Truth, Medicine, and AI
Source transcript: Truth, Medicine, and AI - Raw Transcript
When I was a kid, I remember looking at pictures on the internet and sometimes having this weird feeling that I wanted to be part of the picture.
I would see a mountain. Then I would zoom in on the mountain. I’d keep zooming closer and closer.
And somewhere in that process there was almost this instinctive question: when do I get to be on the mountain?
Of course, i knew that you never do when it’s just a picture. You’re not there. No amount of zooming changes that. You chasing down a mirage. And as you zoom the image just gets blurrier.
I think there is something similar going on in medicine.
Sometimes medicine is pretty well defined. Someone comes in with a bump. It is growing. You image it. It turns out to be a cancer. There is a whole field of medicine, oncology, built around what happens next.
But especially in primary care, a lot of things do not work that way.
Someone comes in with a small painful bump. I can feel it. Maybe I get an ultrasound. The ultrasound says there is nothing there. That is strange, because I can still feel the bump. So maybe we get an MRI. The MRI does not really show anything either. The patient and i are both confused.
At this point, I can probably say the bump does not appear to be cancer. There is no obvious dangerous process going on.
But what is it? I don’t know.
I can feel it. The patient can feel it. The imaging does not see much. There is clearly something happening, but I cannot necessarily tell you exactly what that something is.
Of course, you could keep going.
Eventually you could cut something out and look at it under a microscope. Tissue is probably about as close as medicine gets to ultimate truth.
But I do not deal with tissue very often in primary care.
I mostly deal with the abstract. Someone else gets the tissue.
For me, there is a limit.
Pain is another example.
Someone comes in and says their hand hurts. Or their shoulder. Or their back.
You start looking for the diagnosis. Maybe you check labs for inflammation, normal. You get an X ray. The bones look normal.
You can get an MRI, but now you are looking at tendons, muscles, small structural abnormalities. And even if you find something, the likelihood that it produces a meaningful intervention may be pretty low if the pain is not too significant.
This is one of the strange things about medicine.
First, you can find an abnormality and still not have found the answer. You can show someone an MRI and point to the thing that looks different. But just because it is there does not mean it is why they hurt.
It can be hard to explain that we found something but it does not mean we found the cause of the pain, we just found a new thing to look into. Eventually you reach this point where you have to decide whether to keep going.
Patients are people, and people generally want an ultimate truth.
I want it too.
There is something deeply satisfying about taking a symptom and following it all the way down until eventually you can say: this is what it is.
But sometimes medicine does not give you that.
There are basically two reactions.
One is to say, okay. We looked. We know enough. We ruled out the things we were worried about. We may never know exactly why this hurts or what this bump is, and that is okay.
The other is to keep going.
Another scan. Another specialist. Another test. Another way of looking at the same thing.
Part of my job is figuring out when that search is still useful and when we are just zooming further into the picture.
Because sometimes you do find something important.
But sometimes the picture just gets blurrier.
Interestingly, I think AI has a similar feeling to it.
Before AI, if you wanted to understand something, you could read more. Search more. Ask more people. Follow references. Keep digging.
Now you can feel like you are doing that almost instantly.
You can ask an AI a question, then ask a follow up, then another follow up, then ask it to explain the explanation.
You can mentally zoom in forever.
And there is a strange feeling that if you just keep asking, eventually you will arrive at the truth.
But I am not sure that is how it works.
At some point, the intelligence starts behaving like the picture.
You keep zooming and the image starts to blur.
The words are still there. The sentences still sound intelligent. There is still an answer.
But there may not be any more truth underneath it.
In fact, one of the most important signs of intelligence may be knowing when to say, “I don’t know.”
That sounds simple, but it is actually a pretty sophisticated thing.
“I don’t know” means you know something about the boundary of your own knowledge.
You have reached the wall.
That is different from continuing to generate an answer because an answer is expected.
There is a lot of discussion about artificial general intelligence and the possibility that some future system will simply keep working on a question until it solves it.
And in some areas, I think that will probably work.
There are mysteries that are mysteries because we have not had enough intelligence, enough computation, enough data, or enough time.
AI may unlock these.
But I suspect there are other places where we will eventually discover the same thing medicine discovers every day.
Sometimes you have reached the edge of what can be known from the information you have.
Sometimes the question itself is wrong.
Sometimes the thing you found is real but not causal.
Sometimes more searching does not produce more truth. It just produces more resolution around our uncertainty.
Maybe that is the harder form of truth.
Not the answer, but knowing when you have reached the pixelated mixture at the end of intelligence and admitting that you cannot zoom yourself inside.