Classification of reviews as fake or authentic is my MSc dissertation, so I thought I'd point out a few things. I have not read the paper referenced in the arti

Classification of reviews as fake or authentic is my MSc dissertation, so I thought I'd point out a few things. I have not read the paper referenced in the article yet, though. The article seems fine, given it's a pop-sci sort of thing. a) Humans are generally pretty incompetent at determining if a review is fake or not, so my opinion is it is a crude yardstick at best. b) I personally don't think using AMT as a source of subjects is a great idea unless care is taken to deliberately take into account psychological/social type factors - using AMT sort of turns your research into a psych focused one as well (unless you control who is in your pool of workers consciously and deliberately) The important point I think is that they have gotten to the point where neural nets etc can generate readable text that obviously isn't computer-generated, and so can fool the typical or casual reader. My research is showing promising results for using ensembles to classify reviews, and the state of the art systems use more than just the text which make them fairly useful. But there will certainly be an arms race. I am more terrified of the advances made in faking video and speech - when that really kicks off, Facebook will have to have _so many_ AI systems examining every bit of data "people" put into it.