Neural abstraction of evidence accumulation enables flexible multidimensional perceptual decisions

Human decisions in perception are often tested under simplified conditions, but they must adapt to complex and changing situations in real life. A study has now shown that the human brain can flexibly combine multiple sensory features during decision-making, even when those features come from different sensory modalities. This ability is made possible by the brain's capacity to abstract and filter underlying neural representations, allowing for more efficient and relevant information processing.

Using rapid magnetoencephalographic (MEG) measurements, researchers observed a hierarchy of object representations in the brain. These representations become increasingly abstract and selectively gated based on their importance to the decision-making process. The study found that a shared abstract representation of evidence accumulation occurs across both unidimensional (selection-based) and multidimensional (combination-based) decision contexts.

This means that the brain can efficiently process and integrate information from various sensory inputs to make informed decisions.

To test these findings further, the researchers used recurrent neural networks, which are artificial intelligence models designed to mimic the way human brains process information. These networks were trained under conditions similar to those encountered by humans. The results showed that the networks could acquire and benefit from abstract representations only when operating at human levels of computational noise, which refers to the inherent limitations and variability in human decision-making processes.

Overall, these findings provide evidence for a "flexibility through abstraction" account of human decision-making. By abstracting and filtering relevant information, the brain can adapt to a wider range of everyday perceptual conditions, allowing for more efficient and effective decision-making in complex and dynamic environments.