We use a model with local interaction (a one-dimensional cellular automaton) to study how commuters choose among alternative roads. Commuters have information about their neighbours' most recent experience (local interaction) and they remember their own experiences (memory). We illustrate how a simple, self-organizing system, based on local information and locally rational agents can in some cases outperform the Nash equilibrium. While the social optimum is unenforceable without a central planner, due to the variations in individual travel times (i.e. the social optimum is not individually rational), variations across commuters in the steady-state of our self-organising system are at least equivalent to, and mostly significantly larger than those required for the social equilibrium. Increasing the neighbourhood size illustrates that more information without co-ordination leads to worse overall performance.