Interview
Highlights: Michael Webb on whether AI will soon cause job loss, lower incomes, & higher inequality
- AI exposure is projected to peak for jobs at the 88th salary percentile before declining for top executive roles, a pattern mirroring OpenAI research on GPT-4.
- Highly regulated professions such as doctors, lawyers, and accountants possess the capacity to erect barriers against wage declines, though professionals and unions are expected to succeed in slowing change to suit their interests.
- Automation may offset initial displacement through demand elasticity and complementarity, potentially increasing human employment in modified roles as seen in banking with ATMs.
- In sectors like fast food, full automation could eliminate employees within a year via natural wastage given an average tenure of six months, without requiring active firing.
- Significant and enduring wage declines of approximately 25% are anticipated for older workers in geographically concentrated areas with single employers where jobs are automated.
- The adoption curve for general purpose technologies historically spans roughly 30 years to reach 50% adoption, with US capital stock in software and computer hardware rising from near zero percent in 1970 to two percent by 1990, and eight percent by 2000.
- Historical precedents indicate slow transition speeds, such as the 90-year interval from telephone system invention to the 1980 elimination of the last manual switcher job, following a peak in human operators in 1920.
- Legal frameworks may restrict AI use in sensitive areas like prescription medication within a decade due to powerful interest groups, though such bans could be overturned.
- Regulatory implementation is expected to be suboptimal, occurring either too slowly or too rapidly, while governments face the plausible ability to classify all AI research via executive order overnight.
- Long-term mass unemployment could theoretically reach 50% or 90% of jobs, with a 90% automation scenario likely unfolding over a 100 to 200-year timeframe similar to US agricultural automation.
- Even with 90% of tasks automated, the economy would continue to function due to non-cognitive activities generating high demand for human labor.
- The speed of technological transition is constrained by capital availability and the economic impossibility of halting the entire economy to facilitate retooling.