Prospects for Alignment Automation: Interpretability Case Study — LessWrong

For human-level AI (HLAI) we will need robust control or alignment methods. Assuming short timelines to HLAI, the tractability of automating safety r…

x Prospects for Alignment Automation: Interpretability Case Study — LessWrong AI-Assisted Alignment AI Frontpage 33 Prospects for Alignment Automation: Interpretability Case Study by Jacob Pfau , Geoffrey Irving 21st Mar 2025 AI Alignment Forum 10 min read 5 33 Ω 14 For human-level AI (HLAI) we will need robust control or alignment methods. Assuming short timelines to HLAI, the tractability of automating safety research becomes central. In this post, I will make the case that safety-relevant progress on automated interpretability R&D is likely; however, naive interpretability automation may on

saved by

• Emanuel Ruzak

related reading

• Automation collapse — LessWronglesswrong.com
• Automated Alignment is Harder Than You Think — LessWronglesswrong.com
• Alignment remains a hard, unsolved problem — LessWronglesswrong.com
• Alignment is not solved but it increasingly looks solvablealigned.substack.com
• The Hitchhiker's Guide to Actionable Interpretabilityactionable-interpretability-guide.github.io
• A Pragmatic Vision for Interpretability — AI Alignment Forumalignmentforum.org
• Can we safely automate alignment research? - Joe Carlsmithjoecarlsmith.com
• Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover — LessWronglesswrong.com
• How To Go From Interpretability To Alignment: Just Retarget The Search — LessWronglesswrong.com
• How Can Interpretability Researchers Help AGI Go Well? — AI Alignment Forumalignmentforum.org
• Research Areas in Interpretability (The Alignment Project by UK AISI) — AI Alignment Forumalignmentforum.org
• A Pragmatic Vision for Interpretability — LessWronglesswrong.com