Researcher’s reasoning

Eager to please AI assistants smooth over the gaps in our thinking

Adrian Barnett and Matt Spick recently described a world in which agentic AI can assemble what they call a “thirty minute paper“, essentially churning out formulaic analyses at a scale that, by their estimate, now outnumber legitimate research ten to one in some datasets. Thomas Plumper and Eric Neumayer argue that the larger threat is not fabrication, but what they call “tweaking,” the selective reporting of model specifications that is, in their words, observationally indistinguishable from honest practice. The incentive structure Barnett and Spick describe is the academic version of a problem media industries have been losing to for the past decade. ….[READ]