Human memory is notoriously fallible; it often fails the moment we need it most. Even though we’re drenched in a nonstop stream of sensory experience, what most of us manage to actually retain is spotty at best. To supplement to the details that slip our minds, we’ve learned to rely heavily on technology. From the printing press to the internet, we’ve taken great lengths to ensure our raw knowledge is preserved in an exterior form, as accurately as the tools at our disposal allow. However, new research suggests machines are capable of much more—that they now hold the key to optimized knowledge storing not only externally, but within the mind as well.
The practical applications of machine learning may soon extend deeper into the psyche than targeted Amazon ads, or so argues Michael Kahana, psychologist at the University of Pennsylvania. Kahana’s research team recently managed to demonstrate that machine learning algorithms—when coupled with precisely-timed electric bursts to the brain—can recognize memory patterns and even improve recollection.
Kahana’s study was done in collaboration with 25 epilepsy patients. Each was previously equipped with between 100 and 200 electrodes used to track seizure-related brain activity. Patients were asked to complete a series of tests in which they attempted to memorize and recall lists of words. During the tasks, Kahana analyzed patients’ brain voltage readings, and built a data set of neural patterns related to the acts of remembering and forgetting.
After a few rounds of memorization and retention tests, Kahana compiled the data, and produced a set of algorithms capable of predicting whether a word would be remembered. The tests were then run again, only now the algorithm informed the electrodes not only to record brain activity, but stimulate it—via electric signals—at the opportune time, in an attempt to trigger memory formation. As a result, patients’ memory retention improved by an average of 15%.
These findings seem to solve an issue Kahana encountered during a previous, similar memory study, which found that subjects performed better when stimulated during periods of low brain functionality. Those findings were limited in a practical sense, however, as they couldn’t link the brain’s state to improved memory until after the tests were performed. Now, with the addition of algorithms, Kahana hopes to create a type of “decoder” capable of determining in real time whether the brain is in a state beneficial to learning, and delivering stimulation if it isn’t.
Future studies using a greater number of carefully placed electrodes, as well as a greater training volume for the algorithm, might improve the effectiveness of Kahana’s brain-boosting technology. Until then, Kahara and like-minded scientists are left to wrestle with the fact that as much as we don’t understand the nuances of memory formation, we also don’t fully grasp how sending electric bursts to the brain stimulates remembrance, or how exactly an algorithm discerns which neural patterns are most conducive to memory.