In an Oct. 24 post, Arnold Kling argues that a total addressable market (TAM) calculation may miss industries enabled by AI, making it an incomplete way to measure artificial intelligence’s effect on GDP.
Kling points to changes in U.S. homes across the 20th century. In his account, fewer than 15% of American households had indoor plumbing in 1900; by 1960, 87% had a television. He says water and electrical infrastructure helped bring household products into use that were absent from the 1900 economy. A calculation based on industries and demand at the time, he suggests, might have assigned washing machines no addressable market.
He draws a similar lesson from Ford’s assembly line. Kling says Ford introduced the production method in December 1913, cutting the time to make a car from 12 hours to 90 minutes. He argues the productivity gain would look small when measured against cars’ share of GDP in 1913, even though cars represented a larger part of the economy by the mid-1920s.
Kling describes a startup slide-deck calculation that estimates annual profit using market size, expected market share and profit margin. The AI version, he says, estimates how much of the economy is in scope, how many tasks could benefit and the productivity improvement on those tasks.
In his example, AI applies to 20% of industries and benefits 50% of tasks in those industries, with a 90% productivity improvement on the affected tasks. Kling calculates an overall gain of 9%; if it takes 10 years to arrive, the annual increase is below 1%. He says this approach roughly reflects how Daron Acemoglu arrived at a conservative estimate for AI productivity gains.
Kling argues that calculations based on current economic activity can overlook industries that emerge as technology spreads. He also says GDP measures understate the change indoor plumbing brought to daily life, including having water and a toilet inside the home. His post offers no alternative numerical estimate for AI’s effect on GDP.
