening international inequalities.

*Andrew Berg, Chris Papageorgiou, and Maryam Vaziri. Source: "Technology’s Bifurcated Bite," F&D, December 2023.*

---


_Source: https://www.imf.org/-/media/files/publications/fandd/article/2023/december/34-37-berg-final.pdf_