AI Work Displacement Estimates and Policy Proposals Span Wide Range

Estimates of how much work AI will displace range widely across analysts and institutions. An Anthropic analysis concludes that robots and LLMs together expose roughly 80% of work tasks to automation, with robots already capable of performing 74% of physical tasks in the US, covering 34% of working hours, though robots are cost-competitive for only 0.3% of job tasks today. Acemoglu projects AI will take over roughly 5% of human work within ten years, assuming about 20% of US work is automatable but expecting only a quarter of that to actually be automated within a decade, with a projected GDP gain of about 1.5%. Shane Legg, co-founder of Google DeepMind and its Chief AGI Scientist, stated that remote cognitive jobs done entirely on a laptop are the most exposed to automation or significant reduction by AI.

Benchmark evidence points to near-term displacement in at least one professional field. A study found Claude Opus 5 scored 20 out of 20 on medium-length accounting tasks, completing each in minutes, while 12 licensed CPAs scored between 0% and roughly 90%, with several failing to finish within the three-hour limit. The study concluded that on medium-length, well-defined accounting tasks, frontier AI models are now faster and more accurate than junior accountants, and noted this is a reversal from eighteen months earlier when frontier AI scored well below human accountants on comparable tasks. Blackstone President Jon Gray predicted that any rule-based businesses, including accounting, legal, and finance, will be completely disrupted by AI. JPMorgan dropped proxy advisors for shareholder votes and replaced them with AI.

On the labor creation side, LinkedIn estimates more than 750,000 new AI-related jobs have been created in the US since 2023, led by data annotators at roughly 282,000 new positions, followed by data center jobs at about 117,000 and AI engineers at about 105,000. Data annotator roles carry a median listed salary of about $51K, below the overall non-AI job median, while the median across all AI-related roles is about $180,000 compared to about $80,000 for all jobs.

Public opinion runs well ahead of displacement already occurring. Gallup found 79% of Americans expect AI to reduce US jobs over the next decade, while only 27% trust businesses to use AI responsibly. A Reuters/Ipsos poll found 53% of Americans fear AI could cost them or someone in their household a job, and 73% of adults under 30 believe AI will eventually reduce jobs.

Policy proposals span several frameworks. A joint paper from Google DeepMind, Oxford University, and the University of Chicago concludes that no single policy works in every scenario and recommends governments immediately expand unemployment benefits and low-wage tax credits, with those credits potentially evolving into a guaranteed income floor if joblessness persists, and citizens receiving stakes in a public investment fund if workers' share of the economy continues declining. The paper notes retraining programs only help if AI keeps creating new jobs, otherwise it just reshuffles who gets the jobs left. Bill Gates, speaking on The Ezra Klein Show, argued that companies replacing workers with robots or AI should pay FICA taxes equivalent to what displaced workers would have paid, stating that society has the right to decide whether to prioritize lower costs from AI over broader social considerations.