Brain recordings show goals are blended, not chosen one at a time

A team led from Baylor College of Medicine had people chase moving targets in a continuous prey-pursuit task and asked how the brain handles two goals at once. The answer, published in Nature on 15 August 2026 as “Neural basis of compositional control” (DOI 10.1038/s41586-026-10896-8), is that it does not pick one. Participants ran a blended controller — holding both movement plans simultaneously rather than switching — with a probability of 84% ± 8% across the group.

The study covered 69 adults: 50 healthy volunteers and 19 patients with severe epilepsy whose electrodes, implanted for clinical seizure monitoring, let the researchers record inside the brain while the task ran. Hippocampal activity tracked the current plan during early planning, the left anterior cingulate cortex signalled when to change strategy, and the orbitofrontal cortex encoded how much each option was worth. Attention split between competing targets 36% to 41% of the time when the two were equally valuable.

The finding bears on decoding. A brain-computer interface built on the assumption that intention is a discrete selection is reading for a state the recordings say is rarely occupied; a mixture is a different quantity to estimate.