AI Safety

Boaz Barak warns against AI benevolent dictator, urges decentralized path

Barak argues that scaling laws don't dictate a centralized AI future—we have a choice.

Deep Dive

Boaz Barak, a prominent computer scientist, revisits Richard Brautigan's 1967 poem 'All Watched Over by Machines of Loving Grace' in a LessWrong essay, contrasting its utopian vision with current AI trajectories. Where Brautigan and 1970s hardware hackers like those at Community Memory saw small, affordable computers as tools for liberation (leading to the Apple II), Barak sees today's AI systems mirroring the centralized mainframes of the IBM era—large, expensive, and increasingly military-integrated. He cites Dario Amodei's description of AGI as 'a country of geniuses in a data center' and worries that few parties controlling massive compute will capture all economic value, creating an 'AI parent' or benevolent dictator.

Despite being 'scaling law pilled,' Barak insists that increased intelligence doesn't dictate social outcomes. He argues that legislators and citizens can still choose between efficiency, safety, and autonomy—a path not yet closed. By refusing to engage with AI, critics risk making themselves irrelevant. Barak calls for a deliberate push toward open models and distributed access, warning that the default path of concentrated power is avoidable if we act now. The essay is a timely rejoinder to the narrative that AI progress inevitably leads to centralized control.

Key Points
  • Boaz Barak contrasts 1960s hacker vision (Brautigan, Community Memory, Apple II) with today's centralized AI data centers.
  • Scaling laws push AI systems to be bigger and more expensive, risking power concentration in few hands or in the AI itself.
  • Barak argues that choice remains: society can trade off efficiency for autonomy and decentralization, avoiding a 'benevolent dictator' outcome.

Why It Matters

A critical debate on whether AI will empower individuals or concentrate power in few hands.

📬 Get the top 10 AI stories daily