A technological solution would be to have an AV map out an intersection from different angles, or at least account for other cars. Ideally, it would be able to “look” at the intersection from the data it has, and do some matrix transformations to “see” it from another point of view (like someone trying to make a left turn). While computationally expensive, cars are already expensive and should house a decent amount of hardware natively.
A more “fuzzy logic” approach would be to ask “how many cars are there on the road besides me?” and have different levels of caution based on the answer. So “no cars” could map to “I can do anything that is legal”, “some cars” could map to “I should be ready for potential problems”, many cars to “I should slow down a little, and assume that a problem will come up.”