Multisensory perceptual awareness: Categorical or graded


Neural evidence suggests that mechanisms associated with conscious access (i.e., the ability to report on a conscious state) are “all-or-none”. Upon crossing some threshold, neural signals are globally broadcast throughout the brain and allow conscious reports. However, whether subjective experience (phenomenal consciousness) is categorical (i.e., transitioning abruptly from unconscious to conscious states) or graded (i.e., characterized by multiple intermediate states) remains an open question. To address this issue, we built a series of artificial neural networks containing distinct feedback connectivity from “multisensory” to “unisensory” cortices. In line with consciousness theories, we operationalized perceptual consciousness by the presence of feedback from higher-order nodes back to unisensory nodes which allow ’neural ignition’ - a rapid, non-linear boost in response putatively leading to phenomenal consciousness. When simulating how these networks responded to unisensory and multisensory inputs, we found the fastest responses for multisensory presentations associated with multisensory feedback, and the slowest responses for multisensory presentations without feedback. Most interestingly, despite being built in line with “all-or-none” models of consciousness, multisensory stimuli associated with unisensory feedback (i.e., auditory or visual), and hence consistent with unisensory phenomenology according to theories of consciousness, generated intermediate reaction times. To extend these models to human perception and performance, we conducted extensive psychophysical testing in 29 subjects who each completed 10 h of a multisensory cue-congruency task. Consistent with the modeling results, we found that reaction times to multisensory cues reported as unisensory were intermediate between those of fully aware and fully unaware cues. These results support the existence of graded forms of phenomenological consciousness that can be instantiated by simple neural networks built in line with “all-or-none” models of consciousness.