The Gradient: Perspectives on AI

Joe Edelman: Meaning-Aligned AI

Mar 9, 2023
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Couchsurfing Metrics Sparked A Movement

  • Joe Edelman recounts building metrics at Couchsurfing that measured meaningful human connections rather than raw engagement.
  • That work led him to found Time Well Spent and later the Center for Humane Technology to counter engagement-driven harms.
INSIGHT

Revealed Preferences Are Misleading

  • Revealed preferences are information-poor because choices are made from limited option sets and influenced by system structure.
  • Optimizing for raw clicks amplifies feedback loops that diverge from what genuinely matters to users.
INSIGHT

Decompose Clicks Before Optimizing

  • Click data mixes limited menus, early exploration, and platform-driven behaviors, making it a noisy signal of true values.
  • Systems should decompose revealed preference into components and amplify only the parts tied to genuine meaning.
Get the Snipd Podcast app to discover more snips from this episode
Get the app