In 2012, Brian Nosek and Rachel Riskind published a piece in this journal arguing that policymakers should take the concept of “implicit bias” seriously. The article reviewed contemporary research in the field of implicit social cognition, arguing that automatic associations about social groups can shape behavior outside of conscious consent or conscious control in contexts like the law, medicine, organizations and beyond. Throughout the paper, the authors acknowledged that many of their claims were still tentative.  

Over a decade later, we asked ourselves: now that the area of research has accumulated more evidence, how has the state of the field changed?  

Following the same structure as Nosek and Riskind, we updated every section of the original paper in light of the fourteen additional years of evidence.  

On the one hand, many new findings were consistent with the previous literature: the core claim that implicit biases are real, widespread, and distinct from what people consciously endorse has only grown stronger. On the other hand, claims about the relationship between implicit bias and real-life, discriminatory behavior have become more complicated, and there’s been growing skepticism in the field—and society at large— about whether the concept of implicit bias should be incorporated in discussions about policy. 

This skepticism arises in part due to findings that interventions inspired by the implicit bias literature have largely failed to deliver. Meta-analyses have found interventions that changed people's scores on implicit bias measures like the Implicit Association Test do not seem to translate into changes in actual behavior. Conversely, some evidence suggests that changes in behavior don’t seem to require changes in implicit associations at all.  

The more we sat with the evidence, the more we became convinced that the right framing wasn't "implicit bias is a myth" or that “implicit bias is universally relevant to social issues”. Instead, we think it best to understand that policymakers should take the concept of implicit bias “seriously”—as a real and consequential phenomenon—without taking it too “literally”—that is, tying their interventions or policies too closely to the specific measures researchers use in the lab. 

In our paper, we borrow the metaphor of a “leaky roof” to clarify this distinction. Think of the mind as a house with a leaky roof, where the goal is to keep the floors dry. Much of the prior research in implicit social cognition has explored how to make better mops —interventions that try to reduce the individual-level implicit biases that participants hold. However, evidence increasingly supports that policy should focus more on building a stronger roof: creating structural and contextual changes that make it harder for bias to influence decisions in the first place, regardless of what's happening inside someone's head. 

This conclusion feels especially timely as questions about implicit bias migrate into new domains, including Artificial Intelligence. Algorithms inherit biases from the data and decision environments they're trained on and can in turn amplify them. Even with AI, our logic applies: rather than train people to be more aware of this bias, it is probably more productive to audit datasets and design more equitable AI systems.  

In the fourteen years since Nosek and Riskind’s paper, we now have a clearer picture of what hasn't worked, and a more promising sense of what might. We hope this update is useful to researchers, practitioners, and policymakers working to close that gap. 

Interested in learning more? Check out the full SIPR article: 

Axt, J., Palacio Posada, V., Roy, E., & To, J. (2025). Revisiting the policy implications of implicit social cognition.?Social Issues and Policy Review,?19(1): https://doi.org/10.1111/sipr.70003