Michael Sechman

Current AI-assisted debiasing interventions assume a defeatist premise: biased minds cannot be changed, only contained. These “outcome-focused” tools mask demographic information or correct decisions post hoc, leaving underlying psychological mechanisms untouched. We argue that this pessimism is premature. The disappointing record of “change-based” interventions partly stems from a mismatch between simple intervention designs and the pluralistic, dynamically interacting memory systems that sustain bias. Drawing on 4E cognitive science, Intergroup Contact Theory, multiple memory systems research, and the Heterarchical Control Network model of implicit bias, we articulate six design principles for next-generation interventions and demonstrate how Generative AI and immersive Virtual Reality make it possible to scale optimal debiasing conditions. Using recruitment as our example, we sketch a training module where recruiters engage in intergroup interaction under controlled yet ecologically valid conditions. Finally, we address the ethical risks that arise when AI is used not merely to sanitize outputs, but to transform minds.