No, because it can still measure response to its output and cyclically feed itself the most positive responses as new training data.

It could eventually be rather self supporting.