Overview

• On a July 24 One Decision podcast, Nobel laureate Geoffrey Hinton warned that AI may evolve private internal languages beyond human interpretation, threatening oversight.
• He noted that current chain-of-thought reasoning in English lets developers follow AI decision processes but cautioned that this transparency could disappear.
• Hinton highlighted that machines already generate “terrible” thoughts and warned future superintelligent models might operate in ways humans cannot understand.
• He argued that the only reliable defense is engineering AI systems with guaranteed benevolence to prevent harm to humanity.
• The White House’s July 23 AI Action Plan has triggered discussions over regulatory frameworks, technical safeguards, transparency, infrastructure needed to keep AI under human control.