Summary: People can learn to use the nonconscious content in their brains to make profitable decisions. Findings suggest a novel form of nonconscious metacognition.
Source: ATR Brain Information Communications Research Laboratory Group
Although we experience the world as conscious beings, the majority of brain activity operates outside our awareness. This raises a fundamental question: can people make deliberate, advantageous decisions by tapping into that hidden reservoir of neural information? The challenge is immense because nonconscious brain activity is high-dimensional and complex. How could the brain identify which aspects of such vast unconscious activity are relevant for making decisions when, by definition, that information is not available to awareness? Contemporary artificial intelligence struggles with similarly high-dimensional problems, suggesting no simple computational shortcut exists. This study explores whether human learning can nevertheless exploit nonconscious brain states to guide choices and whether metacognitive signals such as confidence play a role in that process.
An international research team applied a cutting-edge real-time brain decoding method to monitor and detect specific patterns of nonconscious neural activity using fMRI. These multivoxel activity patterns were used, covertly and in real time, to determine the correct choice in a simple reward-based decision task. During learning sessions, participants received a small monetary reward when their selected option matched the choice indicated by their own hidden brain state. Over repeated trials, researchers tested whether participants could learn to select the rewarded option even though the contingencies were based on brain activity that remained below the threshold of conscious awareness.
The results show that people can indeed learn to exploit latent, nonconscious brain representations to maximize reward. Importantly, most participants did not report awareness of the learning process itself. Yet their subjective confidence in individual choices proved informative: when participants reported higher confidence, they were more likely to select the correct option and obtain a reward. This pattern indicates an unexpected form of nonconscious metacognition—participants were not explicitly aware of the relevant brain signals or of learning those signals, but some internal mechanism correlated with confidence appears to track performance and support learning.
Dr. Aurelio Cortese, Senior Researcher at the Advanced Telecommunications Research Institute International, Kyoto, who led the study, summarized the findings:
“Remarkably, participants learned to make rational choices based on their own nonconscious brain activity through a simple trial-and-error procedure. The degree of learning was predicted by their capacity for introspection about decisions. This approach could one day be used to reactivate or enhance dormant skills.”
Dr. Hakwan Lau, Professor in the UCLA Psychology Department and co-author of the study, commented on the broader implications:
“This work is unique in showing how flexibly the human brain can reconfigure and learn under conditions that previously seemed impossible. The association between confidence and the ability to learn from nonconscious signals raises new questions about the functional role of metacognition in guiding behavior.”
Dr. Mitsuo Kawato, Director of the Computational Neuroscience Laboratories at ATR and senior author of the paper, emphasized the theoretical challenge the study addresses:
“One of the biggest problems in neuroscience and artificial intelligence is the ‘curse of dimensionality’: the brain contains billions of neurons, producing highly complex, largely nonconscious activity. How can efficient decision-making occur when time and experience are limited? Our findings offer an initial indication that subjective confidence may help the brain navigate these vast latent representations.”
The experiment involved 18 participants who each completed three separate sessions, yielding a total of 54 neuroimaging sessions. While the sample size is modest, it is comparable to many basic science investigations that use intensive neuroimaging protocols. The study combined behavioral measures, subjective confidence ratings, computational modeling, and multivariate neural analyses to probe how reinforcement learning interacts with metacognitive processes.
About this neuroscience research article
Source:
ATR Brain Information Communications Research Laboratory Group
Contacts:
April Toler – ATR Brain Information Communications Research Laboratory Group
Image Source:
The image is in the public domain.
Original Research: Open access
“Unconscious reinforcement learning of hidden brain states supported by confidence” by Aurelio Cortese, Hakwan Lau & Mitsuo Kawato. Nature Communications.
Abstract
Unconscious reinforcement learning of hidden brain states supported by confidence
Can humans be trained to make strategic use of latent representations in their own brains? The research investigates whether subjects can derive reward-maximizing choices from intrinsic high-dimensional information represented stochastically in neural activity. Reward contingencies were defined in real time by fMRI multivoxel patterns, so optimal actions depended on multidimensional brain activity below the threshold of consciousness. Subjects solved the task within roughly two hundred trials as reinforcement learning interacted with metacognitive functions, quantified by the meaningfulness of decision confidence. Computational modeling and multivariate analyses point to a frontostriatal mechanism: synchronization of confidence representations in prefrontal cortex with reward prediction errors in the basal ganglia supports exploration of latent task representations. These results open a new path for studying unconscious learning and the functions of metacognition.