Donner Lab

✨ This provides key new experimental constraints for mechanistic models of decision and confidence formation. Our results open a new window on the distributed neural dynamics underlying important subjective states such as confidence.

✨ In sum, for the first time, we simultaneously track competing neural decision variables in the frontal cortex and show that both of them predict decision confidence.

It fits the behavioral data better than a range of alternatives, and reproduces a number of behavioral and neural signatures, including the concurrently measured dynamics of the encoding of contrast samples in visual cortex.

Our model combines sensory adaptation, leaky accumulation with feed-forward inhibition, and an asymmetric readout of the DVs in confidence formation.

These observations, combined with several others, constrained a dynamical model of decision and confidence formation. The model identified the intrinsic correlation between the DVs as a signature of inhibition between the sensory-motor pathways supporting the two choices.

We next used single-trial regression to predict confidence from the winning and losing DVs, controlling for evidence strength. Both DVs made a significant, unique contribution to confidence. However, in contrast to standard models, the magnitude of the winning DV was stronger.

Despite the independent inputs, we found intrinsic negative correlations between the two DVs within trials, a signature of inhibition between the sensory-motor pathways supporting the two alternative choices. The strength of this correlation on each trial predicted confidence.

Our task removed that correlation at the level of the input. It required participants to track and compare the mean contrast of two sequences of grating contrasts. Critically, the contrasts of the two sequences fluctuated independently within each trial.

n typical tasks used to study evidence accumulation for decisions, evidence supporting one choice is, by construction, evidence against the alternative. The inputs to the two DVs are anti-correlated, which complicates the identification of competitive dynamics within the brain.

We tracked the competition between neural decision variables (DVs) in the human frontal cortex, combining a novel task with atlas-based multivariate decoding of source-level MEG data to track two neural DVs for alternative choices in left and right premotor/motor cortex (PMd/M1).

While the choice is only dictated by the winning neural population, theoretical considerations indicate that also (or only) the losing population should shape the internal sense of confidence associated with the decision.

⚖️ It has long been held that decisions result from a competition between neural populations encoding different choices. Yet, several technical challenges have so far precluded the direct observation of this so-called neural race🏃

Competing Neural Decision Variables in Human Frontal Cortex Shape Decision Confidence Mounting evidence indicates that decisions emerge from a competition between populations of neurons encoding the different choice options. Theoretical models propose that the outcome of this competiti...

🚨New preprint on @biorxivpreprint.bsky.social:
“Competing Neural Decision Variables in Human Frontal Cortex Shape Decision Confidence”
by Alessandro Toso, @ayeletarazi.bsky.social, @jrochav.bsky.social, @ktsetsos.bsky.social & Tobias H. Donner
🔗 www.biorxiv.org/content/10.1...

It fits the behavioral data better than a range of alternatives, and reproduces a number of behavioral and neural signatures, including the concurrently measured dynamics of the encoding of contrast samples in visual cortex.

Our model combines sensory adaptation, leaky accumulation with feed-forward inhibition, and an asymmetric readout of the DVs in confidence formation.

These observations, combined with several others, constrained a dynamical model of decision and confidence formation. The model identified the intrinsic correlation between the DVs as a signature of inhibition between the sensory-motor pathways supporting the two choices.

We next used single-trial regression to predict confidence from the winning and losing DVs, controlling for evidence strength. Both DVs made a significant, unique contribution to confidence. However, in contrast to standard models, the magnitude of the winning DV was stronger.

Despite the independent inputs, we found intrinsic negative correlations between the two DVs within trials, a signature of inhibition between the sensory-motor pathways supporting the two alternative choices. The strength of this correlation on each trial predicted confidence.

Our task removed that correlation at the level of the input. It required participants to track and compare the mean contrast of two sequences of grating contrasts. Critically, the contrasts of the two sequences fluctuated independently within each trial.

In typical tasks used to study evidence accumulation for decisions, evidence supporting one choice is, by construction, evidence against the alternative. The inputs to the two DVs are anti-correlated, which complicates the identification of competitive dynamics within the brain.

We tracked the competition between neural decision variables (DVs) in the human frontal cortex, combining a novel task with atlas-based multivariate decoding of source-level MEG data to track two neural DVs for alternative choices in left and right premotor/motor cortex (PMd/M1).

While the choice is only dictated by the winning neural population, theoretical considerations indicate that also (or only) the losing population should shape the internal sense of confidence associated with the decision.

⚖️ It has long been held that decisions result from a competition between neural populations encoding different choices. Yet, several technical challenges have so far precluded the direct observation of this so-called neural race🏃

✨ This provides key new experimental constraints for mechanistic models of decision and confidence formation. Our results open a new window on the distributed neural dynamics underlying important subjective states such as confidence.

✨ In sum, for the first time, we simultaneously track competing neural decision variables in the frontal cortex and show that both of them predict decision confidence.

It fits the behavioral data better than a range of alternatives, and reproduces a number of behavioral and neural signatures, including the concurrently measured dynamics of the encoding of contrast samples in visual cortex.

Our model combines sensory adaptation, leaky accumulation with feed-forward inhibition, and an asymmetric readout of the DVs in confidence formation.

These observations, combined with several others, constrained a dynamical model of decision and confidence formation. The model identified the intrinsic correlation between the DVs as a signature of inhibition between the sensory-motor pathways supporting the two choices.

We next used single-trial regression to predict confidence from the winning and losing DVs, controlling for evidence strength. Both DVs made a significant, unique contribution to confidence. However, in contrast to standard models, the magnitude of the winning DV was stronger.

Despite the independent inputs, we found intrinsic negative correlations between the two DVs within trials, a signature of inhibition between the sensory-motor pathways supporting the two alternative choices. The strength of this correlation on each trial predicted confidence.