We describe a virtual agent that learns qualia-free semantics in real time. Such an agent will be useful in computer games. The semantics lack qualia or “raw feels.” We specify syntactic rules for learning. These lead to functional understanding: agents demonstrate that they understand the objects they interact with based on how they influence agent well-being. We build on our previous work modeling the affect-handling system of the mammalian brain. This system provides a flexible, real-time decision system that seeks to maximize affect. We constrain ourselves narrowly to mainstream neuroscience, syntactically specifying nonarbitrary, phylogentically-evolved brain-based rules to compute affect. Agents use these rules to evaluate the objects they interact with, depending on how these objects enhance or reduce their well-being. By modeling these causal interactions with the outside world in terms of affect, agents develop qualia-free semantic understanding which they demonstrate by using objects to maximize their well-being.