After watching the below video on machine learning. Which showed some sort of generation. I imagine that this could be replicated for 3D objects as well. Say you let the matrix-setup train on a biological part. With many variations of this part the setup will have learned what it should look like. Then as we saw in the video. The setup can also generate different shapes of the same object. I don’t know the implication for evolution. I guess however the genetic makeup would also include support data (the learned matrix). To help generate biological elements like the trained part. This thought experiment I think shows how evolution can produce so much variation. Without the energy cost due to error.
Below is a link to the Neural Network video and a would be training set for an arbitrary biological part.
Would be training set for an arbitrary biological part.