Augmenting Research with AI - jonathancolmer/lab-guide GitHub Wiki

Augmenting Research with AI

Artificial intelligence is changing how research is done. The goal is not to replace human thinking, but to work with AI to improve the quality, creativity, and efficiency of our work.

What is AI?

1. Rules-Based AI

  • Follows explicitly programmed instructions and “if–then” logic crafted by experts.
  • Hardcoded, fast, reliable, transparent — but rigid and unable to adapt beyond predefined rules.

2. Predictive AI

  • Uses statistical and machine learning methods to analyze historical data, detect patterns, and forecast outcomes.
  • Trained on labeled datasets to learn relationships between inputs and outputs.
  • Common in risk assessment, demand forecasting, and anomaly detection.

3. Generative AI

  • What most people mean when they say “AI” today.
  • Uses complex neural networks to learn patterns from massive datasets and generate new content (text, code, images, etc.).
  • Powerful for drafting, summarizing, brainstorming — but can hallucinate, miss nuance, or overconfidently be wrong.

A Philosophy for Working with AI

Adopt an augmentation mindset — AI is a collaborator, not a shortcut.

  • Treat AI like a junior RA
    It can scaffold plans, draft text, debug code, or suggest ideas. You guide its work, check its reasoning, and integrate the results into your own thinking.

  • Don’t outsource judgment
    You remain responsible for conclusions, accuracy, and ethics. Always critically evaluate AI’s suggestions.

  • Collaborate
    Ask follow-ups, refine prompts, and push for reasoning. Use AI when it can outperform the best human available for a sub-task.

  • Prompting is thinking
    Clear, specific prompts force you to clarify your goals — improving both AI output and your own understanding.

  • Question everything
    Verify facts, test code, and probe assumptions before using AI output.

  • Document the process
    Keep a record of prompts, iterations, and manual edits for transparency and reproducibility.

  • Train it, then check it
    Provide examples and feedback to guide AI behavior, but always validate results.

AI Reckoning + Human Wisdom

AI Reckoning Human Wisdom
Pattern detection Practical judgment
Calculative prediction Storytelling & narrative
Summarization & synthesis Ethical reasoning
Automating routine tasks Framing complex questions
Deep contextual understanding

The whole should be greater than the sum of the parts.

Best Practices

  1. Think in iterations, not essays – Start small and build up gradually.
  2. Use short, specific prompts – Be direct and clear; ask AI to explain steps.
  3. Use AI to think, not just write – Brainstorm, clarify, iterate, explore.
  4. Treat AI as a collaborator/thought partner – Converse with AI rather than just requesting outputs. Ask follow-ups, give feedback. This back-and-forth can mimic dialogue with a colleague, helping to refine ideas and surface counter-points.
  5. Always review and stay critical – Check all outputs before use.

AI-Augmented RA Task Pipeline

  1. Plan the Task

    • Clarify the goal and break it into modular sub-tasks.
    • Confirm the plan with your PI.
    • State context and objectives clearly in the first prompt.

  2. Generate Initial AI Output

    • Use AI for scaffolding: outlines, code skeletons, analysis plans.
    • Assess alignment with expectations.

  3. Execute

    • Work with AI to complete sub-tasks (writing, coding, analysis).

  4. Review & Critically Assess

    • Check accuracy, completeness, and relevance.
    • Identify gaps or errors needing deeper expertise.

  5. Iterate & Refine

    • Provide targeted feedback and revised prompts.
    • Repeat until the output meets standards.

  6. Finalize & Document

    • Integrate into your project.
    • Document AI’s role and your edits for transparency.

Example Use Cases

Summarizing Literature

  • Prompt:

    "Summarize the core ideas of this paper in 5 bullet points."

  • Check: Skim the paper to confirm key logic and nuance are captured.
  • Iterate: Ask targeted follow-ups for methods, assumptions, or missing points.

Coding Help

  • Prompts:

    "What does this code do?"
    "Please add comments explaining each step."
    "Given this data structure, write Stata/R/Python code to do X."

  • Best Practice:
    • Define your start and end points before prompting.
    • Never copy-paste blindly — read, verify, and test.
    • Ask for refinements (e.g., optimization, error handling).

Risks & Pitfalls

  • Overreliance: Stifles your own critical thinking. Many breakthrough come from wrestling with mess problems. If we use AI to smooth all the friction, we risk losing the discomfort that sparks insight.
  • Laziness: Leads to shallow understanding and a “glass ceiling” in skills. The harder, slower steps are often where expertise is built.
  • Vague prompts: Produce vague, unusable results.
  • Losing the plot: Faster output is not the goal — better thinking is. Deep thinking is no longer the default, it's a deliberate choice. Don't lose sight of that.

Ethical & Responsible Use

  • Always verify facts and outputs. AI can sound confident while being very wrong.
  • Be aware of bias in training data.
  • Respect privacy and intellectual property.
  • Keep AI contributions transparent. Colleagues should know which ideas were yours and which were AI-assisted.
  • Share use-cases and lessons learned so others can build on what works and avoid dead ends.
  • Remember, AI can shape how you think, not just what you produce. Be intentional about how you use it.

Core Skills Going Forward

AI can reduce the cognitive load of doing hard things — but you still need to build the ability and endurance to do them yourself. Become comfortable with discomfort. Thinking deeply is critical to research success, and it’s also demanding.

Treat "deep work" as a skill to be trained. Use a "time under tension" approach: deliberately spend sustained periods wrestling with complex ideas, arguments, and problems without rushing to a shortcut. This preserves the necessary friction which leads to learning. Over time, you’ll increase your capacity to focus, reason, and connect ideas — skills that AI can support but never replace.

Continue to develop your ability to:

  • Read critically
  • Write clearly
  • Structure arguments
  • Learn new tools and concepts
  • Focus and persist on challenging tasks
  • Connect ideas across domains
  • Relate to and collaborate with others
  • Stay curious and explore
  • Use AI as a thinking partner -- knowing when to work with it and when to work independently