PC
Paul Christiano
Also known as: Paul
Pioneer of RLHF and AI alignment research; founder of Alignment Research Center (ARC); key theorist of iterated amplification and eliciting latent knowledge
Current Role
Head of AI Safety, US AI Safety Institute
Organization
US AI Safety Institute
Born
1992
Age ~34
Expert Positions12 topics
| Topic | View | Estimate | Confidence | Date |
|---|---|---|---|---|
| AGI Timelines | Medium | 2035-2045 | low | 2023 |
| P(doom) | Significant | 10-20% | medium | 2023 |
| How Hard Is Alignment? | Hard but tractable | 50% | medium | 2023 |
| Current Approaches Scale | Uncertain | 40% | medium | 2023 |
| Inner Alignment Solvability | Hard but tractable | Solvable with sufficient investment | medium | 2023 |
| Likelihood of Deceptive Alignment | Significant concern | 50% | medium | 2023 |
| P(AI X-Risk This Century) | Moderate | ~20-50% | medium | 2023 |
| Would Misalignment Be Catastrophic? | Uncertain, depends on scenario | 20-40% → catastrophic | low | 2023 |
| P(AI Catastrophe) | Significant | 10-20% | medium | 2023 |
| Takeoff Speed | Slow | 5-15 years | medium | 2023 |
| Will Advanced AI Be Deceptive? | Possibly detectable | 40% | medium | 2023 |
| Will We Get Adequate Warning? | Likely | 70% | medium | 2023 |
Sources: ARC Research (2023) · Various posts (2022-2023)
Career History3
Education
PhD in Computer Science, UC Berkeley; BS in Mathematics, MIT
Publications & Resources5
Eliciting Latent Knowledge (ELK)
→2021Technical ReportAlignment Research
AI Safety via Debate
→2018PaperAlignment Research
Iterated Amplification and Distillation
→2018Blog PostAlignment Research
Deep Reinforcement Learning from Human Preferences
→2017PaperTechnical Safety
Concrete Problems in AI Safety
→2016PaperTechnical Safety
No funding connections recorded.
Links
Organization Roles1
Facts9
People
Role / TitleHead of AI Safety, US AI Safety Institute
Employed ByUS AI Safety Institute
Biographical
Birth Year1992
EducationPhD in Computer Science, UC Berkeley; BS in Mathematics, MIT
Notable ForPioneer of RLHF and AI alignment research; founder of Alignment Research Center (ARC); key theorist of iterated amplification and eliciting latent knowledge
Social Media@paulfchristiano
Wikipediahttps://en.wikipedia.org/wiki/Paul_Christiano
Google Scholarhttps://scholar.google.com/citations?user=6gHkYDgAAAAJ
General
Websitehttps://paulfchristiano.com