AI Policy and Resources
The SAGE Lab encourages thoughtful use of AI for research and learning. Use it to improve the work and develop your understanding. You remain responsible for what you submit, share, or publish.
Raise the standard
Our aim is higher standards at the same level of effort, rather than the same standards with less effort. Reinvest time saved in learning, checking sources, testing analyses, and considering alternative explanations. Verification is part of the research effort. This means making better use of a sustainable workload, not extending working hours.
Use AI with purpose
AI can help explain methods, explore ideas, write and debug code, improve prose, and critique a study or manuscript. Choose tools that fit the task and discuss substantial uses with your supervisor or collaborators. Agree on access and costs before committing to paid tools; do not assume a personal subscription is required.
Build understanding you can use
Use AI to learn concepts and methods that would otherwise be too difficult or time-consuming to approach. Ask for worked examples, competing explanations, practice problems, and feedback on your reasoning. Develop your foundations in statistics, research design, and critical thinking so you can judge whether an AI’s choices and conclusions are justified.
The goal is a virtuous cycle: AI supports deeper learning; deeper understanding improves your ability to question and check AI; better checks strengthen both your research and your understanding.
Can you explain why a method fits the question, identify its assumptions, recognize a misleading result, and check it against data or original sources? If not, use that gap to guide further learning and supervision. For training tasks, agree on what to attempt independently and what assistance is appropriate. Read the original papers you rely on, and bring uncertainties and errors to supervision early.
Verify before sharing
- Sources: open the original reference and check that it supports the claim. Never cite a source solely because an AI supplied it.
- Analyses: run the code, inspect the data transformations, and check assumptions and results against the research question.
- Writing: check facts, numbers, quotations, and the strength of conclusions. You should be able to explain and defend the work.
- AI critiques: assess each suggestion against the evidence. A confident review—or agreement among several models—is not independent confirmation.
Before circulating a manuscript, consider an AI-assisted critique or reproducibility check when permissions and confidentiality allow. Use the findings to guide checks and discussion with coauthors.
Tools to explore at this step:
- Stanford Agentic Reviewer (PaperReview.ai): manuscript feedback informed by related work on arXiv. Its documentation notes that coverage may be weaker outside arXiv-heavy fields; it currently supports English papers.
- statcheck: a complementary non-AI tool that recalculates p-values from supported statistical tests reported in APA style. It checks reporting consistency, not whether the analysis or design is appropriate.
A manuscript review does not establish computational reproducibility: rerun the analysis code against the appropriate data and compare outputs with the reported results. Check permissions, data handling, and costs before uploading material to any service; these links are options to evaluate, not required purchases or guarantees of accuracy.
For examples of AI-assisted research reports, explore The File Drawer, an archive of studies and explorations with provenance shown per paper. Treat its reports as starting points for examining the underlying work, rather than established findings or a manuscript-review service.
Keep code under version control, test reusable functions, and maintain clear documentation. AI-generated code needs the same review as other research code.
Protect data and confidential work
Do not put identifiable participant data, confidential records, credentials, or material received under peer-review confidentiality into an AI service without explicit authorization and suitable safeguards. Removing names alone may not make data anonymous. Check institutional requirements, ethics approvals, consent, agreements, and the tool’s data handling before uploading research material. When uncertain, ask before sharing; use public or synthetic examples while resolving the question.
Make substantial use traceable
Keep enough information to understand and reproduce AI-assisted research: the tool and model, date or version, relevant prompts and settings, code, outputs, and human checks. Record consequential decisions. For AI used as a research method, document the procedure in the methods and preserve appropriate materials.
Discuss substantial assistance with coauthors and disclose it as required by the journal, funder, institution, or course. Human authors retain responsibility; AI tools do not receive authorship. Course and peer-review rules may differ from everyday lab practice.
Respect readers’ time
When sharing AI-generated suggestions or critiques, label them clearly. Read and edit material before circulating it; send colleagues a concise account of what matters and what you need from them.
Resources for discussion
- King & Worthington (2026): AI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices. Research on AI-generated reviews and journal policy; its findings should be interpreted within the study’s scope.
- Michael Frank: My lab’s AI policies and norms. Practical guidance on AI review, disclosure, data protection, code quality, and learning. Its manuscript-writing restrictions are specific to that lab.
These practices will evolve as tools, evidence, and institutional requirements change.