My problem with cellphone cameras is the amount of post-processing they perform to make the picture look more attractive. They can be made to trick our neural networks into classifying the image as high quality, and they may do so with the artificial NNs present on the phone; but this doesn’t make the image more realistic. Some of the post-processing is done to make faces look different, in accordance to some societal ideals, but that doesn’t bring us closer to seeing things as they are.
We are veering into a world where it is way cheaper to trick each other about reality than to pursue gaining accuracy in measuring it. GPT is another instance of a system that seems to satisfy our perception of correctness; but when more cognitive cycles spent on its results, holes and contradictions emerge. If not enough effort is spent on interrogating these results, we end up with short feedback loops that converge on the artifacts created by these models. We then get a reality tunnel formed by models and machines, with blindspots that span most of society through its dependence on these machines to acquire information. These blindspots feed back into future iterations of machinery and societal structures / policies.