Data, at least until the advent of machine learning, was usually taken as the alphanumerical representation of a physical state or process, for example, as a measurement or categorical designation.
While there has been a lot of discussion about the accuracy or politics of such representations, the implicit assumption has always been that the physical state or event comes first, and the data second.
Quite arguably, this has never been the complete story. Indeed, many anthropologists claim it to be a uniquely human capacity to make up and believe in stories that have no direct link to physical reality.
As we move deeper and deeper into digital environments, data functions less as a camera, and more as an engine, not recording the world, but bringing it into existence. Because of the growing importance of post-representational data, we should ask less what data is, and more what is does. Because also unreal data is real in its consequences.