Macroeconomic implications of advances in AI
Daron Acemoglu in this paper evaluates the macroeconomic implications of advances in artificial intelligence (AI). Here are the main findings:
- Productivity Gains: Acemoglu uses a task-based model to estimate the productivity gains from AI. He finds that the macroeconomic effects of AI, driven by cost savings and productivity improvements at the task level, are modest. The estimated increase in total factor productivity (TFP) over the next 10 years is no more than 0.66%, and potentially less than 0.53% when considering the difficulty of automating hard-to-learn tasks.
- Wage and Inequality Effects: The paper explores the impact of AI on wages and inequality. It theoretically shows that even if AI improves the productivity of low-skill workers in certain tasks, it may not reduce inequality. Empirical findings suggest that AI advances are unlikely to increase inequality as much as previous automation technologies, but there is no evidence that AI will reduce labor income inequality. Instead, AI is predicted to widen the gap between capital and labor income.
- New Tasks and Social Value: The paper discusses the creation of new tasks by AI, noting that some may have negative social value, such as those designed for online manipulation. It emphasizes the importance of considering the macroeconomic effects of these new tasks.
- Investment and GDP Growth: The paper estimates that the overall impact of AI on GDP growth is modest. With investment responses incorporated, GDP is estimated to grow by 0.93% to 1.16% over the next 10 years. If the investment response is similar to earlier automation technologies, the upper bound on GDP effects rises to around 1.4% to 1.56%.
Overall, the paper concludes that while AI has the potential to impact productivity, wages, and inequality, the macroeconomic gains are likely to be modest, and the distributional effects may not favor reducing inequality.
The Simple Macroeconomics of AI
Author: Daron Acemoglu
From: MIT
More on AI and its impact on wages and output
In this paper, Korinek and Suh develop an economic framework to analyze how the transition to artificial general intelligence (AGI) – AI systems that can perform all cognitive tasks at human levels – may impact output and wages. Their basic assumption is that advances in technology make ever more complex tasks amenable to automation. They find that:
- If the distribution of task complexity is unbounded (has an infinite Pareto tail), wages can rise forever as automation complements abundant capital by taking over more complex tasks, but there are always remaining tasks for labor. However, if the tail is too thin, wages eventually collapse as automation outpaces capital accumulation.
- If the complexity of tasks humans can perform is bounded, mirroring limits of human cognition, wages first surge as machines automate tasks but then collapse even before full AGI, as labor loses its scarcity value.
- Adding fixed factors like land or minerals causes wages to decline in the long run, as these pose bottlenecks that prevent sufficient capital accumulation to complement labor.
- However, automating technological progress itself can restart wage growth after an initial collapse by triggering an intelligence explosion and economic singularity.
- The paper analyzes various scenarios numerically, showing for example that under plausible “AGI within 20 years” assumptions, wages first rise but then collapse, with output temporarily growing 10x faster than usual before settling at a higher steady-state growth rate.
- Slowing the pace of automation can maximize wage growth, albeit at the cost of foregone output. Heterogeneous worker skills imply a declining labor share over time with an ever-smaller elite earning rising wages.
In summary, the paper provides a rigorous economic framework to study the impacts of AGI under different assumptions, highlighting that the transition could lead to very different scenarios for wages depending on the nature of tasks and capital accumulation.
Scenarios for the Transition to AGI
Authors: Anton Korinek, Donghyun Suh
From: University of Virginia
