After attending a machine learning presentation I got the chance to talk to the presenter about AutoML. He said that it is out-performing a lot of data scientists at Google. Which really peaked my interest. I asked him if the software was similar to the software Daniel Hillis wrote back in the '90s to generate code that was out-performing human coders (see The Pattern on the Stone). Sadly, he had no idea who Daniel Hillis is. I think the idea of software generated code never caught on because no one could understand why it was faster but everyone could agree that it wasn't safe because it was so difficult to decide if it would always work correctly. Which brings to mind the question - is AutoML safe? Will it create bias because the data is formatted a certain way?
I look forward to learning more about how AutoML works. I'm also interested in learning how to use AutoML but am vastly more interested in its internals.
