I found the above browsing about Mastodon (my favorite social media site given Musk's destruction of the old Twitter). I've been too lazy to look for confirmation. Maybe this calls for some philosophical exegesis BEFORE confirmation!
The above is a screenshot of a post on a social media site that says that in 2022 medical researchers tried to make a machine-learning tool for identifying skin cancer. The idea, as always, was that the Algo would learn from large quantities of online photos of skin lesions that turned out to be cancerous and other skin lesions that did not, until the algorithm would teach itself how to tell the difference maybe better than humans would ever be able to.
Didn't work out. The Algo ended up focusing on the presence or absence of a ruler. (Doctors routinely use a ruler to measure a potential tumor for scale.)
The social media post ends, "Basically they tried to build a cancer finding machine and instead built a ruler finding machine."
I find that amusing in a humans-laugh-at-robots sort of way.
Post-script: I have done some checking despite my indolence. The only thing wrong in the post displayed above is the date. The experiment in question, and its amusing result, take us back to 2018. It was written up in the JOURNAL OF INVESTIGATIVE DERMATOLOGY in October of that year.
Comments
Post a Comment