AI-Enhanced Research: Low Cost to Write, High Cost to Read
AI-Augmented Research: Free to Write, Costly to Read
The paper builds a model to study the impact of AI reducing writing costs faster than reading costs, finding that as writing costs fall, the range of verifiable papers shrinks: papers by external authors are no longer read, and papers by well-known authors are no longer checked. In the extreme state of zero-cost writing, the proportion of verifiable papers approaches zero, and paper impact depends solely on author reputation. Readers can use AI to summarize literature, but when the proportion of papers written using the same AI exceeds a threshold, AI-generated summaries inherit that AI's errors, and AI reviewers filter out original research that could correct these errors.
Source:arXiv q-fin · arxiv.org