I think that plenty users here already highlighted the main points (survival bias, lack of reinforcement with steel, optimisation for other characteristics). I'll focus on the chemistry instead.

Think on a tea strainer, a chicken wire, and some chain link fence. Sure, they might be made of the same steel, and they're all meshes. But they're all linked in different ways, with different properties, and they will serve different purposes. Aluminosilicates are also like this; even if you have the exact same composition, it's perfectly possible that some are more resistant than others, based on their structure.

Studying Roman concrete might reveal something about the aluminosilicate of the surviving buildings that might become useful later on. With that knowledge, even if you believe (as I do) that modern concrete already surpassed Roman concrete in plenty attributes, we can make it even better.