Debate: Offense-Defense Balance Governs AI Centralization
deedydas · x · 2026-08-16
This post summarizes a high-quality debate between Gavin Baker and Sholto Douglas on whether AI power should be concentrated or distributed. The core framework is the "offense-defense balance" from international relations: whether it is cheaper to attack or defend with a given technology.
Key Arguments:
- Biosecurity: Offense-dominant well into the 2030s. Synthesizing pathogens costs $10k, while vaccines take billions and years.
- Cybersecurity: Defense-dominant eventually, but requires 2 years of hardening critical infra before open distribution is safe.
- Compute Bottleneck: The real bottleneck is compute. By 2030, 65-75% of >100GW compute will be owned by 6-7 US entities, meaning frontier capabilities will remain centralized regardless of distribution ideals.
Nuance: Domains like propaganda and surveillance are already offense-dominant and widely distributed. The conclusion is a mixed strategy: distribute in defense-dominant domains and gate in offense-dominant ones.
More from AGI Musings
- Why you shouldn't press the 'replace this universe' button: an alignment debate — jd_pressman · 2026-10-16
- Meta scales back AI agent plans as staff clean up after the machines — SumitGup · 2026-10-16
- RFdiffusion3 debuts in Science: generative AI designs proteins and nucleic acids atom by atom — AllThingsApx · 2026-10-16
- Researcher: 'Guru-like' rhetoric fuels claims that AI safety is a cult — jd_pressman · 2026-10-16
- Comparing AI data centers to 1900s power buildout: the anti-DC craze will look silly in a decade — GabGarrett · 2026-10-16
- Commentary: it's unfair to expect Yudkowsky to update on a safety announcement without real mitigations — jd_pressman · 2026-10-16