A new theoretical paper shows that helping drivers remember more routes can backfire, making traffic worse instead of better.
The researchers model routing not as travelers picking from every possible path, but as picking from whatever routes they actually remember or get shown. In their setup, each traveler carries a limited memory, receives a set of surfaced route options, chooses probabilistically via a logit rule, and updates memory using a policy like least-recently-used. They prove this settles into a stable equilibrium they call a Forgetful Wardrop Equilibrium, and connect it to a simpler model where each route just carries a weight reflecting how often it gets remembered or surfaced. For travelers who only recall their single most recent route, the two models are mathematically identical; for longer memories, the paper builds an approximation pipeline with error bounds.
The key finding is what the authors call a Recall Braess Paradox: improving recall, with zero change to road capacity, can increase average delay at equilibrium, and they show this can happen on any network with more than one route between start and end points. That is a memory-based cousin of the classic Braess paradox, where adding a road counterintuitively slows everyone down. It is a pointed caveat for anyone building navigation apps or traffic systems around the assumption that showing people more route options is a straightforward win.
Worth remembering next time an app update promises smarter route suggestions: smarter is not the same as faster for everyone.