VUMC statement on the use of generative AIJun. 27, 2023, 8:29 AM
The recent public release of generative artificial intelligence (AI) tools that are capable of creating new, realistic content from training data, like OpenAI’s latest multimodal large language model GPT-4, has generated a lot of excitement.
While still experimental, these tools have the potential to change the way we do many of our daily work tasks and might be able to improve clinical care, operational efficiency, and advance research. While enthusiasm for these tools is understandable, they also have limitations, and use of publicly available versions of these tools can pose risks to organizations like ours.
VUMC policies forbid the sharing of VUMC data with third parties, including companies that provide access to public-facing generative AI tools like OpenAI’s ChatGPT, Google’s BARD.
No VUMC-derived data of any kind should be shared (e.g. manually entered, via APIs, etc.) with such public-facing AI tools, including, but not limited to:
- VUMC patient or clinical data of any kind, including de-identified patient data
- VUMC research data
- Proprietary VUMC operational data
- Intellectual property of any kind owned by the institution or its employees
If any doubt, no VUMC-derived data should be shared with such AI tools.
In an abundance of caution, given the concerns about using publicly accessible generative AI tools, effective immediately, VUMC personnel are no longer able to access or use such tools from our networks. Further, regardless of what network an employee is using to access such tools and in accordance with VUMC policies, VUMC employees are not to share any VUMC-derived data with any such tools.
Recognizing that these tools can also be beneficial, a multi-disciplinary team of Informatics, IT, operational and research professionals at VUMC have been working on alternatives to publicly accessible tools. Working with Microsoft, we have established private, institutionally managed instances of tools like GPT-4 that are secure and governed by appropriate legal and business agreements. While testing is ongoing for more advanced capabilities, a simple, secure interface to ChatGPT for VUMC personnel is available for use starting today.
The Department of Biomedical Informatics has created a new landing page with information about Generative Artificial Intelligence (AI) tools and capabilities at VUMC, and related information for the VUMC community. As these tools and capabilities change, this page will be updated, so check back often.