Even though most everyday decisions are among more than two alternative, the optimal policy for such decisions was unknown. Together with Satohiro Tajima, Nisheet Patel, and Alexandre Pouget we derived this optimal policy and showed that it is significantly more complex than a simple scaling-up of decisions among two alternatives. However, a simple network implementation with features inhibitory cross-talk among the accumulators that collect evidence for the different alternatives turns out to feature close-to-optimal performance. This network has several cortex-like features, like an urgency signal, and independent decision boundaries on individual accumulators. Furthermore, it replicates multiple features of such decisions, such as Hick's law. In particular, with normalized inputs and additional accumulator noise, it replicates several 'irrational' behaviors, like the similarity effect, and the violations of both the independence of irrelevant alternatives as well as the regularity principle.

For more details, check out our article in Nature Neuroscience or get the pdf here.