Scientific theories of subjective experience, such as such as Global Workspace theory and Attention Schema theory, are being co-opted as architectural proposals for artificial systems. A lack of consensus on what a scientific theory of subjectivity should actually explain will make it hard to evaluate any such artificial systems, however. In this paper, we adopt a naturalistic starting point, rejecting the idea that science must explain why we have subjective experiences, and instead suggest that science need only explain why we take ourselves to be having subjective experiences. But we resist the idea that this move leads to some kind of illusionism; instead we propose a more functionalist approach that saves certain key realist intuitions. Specifically, we argue that an idea from computer science, namely virtualisation, and the associated distinction between abstract content and implementation mechanism, sheds light on two characteristic features of subjectivity: first, the fact that we seem to have dense (everywhere rich and contentful) subjective lives, in contrast to the sparser view emerging from cognitive neuroscience; and second, the fact that we take our subjective lives to be broadly private. To the extent that these are robust (functional) features of our self-theorising, they need to be accounted for by any naturalistic theory of subjectivity, and are likely to be important characteristics of any agents with artificial subjectivity.