← Back to the overview

Critical Evaluation of AI Experimental Predictions

Develop students' ability to assess the scientific reasoning, mechanistic logic, and plausibility underlying experimental predictions.

Disciplinary ExpertiseLevel 2Complexity: AdvancedDuration: 30-60 min

Recommended for: Upper-level research-focused courses

Procedure

  1. Students design a specific experimental scenario or receive one from the instructor (e.g., "Predict the phenotypic outcome of treating cultured cells with rapamycin while simultaneously inhibiting mitophagy" or "Predict the behavior of a pendulum system with non-linear damping")
  2. Students request that an AI system predict the experimental outcome and provide a detailed explanation of the underlying mechanisms
  3. Students systematically evaluate the AI's prediction and reasoning:
    • Plausibility: Is the predicted outcome consistent with established principles, laws, or empirical patterns in the field?
    • Mechanistic logic: Is the causal explanation mechanistically sound and free from logical inconsistencies?
    • Completeness: Are all relevant mechanisms, pathways, or factors considered?
    • Overlooked factors: What has the AI missed, oversimplified, or failed to integrate?
    • Literature support: Can the prediction be verified or refuted using published empirical research?
    • Quantitative reasoning: Where applicable, are any quantitative predictions mathematically sound?
  4. Students propose alternative experimental outcomes with mechanistic justifications
  5. Students identify what additional information, controls, or measurements would be needed for a reliable prediction

Extension options

  • Students conduct systematic literature searches to empirically verify or refute AI predictions
  • Compare predictions for identical scenarios across different AI systems
  • When feasible, compare predictions to actual experimental results from published literature or laboratory work

Discussion questions

  • What factors made certain predictions more or less scientifically plausible?
  • How did the AI handle experimental complexity, uncertainty, and the integration of multiple mechanisms?
  • What depth of domain expertise was required to adequately evaluate the predictions and reasoning?
  • How should researchers appropriately integrate AI predictions into experimental planning and hypothesis development?

Assessment opportunity

Students produce a scientific evaluation of the AI prediction, including mechanistic critique, alternative hypotheses, and a proposed experimental approach to test competing predictions.