Search across guide titles, topics, keywords, and content throughout the Researcher's Field Guide.
Showing 26–50 of 65 results
AI model versions should be documented as research dependencies, much like software versions. However, model names, snapshots, and hosted services introduce additional complications that researchers need to understand.
Read Guide →Generative AI can help researchers without programming experience write scripts, process data, and understand computational workflows. However, producing executable code is not the same as knowing whether it is correct. Learn how to use AI programming assistance while developing the knowledge needed to verify research results.
Read Guide →AI-generated code can reduce the effort required to write research scripts, but it cannot replace the understanding needed to evaluate their correctness. Learn why programming literacy remains important, which skills researchers should develop, and how to use AI as a learning aid rather than an unquestioned substitute.
Read Guide →Generative AI can explain unfamiliar research scripts, describe functions, and help researchers understand data transformations. However, AI explanations may misrepresent actual code behavior. Learn how to verify explanations through testing, documentation, and careful examination of computational outputs.
Read Guide →Generative AI can help identify programming errors, explain error messages, and suggest corrections to research scripts. However, a fix that makes code run may still alter the analysis. Learn how to test AI-generated debugging suggestions and preserve the validity of research results.
Read Guide →Generative AI can help restructure research code for readability and maintainability, but refactoring may unintentionally change calculations, data handling, or analytical results. Learn how to verify behavioral equivalence and preserve scientific reproducibility.
Read Guide →AI-generated research code can contain insecure dependencies, unsafe file operations, weak authentication, and vulnerabilities that expose confidential data or research systems. Learn how researchers can identify security risks, test generated scripts safely, and apply appropriate safeguards.
Read Guide →Generative AI can accidentally expose API keys, passwords, and access tokens through generated scripts, debugging logs, or shared code. Learn how credential leaks occur, why deleting an exposed key may not be enough, and how researchers can protect access to research databases and institutional systems.
Read Guide →Running AI-generated code without understanding its behavior can expose research data, modify files, or produce misleading results. Learn when controlled experimentation may be appropriate, which operations require careful review, and how researchers can verify code before relying on it.
Read Guide →Researchers do not necessarily need advanced programming expertise to use AI-generated code, but they must understand enough to evaluate its behavior and results. Learn which programming skills matter, how requirements differ by task complexity, and when specialist assistance becomes necessary.
Read Guide →AI-generated research code can be reproducible, but generating working code is only the beginning. Researchers must preserve the code, computational environment, dependencies, data-processing decisions, and verification evidence needed for others to reproduce the analysis.
Read Guide →Not every AI-generated suggestion needs its own commit, but changes that affect research code should be traceable. Learn how researchers can use version control to preserve analytical decisions, investigate errors, and reproduce published results.
Read Guide →Saving AI prompts can help explain how research code was developed, but prompt records are not always necessary for computational reproducibility. Learn which interactions deserve preservation, what information to record, and how to protect sensitive research material.
Read Guide →Identifying the AI model used to generate research code can improve transparency, but a model name alone is not enough to reproduce an analysis. Learn which model details researchers should preserve, when precise version information matters, and how to connect AI assistance to verified research code.
Read Guide →Research code often combines human-written functions, AI-generated suggestions, and subsequent revisions. Learn how to document these contributions accurately without relying on unreliable AI-code detection or confusing code provenance with scientific responsibility.
Read Guide →If AI-generated research code has been thoroughly verified, does its origin still matter? Understand why code correctness should be assessed independently of authorship, while provenance, accountability, and disclosure remain relevant.
Read Guide →AI-generated research code should generally have unit tests when its functions affect data processing, calculations, or analytical results. Learn how to design meaningful tests, establish independent expected outcomes, and recognize the limits of passing test suites.
Read Guide →Generative AI can write useful tests for its own code, but passing those tests does not establish that the implementation is correct. Learn how shared assumptions, inadequate test coverage, and circular verification can undermine AI-generated tests.
Read Guide →Researchers should define the expected behavior of AI-generated code rather than rely entirely on AI-proposed tests. Learn how independent test cases connect computational verification to research methods and help detect plausible but consequential coding errors.
Read Guide →Generative AI can help draft software environment files for reproducible research, but it cannot reliably infer every dependency or version from source code alone. Learn how to generate, verify, and test environment specifications that support the reconstruction of research analyses.
Read Guide →Generative AI can help researchers identify, organize, and explain software dependencies, but source-code inspection alone cannot reveal every requirement. Learn how to create accurate dependency documentation that supports maintenance, collaboration, and computational reproducibility.
Read Guide →Saving an AI-generated script is not enough to ensure that others can reproduce a research analysis years later. Learn which code, data, software environments, verification records, and AI-assistance documentation researchers should preserve, and how to create an identifiable archival release.
Read Guide →Generative AI can write code for scientific figures using Python, R, and other tools, but a chart that renders successfully may still contain analytical errors. Learn how to verify data mappings, statistical calculations, scales, and reproducibility before using AI-generated visualizations in research.
Read Guide →Generative AI may provide less appropriate research advice for underrepresented countries or populations when its recommendations rely on dominant scholarly contexts, incomplete knowledge of local institutions, or unrealistic assumptions about research resources. Researchers should evaluate advice against local requirements, methodological evidence, and the practical conditions of their study.
Read Guide →Generative AI may recommend research designs that require unavailable datasets, expensive software, extensive funding, or institutional infrastructure. Researchers should evaluate methodological appropriateness and feasibility separately, adapting research questions and procedures without making unsupported compromises in scientific rigor.
Read Guide →