The faculty-led Generative Artificial Intelligence Advisory Committee compiled the following information to approach the evaluation, use, and development of emergent artificial intelligence tools and services in research.
GenAI will create new opportunities across the entire spectrum of research activities, ranging from the natural sciences, social science, economics and political sciences, humanities, engineering, biomedical and clinical sciences, as well as the creative arts. GenAI will serve as a catalyst, instigating profound transformations across various domains. Concurrently, it will foster the emergence of novel fields of study, while potentially diminishing the prominence of others.
GenAI Potential and Research
GenAI has great potential to assist different kinds of researchers at various stages of their careers. Recently, it has been demonstrated that GPT-4 can act as a data analyst, capable of autonomously handling datasets, developing analytical strategies, cleaning data, running tests, and interpreting results. In addition, it can also conduct visualizations, descriptive analyses, regression analyses, and even write sections of academic papers based on findings, showcasing its potential in revolutionizing academic publishing and data analysis.
In addition, GenAI potentially challenges many current support mechanisms and personnel in research practice (including roles of scientific writers, coding support, library assistance, etc.). In addition, GenAI has also been shown to be prone to hallucination, bias, and sometimes, outright fabrication. Since the technology is constantly in development and in flux, response to these challenges will have to be nimble.
Despite all of this, there is no question that GenAI will create new research opportunities across the entire spectrum, and will transform most fields of research, methods, and training.
Find additional resources for using GenAI for research
Principles Related to using GenAI in Research
Responsible Use
Researchers should utilize GenAI systems in research only where they perform well and exhibit few hallucinations. Researchers should verify all outputs for accuracy and attribution and attest that this has been done in all cases, detailing the methods used to do so.
Documentation
All use of GenAI in research should be fully documented and reported in detail to sponsors for grants, editors for manuscripts and publications, reviews for conferences, and audiences for invited talks and presentations. Of note, sponsors, journals, and professional societies are generating their own best practices and policies in tandem so researchers should be aware to look for additional requirements and that adherence to only U-M policy might not be satisfactory for external entities.
Account for and Limit Bias
Bias in AI systems is understood to be a major problem that should be understood, quantified, described, and mitigated. This is especially true with applying outputs to research involving human participants or using data from such studies.
Privacy Protection
Researchers should not input personal, private, or HIPAA, FERPA, Common Rule, and Export Control protected-information into a unsecure GenAI system. Researchers should follow University of Michigan Guidelines and Policies, which should be immediately updated to address changes brought on by GenAI.