An additional smart route may help each individual choose while still making the overall system worse.
Braess’s Paradox shows that more capacity or an additional route in a network does not automatically improve system performance. When users adapt their routes, a new option may create a worse equilibrium.
An additional smart route may help each individual choose while still making the overall system worse.
In the classic traffic network, every driver gains an attractive additional route. Because everyone chooses the same local improvement, congestion shifts so that everyone’s travel time may ultimately be longer than before the new connection.
Braess, Nagurney and Wakolbinger give a mathematical traffic example in which an added connection increases equilibrium travel time. Demonstrations in other domains require separate source verification. Organisational AI routing remains a transfer hypothesis.
An option set changes choices; choices change shared load; shared load changes payoffs. The paradox arises only when this endogenous redistribution and equilibrium formation cause the worse system outcome.
AI transfer to investigate: AI lowers local routing costs and may make responses faster and more uniform. The causal principle remains classical congestion/equilibrium theory.
Proposed test, not yet conducted: Create a workflow with shared bottlenecks and add an AI-routing shortcut. Look for pre-specified demand/capacity regions in which travel or waiting time after route choices stabilise worsens because routes are redistributed.
More options do not guarantee greater resilience. Where capacity is shared, a local improvement must always be evaluated together with the system’s response.
At user equilibrium, no individual traveller can shorten travel time by changing route alone; stable route shares alone are insufficient. The paradox requires specific costs, demand and incentives; an extra connection is not generally harmful.
Braess, Nagurney & Wakolbinger (2005), §§3–5 — pp. 448–450
Braess, Nagurney and Wakolbinger describe the traffic model and user equilibrium. The conclusions depend on that model’s assumptions; applications in other domains and effects of AI require separate research.
Help improve this case
We review your contribution before updating the text or sources.