Don’t worry- the spelling error in the title is not due to human oversight. In fact, it is intentional. But why would scientists be interested in deliberate errors? When we invite individuals to take part in empirical studies, we ask them to approach the task with a specific goal in mind. Specifically, we frame experiments in terms of doing one’s best: a memory test might require participants to try to remember as much detail as possible, or, a perceptual test might insist that performance is both as fast and as accurate. We also frame certain real-world tasks in this way, such as the hope of leaving a casino with as much money as possible.
We describe this goal as win maximization. Therefore, it should not be surprising that when our memory is inaccurate, our perception is wrong, or, our money is lost, these negative outcomes make future performance worse. Here, we can cite the decreased flexibility in responding observed following losses relative to wins (eg, Forder & Dyson, 2016), and, the observation of impulsive ‘tilting’ behaviour caused by losses in gambling environments (eg, Torrance, Roderique-Davies, Greville, O'Hanrahan, Davies, Sabolova & John, 2022).
Such observations are, however, limited to this one specific type of goal. Therefore, it is important to study goals other than win maximization, as this helps us to understand whether there is something fundamentally negative about ‘negative’ feedback. The figure below reminds us how fixated the domains of Psychology and Economics can be on this particular goal.
In viewing wins versus losses, and, maximization versus minimization as independent dimensions, then there are three other goals that we may have. For example, one regular example of loss minimization is the tracking of our bank account: we are paid once a month and then spend the rest of our time in pursuit of loss minimization (e.g., groceries -$339; utilities -$218). In our experiments, we were inspired by the unique goal of loss maximization, as depicted in the film Brewster’s Millions and the board game Go For Broke, where players must be the first to lose $1 million dollars.
We carried out four experiments where participants played simple games against computerized opponents using the traditional goal of win maximization and the more novel goal of loss maximization. In the case of win maximization, individuals are given the goal to play ‘as well as possible’, whereas in the case of loss maximization, individuals are given the goal to play ‘as poorly as possible.’ Our computers played in predictable ways, allowing participants to increase their win rates (win maximization) or increase their loss rates (loss maximization) by selecting the appropriate action at each trial. The scatterplots below evaluate this behavioural consistency across participants under four conditions.
To begin, let us compare the two left-side panels, where the goal is to perform as well as possible and maximize the number of wins. The top-left panel plots win rate against the ability of participants to select the appropriate action immediately following a win. In contrast, the bottom-left panel plots win rate against the ability of participants to select the appropriate action immediately following a loss. As is clear from the correlations, behavioural consistency was stronger for actions following wins (r = .683) rather than losses (r = .390). This represents the replication of a standard effect that performance is sub-optimal following losses under conditions of win maximization.
Now let us consider the two right-side panels, where participants now must perform as badly as possible and maximize their number of losses. First, it is worth noting that the line of best fit is lower for loss maximization relative to win maximization. This gives us some confidence that participants were responsive to the unique goal of loss maximization. Critically, we observe the opposite pattern of results in terms of correlation strength. Now, behavioural consistency is stronger for actions following losses (r = .789) rather than wins (r = .495). This represents the novel finding that performance is now sub-optimal following wins under conditions of loss maximization.
Across all our experiments, we replicate this double dissociation where actions following wins were more consistent during win maximization, but actions following losses were more consistent during loss maximization. What does this mean? This suggests that a mismatch between a goal and your current state creates problems for future performance, rather than ‘losing’ per se. In other words, there is nothing inherently damaging about the cognitive impact of losing, as long as one has the goal of losing in mind. Understanding how individuals are able to change their reactions to loss remains a central aspect of understanding industrial, educational and gambling behaviour.
Our experiments also show how performing as badly as possible serves as an additional expression of expertise. That is, in order to maximize losses, one must know how to win and then do the opposite. For another example of cognitive proficiency demonstrated via the act of intentional self-sabotage, the interested reader may wish to review the intentionally terrible piano playing of Les Dawson (https://www.youtube.com/watch?v=9nNGlaiVypU). Our continued interest in how individuals deliberately make miskates represents steps towards normalizing the experience of negative outcomes, and may lead us to re-evaluate the extent to which our understanding of human and animal behavioural sciences has been limited by our focus on win maximization.