Hacker News Reader

Comment by Inhibit | Hacker News Reader

Am I missing something or does this look like poor uptake for (possibly) reasons other than performance? The article cites a lack of use as justification for the assumption that the model doesn't work. It might just take time.

Especially in medical. A century (+) ago hand-washing took time to adopt unless I'm mis-remembering.

View on HN

Both that and "people refuse to use technology that makes them redundant" indeed seem to be hinted at. However, one of the quotes says that poor performance is the reason and if you click through to the original article, it seems fairly clear that there is a problem with the existing systems adapting poorly to different setups.

It's not about "poor performance", in the same way as poor IDE performance isn't really the root cause of my laptop's stubborn inability to write good Python code. Radiology is about using a generalist medical education to diagnose (or be instrumental in diagnosing) patients. Pattern matching or statistical information are a rather modest subset of that skillset.

but the article is right in saying that AI advocates said no need for any new radiologists.

I know this in the field of vehicle damages, AI is good but not good enough to take over... and that has been the case for a long while yet every now and then a new company / product comes saying that the future is AI and no need for human effort.

True. I was internalizing "no new need" as hyperbole because it didn't make sense (given the reality of medical) but that's my mistake.

On re-reading, a highly specialized and entrenched workforce having a 30% uptake on a new technology in only a few years seems phenomenal.

Advancements in Medicine happen in spurts because of the regulatory review process and the risk aversion.

The advance must either be VERY significantly better to warrant the approval process, OR an extremely low risk incremental change.

So what we end up with is this sputtering of tiny and big advances.

Reply on news.ycombinator.com