Updated
Updated · The New York Times · Sep 16
Inherent Develops Faraday AI to Build Better Versions of Itself, Chasing Recursive Self-Improvement
Updated
Updated · The New York Times · Sep 16

Inherent Develops Faraday AI to Build Better Versions of Itself, Chasing Recursive Self-Improvement

3 articles · Updated · The New York Times · Sep 16

Summary

  • London start-up Inherent is training a prototype called Faraday on the company’s own work so it can help create improved versions of itself.
  • December departures by Edward Hughes and Louis Kirsch from Google launched the effort, with the pair aiming to build an A.I. system smart enough to design a better A.I. system.
  • Faraday ingests emails, instant messages, meeting transcripts and chats with the system itself, giving it access to nearly everything happening inside the company.
  • Inherent says humans will stay involved, but the project targets recursive self-improvement — a goal many researchers see as a path toward A.I. systems that can upgrade themselves with little or no human help.

Insights

Could feeding an AI its creators' emails unlock the secret to autonomous superintelligence, or just an echo chamber?
Will the race for recursive self-improvement trigger an intelligence explosion, or a catastrophic loop of self-confirming errors?
If an AI successfully learns to rewrite its own code, who guarantees its new goals remain safe?