Will A.I. Trap You in the "Permanent Underclass"? (www.newyorker.com)

Tech commentary and Silicon Valley memes have crystallized a fear that accelerating AI could create a “permanent underclass”: a mass of people economically displaced as those with capital buy compute and deploy AI to do everything from coding and marketing to managing factories. The idea gained traction after researcher Leopold Aschenbrenner suggested models could match human researchers by 2027, and pessimists like Nate Soares warn of a self‑improving, runaway loop where AI builds ever more capable AI. Evidence cited includes AI-generated video feeds from OpenAI and Meta, AI “agents” embedded in corporate software, autonomous vehicles like Waymo, and early labor-market signals such as higher unemployment for recent graduates and shrinking demand for entry‑level engineers. For AI/ML practitioners the stakes are both technical and social: if models reach or exceed human capability in research and engineering, the economics of work change fundamentally, concentrating power with those who control compute and data. That raises urgent policy questions — redistribution, UBI, regulation of compute and model development — that currently lack consensus. Culturally, the response is bifurcation: a hyperproductive “cracked” elite who collaborate closely with machines and a marginalized majority pushed toward gigged, AI‑mediated labor or human‑only niches. The piece warns this dynamic could entrench inequality and political instability unless technical progress is paired with deliberate economic and governance strategies.