They do. It's just easier to add extra CPU cores to a ML pipeline than another person.

As a less cynical observation than a sister comment, I think this is partly due to how current tools work, or possibly even inherent to the technology. Just look at the number of ML examples where you never even get to see a single example of the data you are working with. Interactivity is even rarer, certainly because it doesn’t scale, but maybe also for the somewhat pedestrian reason that interactive UIs doesn’t come natural for either the tools and/or the people using them.

Then, there’s this issue of “explainability”: if you want to direct some generator you need to find out how the concepts you want to work with are encoded in intermediary layers.

To be fair, all this isn’t much of a secret, and there are quite a few projects doing interesting things. Magenta comes to mind, or GANBreeder.

One challenge here is seen in the types of articles researhers in comp sci have published over the years. I believe that there has been a massive decline in case studies and field trials of software and a rise in analytic and empirical algorithm development. To unlock the super power for people we will need more good hci and the wheel will have to turn in research practice once more.