Skip to content

Responsible use of AI

Adapted from fastplotlib/claude-skills: Scope, Responsibility & Scientific Integrity.

Review over generation

AI tools are cheap at producing code and expensive to review. Generated code is a draft: the value of a contribution lies in the review and validation behind it. Prefer small, focused changes you can read in full over large generated ones, and do not submit code you cannot explain.

Scope

Use AI for implementation, documentation and routine coding whose expected behaviour you can check. It cannot decide whether a model, analysis or plot answers the scientific question. Without a meaningful correctness check, seek domain expertise.

Responsibility

You are responsible for every change you submit. Review it, verify assumptions and run the tests. Fluent output can hide invented APIs, changed requirements or unsupported claims.

Scientific integrity

Code that runs and plots that look right can still misrepresent an experiment. Check against independent expectations:

  • array shapes, physical units, sampling rates and time alignment;
  • neuron identifiers across imported data, stimulus arrays and recordings; stimulation amplitudes and durations;
  • axis labels, colour scales and display limits;
  • simulation and analysis output against known cases, ideally as a scientific validation test.

Record the assumptions and validation evidence. If you cannot explain how the result was obtained and verify that it is correct, get expert review before drawing a scientific conclusion.