When Being Wrong Is Expensive: Non-Punitive Choice Architecture in Interactive Instructional Media

Abstract

Interactive instructional media teach through consequence. A learner chooses, the system responds, and the difference between responses is where the learning is supposed to happen. This works because being wrong is informative. It assumes, however, that being wrong is affordable. For some learner groups it is not: where a history of correction, social failure or elevated anxiety makes the experience of error costly in itself, an artefact that dramatises the learner’s mistake may teach avoidance of the artefact rather than the skill. This paper develops a conceptual framework for that situation. We first set out the case for choice-based instruction and the case for error, including the productive failure literature and its boundary conditions — notably the finding that younger learners and those with limited prior knowledge do not reliably benefit from failing first. We then characterise failure sensitivity as a population-level design constraint rather than an individual fragility, drawing on the elevated prevalence of anxiety in autistic children and on the errorless teaching tradition in developmental disability education. From this we propose a four-part typology of branch topologies — punitive, corrective, differentiated and equivalued — and identify the fidelity–safety trade-off as the central design tension: consequence carries information, and removing it removes information along with the harm. We specify five conditions under which non-punitive branching should be preferred, four design moves that preserve informativeness without punishment, and the objections that a designer adopting this position must be able to answer. The framework is illustrated briefly with reference to one instantiation, an animated social skills curriculum for autistic children, but is stated to apply wherever choice-based instruction meets a failure-sensitive learner group. This is a conceptual contribution; no empirical claims are made and none should be inferred.