Reproducibility is the foundation of scientific trust. It requires publishing the model code, training parameters, and, where possible, the dataset (or a synthetic version). When a result is not reproducible, it remains anecdotal and does not solidify into validated knowledge.

Adhering to the FAIR principles means documenting, sharing, and reusing data responsibly. It involves depositing datasets in public repositories, using open licenses, and applying robust anonymization techniques. This protects privacy and accelerates scientific and social innovation.

AI methods must undergo rigorous evaluation by the community. This requires clearly documenting modeling decisions, metrics used, and the rationale behind the architecture. It allows for review of the algorithmic method, not just the results, and mitigates publication bias.

Ensuring that contributions of data, infrastructure, code, and models are properly cited strengthens the research ecosystem. This involves using persistent identifiers, such as DOIs, for all assets. In this way, the diversity of scientific contributions is recognized and valued.

Making AI tools usable for researchers with varying levels of resources and skills is a matter of equity. Results should be published in open access formats, avoiding paywalls, so that the resulting knowledge benefits society as a whole.

AI research must seek solutions with a positive social impact and contribute to educating the next generation. It requires developing open resources, establishing diverse collaborations, and actively participating in public dialogue, so that policies are evidence-based and serve the common good.