## SCENARIO PLANNING FOR AN A(G)I FUTURE

## Source details

**Canonical URL:** [SCENARIO PLANNING FOR AN A(G)I FUTURE](https://www.imf.org/-/media/files/publications/fandd/article/2023/december/30-33-korinek-final.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/fandd/article/2023/december/30-33-korinek-final.pdf.md)
- [Structured JSON version](/-/media/files/publications/fandd/article/2023/december/30-33-korinek-final.pdf.json)

---

### Recent advances and compute trends
- ChatGPT was released in November 2022 and demonstrated human-quality text and code generation, seamless translation, creative writing, and informative question answering.
- The amount of computational resources used to train the most cutting-edge AI systems has doubled every six months over the past decade.
- Foundation models underlying generative AI have been advancing rapidly for more than a decade.

### Competing perspectives on the human brain and automation
- Two perspectives on the complexity distribution of tasks the human brain can perform:
  - Panel 1 perspective: human brain capabilities are unbounded; as automation advances, humans reallocate to ever more complex tasks (continuation of historical pattern since the Industrial Revolution).
  - Panel 2 perspective: there is an upper bound to human-brain task complexity; the brain is a computational entity with biological limits and slower information transmission compared with AI.
- If the second perspective is correct, modern AI systems are catching up fast; many measures of computational complexity of cutting-edge foundation models are already close to those of the human brain.
- AI systems can exchange information at speeds that are significantly faster than human senses and language.

### Scenario framework (three technological scenarios)
- Rationale:
  - Given starkly differing expert perspectives and deep uncertainty, use a portfolio/scenario approach and stress-test policies across multiple futures.
  - The author estimates each scenario has a greater than 10 percent probability of materializing.
- Scenario I (traditional, business as usual):
  - Advances in AI boost productivity and automate a range of cognitive tasks, but displaced workers move into new, on-average more productive jobs.
  - Corresponds to Panel 1 of Chart 1.
- Scenario II (baseline, AGI in 20 years):
  - Over the next 20 years, AI gradually advances to AGI, able to perform all human work tasks by the end of the period, devaluing labor.
  - Corresponds to Panel 2 of Chart 1 and assumes 20 years for the most complex cognitive tasks to become accessible to AI.
- Scenario III (aggressive, AGI in five years):
  - Replicates Scenario II but with AGI reached within five years.

### Macroeconomic implications (Korinek and Suh 2023)
- Korinek and Suh (2023) analyze output and wages in a mainstream macroeconomic model of automation; Chart 2 illustrates paths of output (panel 1) and competitive market wages (panel 2) for the three scenarios.
- Three core insights:
  - First: In the business-as-usual scenario, growth continues along historical trajectories; in the two AGI scenarios, output growth is much faster because labor scarcity is no longer a constraint on output.
  - Second: Wages initially rise in all three scenarios while labor is scarce, but they plummet as the economy approaches AGI.
  - Third: Both the takeoff in output and the collapse in wages in the AGI scenarios are driven by substitution of scarce labor with comparatively more abundant machines.
- Implication: It should be possible to design institutions that compensate workers for income losses and ensure AGI gains lead to shared prosperity.
- Caveats noted about the model and projections:
  - The model assumes an efficient economy in which labor earns competitive returns.
  - Factors that may slow AGI rollout include organizational frictions, regulations, constraints on capital accumulation (such as chip supply chain bottlenecks), and societal choices to retain humans in certain functions (for example, priests, judges, or lawmakers).
  - “Nostalgic” jobs could sustain demand for human labor in perpetuity.

### Monitoring indicators to assess which scenario is unfolding
- Policymakers should track complementary indicators across multiple domains:
  - Technological benchmarks: how well AI systems perform a wide range of labor tasks (most direct measure).
  - Levels of investment: investment in research and development, talent, and computer chips that capture resource flows into AI development.
  - Adoption signals: indicators of growing AI deployment across sectors.
  - Macroeconomic and labor market trends: productivity statistics and labor market data that reveal realized economic impacts.
- Use of these indicators allows tailoring policy responses to observed developments while recognizing the future is likely to surprise us.

### Policy recommendations and institutional readiness
- Adopt adaptive, scenario-based policy frameworks and iterative scenario planning by expert teams to update probabilities and responses over time.
- Stress-test existing institutions against each scenario and reform where necessary to ensure resilience.
- Potential policy actions to consider (scalable and contingent on observed developments):
  - Reform systems of taxation.
  - Expand social safety nets.
  - Introduce small basic incomes that can be scaled up when necessary.
- Policymakers should prepare contingency plans across monetary policy, financial regulation, industrial and development strategies, fiscal sustainability, inequality mitigation, and political stability concerns.

*Anton Korinek, "SCENARIO PLANNING FOR AN A(G)I FUTURE" (December 2023).*

---


_Source: https://www.imf.org/-/media/files/publications/fandd/article/2023/december/30-33-korinek-final.pdf_