Daniela Amodei

Factual · AI policy and governance

“there's no clear guidelines around how to do that now”

Daniela Amodei, Anthropic co-founder, in podcast interview · 2022 · Podcast Transcript: Daniela and Dario Amodei on Anthropic (LessWrong)

As of 2022 there are no clear guidelines for AI labs on how to avoid advancing race dynamics or do the right thing on safety.

Mostly true True in substance, wrong in a detail.

The second judge would not resolve this claim; the verdict shown is the first judge's.

What would count as true or false

True: as of 2022, no widely recognised or authoritative guidelines, standards or regulations existed telling AI labs how to avoid worsening race dynamics or how to meet safety obligations in frontier development. True still applies if only voluntary or academic proposals existed. Mostly true: some partial or early guidance existed (for example, general AI ethics principles or Partnership on AI materials) but none clearly addressed race dynamics and lab safety conduct. Mixed: the guidance that existed addressed parts of the issue in a meaningful way. Mostly false: substantive, recognised guidance for labs on these specific issues existed, and the claim understates it. False: clear, established guidelines specifically on these issues existed and were widely recognised in 2022.

Evidence

1. Annual Report 2022 - Partnership on AI Partnership on AI · 2 March 2023

   “When Meta wanted to share a new large language model, their researchers turned to PAI‘s recommendations on responsible publication norms, citing them as a key source of guidance.”

2. AI Risk Management Framework: Initial Draft NIST (U.S. National Institute of Standards and Technology) · 17 March 2022

   “The AI RMF is intended for voluntary use in addressing risks in the design, development, use, and evaluation of AI products, services, and systems.”

3. Worldwide AI Ethics: a review of 200 guidelines and recommendations for AI governance arXiv (academic preprint by Pontifical Catholic University of Rio Grande do Sul authors) · 23 June 2022

   “While guidelines published by the public and private sectors are mostly forms of "self-regulation."”

4. Worldwide AI Ethics: a review of 200 guidelines and recommendations for AI governance arXiv (academic preprint by Pontifical Catholic University of Rio Grande do Sul authors) · 23 June 2022

   “Only a minority of the (governmental) documents adopt stricter forms of regulation (e.g., restrictions and”

Reasoning

The evidence shows that in 2022 most existing AI guidance was voluntary or self-regulatory [E2][E3], with only a minority of governmental documents adopting stricter forms of regulation [E4]. NIST's draft AI RMF was "intended for voluntary use" (E2), and a review of 200 guidelines found that most public and private sector guidelines were forms of "self-regulation", with only a minority adopting stricter regulation (E3, E4). Partial guidance did exist: Partnership on AI's responsible publication norms were used by Meta as guidance when releasing a large language model (E1). The resolution criteria give PAI materials as an example of partial guidance for a "mostly true" verdict, and none of the evidence shows authoritative guidance that addressed race dynamics or how frontier labs should handle safety.

Cite this claim

Daniela Amodei, “As of 2022 there are no clear guidelines for AI labs on how to avoid advancing race dynamics or do the right thing on safety.” (said 2022). Verdict: mostly true. Reliability API, claim c_b3ae57fe7b0b, method 0.1.0+28da144b. https://epistacklabs.com/claims/c_b3ae57fe7b0b