Assembled Chaos

Narrowing the translational gap with highly interactive consortia

AI for translational research: Video

 

How should we apply AI in translational research?
The other day I was talking with James Schofield, one of the founders of TopMD a startup that I advise which is using AI to identify pathway biomarkers.

Having an interest in sorting out how to apply AI in translational research, I asked Jim what the difference was between supervised and unsupervised learning:

"You use unsupervised learning when you don't know what you are looking for."

I then remembered reading in Eric Topol, MD's Nature Medicine review on AI that unsupervised learning is rarely applied.

There is a reluctance to use 'black box' approaches where you can't easily describe how an AI. is making its discrimination.

In 'supervised learning' labels guide an AI. However, having to apply labels limits the ability to move beyond lumping of patients into arbitrary disease categories.

Should we be bold and do the complex and uncertain work of figuring out how to make use of unsupervised learning?