How to Poison the A.I. Machine - Freakonomics Radio
Source: freakonomics.com
This Freakonomics episode features a conversation with Ben Zhao, whose research lab has been doing some pretty pioneering work to prevent AI misuse and protect the rights of content creators from having their work eaten up by AI training models. One approach that's covered in this episode is the idea of "poisoning" the material on which the AI is trained, so that when it's asked to reproduce something similar the AI application consistently ends up generating something that's totally incorrect.
Art is interesting when it has intention, when there's meaning and context. So when AI tries to replace that, it has no context and meaning. Art replicated by AI, generally speaking, loses the point. It is not about automation. I think that it is a mistaken analogy that people will oftentimes bring up. They say, well you know what about the horse and buggy and the automobile? No, this is actually no about that at all. AI does not reproduce human art at a faster rate. What AI does is takes past samples of human art, shakes it in a kaleidoscope, and gives you a mixture of what has already existed before.