AI Models Are Using Material From Retracted Scientific Papers

"The chatbot’s answers referenced retracted papers in five cases but advised caution in only three."

I think this may actually be positive, it just needs to get better at advising caution on the redacted papers.

It is important that negative or null responses be understood. In some ways redacted papers lead to better understanding of what didn't work, why not, what were the errors in the paper?

Was it complete fraud, ok, good to know. Was a mistake made, or have we learned something else as the science has progressed?

I wrote about this recently on our blog Beyond Cheerypicking Data[1].

We work in neurotech/sleeptech we increase slow-wave activity during sleep or what we refer to as enhancing restorative function.

We are building on more than a decade of research and 50+ published peer reviewed papers. We haven't seen retractions of papers, but we've seen a few null results, which helped us understand where the implementation of the technology has struggled in the past.

The article focuses on scientific papers for public consumption, but that doesn't really matter to the AI atm. There is so much junk pretend science on the internet already that has no citations. I see this as being something that can help the research community as it improves. Maybe with that background AI can be better at understanding when junk science is being promoted.

[1]https://www.affectablesleep.com/blog/beyond-cherry-picking-m...