294. Using AI in the presence of congestion
Invited abstract in session TC-61: Behaviour in information systems, stream Behavioural OR.
Tuesday, 12:30-14:00Room: Maurice Keyworth G.31
Authors (first author is the speaker)
| 1. | Benjamin Legros
|
| Skema | |
| 2. | Francis de Véricourt
|
| Management Science, ESMT Berlin |
Abstract
This study examines the application of artificial intelligence (AI) in service systems where a decision maker (DM) must diagnose a patient. Drawing on a rational inattention framework, we model the DM’s diagnostic accuracy and associated cognitive cost. While incorporating AI always enhances diagnostic accuracy, it can simultaneously increase cognitive burden. This elevated cognitive cost is particularly concerning in congested service settings, where decision-making delays translate into extended waiting times for patients.
Moreover, prolonged waiting creates added pressure on the DM, raising the marginal cognitive cost of allocating attention. In turn, the DM may opt for less cognitively demanding (and therefore less accurate) decisions. Despite this potential drawback, we prove that AI usage consistently improves diagnostic accuracy, even under conditions of heightened marginal cognitive cost. We further show that the service value—defined as the difference between accuracy and cognitive cost—also benefits from AI adoption.
However, when the arrival rate of patients surpasses a certain threshold, leveraging AI leads to longer waiting times. We define the payoff as the difference between accuracy and a cost that is proportional to waiting time. Adopting AI proves beneficial primarily in settings with very low or very high congestion levels. By contrast, scenarios with moderate patient arrival rates may be better served without AI.
Keywords
- Behavioural OR
- Queuing Systems
- Artificial Intelligence
Status: accepted
Back to the list of papers