16/07/2026
Today, I asked Chatgpt to create an anatomically accurate picture of the diaphragm - out of pure curiosity. The things I learned from that were quite interesting.
The biggest (but not only) issue I see is the central tendon - almost invisible in the first image from below and still inaccurate on both "corrected" images - see the last one for reference that I took from the Thieme Atlas of Anatomy (Schuenke et al., 2006).
As I complained to Chatrick (yes, my Chatgpt has a name...) that it seems impossible to get an accurate picture from it, it said something valuable:
"Even AI can't draw the diaphragm correctly.
It got close but kept making the same mistake: the central tendon was too circular instead of its characteristic trefoil (heart-like) shape extending posteriorly.
Why?
Because AI learns from us.
If most textbooks, illustrations, animations, and online diagrams simplify or misrepresent a structure, AI learns those simplifications too.
It doesn't understand anatomy. It recognizes patterns.
And honestly... humans aren't that different.
Many of our beliefs about singing come from diagrams we've seen hundreds of times rather than from anatomy itself.
This is exactly why we should remain curious—even about ideas we've believed for years.
Evidence beats repetition."
Honestly would, you have believed that first image to be accurate? Years ago, I definitely would have. Today, I spent a considerable amount of time trying to make AI reconsider its patterns, but failed miserably because the majority of information available to it is inaccurate. I believe that AI can be invaluable for computing and organizing data, that would take us weeks, months, years to get organized - but I am worried that the quality of that very data is not always sufficient to give us something credible - yet. And it will only get better if WE think critically, ask very specific questions and don't believe everything we see...
What are your thoughts about this?
