BCI Designed to Decrypt Human Internal Speech

This piece, by Onno Berkan, was published on 9/11/24. The original text, by Miryam Naddaf, was published by Nature on 06/13/2024. 

Researchers from CalTech are in the early stages of creating a BCI (brain-computer interface) that is meant to “decode internal speech”, identifying what the user is thinking without them ever having to speak it out.

The researchers recorded invasively from the supramarginal gyrus (SMG) of two patients with spinal cord injuries, having the participants “imagine speaking” several real words and a couple control words (gibberish.) The BCI was then trained on this data and tried to figure out which word each patient was thinking about.

With one participant, the BCI was able to identify the word with a 79% accuracy rate. For the other, however, its accuracy dropped down to 23%, who’s representation of words like “spoon” and “swimming” were too similar for the BCI to tell apart. One explanation for this suggests that everyone’s SMG represents words differently, while a related explanation suggests that one’s ability to imagine the words, as well as their history with them, impacts how their brains light up when they think of the words.

This is not as big a step forward as one may imagine. First, the neurons that light up in the SMG are more or less the same that are activated when you speak. This, paired with how the experimental task was to essentially say a word without sounding it out, brings into question the extent to which these findings can be generalized to all of internal speech. Could this system, trained with enough words, be able to decode your thoughts as you have them or would it just be able to sound out words for those who can’t speak?

As it stands right now, this technology is geared towards helping those with speech impairments– people suffering from locked-in syndrome, for example. For now, we don’t know enough about internal speech to fully decode it. It is also unknown if this technology can be generalized to those who are unable to speak. Alas, researchers remain optimistic.

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. 

All submissions to berkan@usc.edu with the header “Byte Sized Submission” please. Thank you!