AI makes the production of plausible items cheaper than their evaluation. Hence the proposal to capitalize on the audits that researchers are already doing by reading, citing or reproducing.
Haotian Zong and Luhao Zhang's observation is simple: AI lowers the cost of producing articles and results that appear credible much faster than that of a serious evaluation. However, much of the control already carried out by researchers, when they read, quote, reimplement or extend a work, is never preserved or reusable. They therefore propose a Community Verification Layer, an interoperable and rule-based infrastructure that would transform these ordinary gestures into lasting pieces of evidence, attached to a specific statement and a given version of an article.
The system would take the form of a graduated scale, from comprehension notes to reproduction, replication and downstream validation, including verification of claims, audits and code execution. It would collect opinions before displaying social signals, to avoid the herd effect, keep track of disagreements, verify the identity of contributors while giving them the choice to reveal themselves, and direct rare attention to uncertain, high-stakes or neglected work. Above all, this verification work would become citable, with a credit proportionate to the quality of the elements provided. The authors distinguish it from existing forms of open or post-publication evaluation.
Read the full article on the original source
Go to the source →