newsfilter.io

AI May Not Take Over. But It Could Let A Few Humans Do Exactly That. (article by Rose Hadshar)

80,000 HoursRose Hadshar, Dominic Armstrong
Watch on YouTube →
  • By 2047, there is a 50% probability that AI could automate all human tasks, with some experts and CEOs anticipating AGI or intelligence explosions occurring significantly sooner or later.
  • If advanced AI is deployed, predictions suggest that no single country will control more than 25% of global GDP and no company will earn over 1% of GDP, potentially leading to extreme power concentration where a tiny number of people control systems thinking millions of times faster than humans.
  • Plans and potential outcomes include the displacement of human workers, a collapse in labor value, the erosion of democratic checks and balances, and the possibility of a single entity controlling over 99% of resources or establishing permanent hegemony in space.
  • Significant risks involve military coups via loyal AI, secret loyalties in algorithms, epistemic interference through biased advice or obfuscation, and the inability of society to adapt to rapid technological shifts, which could lead to the permanent preservation of a single worldview.
  • The tractability of preventing these outcomes is currently viewed as uncertain; while technical mitigations like alignment audits are considered feasible, structural forces like competitive dynamics and capability gaps are seen as difficult to reverse.
  • The authors identify a severe funding gap as of September 2025, with only one known public grant program totaling $4 million and very few full-time researchers working on the issue, despite it being classified as one of the world's most pressing problems.
  • Counter-arguments and potential benefits suggest that if power becomes concentrated, leaders might have no incentive to oppress the populace due to guaranteed tenure, and massive wealth generation could improve material living conditions even if political power is lost.
  • Strategic advice recommends that most people bear the risk in mind rather than acting directly, while those well-positioned in AI companies or governance should focus on law-following AI, alignment audits, reasoning tools, and policies that distribute power.