The problem is real, and the solutions offered are simple and easy to apply. Interesting how one of the third-party sites is a .net facebook domain rather than .com. The generation of false addresses seems to make a lot of sense for avoiding phishing attempts. Kind of similar to how we can poison data in AI which may be used for both good and nefarious purposes.
I’d say poisoning the data is good if the model is being used for nefarious purposes such as in the case of throwing off the models used for unwanted surveillance. Kind of on a tangent - I remember in the video game “Watchdogs 2” they were trying to use mass surveillance AI to find Marcus, and the hackers threw off the surveillance by making 40 different Marcus’s walking just around the SF bay area. Naturally, data poisoning can also be bad if used to provide false authentication, or for things like deepfakes. To me, it’s slightly surprising that there are no universally used methods for proving the legitimacy of data. This is something that blockchain has largely already made possible. We can use blockchain for verifying whether a video was actually recorded in a specific place and, with the timestamp, at a specific time, thereby preventing deepfakes.
@zube.paul @Larry_Bates tagging for if you’d like to carry our discussion last month to the forum