My work focuses on understanding and undermining the broad reach of narrow thinking. Specifically, I study cultural, social, cognitive, and neural mechanisms that perpetuate stereotypes and prejudice, and leverage basic science about those mechanisms to develop and refine interventions to reduce the expression of stereotyping and prejudice.


One line of our work explores how Hebbian learning, or activation-based learning, leads to stereotype perpetuation. Hebbian learning contributes to the perpetuation of stereotypes; for example, seeing a Black person activates Black, which then activates criminal because of the Black criminality stereotype. The mere activation of this cognitive link should strengthen the link — even without any evidence that the man is actually criminal. In other words, untested stereotypic assumptions strengthen stereotypes.

In addition to Hebbian learning, hedonic learning processes (e.g., reward/aversion) also may play a role in stereotype perpetuation. Stereotypes operate with a large degree of probabilistic uncertainty — no stereotype is incorrect in 100% of cases, and no stereotype is correct in 100% of cases. As a result, probabilistic uncertainty engages the brain’s reward system. If an uncertain prediction ends up being correct, it is rewarding; if an uncertain prediction ends up being incorrect, it is aversive. Therefore, we hypothesize that stereotype confirmation is rewarding and stereotype disconfirmation is aversive using an affective transfer (classical/Pavlovian conditioning) paradigm.


We examine how stereotype-based humor perpetuates stereotypes, specifically in the form of internet image memes. Stereotype-relevant memes and stereotype-irrelevant memes are collected and coded for preliminary analysis and rated on their funniness, offensiveness, etc. The data will be used to test the hypothesis that joke images that affirm stereotypes will be more popular, less complex, and spread farther than those opposing stereotypes and those that are stereotype-irrelevant.

We hypothesize that when messages in the media support stereotypes (e.g., if an article says men are better at math than women), people will remember them better. In addition, we hypothesize that they will overestimate these differences relative to what is warranted by the actual statistics. When messages disconfirm stereotypes (e.g., women are better than men at math), people won’t remember them as well and are less likely to overestimate the differences.


The bias habit breaking intervention was the first, and remains the only, intervention that has been shown experimentally to produce long-term changes related to personal biases (Devine, Forscher, Austin, & Cox, 2012, cited 268 times; Forscher, Mitamura, Dix, Cox, & Devine, 2017). This intervention compares bias reduction to the process of "breaking the bias habit" in that it requires certain elements to be successful. Specifically, it requires awareness and concern about one’s personal biases and its effects, motivation to overcome bias, and strategies to aid or guide one’s efforts to reduce bias, and sustained effort over time. Whereas previous models of prejudice and bias suggested that prospects for true change were dim, this evidence-based approach offers encouraging prospects for reductions in bias.

We have administered this intervention with many different audiences, including public school teachers, professors, graduate students, lawyers, judges, doctors, police officers, tech companies, and more. A gender version of this intervention directed at STEM faculty (Carnes et al., 2012) caused UW-Madison science departments to have a 15 percentage point increase in hiring women as faculty (Devine, Forscher, Cox, Kaatz, Sheridan, & Carnes, 2017). In several randomized-controlled studies, our team has tested this intervention’s replicability and long-term effectiveness, with effects lasting up to at least 2-3 years post-intervention.

Currently, we are running a follow-up study to an online version of our intervention, that is focused primarily on race, to see if it has long term effects on people’s attitudes and actions. We hypothesize that individuals who received our bias habit breaking intervention will show more concern about bias topics, be more likely to choose to engage with bias topics, and be more likely confront bias messages in comparison to our control groups.