@shoq That's exactly the problem. Machine learning works by training the machine with the contrast of some low-quality data and high quality data. The problem with anything under study or with controversy is there is no high quality data. Sometimes there's no consensus, and even consensus is often wrong.
And insofar as you have accurate info to train the machine with, you'd be better off just releasing that rather than a complex AI that will produce somewhat less accurate paraphrases.