The “AI Scientist” Hides Money
Why will nobody divulge who paid to publish this commercial OA paper?
Last week, a weirdly outdated paper touting a demonstration of what the authors called an “AI scientist” made the rounds. It was published OA in Nature, which commands an APC of nearly US$13,000.
The paper seems outdated because we have been dealing with systems that claim to be able to function as end-to-end science publishing bots for years now — from Rachel So and her coterie of fake scientists associated with aiXiv to systems that can write papers in 20 minutes with only the most basic prompt.
It also seems like a press release.
So I wasn’t surprised to find that the core authors had posted a preprint in 2024. (Two authors were added to the Nature paper.) The preprint contained even more overt commercial claims like showing how a lot of papers could be generated for $15 each. Once again, we have a preprint as a press release, followed by a paid paper as an advertisement. We have created terms for being used.
The Nature paper claims to show that papers generated by their “AI scientist” passed screening for a conference proceedings, but if the Nature editors don’t know the history of SciGen and how fake machine-generated language papers were passing into conference proceedings nearly two decades ago, shame on them.
- The premise is fundamentally flawed because this isn’t science, it’s manipulative science cosplay.
The two entities with commercial angles the researchers come from are:
- Sakana AI, a Japanese AI company with about $230 million in funding
- Vector Institute, a private, non-profit research institute in Canada
- One of the goals of the institute is to support AI adoption in industries. Private funders include Uber, Google, and Shopify.