Why do humans decide to switch or repeat a task? Suppose that the purpose of decision making is to maximize reward and minimize potential costs. Furthermore, suppose that agents learn about potential rewards and costs through trial-and-error. For example, incorrect responses may indicate that the current task is too difficult, suggesting to switch tasks, or, alternatively, wrong responses may indicate a lapse of attention unrelated to the task, suggesting to repeat. Here, we focus on the effects of errors on voluntary task choices. We developed a novel adaptive voluntary task switching paradigm and hypothesize that errors are cues which inform task choices. Based on theories of error processing, we derived five statistical models of how errors might influence voluntary task choices. We show that choice behavior was best explained by a combination of error probability of the performed task, error probability of the alternative task, and whether the previous response was correct or incorrect. This suggests that agents can integrate potential costs in a joint utility function guiding decision making.