Gary Marcus

And part of reliably anticipating instances of unreliability is being able to understand the thought processes behind deep learning algorithms. We can’t do that. But we can usually explain why a lab test would break down under certain circumstances and therefore anticipate that.

@MaxALittle @DrHughHarvey @DrLukeOR developing thread on deep learning in medicine, feel free to pipe in...

I've actually got what I think is a very relevant blog post coming out in the next few days. Preview of core idea: we don't need to reboot AI to make medical applications much safer than they are today, we need to use human smarts to much more effectively validate these models.

Wish we’d stop using “AI” to refer to human-level AI, which @GaryMarcus talks about, & machine learning. Former is fascinating (enjoying the book!) but the latter in its CURRENT state could be hugely helpful in medicine (& many other areas). Analogous to advent of spreadsheets.

to be clear, this whole thread is largely about how and whether current (marrow) AI can help, and some challenges in applying even current techniques in practice. i am long term optimistic, short term pessimistic.

Why short term pessimistic? Focusing on only medical or diagnostic bits of healthcare? Much of hc is like any other biz, with supply chain, scheduling, comms— and many of these processes are a mess (scheduling, esp). Those processes are ripe for improvement with workaday AI.