This paper needs to be spread. Good post thank you. As someone who works with AI there are a few concerning things to me regarding this paper. First: Absolutely no where does it give a baseline classification or an example of the control "statements" used to train the AI in terms of what is considered factual or conspiratory. Therefore it can be assumed that someone at google is using their opinion on what is considered fact. This plainly violates its own justification of the project "blurring the lines of fact and opinion". Second: the way it identifies conspiratory language in its "hybrid algorithm" is not defined to the reader. I can assume based off the language used it is based off of the noman chomsky identification of propaganda. However while his linguistic analysis is very good it is not perfect. He is an admitted socialist and has viewpoints that are biased. Liberal pseudo intellectuals love to cite him as god. Third: the AI training algorithm is not identified. Is is manually trained? Is it trained using the hybrid algorithm? Every AI model has an inherent bias of the programmer. Trusting the "fact" from an algorithm basically the same as the person who wrote the algorithm telling you the fact. Stop pretending AI is omnipotent and all knowing. It isnt. People need to realize AI is just advanced pattern recognition. Modern day AI does not have the ability for sentience. Yea it can pass the turing test but there is no such thing as a universal truth.