What would happen if you trained a neural network to publish new papers in science and mathematics by feeding it existing papers from journals like Science, Nature, etc.?

Could this be a way to make new discoveries without paying researchers' salaries and carrying out actual experiments?