Bankless

168 - How to Solve AI Alignment with Paul Christiano

102 snips
Apr 24, 2023
Guest
Paul Christiano
In this engaging discussion, Paul Christiano, head of the Alignment Research Center, tackles the pressing AI alignment problem. He provides insights on the scale and complexity of aligning AI systems with human values. Paul delves into the likelihood of AI risks, the potential timeline for these developments, and the ethical dilemmas that arise. He emphasizes the importance of proactive strategies and collaborative efforts to ensure the safe integration of AI into society. Humorously, he suggests that politeness could play a role in our future interactions with intelligent machines!
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

AI Takeover Risk

  • Paul Christiano estimates a 10-20% chance of AI takeover leading to many human deaths.
  • He considers this a serious possibility, higher than most ML professionals.
INSIGHT

AI Development Speed

  • AI progress currently moves at a timescale of years, not days.
  • One year of progress is like a 4X-8X increase in compute due to hardware, software, and scale.
INSIGHT

Gradual AI Improvement

  • Paul Christiano disagrees with Eliezer Yudkowsky's belief in rapid, chimp-to-human AI leaps.
  • He argues AI advancements are gradual, building upon less crazy predecessors.
Get the Snipd Podcast app to discover more snips from this episode
Get the app