Amazon security VP argues human-in-the-loop AI oversight fails due to human inconsistency

Summary

Amazon's VP of Security, Eric Brandwine, argues that human-in-the-loop AI oversight is ineffective because humans are inconsistent and prone to "normalization of deviance" — a phenomenon where people gradually accept lower standards over time. He suggests that relying on humans to catch AI errors is flawed, as human attention and vigilance naturally degrade, making the widely accepted governance principle less reliable than commonly believed.

Source

bskyAmazon security VP argues human-in-the-loop AI oversight fails due to human inconsistencythenextweb.com