In a recent interview Mark Zuckerberg said he is confident that withing five years you will be able to send messages with your mind seamlessly.
Link to the interview: https://youtu.be/5wk5WPONF7c
For this market to resolve true Meta must claim that they can:
Type and interact with a computer of any form (phone, PC, AR, VR etc) using their brain-computer interface.
The typing must be around the average speed of typing of 40 words per minute. (I'll count anything greater than 35 words per minutes).
The brain-computer interface must be non-invasive.
This can happen in a lab environment with equipment that is not yet consumer friendly.
New case study (not peer-reviewed yet), presenting the results from an invasive (“four intracortical microelectrode arrays”), and hardly seamless (~26 days of multi-hour work to create person-specific data for RNN training), but very promising for motor-impaired people, approach:
23.8% word error rate on a 125,000 word vocabulary [… and] 62 words per minute, which is 3.4 times faster than the prior record for any kind of BCI
https://twitter.com/WillettNeuro/status/1617245600898248704
Before you read this as me warming up to Zuckerberg’s claim, I believe this schematic is accurate enough in representing the difference in data quality between intracortical microelectrode arrays (MEA) and scalp EEG, and that it is very difficult to improve.
AFAICT that’s two/three orders of magnitude in throughput more than ca. 2020 academic results (from a few to a few thousand bits per minute, predictive typing might cut that down a little). If it happens in time, I doubt it’ll be on a consumer-friendly device, rather still something like a goofy whole-scalp EEG cap with careful gel and electrode application.