Source: Cambodianess

Governments adopting AI to search, summarize, sort and draft face a new bottleneck: finite human attention. The piece documents how AI-generated writing has surged — citing rises in pro se federal filings and machine-written civil complaints in the United States, a July warning from the U.S. Government Accountability Office that the IRS could be swamped by AI-produced public comments, and a 57% rise in grant applications across 12 UK and European funders tied to AI assistance. An OECD survey of 58 national tax authorities shows AI use jumping from 9% in 2016 to 69% in 2023, with tax administrations applying AI to analytics, case selection, virtual assistants, and high-volume automation.

Scholars and rulings create legal constraints and oversight challenges: the EU AI Act requires designs enabling effective human oversight for high-risk systems, and EU data-protection rules bar decisions based solely on automated processing. The European Court of Justice’s 2023 SCHUFA decision is cited to show that formal human sign-off does not equal meaningful oversight when outcomes depend on automated assessments. The author cites David Woods’s “law of stretched systems” and Lisanne Bainbridge’s “ironies of automation,” and an engineering study modeling nuclear-reactor licensing that found senior reviewers (about three approvals per day) determined timelines regardless of AI capability. The recommended approach is to use AI to reduce cognitive burden and present trade-offs and uncertainties so humans can exercise judgment rather than delegating consequential decisions to machines.