Introduction

Identify where generative AI may support research.

For each task, identify the AI role, the researcher’s decision, and the required check.

AI in context

How research software has changed.

• Machine learning finds patterns in examples.
• Generative models assemble new text, code, images, and structure from context.
• Check each output against evidence and researcher review.

Model limitations

Generative models produce plausible output that requires review.

They can explore language, organize information, and produce drafts. They cannot determine whether a source, method, permission, or claim is valid for your study.

Three AI task types

Steps in an AI-assisted task.

• Explore options, questions, and tradeoffs.
• Transform an approved input into a clearer artifact.
• Check outputs against sources, calculations, and constraints.

Before each workflow

Name the input, the decision, and the check.

• What information is safe and appropriate to use?
• Which choices still belong to the researcher?
• What evidence will show whether the output is usable?

Researcher responsibility

Researchers remain responsible for consequential decisions.

The researcher remains responsible for questions, permissions, methods, interpretation, authorship, and release. Preserve provenance and keep claims inside what the design can support.

Arcade recording placeholder

Prepared example: document one research task across tools.

Recording pending: research idea, tools used, and saved evidence record.

Setup

Complete the setup and one readiness test.

Use the approved account or no-install option. Confirm that you can start a task, submit a request, and save the result.