electron micrograph myelin in mouse brain
This result showed that the AI had no idea of scale of microscopy and composition of micrographs. Very pretty, but diverging from every term except mammalian brain.
electron micrograph myelin in mouse brain
This one was comedic.
cytokinetic bridge
This software was definitely programmed with a fantasy images and no knowledge of cell biology. The AI bot has a specific style or vision.
Regardless of this failure, I pressed on with more specific terminology.
micrograph of tubulin and actin in cytokinetic bridge
The images suggest that tubulin &/or f-actin are known.
But the next one I consider to be another failure. It looks like it may know cytokinesis results in pairs, but knows nothing about cytoskeletal structure. On the other hand, compared to technology a few years ago, this is impressive. (The cell in the lower right looks like taxol treatment of microtubules, which, as we know, inhibits cytokinesis.)
micrograph of cytokinesis, tubulin is magenta, actin is cyan, cell structure mulicolors
The next two are beauties, but stray far from science. They are
real science micrograph of cytokinesis
and
real science micrograph of dividing cells
The NYTimes article interviews artists who say that they are not scared of their jobs going away. I spent less than two hours with this software and I have to say, illustrators should be very scared.
I also watched someone design embroidered leather jackets. Wow, they got instant variations and updates which probably resulted in final designs that I thought would sell. They were gorgeous.
The playform is being used for a lot of sketching, getting composition ideas, fantasy pornography, various illustrations. This platform shows images publicly as they are generated, so interspersed with mine were a diverity of other ideas. But what they had in common was a style. Bright colors, smooth rendering, fantasy novel covers. But the point of this exercise was to get an initial feeling for what gets generated with science terms.
More training with real science images looks like there is promise for finding categories in the images we have not noticed before. But more threatening, an ability to fake results.
And, of course, I only tried one AI and did not dig deep.
first written 2022-10-24 22:00
mcammer@gmail.com