I have an search plug-in for Photoshop (Excire) that supposedly uses machine learning (AI) to analyze images and figure out what's in them. It can be quite helpful because I haven't taken time to add keywords to each image. To date, my image catalog contains almost 25,000 images.
I wanted to find an image of a sandwich, even though I wasn't sure if I have ever photographed one. The closest search term available was food, for which I got many, many results. Most were food, but this image definitely isn't.
So what is it? During the first year of the pandemic I undertook many homey tasks--cooking, gardening, and so on--to pass the time. One day I decided to make some fire starter for my wood stove to replace the turpentine wood sticks I had been using. I took pellets from my barbecue, placed them in a used egg carton, and doused them with hot wax. Once solidified, I could simply light the egg carton, which would catch the wax, which would light the pellest. Voila! Long acting fire starter. But definitely not anything I would eat.
This wrong categorization by the "smart" search engine is making me consider to enter a few keywords with all new images. Then the "smart" search can work hand-in-hand with my real intelligence. It has been shown that in the medical world, results are better when AI teams up with real physicians. The AI identifies some of the cases missed by the physician, and vice versa.