As we observe and act in the world, the brain integrates recurring relationships among events and associations with rewards into cognitive maps to enable flexible planning. Furthermore, as humans, we influence each others’ memories, beliefs, and models of the world in conversation networks. Momennejad’s first line of research asks what algorithms the brain uses to encode, update, and retrieve cognitive maps. Using computational, behavioral, and neuroimaging methods, Momennejad shows that the brain stores cognitive maps as multi-step predictive representations in memory, updates and generalizes them via offline replay, and retrieves them for flexible learning and planning. The second line of Momennejad’s research concerns what structures of social interactions lead to more convergent memories after conversations, and when “memory bubbles” are formed (where subnetworks are isolated and do not influence each others’ memories). Momennejad uses graph theory, propagation models, and empirical studies to show that if weak social ties interact early, they can break “memory bubbles”. In ongoing and future directions, Momennejad studies how the volatility of the world and disparity in social networks sculpt memories in ways that can sometimes be maladaptive. Momennejad proposes computational approaches to investigating the spread and generalization of memories at different hierarchical scales (hierarchical and convolutional cognitive maps), re-sculpting maladaptive semantic memory networks (computational psychiatry), and designing interventions that lead to more just social structures (computational justice).