This post is in response to Critiques of Prominent AI Safety Labs.

Two points are specifically advocated for:

  1. Specifically, it is requested that "Omega" - the anonymous account behind the critiques - immediately discloses all of their authors' funding, affiliations, and conflicts-of-interest around organizations they've been anonymously lampooning.
  2. Generally, it is recommended that the AI Safety and Effective Altruism communities discuss, debate, and adopt principles around anonymous critique.

Generally, there are multiple issues at play. Anonymous critique can be useful for whistleblowing or discussion of very sensitive issues. However, there is a general presumption of a "right to face one's accuser" in the enlightenment tradition. To make accusations that a person or group is ineffective in research, has bad character or culture, or to go as far as to make suggestions that an organization should cease hiring and funding activities until anonymous critiques are addressed...

... this is a very obvious vector for abuse.

Specifically, the anonymous Omega authors wrote:

Readers should not assume that we are completely unbiased or don’t have anything to personally or professionally gain from publishing these critiques.

Reasonable minds would like to know what those biases and personal and professional gains are, as well as the credibility of anonymous authors in inspecting and judging the quality of research output, staffing, progress towards goals, etc.

There are ways to balance anonymity or pseudonymity with transparency and ability to respond to one's accusers. For instance, a small panel of highly trustworthy people from different organizations and backgrounds might be selected to review credentials, funding, and personal and professional biases.

As it stands, the "Omega" account has spent tremendous time and effort mixing objective facts, subjective judgments, and - frankly - gossip and slander in their accounts.

The author of this counter-critique  is not a member of Redwood Research, Conjecture, or Anthropic, not a funder of said organizations, and is not a friend or close associate of any of their staff or managers. The author, likewise, is not running an AI Safety Lab with over $10 million in funding.

The author of this piece simply thinks that anonymous hit pieces have material negative consequences, and that we as a community should have stronger safeguards for allowing anonymous and pseudonymous critique without it becoming a vector for abuse.




Sorted by Click to highlight new comments since: Today at 12:42 PM

I note that the author of this post apparently felt it necessary to create a burner to criticize a psuedonymous account with a few posts. I'm not criticizing that choice . . . but if we're at the point where people feel the need to use burners to criticize psuedonymous accounts like Omega, then it seems the fear of adverse consequences from posting criticism runs deeper than I had expected.

And that extent of fear would seem to validate Omega's choice to remain psuedonymous.

For posts like this, it seems valuable to decouple voting dimensions for karma and agreement.