Using AI in Plotting and Analysis

This page provides guidance on how to use AI tools (such as ChatGPT and coding assistants) responsibly and effectively when developing figures, analyzing data, and writing code.

AI is treated here as a support tool, not a replacement for scientific judgment or statistical reasoning.

What AI is good at

AI tools are particularly useful for:

Code assistance

  • generating boilerplate Python/R code
  • explaining unfamiliar functions or syntax
  • converting logic between languages (e.g., R ↔︎ Python)
  • writing helper functions
  • debugging simple errors

Visualization prototyping

  • generating first drafts of plots
  • suggesting plot types for a dataset
  • translating conceptual ideas into code
  • exploring alternative encodings (e.g., color vs shape vs faceting)

Documentation and writing

  • drafting README files or documentation pages
  • improving clarity of plot explanations
  • summarizing code behavior
  • generating examples and templates

What AI is not good at

AI should not be trusted blindly for:

Statistical correctness

  • model selection
  • interpreting results
  • validating assumptions
  • error analysis

Always verify statistical reasoning independently.

Scientific interpretation

AI does not understand your domain.
It cannot judge whether a result is scientifically meaningful.

Data privacy and ethics

Do not upload:

  • confidential data
  • personally identifiable information
  • unpublished results
  • sensitive datasets

When in doubt, assume the data should not be shared.

Good workflow: how to integrate AI into plotting

A recommended workflow:

  1. You define the question.
  2. You clean and structure the data.
  3. You ask AI for plotting ideas or code templates.
  4. You review and adapt the code.
  5. You validate the output visually and statistically.
  6. You refine styling and labeling manually.
  7. You document the final reasoning in your own words.

AI is a drafting assistant, not an author.

Asking better questions

AI works best when you provide:

  • a small reproducible example
  • a clear description of the outcome you want
  • context about your dataset
  • constraints (e.g., publication quality, grayscale safe, accessible colors)

Example:

“I have a dataframe with time, temperature, and TSV.
I want a plot that shows trend and uncertainty bands, in a style suitable for a journal paper.
I am using Python/matplotlib.”

Poor prompts are vague; good prompts are specific.

Validation checklist

Before trusting AI-generated code or plots:

  • Does the chart type match my data?
  • Are axes correctly labeled and scaled?
  • Are units correct?
  • Are categories ordered meaningfully?
  • Are colors accessible?
  • Does the code reproduce the same figure on rerun?
  • Do I understand what every major line of code does?

If the answer is “no” to any of these, treat the output as a draft, not final work.

Transparency and acknowledgment

If AI tools made a substantial contribution to:

  • code generation
  • writing
  • figure design

consider:

  • documenting this in internal project notes
  • acknowledging AI assistance when appropriate (e.g., teaching material, tutorials)

Do not list AI as an author of scientific work.

How this page will evolve

Over time, this page may include:

  • examples of good prompts
  • case studies
  • AI-assisted figure walkthroughs
  • lab policies on AI use
  • reproducibility guidelines

Treat AI as a powerful tool that improves productivity — not as a substitute for expertise.