Welcome — use AI without losing your science

Many research teams want the speed and insight AI can provide, but worry about opaque models, unreproducible pipelines, and accidental harm. This resource helps you decide where AI genuinely adds value for your research, validate models so results can be trusted and reproduced, document decisions for collaborators and reviewers, and embed models into reproducible workflows that survive staff changes and model drift.

If you only do three things today

1. Run the AI Use Risk Assessment (this resource) to triage potential harms and required validation rigor.
2. Fill the Model Validation Checklist for any model you plan to publish or use in decision-making—capture answers so you can reproduce later.
3. Write a short Model Card that documents purpose, data, performance, and limitations before sharing results.

Who this helps

Individual researchers, lab leads, small research teams, translational groups, and research-support staff who must balance speed with rigour. Examples include a graduate student running an ML analysis, a lab team piloting image analysis automation, and a hospital research unit exploring diagnostic decision aids.

How to use this resource

• Start with the risk assessment to set your validation bar.
• Use the interactive checklist to collect reproducible evidence for that bar.
• Follow the Model Card guide to communicate limitations before you publish or deploy.

Every item stands alone, so you can jump to the checklist or model card template as needed. If you have an existing reproducibility audit, use that report to complete items in the checklist rather than repeating work.