Fastest communication speed using a Brain-Computer Interface

The fastest human communication speed enabled by a brain-computer interface (or BCI) is 78 words per minute, achieved by a volunteer called "Ann" with the help of a team of researchers at UC San Francisco in 2022. The details of the research project, which involved using a deep-learning AI to interpret electrical activity in the brain, was published in Nature on 23 Aug 2023.

Scientists have long dreamed of creating devices – "neuroprostheses" – that could be used to read and interpret neural activity, and to map it to specific intentions or actions. Such technology would enable, for example, the direct unconscious control of a prosthetic limb, or the transcription of ideas into language.

The primary obstacle to the implementation of this concept is the sheer complexity and variability of the electrical activity in the brain. Researchers have struggled to reliably associate these intricate patterns with specific actions using traditional software architectures.

Recently, two groups in California, USA – one at UC San Fransisco and one at Stanford University – have made significant advances using deep-learning networks to transcribe imagined speech into written language, which can then be displayed or spoken by a text-to-speech system.

In Aug 2023, both teams published their latest results in the same issue of <em>Nature</em>. The Stanford group, led by Frank Willett, achieved a communication speed of 62 words per minute, which was three times faster than the previous record. However, the UC San Francisco group, led by Edward Chang, managed an even quicker 78 words per minute. The two teams use quite different methodologies and equipment, with the most significant difference being the type of neural interface used – the Stanford group's approach uses implanted electrodes, while the UC San Francisco group uses a surface electrocorticogram.

With these advances, users of both experimental systems can communicate at about the same rate as an person typing on a keyboard, but are still short of the roughly 160 words-per-minute that people typically speak. The reliability of the systems also need work, as they're currently misunderstanding around 20% of the words users try to say.

The volunteer who tested the UCSF prototype was a woman identified only a "Ann", who suffered a debilitating stroke in her early 30s. For the last 18 years she has been using a eye-tracking pointer system to write things out. She hopes that the system will one day make it possible for her to begin a new career as a counselor.