In 2017, the team determined that the brain effectively operates in two different “memory states,” one that allows memories to be effectively stored, and another that does not. They also showed that they could flip the switch using direct stimulation of the brain, toggling between the “forgetting” and “remembering” brain states.
To accomplish this, the team first trained machine-learning classifiers to identify these distinct "memory states" from brain activity that predicted whether a word would be remembered or forgotten during learning. They then applied electrical stimulation in real-time based on these predictions, finding that stimulation delivered during predicted "poor" encoding states significantly improved memory recall. Conversely, stimulation applied during predicted "good" encoding states tended to disrupt memory, demonstrating that the effect of stimulation is highly dependent on the brain's momentary cognitive state.
