(Sorry if I'm spamming your board with Popular Science speculations, but I just thought this could be an interesting thought experiment).
There are quite a few videos on youtube about Bell's Experiment. This experiment supposedly disproves the hypothesis of local hidden variables. We split entangled photons, measure them separately and while our results look individually random, they are somehow align if we compare them afterwards. This supposed to prove that photon "decided" its orientation at the time of measurement.
This heavily relies on our ability to "surprise" the photon with a random direction we're going to measure it. But what if that photon knew all along in which direction it was going to be measured. Since photons travel at the speed of light, they don't experience time, hence they should know the future, and there shouldn't be a way for us to "surprise" it at all.
Best Answer
Bell's inequalities hold more generally. You can verify the existence of nonclassical correlations in a variety of platforms, including those that have nothing to do with photons or light in general.
"This heavily relies on our ability to "surprise" the photon with a random direction we're going to measure it. But what if that photon knew all along in which direction it was going to be measured":
I don't think this means anything. Bell's inequalities are a way to certify the existence of a specific type of correlations between measurement outcomes. It's not about the photon being "surprised" or not. It's about the way you interact with the photon during measurement. In other words, it's about the types of "questions" you "ask" the two photons.
Any kind of classical correlation between the two parties isn't sufficient to violate Bell's inequalities. If you assume that the systems "know the direction along which they will be measured", then yes, you can violate the inequalities. This is often referred to as the superdeterminism loophole. However, it's worth remarking that any probability distribution can be obtained with such an assumption. In other words, you get a model which is not falsifiable, as it can "explain" any observation. Furthermore, you'd have to come up with a physical mechanism to explain how the information about the measurement choices somehow leaks to the detectors. In many situations, such a mechanism would be really weird and far-fetched.