Radiology & AI
Jenny Melcher writes:
I just attended a lecture about AI in the field of radiology. I was eager for this lecture because as cool as I think radiology is, it is hard for me to believe that the radiology workforce won't be drastically devalued by AI assistance and specifically computer vision.
Because the talk was being given by a radiologist, I was feeling somewhat skeptical, but I tried to go into it with an open mind. However, the talk essentially consisted of examples of tasks that AI fails (not related to radiology). For example, we saw this classic example of AI failing to distinguish between blueberry muffins and dog faces, chat GPT citing fake sources, AI not understanding negation in text, etc.
I thought this was really disappointing because the talk skewed so far towards AI's failings and assurances that radiology is still a good field to go into. Other arguments that the lecturer made included things like - people have been saying AI is going to take over for years, but it still hasn't happened.
When I read about this on my own, it seems like there is a need for more training data and continued oversight of AI in radiology, but that there is potential for AI to take on a lot of the workload in this field.
I'm curious to hear if you have any thoughts about this.
In my response I mentioned a few aphorisms that come to mind:
"Slowly at first, and then all at once."
AI advances in medical imaging have been coming in fits and starts, but the progress is gradual and increasing.
"Things are never as bad as we fear nor as good as we hope."
There's lots of hype and doom around AI, but both are extreme positions that don't reflect the underlying reality properly. The reality is that there are real improvements, but it's also not the end of the world.
Amara's Law: "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."
In some domains, we're approaching the limits of human ability; some of those blueberry muffin / dog photos are confusing even for humans. On the other hand, people have been predicting the end of radiologists for a while and it hasn't materialized in a noticeable way.
In summary, I think Radiology as a field isn't quite done yet, but short of union-style regulations that say that only a human can look at the images, it does seem likely that eventually machines will do that work.
