Ultrasound neuromodulation reveals distinct roles of the dorsal anterior cingulate cortex and anterior insula in learning.

Pavlovian biases reflect how evolutionarily hard-wired tendencies-automatic approach toward reward cues and withdrawal from threat cues-can interfere with flexible, goal-directed action. Such biases arise through three mechanisms: (a) anticipated rewards energize action while anticipated punishments suppress it (response bias), (b) agents learn differently from actions than from inactions (learning bias), and (c) reward/punishment cues themselves drive repetitive behavior, independent of outcomes (perseveration bias). The neural origin of these biases is unclear. Past evidence suggests dorsal anterior cingulate cortex (dACC) and anterior insula (aIns) as part of a "reset network" that rapidly responds to salient information and might contribute to these biases. We used transcranial ultrasonic stimulation (TUS) in 29 healthy participants to interfere with neural activity in these regions and test their causal role in a within-subject, counter-balanced design across three sessions (sham, TUS-dACC, TUS-aIns). Computational modeling revealed a functional differentiation of both regions in Pavlovian biases: while TUS to either region did not affect the response bias, TUS to the aIns decreased people's learning bias, while TUS to dACC increased participants' perseveration bias. Although the dACC and aIns are part of the same network and often co-activate during decision-making tasks, TUS interference reveals their distinct roles: the dACC mediates cue-dependent persistence while the aIns is critical for inferring whether outcomes are self-caused.