The Urgency for Institutional Support in Adopting Transformative Technologies
The scale and speed of the generative AI (artificial intelligence) revolution has presented unprecedented opportunities to advance science, as well as significant challenges to the traditional academic research model. Exciting new AI capabilities are leading to a rapid adoption by researchers, many of whom lack the necessary technical expertise and resources to implement these technologies responsibly. This problem extends beyond generative AI and could apply to other disruptive technologies that emerge in the future.
The current norm of relying on individual researchers to adopt new technologies is no longer adequate. Academic institutions and their research organizations must play a much more active role in enabling researchers to develop new skills and implement new research methods in ethical, responsible, and effective ways. This is essential for helping academic researchers stay at the forefront of research and discovery, while preserving the validity and trustworthiness of science.
Opportunities and Concerns with Generative AI in Academic Research
Generative AI has the potential to revolutionize academic research in multiple ways:
- Improving Productivity: Generative tools can assist with routine tasks such as drafting and editing documents, checking for compliance, and communicating with diverse audiences.
- Enhancing Research Expertise: Generative AI can help summarize and represent knowledge, gather interdisciplinary insights, and support collaboration.
- Accelerating the Research Process: Generative models can automate tasks like data cleaning, formatting, and analysis, as well as suggest hypotheses and experimental parameters.
- Enabling New Research Paths: Domain-specific generative models can open up new avenues for research discovery in fields like aerospace engineering or biology.
However, this enthusiasm for generative AI in research is accompanied by a lack of preparedness among many researchers. A survey conducted by the Michigan Institute for Data Science (MIDAS) found that:
- Only 12% of respondents have the expertise to train their own generative AI models.
- Fewer than one-third can run existing models or fine-tune them.
- Even after the release of ChatGPT, half of the respondents are still not able to use prompts effectively.
This unpreparedness is accompanied by various concerns that researchers have, including:
- Data Privacy and Confidentiality: Generative AI models may inadvertently reveal sensitive information from their training data.
- Biases and Hallucination: Generative AI models can inherit biases from their training data and may produce inaccurate or fabricated information.
- Lack of Transparency: The opaque nature of generative AI models and their training data makes it challenging to assess their appropriateness for specific research uses.
- Rigor and Reproducibility: The use of generative AI in the research workflow poses fundamental challenges to ensuring the validity and reproducibility of research outcomes.
The Need for Institutional Transformation
The typical academic research model, where individual researchers are expected to learn and adopt new technologies on their own, is no longer sufficient to address the challenges posed by rapidly advancing technologies like generative AI. Universities and their research institutes must play a much more active role in enabling the responsible adoption of these technologies.
However, universities themselves are often not designed to be nimble in responding to new technology developments. This is where research institutes within universities can play a critical role in driving institutional transformation.
The Expanded Role of Academic Research Institutes
Research institutes at universities can serve as knowledge bases and facilitators for the adoption of new, potentially transformative research methods. Their responsibilities should include:
- Developing and Disseminating Best Practices: Compiling guidelines, processes, and tools to help researchers address issues like data privacy, bias, transparency, and rigor when using generative AI.
- Providing Training and Support: Offering systematic training programs to help faculty and staff researchers develop the necessary skills to use generative AI and other emerging technologies responsibly.
- Coordinating Exploration and Collaboration: Organizing workshops, webinars, and hackathons to expose researchers to successful examples of generative AI implementation and foster cross-disciplinary collaboration.
- Assessing and Validating Models: Developing methods and resources to help researchers choose appropriate generative AI models, evaluate their transparency and bias, and validate their outputs.
- Staying Ahead of the Curve: Continuously monitoring the rapid advancements in generative AI and other transformative technologies, and updating institutional guidelines and support accordingly.
By taking on these expanded responsibilities, research institutes can help bridge the gap between the availability of new technologies and the ability of individual researchers to implement them effectively and responsibly.
The Michigan Institute for Data Science (MIDAS) Leading the Way
The Michigan Institute for Data Science (MIDAS) at the University of Michigan has been at the forefront of these institutional efforts to enable responsible adoption of new research methods.
Developing Best Practices and Guidelines
MIDAS has compiled a set of guidelines to help researchers navigate the use of generative AI in their work, covering topics such as:
- Writing with generative AI (e.g., using it for papers, grants, or literature reviews)
- Improving productivity with generative AI (e.g., reviewing proposals, writing support letters)
- Using generative AI for data generation and analysis
- Reporting the use of generative AI in research
- Considerations for choosing appropriate generative AI models
These guidelines are regularly updated as new policies and regulations emerge from funding agencies, journals, and professional societies.
Providing Training and Support
MIDAS has offered a series of hands-on tutorials to help researchers develop skills in using generative AI for domain-agnostic tasks, such as:
- Enhancing productivity with generative AI (e.g., writing, planning, literature review)
- Integrating generative AI into research workflows (e.g., using image generation)
- Fine-tuning and experimenting with generative AI models for custom research solutions
These training sessions have been well-received, but MIDAS has also recognized the need to start from an even more basic level, helping researchers master the use of tools like ChatGPT before exploring more advanced applications.
Coordinating Exploration and Collaboration
MIDAS has organized workshops and webinar series to expose researchers to successful examples of generative AI implementation in various fields, such as healthcare, chemistry, and social science. These events have also provided a platform for discussing ethical and technical considerations, as well as infrastructure challenges.
Assessing and Validating Models
While MIDAS has made significant progress in supporting researchers, the team recognizes that the next critical step is to develop processes and resources to help researchers choose appropriate generative AI models, assess their transparency and bias, and validate their outputs. This is an area that MIDAS is actively investing in, as the rapid advancements in generative AI make it increasingly challenging for individual researchers to keep up.
Conclusion: The Path Forward for Institutional Support
The emergence of transformative technologies like generative AI is a call for academic institutions to play a much more proactive role in enabling researchers to adopt new methods and tools responsibly. By leveraging the agility of research institutes within universities, institutions can develop and disseminate best practices, provide targeted training and support, coordinate exploration and collaboration, and help researchers assess and validate the use of these technologies.
The work of MIDAS serves as a model for how research institutes can drive institutional transformation and ensure that academic researchers stay at the forefront of research and discovery, while upholding the integrity and trustworthiness of science. As new waves of disruptive technologies continue to emerge, the need for this type of institutional support will only become more critical.
