Aligning transformer circuit mechanisms to neural representations in relational reasoning

Relational reasoning--the capacity to understand how elements relate to one another--is a defining feature of human intelligence, yet its computational basis remains unclear. Here, we combined human neuroimaging (7T fMRI) with artificial neural network modeling to identify circuit-level analogues of human reasoning computations. Using the Latin Square Task, we found that humans and transformers were able to generalize the task reliably, while standard architectures used in cognitive neuroscience could not. Analysing the transformer components revealed distinct computational roles: positional encoding captured the spatial structure of the task and aligned with representations in visual cortex