QUANTIFYING EXTERNAL INFLUENCES ON NEURAL SIGNALS UNDERLYING BEHAVIOUR

Neural signals that correlate with subjective reports provide a window into the neural bases of perception, memory, and phenomenal consciousness. Such signals can be driven primarily by exogenous input or by endogenous dynamics. Yet, existing approaches do not directly reveal the relative impact of external input on neural signals that encode subjective reports. Here, we present a computational framework that treats report-related neural signals as a mixture of internally generated and externally driven components, and we develop an algorithm to quantify their respective contributions. We apply this method to estimate the input contribution to subjective representations in multiple large-scale datasets collected from mice, macaque monkeys, and humans. Across species and task contexts, we find converging evidence that neural report signals are not purely internally generated, but contain a substantial input contribution consistent with sensory integration. While the spatiotemporal profile of this input contribution is initially heterogeneous, it later converges toward a globally uniform level. This pattern suggests that mammalian brains produce behaviour through widespread but functionally differentiated networks, followed by a global broadcast of the sensory-informed behavioural outcome. Our framework provides a new quantitative account of the neural computations that underlie behaviour.