The Journal of Artificial Knowledge (JOAK) explores how AI can contribute to producing, testing, and communicating scholarly knowledge. Its articles make methods, evidence, AI contributions, and editorial decisions inspectable. This first issue contains one article, “What Must a Model Preserve? Synthetic Tests Toward Evaluating Artificial Knowledge.”
The article asks a question that arises whenever a model is used to support a scientific claim: what must the model preserve for the claim to be warranted? Starting from Rosen's modelling relation, it distinguishes predictive, structural, and purpose-relative adequacy. Synthetic studies show how a representation can produce a correct or incorrect prediction, hide consequential differences among possible transitions, or preserve a final state while losing information about the path to it. These bounded results do not demonstrate that an AI system produces knowledge. They establish explicit questions and failure conditions for further experiments.
JOAK itself is also an experiment in scholarly work. The article was developed with substantive AI assistance in formalization, implementation, analysis, writing, and criticism. The Sponsor set the research direction, approved the material decisions, and accepts responsibility for the article. The article's AI-assistance disclosure describes the contributions and their limits in detail.
The article underwent AI-based scholarly review coordinated through AIMan, an experimental system initiated by the Sponsor to organize AI contributors and preserve inspectable records of their work. The Sponsor, acting as Founding Editor, made the final acceptance decision. The Review history page gives a concise account of the review, revisions, and decision for this article.
The accompanying computational supplement is intended to let a reader rerun the reported synthetic calculations and compare the outputs with the article. It does not reproduce the historical reasoning of AI systems or the human editorial decisions. The article and supplement state those limits explicitly.
Future issues may test the three adequacy criteria through new synthetic studies. Their questions, methods, results, review outcomes, and publication decisions remain open. This issue offers a starting set of distinctions that subsequent work can challenge, refine, or reject.