The new paradigm of artificial societies aims to simulate human societies and to study collective social phenomena from the bottom up. The idea that natural human societies are almost certainly computationally irreducible, so the only way to model it is by direct simulation forms the motivation driving artificial society researchers. The risk is that the simulation may then become as complex as the real world and therefore just as difficult to analyze as the phenomena being simulated, providing no explanatory power. Limited effort has been devoted to fighting uncertainty and delineating the limits of knowledge in artificial societies. Broadening the discussion of uncertainty in the social simulation domain by shifting focus from information to knowledge, we argue that there are a number of things that we cannot know (or questions we cannot answer) that are not the result of imperfect information, and thus help to explore how these realms of ignorance may affect our simulation goals and efforts. The paper reviews different modes or forms of social knowledge production in artificial societies, and investigates their built-in imperfections - for reasons of logical principle and not just empirical fact – and then situates ongoing work on an agent-based model for exploring the role of environment on health inequalities under the framework of the discussion. We argue for such an exercise to be part of every social simulation project as it aids artificial society researchers to be clear why simulation should be used and what it is intended to achieve.