Only an unreliable warning system is a good system – but which errors help?
Year: 2025
Type: article
Abstract: Studies on automation systems reliability found that even systems with up to 30% false reactions can still be beneficial. However, there may be a cry-wolf effect (missing reactions if there had been too many false warnings), but also a reduction of overtrust after experiencing unnecessary warnings. However, most studies focused either on only one of these errors (missing critical objects or providing false alarms) or did not distinguish between them. Thus, it is not clear which part of system reliability is beneficial.A driving simulator study with 47 participants was done with three independent groups each experiencing a differently reliable warning system. The critical events were 100 pedestrians either standing at the side of the road or running towards that and then either stopping (90% of the time) or crossing the road (10% of the time).The group with a 70% hit rate and no unnecessary warnings exhibited slower or no braking responses during some critical events, suggesting an overtrust to the system. This was reduced in the group with 70% hits, but 33% unnecessary warnings. This appeared to even improve reaction times in subsequent critical trials where no warning was issued. Contrary to the cry-wolf effect, drivers continued to react to warnings even after numerous unnecessary warnings.However, the study lacks a comparison with a control group (no warnings) to fully assess the warning systems' overall usefulness. Furthermore, including a system with a very high detection rate and no unnecessary warnings could further illuminate these dynamics.
Authors: Mark Vollrath, Julie Morawietz
Institution: Technische Universität Braunschweig
Language: English
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
Estimated APC paid: $2,780
OpenAlex ID: W4416421586