AI might be creating underclass, but makers of tech bubble are replaceable (www.theguardian.com)

The New Yorker piece spotlights a stark warning from OpenAI employee Leopold Aschenbrenner that AI could “reach or exceed human capacity” by 2027, potentially creating a lasting economic underclass for people not embedded in the AI economy. Critics like Nate Soares urge people to stop banking on long-term job security. The column situates those fears in historical context—mechanisation repeatedly dislocated workers—and points to contemporary pushback: the Writers Guild agreement limiting forced AI use and preventing generative models from replacing writers’ source material, plus older examples of union wins in Japanese auto manufacturing (“cooperative modernisation”) and Australian nursing retraining programs. Beyond social consequences, the report flags technical and economic fragilities undermining the narrative of inevitable AI-driven prosperity. Researchers and surveys show weak returns and operational failures: MIT finds 95% of corporate AI pilots deliver negligible ROI, Carnegie Mellon reports agents fail basic office tasks ~70% of the time, and analysts warn of rapid data‑centre depreciation (as fast as three years). Market signals—Bank of England and investor concerns, firms regretting automation cuts, and analyses showing layoffs often cost more than they save—suggest the current AI investment boom may be a bubble. The takeaway: displacement isn’t unchangeable; collective bargaining, regulation, and realistic appraisal of AI’s limits are central levers for steering technological change more equitably.