"The Simple Macroeconomics of AI" - Marginal REVOLUTION

🌈 Abstract

The article discusses Daron Acemoglu's paper "The Simple Macroeconomics of AI", which is skeptical about the overall economic effects of AI. The key points are:

• Acemoglu estimates that the increase in total factor productivity (TFP) due to AI over 10 years will be no more than 0.71%, or less than 0.55% when accounting for the challenges of automating hard-to-learn tasks.
• The author disagrees with Acemoglu's methodology and conclusions, arguing that:
  • The benefits of AI will come from eliminating the least productive firms and introducing new goods, which are not fully captured in Acemoglu's model.
  • Acemoglu's model assumes AI will not enable new capabilities and tasks, which the author believes is incorrect.
  • Acemoglu's welfare economics arguments are questionable.

🙋 Q&A

[01] Acemoglu's Estimates and Methodology

1. What are Acemoglu's estimates for the impact of AI on total factor productivity (TFP)? According to the article, Acemoglu estimates that the increase in TFP due to AI over a 10-year period will be no more than 0.71%, or less than 0.55% when accounting for the challenges of automating hard-to-learn tasks.

2. How does Acemoglu derive these estimates? Acemoglu uses a version of Hulten's theorem, which states that GDP and aggregate productivity gains can be estimated by the fraction of tasks impacted by AI and the average task-level cost savings. However, the author argues that this methodology is flawed, as it does not account for the benefits of eliminating the least productive firms and introducing new goods.

3. What are the author's criticisms of Acemoglu's methodology? The author believes Acemoglu's methodology is misleading, as it:

• Does not consider the benefits of getting rid of the least productive firms, as seen with international trade.
• Fails to account for the gains from "new goods" enabled by AI, where the previous prices were effectively infinite.
• Assumes AI will not enable new capabilities and tasks, which the author believes is incorrect.

[02] Welfare Economics and New Goods

1. How does the author criticize Acemoglu's approach to welfare economics? The author finds Acemoglu's statement that "what is relevant for consumer welfare is TFP, rather than GDP" to be odd and a questionable "redefinition of all welfare economics in a single sentence". The author argues that consumer welfare should be based on the sum of consumer and producer surpluses, not just TFP.

2. What is the author's view on the potential gains from new goods enabled by AI? The author believes the gains from new goods, such as services like Character.ai, could be much higher than the incremental productivity improvements considered by Acemoglu. The author argues that Acemoglu's model fails to adequately account for these potential gains from new AI-enabled goods and capabilities.

3. How does the author respond to Acemoglu's dismissal of gains from "new science" within the 10-year timeframe? The author questions whether Acemoglu's distinction between "new science" and "TFP" is clear, and suggests that improvements in software engineering, which could be considered "new science", may in fact occur within the 10-year period and provide significant gains.