I've wondered for a while if it would be possible to train an AI to create visualisations of dinosaurs and extinct animals.

We've got sample data from living creatures and their skeletons, so if we could somehow traing a model that takes data from the skeletons (3d models? pictures?) as input and the animal as output, we could then apply that to skeletons found from past eras.

Granted, I only have a very basic understanding of machine learning so this is probably an absurd idea for a multitude of reasons I'm ignorant of.

I think that is a fascinating idea.

Skeletons give you a lot more to go on than the meme implies. We know, for example, where the muscles attach, and how strong those attachments are, giving a good proxy for how large the muscle is. It also reveals the angle of attachment, which tells you a lot about its posture, how it moved, and what kind of musculature would be required to support it.

That is calibrated against lots of existing species, and some rare fossils give even more evidence of body shape, giving opportunities to calibrate it further.

There's a lot of delicate inference and opportunities to go wrong, but it's not as haphazard as the meme suggests. The reconstructions are based on much finer details of the skeleton, including microscopic ones.

link?

Probably this: https://i.chzbgr.com/full/9429647872/h73AD2FFA/twittercom41s...

Or an addition to it: https://img.ifunny.co/images/458ec9c84c00a2ec331b511115f302c...