📄 other

Sensory Noise Traits Predict Illusion Susceptibility and Multisensory Mismatch Negativity

This study demonstrates that individual differences in sensory noise traits and common-cause priors, derived from a novel bidirectional audiovisual rabbit task and Bayesian causal inference modeling, predict both behavioral susceptibility to multisensory illusions and specific neural signatures of multisensory mismatch processing in the occipital cortex.

Original authors: Hexin Xu, Amit Yaron, Yiyuan Teresa Huang, Zenas C. Chao, Tomoyo Isoguchi Shiramatsu, Hirokazu Takahashi

Published 2026-07-25

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Hexin Xu, Amit Yaron, Yiyuan Teresa Huang, Zenas C. Chao, Tomoyo Isoguchi Shiramatsu, Hirokazu Takahashi

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your brain as a super-smart detective trying to solve the mystery of what's happening in the world. Every second, your eyes, ears, and skin send in a flood of clues. Sometimes these clues are crystal clear, but other times they are fuzzy, like a radio station with static or a photo taken in the dark. To make sense of this messy data, your brain has to guess: "Are these two clues coming from the same source, or are they just two different things happening at once?" This process is called multisensory integration. It's how your brain decides if a sound and a flash of light belong together (like a firework exploding) or if they are unrelated (like a car honking while a bird chirps).

Scientists have long known that people are different at this detective work. Some people are easily tricked by illusions where a sound makes them see something that isn't there, while others see the world exactly as it is. But for a long time, we didn't know why. Was it because their ears were less precise? Because their eyes were blurry? Or because their brain just had a different "rulebook" for deciding what counts as a match? This new research dives into that mystery, using a mix of brain games and brain scans to figure out if we can measure these hidden "detective skills" and see how they show up in the brain's electrical signals.

The Great Rabbit Illusion Game

In this study, researchers from the University of Tokyo set up a playful experiment to trick people's brains. They used a classic illusion called the "audiovisual rabbit." Imagine you see two flashes of light on a screen, but you hear three beeps. Your brain, trying to make sense of the mismatch, might convince you that you saw a third flash in the middle, even though it wasn't there! It's like your brain drawing a rabbit hopping between two spots just because the sound suggested it.

But here's the twist: the researchers didn't just look at who got tricked. They wanted to know why. To do this, they created a "bidirectional" version of the game. Sometimes the sound would trick the vision, and sometimes the vision would trick the sound. They also added white noise to make the beeps harder to hear, turning up the "static" on the radio. By watching how 28 people played this game, the researchers used a fancy math model (called Bayesian Causal Inference) to break down each person's brain into four secret ingredients:

Auditory Noise: How fuzzy their hearing is.
Visual Noise: How fuzzy their vision is.
Sound-Level Sensitivity: How much their hearing gets better when the sound gets louder.
Common-Cause Prior: How much a person expects sounds and sights to come from the same source.

The Detective's Report: What They Found

The researchers discovered that these four ingredients perfectly explained why some people were more easily fooled by the rabbit illusion than others. If your "auditory noise" was high (your hearing was fuzzy), you were more likely to let the sound trick your eyes. If your "common-cause prior" was strong (you really believed everything was connected), you were more likely to see the whole illusion.

But the real magic happened next. The researchers wanted to see if these "detective skills" showed up in the brain's electrical activity. They hooked 15 of the participants up to EEG caps (which look like swim caps with wires) and played a different game: a passive oddball task. In this game, standard sounds and lights flashed on and off, but occasionally, a "deviant" (a weird mismatch) would appear. The brain usually reacts to these mismatches with a tiny electrical spike called Mismatch Negativity (MMN). Think of MMN as the brain's "Wait a minute!" signal.

Here is the big discovery: The researchers found a direct link between the "fuzziness" of a person's hearing and their brain's reaction to the mismatch.

People with higher auditory noise (fuzzier hearing) showed a weaker "Wait a minute!" signal in the back of their brain (the occipital lobe, which handles vision) when they saw the audiovisual mismatch.
Specifically, the correlation was strong: the fuzzier the hearing, the less negative the brain wave was (meaning a weaker reaction).
This link was specific. It only happened in the back of the brain during the combined sound-and-light task. It didn't happen for just sounds alone, just lights alone, or in the front of the brain.

What This Means (And What It Doesn't)

This study suggests that the way your brain handles "fuzzy" hearing changes how your visual part of the brain reacts when things don't match up. It's like saying that if your microphone is staticky, your brain's video editor gets confused and doesn't flag the error as strongly.

However, the authors are careful not to overhype this. They note that their group of people with brain scans was small (only 15 people), so while the link is strong and survived statistical checks, it needs to be tested on more people to be absolutely sure. They also point out that this doesn't mean high auditory noise causes the weak brain signal in a simple way; it just suggests they are connected.

Crucially, the study rules out the idea that these differences are just random mistakes. Instead, they seem to be stable "traits" of how a person's brain works. The research also shows that looking at just one number (like "how many illusions did you see?") isn't enough. You have to dig deeper to find the specific ingredients—like auditory noise—that make up that number.

In short, this paper suggests that our individual differences in being tricked by illusions aren't just random quirks. They are rooted in specific, measurable traits of our sensory systems, and these traits leave a fingerprint on our brain waves. It's a step toward understanding that everyone's brain is a unique detective, with its own set of tools and rules for solving the mystery of the world.

Technical Summary: Sensory Noise Traits Predict Illusion Susceptibility and Multisensory Mismatch Negativity

Problem Statement
While individual differences in susceptibility to audiovisual (AV) illusions are well-documented, the computational origins of these variations remain unclear. Standard AV illusion paradigms typically measure susceptibility via a single behavioral index, which confounds two distinct components: sensory precision (noise) and causal inference (the tendency to bind signals). Consequently, similar overt illusion rates may arise from different latent combinations of sensory reliability and binding tendencies. Furthermore, it is unknown whether these latent computational traits can predict neural signatures of multisensory processing, specifically the mismatch negativity (MMN), a component of event-related potentials (ERPs) interpreted as a precision-weighted prediction-error signal.

Methodology
The study employed a two-stage experimental design involving a behavioral task and a passive EEG oddball paradigm.

Participants: The study included 28 healthy adults for the behavioral experiment. Of these, 26 were retained for Bayesian Causal Inference (BCI) modeling, and a subset of 15 completed the EEG session.
Bidirectional Audiovisual Rabbit Task: Participants performed a task where sequences of beeps and flashes created illusory intermediate flashes or suppressed real ones. The task was "bidirectional," allowing auditory signals to capture visual perception and vice versa. Crucially, auditory signal reliability was manipulated by varying sound volume (0–40 dB above hearing level) and introducing white noise. This design allowed for the dissociation of sensory noise from causal inference.
Computational Modeling: Perceptual reports were modeled using a Bayesian Causal Inference (BCI) framework. Four model families were compared: segregation, forced fusion, classical BCI, and postdictive BCI. The postdictive BCI model was selected as the best fit based on Pareto-smoothed importance-sampling leave-one-out cross-validation (PSIS-LOO). This model estimated four latent traits for each participant:
- Auditory noise ( $\sigma_a$ )
- Visual noise ( $\sigma_v$ )
- Sound-level sensitivity of auditory noise ( $\sigma_{a,level}$ )
- Common-cause prior ( $p_{common}$ )
EEG Paradigm: A passive oddball paradigm was used to record MMN responses. Conditions included unimodal auditory, unimodal visual, and congruent audiovisual oddballs. EEG data were recorded using a 64-channel cap, transformed to Current Source Density (CSD) to improve spatial specificity, and analyzed for MMN amplitudes in frontocentral (auditory) and occipital (visual) regions of interest (ROIs).
Analysis: Correlations were computed between the BCI-derived latent traits and (a) behavioral illusion intensity measures and (b) MMN amplitudes in the EEG subsample. Multiple comparison corrections (BH-FDR) were applied.

Key Results

Model Selection: The postdictive BCI model significantly outperformed segregation, forced fusion, and classical BCI models in predicting behavioral reports, validating the use of postdictive causal inference for the rabbit illusion.
Behavioral Correlates: In the behavioral sample ( $N=26$ $N = 26$ ), the latent traits robustly predicted illusion susceptibility:
- Higher auditory noise ( $\sigma_a$ ) correlated with greater auditory illusion intensity.
- Higher visual noise ( $\sigma_v$ ) correlated with greater visual illusion intensity.
- A stronger common-cause prior ( $p_{common}$ ) correlated with greater total illusion intensity.
- Sound-level sensitivity showed a weaker, less robust association.
Neural Correlates (EEG): In the EEG subsample ( $N=15$ $N = 15$ ), a specific cross-paradigm association was found:
- Auditory noise ( $\sigma_a$ ) positively predicted occipital MMN amplitude in the congruent audiovisual condition (AVOC). Since MMN is a negative-going potential, higher auditory noise was associated with less negative (weaker) occipital mismatch responses.
- This association was specific to the occipital component of the audiovisual response. No significant correlations were found between auditory noise and frontocentral audiovisual MMN (AVFC), unimodal auditory MMN, or unimodal visual MMN.
- Visual noise ( $\sigma_v$ ) did not show comparable associations with audiovisual MMN.
- The finding remained robust across different EEG reference schemes (FCz, average, mastoid, and CSD) and was confirmed by nonparametric permutation tests.

Key Contributions

Computational Decomposition: The study successfully dissociated sensory noise from causal inference in AV illusion susceptibility using a bidirectional rabbit task and a postdictive BCI model, moving beyond descriptive illusion rates to interpretable latent traits.
Cross-Paradigm Prediction: It demonstrated that behavior-derived sensory-noise traits (specifically auditory noise) can predict neural markers (MMN) in a completely different, passive task, suggesting these traits capture stable individual differences in multisensory processing.
Specificity of Neural Mechanisms: The findings revealed that the link between auditory precision and neural mismatch processing is specific to the occipital (visual) component of the multisensory response, rather than a general reduction in neural responsiveness or a unimodal effect.

Significance and Claims
The authors claim that these findings support a computational approach where individual differences in multisensory perception are not random but reflect stable, trait-like variations in sensory precision and binding tendencies. Specifically, the study suggests that auditory precision contributes to individual variability in multisensory mismatch processing.

Within a predictive-coding framework, the authors interpret the result as follows: if auditory input helps establish predictions about audiovisual events, reduced auditory precision (higher noise) may weaken the cross-modal predictive signal available to visual cortical processing, resulting in a weaker occipital mismatch response to deviant audiovisual events.

The paper modestly concludes that while these results do not establish clinical utility or prove strict mechanistic causality, they provide a framework for linking computationally defined sensory-noise traits to neural measures. The authors note limitations, including the modest EEG sample size and the need for independent replication, but emphasize that the strategy of decomposing illusion susceptibility into latent traits offers a principled way to relate behavior to neural signatures of multisensory integration.

The Great Rabbit Illusion Game

The Detective's Report: What They Found

What This Means (And What It Doesn't)

More like this