Debate: superintelligence could exploit human vision even if predators never did
basedjensen · x · 2026-09-18
- Responding to the claim that our visual cortex evolved to resist manipulation (otherwise predators would exploit it), the reposter argues predators weren't superintelligences with pixel-level control, unlimited experimentation, and a working model of human vision.
- Core argument: if the channel is physically exploitable, a sufficiently capable ASI should bridge the evolutionary gap and exploit it fully — "the tiger never discovered adversarial examples; that doesn't mean the attack surface isn't there."
- A notable AI-safety debate on whether ASI could manipulate humans via the visual channel.
More from AGI Musings
- banteg flags: $1B of AI-safety spending assigned benefit of 10 quintillion future lives under main estimate — banteg · 2026-09-18
- AI risk debates miss the point: skeptics underestimate wet-lab automation — JacquesThibs · 2026-09-18
- UCSD math dept split courses into real and fake versions with identical degree credit — sytelus · 2026-09-18
- Chart by Economist Ernie Tedeschi Challenges the "AI Arms Race with China" Narrative — birchlse · 2026-09-18
- Musk: Starship may be one of the last massive engineering feats designed before AGI — XFreeze · 2026-09-18
- ControlAI proposes three stopgap measures for immediate AI security threats — DavidSKrueger · 2026-09-18