Dissociating volatility and stochasticity reveals transdiagnostic computational signatures of psychopathology

This study identifies distinct transdiagnostic computational signatures of psychopathology by demonstrating that individuals with internalizing disorders tend to misinterpret environmental noise as volatility (over-updating), while those with externalizing disorders tend to misinterpret volatility as noise (under-updating), revealing that learning deficits in psychopathology are selective failures in distinguishing uncertainty types rather than generalized impairments.

The Gist

Imagine your brain is a driver trying to navigate a road where the rules keep changing. To drive well, you need to figure out two very different things:

1. The Road is Slippery (Stochasticity): Sometimes, the car bumps because of a random pothole or a gust of wind. This is just "noise." If you treat every bump as a sign that the road has changed, you'll overreact, swerving wildly and never getting anywhere. You need to ignore these small bumps and keep your speed steady.
2. The Road is Changing (Volatility): Sometimes, the road itself is being repaved, or a new detour is added. The rules have genuinely shifted. If you ignore these big changes and keep driving as if nothing happened, you'll crash. You need to speed up your learning and adapt immediately.

The problem is that both a random pothole and a new detour look the same to your eyes: they both make the car shake. It's very hard to tell the difference.

The Study's Discovery
This paper suggests that people's brains handle this confusion in three different ways, and these ways are linked to different types of mental health struggles:

• The Balanced Driver (Intact Learners): These people can tell the difference between a random bump and a real road change. They adjust their driving speed perfectly—slowing down when there's just noise, and speeding up when the road changes.
• The Over-Reactive Driver (Stochasticity-Blind): These people can't ignore the noise. They think every little bump means the road has changed. So, they constantly swerve and panic, trying to learn from every tiny mistake. The study found that people who drive this way tend to struggle more with Internalizing issues, like anxiety and depression. They are essentially "over-updating" their world view based on random noise.
• The Under-Reactive Driver (Volatility-Blind): These people are the opposite. They think the road is always the same, even when it's clearly changing. They ignore the big detours and keep driving straight, assuming the shaking is just a temporary glitch. The study found that people who drive this way tend to struggle more with Externalizing issues, like behavioral addictions and compulsivity. They are "under-updating," sticking to old habits even when the environment demands a change.

The Big Picture
The researchers tested this on thousands of people using computer games that simulated these tricky road conditions. They discovered a clear split:

• If you are prone to anxiety and depression, your brain might be too sensitive to random noise, making you change your mind too often.
• If you are prone to addiction or compulsive behaviors, your brain might be too stubborn, making you ignore real changes in your environment.

The main takeaway is that mental health struggles aren't just about "bad learning" in general. Instead, different problems come from specific, opposite mistakes in how we interpret uncertainty. Some people learn too fast because they can't filter out the noise, while others learn too slow because they can't see the real changes.

Technical Summary

Technical Summary: Dissociating Volatility and Stochasticity in Psychopathology

Problem Statement
Adaptive learning in dynamic environments requires the cognitive system to distinguish between two distinct sources of uncertainty: volatility (changes in the latent state of the environment) and stochasticity (moment-to-moment noise in observations). These two factors necessitate opposite adjustments to the learning rate: high volatility demands faster updating to track environmental shifts, whereas high stochasticity requires slower updating to filter out noise. The core computational challenge lies in disentangling these factors, as both inflate the experienced variance of observations. This ambiguity creates a fertile ground for systematic individual differences in inference, which the authors hypothesize may underlie specific mechanisms of psychopathology.

Methodology
The study employed a computational phenotyping approach across two large online samples and three distinct tasks. The researchers utilized a framework that models learning under uncertainty to identify specific computational signatures. They defined three distinct learner phenotypes based on how individuals infer uncertainty:

1. Intact learners: Those who correctly distinguish between volatility and stochasticity.
2. Stochasticity-blind learners: Individuals who fail to recognize noise, leading them to over-update their beliefs by treating random fluctuations as environmental changes.
3. Volatility-blind learners: Individuals who fail to recognize environmental shifts, leading them to under-update their beliefs by treating genuine changes as mere noise.

These phenotypes were then correlated with transdiagnostic psychiatric dimensions, specifically Internalizing (anxiety, depression) and Externalizing (behavioral addiction, compulsivity) spectra.

Key Contributions
The primary contribution of this work is the successful double dissociation between specific computational inference failures and transdiagnostic symptom dimensions. The study moves beyond a generalized view of learning deficits in mental health, proposing a selective account where distinct symptom clusters map onto distinct computational errors regarding uncertainty.

Results
The analysis revealed a clear divergence in computational profiles:

• Stochasticity-blind learners (who over-update) were significantly associated with higher scores on Internalizing dimensions, including anxiety and depression.
• Volatility-blind learners (who under-update) were significantly associated with higher scores on Externalizing dimensions, including behavioral addiction and compulsivity.

These findings indicate that the failure to correctly infer the source of variance is not a uniform deficit across psychopathology but is instead specific to the nature of the disorder.

Significance
The paper claims that these results support a selective rather than generalized account of learning-under-uncertainty deficits in psychopathology. By demonstrating that distinct symptom dimensions correspond to distinct failures of inference (specifically regarding volatility versus stochasticity), the study suggests that psychopathology may arise from specific computational maladaptations in how the brain processes uncertainty, rather than a global impairment in learning mechanisms.