Summary: Researchers have discovered a distinct class of neurons — called prediction-error neurons — that remain inactive for ordinary sounds but activate specifically when an auditory outcome violates expectation. These cells signal that an anticipated sound did not match reality, providing a neural marker of unexpected auditory events.
In experiments with mice, neuroscientists at New York University observed neurons in the auditory cortex that are silent during normal, expected sounds but become active only when a sound deviates from what the animal predicted. The study shows these neurons do more than flag an error: different subsets encode different kinds of violations, such as an unexpected change in volume or timing. These findings have implications for understanding how the brain learns from mistakes and how auditory-driven skills like speech and music are acquired and refined.
Key Facts:
- Dedicated error signaling: Prediction-error neurons remain quiescent during expected sounds and respond only when expectations are broken, signaling an auditory prediction error.
- Stimulus-specific responses: Different groups of these neurons encode distinct types of violations — for example, one group responds when a sound is too quiet, while another responds when a sound arrives late.
- Learning and behavior: Because these neurons highlight mismatches between expectation and outcome, they may play a central role in trial-and-error learning for behaviors that rely on sound, such as speaking or playing musical instruments.
Source: NYU
From everyday errors to neural signals: When a car door does not close with the expected thump or a kicked soccer ball misses the target, the sounds we hear differ from what we anticipated. The brain detects these mismatches, and this study explores whether there are neurons whose sole role is to report that a sound is “off.” The NYU team identifies such cells and shows how they encode specific types of auditory errors.
“Brains are excellent at detecting events in the world, and even better at judging whether those events match expectations,” says David Schneider, assistant professor at NYU’s Center for Neural Science and senior author of the study published in JNeurosci. Rather than reporting the identity of a sound, the prediction-error neurons communicate that the predicted sensory outcome was incorrect.
Lead author Nicholas Audette, a postdoctoral fellow at NYU’s Center for Neural Science, notes that these neurons could be fundamental to learning by error: “Speaking and playing an instrument require extensive trial and error, repeated mistakes, and learning from those mistakes. Neurons that mark when auditory expectations fail could guide that learning.” Schneider adds that future work might test whether experts, such as accomplished musicians, show enhanced or more precise prediction-error neuron responses compared with novices, and whether deficits in these neurons relate to developmental speech disorders.
Animals, including mice and primates, learn to predict the sounds their actions will produce and the timing of those sounds. Earlier studies showed that sensory cortex responses are enhanced when self-generated sounds violate expectation and reduced when sounds match expectation. What remained unclear was whether neurons exist that act exclusively as prediction-error detectors and whether those detectors carry general error signals or encode specific kinds of violations.
To investigate, the researchers trained mice to associate a specific sound with pressing a lever. After the association formed, the team introduced deviations in the sound’s properties — variations in volume, timing, or identity — mirroring the kinds of unexpected auditory outcomes animals (and humans) experience when actions don’t go as planned. The authors recorded neuronal activity in the auditory cortex and analyzed how cells responded to expected versus unexpected self-generated sounds.
The results reveal a distinct population of neurons that are largely silent during passive listening and when the expected sound occurs, yet become active when a learned expectation is violated. Importantly, these prediction-error neurons are stimulus-specific: many encode one or two particular dimensions of mismatch rather than signaling a generic error. For example, one group consistently responded to unexpectedly low-volume sounds, while another group responded when the timing of the sound was off.
These findings indicate that the auditory cortex suppresses responses to predictable self-generated sounds across multiple acoustic dimensions and that cortical prediction-error neurons encode specific, not generic, violations of expectation. The specificity of these error signals suggests a nuanced neural mechanism that can inform how the brain corrects behavior and refines sensory-motor skills.
Funding: This research was supported by the National Institutes of Health (R01-DC018802).
About this neuroscience research news
Author: James Devitt
Source: NYU
Contact: James Devitt – NYU
Image credit: Neuroscience News
Original Research: Closed access. “Stimulus-specific prediction error neurons in mouse auditory cortex” by David Schneider et al., Journal of Neuroscience.
Abstract
Stimulus-specific prediction error neurons in mouse auditory cortex
Comparing expectation with sensory experience is a core neural computation and central to predictive processing. When an animal alters the sensory outcome of a behavior, neurons in sensory cortex often show enhanced responses to unexpected self-generated stimuli, suggesting the presence of prediction error signals. However, such enhancements could also arise from non-predictive factors like movement-related amplification of inherent sound responsiveness.
To determine whether true sensory prediction error neurons exist and whether their responses are general or stimulus-specific, researchers trained mice to expect a particular sound following a simple action and recorded auditory cortex activity while presenting either the expected sound or sounds that differed in distinct acoustic dimensions. The data show that auditory cortex learns to suppress responses to predictable self-generated sounds across multiple dimensions. The study identifies a distinct population of neurons that are unresponsive to passive sounds and the expected outcome but that encode prediction errors. These neurons appear only after animals acquire a learned motor-sensory expectation and typically signal one or two specific violations rather than a broad error signal, demonstrating that cortical prediction errors can carry dimension-specific information.