An uncontroversial cancel culture: Cancelling aware and unaware errors
The detection of errors we commit is crucial. If we detect an error, we can often rectify it ourselves or ensure that it is dealt with otherwise. If errors go undetected, however, they can have serious future consequences. For example, past undetected errors have led to serious train and plane accidents as well as fatalities of patients in medical care. A field where minimising error rates is highly relevant is that of data entry. For example, data entry clerks work for the NHS or banks and enter large amounts of data into databases. However, humans cannot be highly accurate all of the time. Thus, the question arises if there might be technical means that could help humans to identify errors. A recently discovered behavioural effect might open a new path towards achieving this aim. The effect has been termed error cancellation as it was found that participants’ response durations—i.e., the time from pressing to releasing a key—are significantly shorter for errors relative to correct responses (Foerster et al., 2021, 2022). That is, individuals appear to try to 'cancel' their erroneous responses by releasing the response key faster than for correct responses. That error cancellation manifests itself while the erroneous response is still ongoing suggests that it might be an automatic response of our cognitive system to the occurrence of an error.
A crucial open question is whether the error cancellation effect only occurs with errors that are consciously detected or whether the same effect can be observed if the error remains undetected. Observing error cancellation also for undetected errors would pave the way for enabling automated error detection in many real-world applications. In the proposed project, we will therefore investigate error cancellation in different error awareness tasks. In a second step, classifiers will be trained to investigate if it is possible to detect errors automatically based on response durations, even if the actually correct response is unknown to the classifier. The classifier will then be refined with other predictors for errors that can be extracted from response patterns during the task. The results will be of interest to stakeholders who aim to improve accuracy in data entry processes by detecting errors as early as possible. On a theoretical level, the results would inform cognitive control models of error monitoring and refine theories of error awareness.
The project will be in collaboration with the Cognition & Behavior group at the University of Würzburg in Germany. Informal enquiries about this project can be directed to Jan Derrfuss (jan.derrfuss@nottingham.ac.uk) or to Claudia Danielmeier (claudia.danielmeier@nottingham.ac.uk).