Work by Tversky and others in psychology have shown that humans utilize a number of cognitive biases and heuristics to simplify their decision-making processes under conditions of uncertainty. Here, we develop a model inspired by the representative heuristic, which we call cognitive probabilities. We show that these cognitive probabilities lead to estimated probabilities based solely on local data and are sufficiently good to work from. The set up involves distinguishing objects (which we organize into a partially ordered set) and corresponding descriptors (which we organize into a complete lattice). It is the interaction between these two lattices that defines the process of identifying objects based on sensed descriptors. The probabilities are entirely calculated using the descriptor lattice. Because these are done locally, we can define such a lattice based on context rather than dealing with the entire universe of possible descriptors. A byproduct of local computation (which translates into a loss of information) is that probabilities may be superadditive, which parallels some of the results in psychology.