I was reading this report: https://www.theverge.com/2017/4/12/15271874/ai-adversarial-images-fooling-attacks-artificial-intelligence
Researchers used noise to trick machine learning algorithms to misidentify or misclassify an image of a fish as a cat. I was wondering if something like that can be used to create subliminals.
What I mean by subliminals: United Nations has defined subliminal messages as perceiving messages without being aware of them, it is unconscious perception, or perception without awareness. Like you may be aware of a message but cannot consciously perceive that message in the form of text, etc.
All the reports about the noise trick said the noise was so transparent that humans couldn't detect it. This can be changed to make it noticeable unconsciously but unnoticeable at a conscious level so a human can register the subliminal but not be aware of it.
Is it possible to take an output from a hidden layer to construct such subliminal for humans, with trial and error one can find right combination? Can it be possible to come up with a pixel pattern or noise with ML which allows one to impose subliminals?