The FDA nutrition label is a paragon of simplicity. Established in 1990, there have been 30 years of standardization and improvements and is now a mandatory part of food labeling.
Similarly, we are in the wild West days of artificial intelligence. With ChatGPT and similar generative AI tools, along with a rapidly growing cohort of predictive AI tools, there is growing public interest, and skepticism about the value of AI, particularly in healthcare.
We face questions from patients as well as healthcare providers:
- Who built this?
- Where was this model trained? On what population of patients?
- How are you accounting for potential bias?
- Who is running this program?
- How are you analyzing performance and working to improve the tool?
These are difficult questions, and even more difficult to explain in an easily digestible way.
Why not Steal Like an Artist and look at the FDA nutrition label as an example?
The FDA label is small, very concise, highly readable and information-dense.
Here is my attempt to recreate this label for AI projects.
I included Purpose, Used by, Used with, Developed by, Monitored by, Risks, Benefits, HIPAA status, Local Fine-tuning status, Bias review status.
I am quite certain this will not be enough, and is also not simple and clear enough. Won’t you help suggest improvements? This needs to be simpler, more readable, more comprehensive, and address both patient and healthcare provider concerns. Easy, right?
Here are more examples based on this model: our generative AI projects we’re working on with partners:
Our predictive AI models we have developed:
CMIO’s take? What is YOUR take? Use make great old ideas new again. Those who don’t know the past are doomed to repeat it. On the other hand, I’m doing it on purpose.
