Uncovering internal evidence representations reveals the nature of confidence computation in multi-alternative perceptual decision making

A central goal of perceptual decision making is to determine the computations underlying choice and confidence in complex, multi-alternative tasks. However, revealing these computations requires knowledge of the internal evidence representation upon which the computations operate. Unfortunately, unlike for 2-choice tasks, there is no established method in multi-choice tasks to determine how stimuli are transformed into internal evidence. Here we address this critical gap by developing a novel modeling paradigm that reveals how the brain transforms external stimuli into internal evidence representations during multi-alternative decision making. Across three experiments that feature 3- and 5-choice conditions, we show that the internal evidence representation for these stimuli can be described with a simple, two-parameter equation that fits data for up to 12 conditions. Critically, the ability to model the internal evidence representations allowed us to test six theories of confidence computation. Our results support a new model where confidence reflects the raw evidence difference between the top two alternatives. These results establish a new paradigm in which a two-parameter equation can be used to determine the internal evidence representation for potentially unlimited number of multi-alternative conditions and put forward a new model of confidence computation in multi-alternative tasks.