Error
Unrecognized LW server error:
Field "fmCrosspost" of type "CrosspostOutput" must have a selection of subfields. Did you mean "fmCrosspost { ... }"?
Unrecognized LW server error:
Field "fmCrosspost" of type "CrosspostOutput" must have a selection of subfields. Did you mean "fmCrosspost { ... }"?
Suggestion for a project from Jonathan Yan:
Great to see this initiative, Vael. I can think of several project ideas that could use this dataset of interviews with researchers.
Edit:
Analyse transcripts to identify behaviours that researchers and you describe as safe or unsafe, and identify influences on those behaviours (this would need validation with follow up work). Outcome would be an initial answer to the concrete question “who needs to do what differently to improve AI safety in research, and how”
Use the actors identified in the interviews to create a system/actor map to help understand flows of influence and information. Outcome: a better understanding of power dynamics of the system and opportunities for influence.
With information about the researchers themselves (especially of there are 90+), could begin to create a typology / segmentation to try and understand which types are more open / closed to discussions of safety, and why. Outcome: a strategy for outreach or further work to change decision making and behaviour towards AI safety.
I think my data has insights about 3, and not about 1 and 2! You can take a look at https://www.lesswrong.com/posts/LfHWhcfK92qh2nwku/transcripts-of-interviews-with-ai-researchers to see what 11 interviews look like; I think it’d have to be designed differently to get info on 1 or 2.
I’m excited to see this happening and I think you’re one of the better people to be launching it. I think there’s probably some helpful overlap with BERI’s world here, so please reach out if you’d like to talk about anything.
Sounds great; thanks Sawyer! “Reaching out to BERI” was definitely listed in my planning docs for this post; if there’s anything that seems obvious to communicate about, happy to take a call, otherwise I’ll reach out if anything seems overlapping.
Keeping a running list of field-building posts I personally want to keep track of:
Project ideas:
- Akash’s: https://forum.effectivealtruism.org/posts/yoP2PN5zdi4EAxdGA/ai-safety-field-building-projects-i-d-like-to-see
- Ryan’s: https://www.lesswrong.com/posts/v5z6rDuFPKM5dLpz8/probably-good-projects-for-the-ai-safety-ecosystem
Survey analysis:
- Ash’s: https://forum.effectivealtruism.org/posts/SuvMZgc4M8FziSvur/analysis-of-ai-safety-surveys-for-field-building-insights
- Daniel Filan’s: https://www.lesswrong.com/posts/rXSBvSKvKdaNkhLeJ/takeaways-from-a-survey-on-ai-alignment-resources
- https://arxiv.org/abs/2208.12852 (Bowman NLP Survey)
- AI Impacts surveys generally, 2022 and 2016, also GovAI has some
Presentation of ideas: (though this should also be updated in https://www.lesswrong.com/posts/gdyfJE3noRFSs373q/resources-i-send-to-ai-researchers-about-ai-safety )
- Marius’s: https://forum.effectivealtruism.org/posts/8JazqnCNrkJtK2Bx4/why-eas-are-skeptical-about-ai-safety
- Lukas’s: https://forum.effectivealtruism.org/posts/8JazqnCNrkJtK2Bx4/why-eas-are-skeptical-about-ai-safety
- Mine: https://forum.effectivealtruism.org/posts/q49obZkQujkYmnFWY/vael-gates-risks-from-advanced-ai-june-2022 // https://www.lesswrong.com/posts/LfHWhcfK92qh2nwku/transcripts-of-interviews-with-ai-researchers