AI Defends the Status Quo

Concerns and disappointments around AI deployments are mounting, which is not a surprise given how the capabilities of the systems never justified the overheated hype used to push stock valuations to stratospheric levels.

Recently, an ambitious effort to move peer review into the AI space involving AIP, IOP, and APS seems to have gone down in flames, with no amount of happy talk in the press release able to conceal the impression that this was gently shelved after underwhelming if not worrisome outcomes.

• Why they even issued a press release is beyond me.
• I’d also argue that the “repetitive administrative work” casually disparaged by the technologists is often where the magic lies in discovering problems, crafting solutions, or gaining insights.
  • Catering to a heightened sense of “above it all” importance among editors instead of requiring them do their jobs — drudgery and all — is one way to have your editorial systems become less effective.
    • The devil is in the details.

But the devil is also in the AI systems, and a recent preprint is making waves showing briefly and elegantly why this seems to be the case as more and more tea spills out from user experiences.

The preprint echoes a recent story in the New York Times about how OpenAI is now facing five wrongful death lawsuits because it tuned its AI to be engaging and affirming, which may have led to multiple people being led down fatal rabbit holes. An even greater number have been led into long-term delusions.

The author of the preprint is an interesting character. Hiroko Konishi is a Japanese voice actor and musician who has become an independent AI researcher. Programming since elementary school and now 50 years old, Konishi has been doing independent research into AI systems for some time. However, a trio of preprints recently caused her research to gain worldwide attention, far beyond what it had received earlier when limited mostly to Japan.