Boom. Timpani-sound. “enshittification” is now a word. Those that have been around computers and the internet for a while, are increasingly complaining about decreasing performance, no longer meeting expectations etc. Google the term, and you find a number of videos on the subject (including the degrading experience of Google and YouTube itself).
Are we there already for AI as well? Surely not. It was only last year (in 2023) that we saw memes like this:
But this is 2024. And for an upcoming presentation, I wanted an original image to illustrate the concept of digital pathology. So I went to Stable Diffusion, and I entered my prompt: “microscope with a computer to illustrate digital pathology”
This is the resulting image I got:
Don’t scroll down yet! Can you tell what’s wrong with this?
Ok, now scroll down…
There are probably finer points to comment on, as well. I’m not a seasoned microscope technician.
Are you laughing yet? Shaking your head? Thinking “oh, the humanity”?…
Maybe the computer-reference confused the rendering engine. So I decided to enter a different prompt: “a microscope in a pathology lab”:
Nope, still not quite there yet:
Maybe you’re thinking “oh, but that’s a scientific piece of equipment”…
It’s a pretty standard piece of equipment, isn’t it? And let science (and radiology, and pathology…) be exactly a highly active area of research for AI!
My son is 9 years old, and he uses microscopes occasionally at school already. The microscope was invented by Antoni Van Leeuwenhoek around 1670. That’s 350 years ago (if you care about what its evolution looked like, the Boerhaave museum in Leiden (NL) has an excellent exhibit on it). That predates trains, cars, and… maybe not boats… we’ve had boats for a while now. My son also uses modern modes of transportation, like cars, planes, and the occasional boat.
Just saying: a microscope is not an particularly esoteric piece of equipment. I didn’t ask for anything special either, like confocal, or atomic-force. It’s not like I asked for self-sealing stembolts.
And yet, a simple run-of-the-mill microscope is not what I got.
So here’s a provocative question: why should I hand over medical problems to AI models, engines, foundation models, LLMs, LMMs,… if they’re not able to reproduce simple stuff? And why are we so convinced that AI is going to change everything? Is AI going to be able to come up with something like Pioneering Spirit?
Why should I hand over medical problems to AI if it’s not able to handle even simple tasks?
Oh, and about those stembolts; I did ask Stable Diffusion to generate them for me. Here’s what I got. See if you can make out which ones are real and which ones are generated.
Let’s just say we’re not quite there yet with all this AI stuff.
🖖