The scientific paper needs an uncertainty layer
Kaeli Rizzo*, Caroline M. Weisman*, Richard Sever, Peter K. Koo
*Equal contribution
Abstract:
AI assistants and scientific agents increasingly shape how biologists navigate a literature no one can keep up with. They can extract results at scale, but judging what those results establish depends on how the evidence was generated, what alternatives remain, and how far the findings generalize. We argue that AI-ready papers should be machine-assessable, not only machine-readable. We propose an uncertainty layer that keeps these qualifications attached as scientific outputs move beyond the paper. By preserving the boundaries of what the evidence can support, this layer could help scientists use literature-scale AI for more reliable scientific judgment.