 |
| Facebook is abandoning the idea of writing with thought |
A Facebook-sponsored project that enables people to think and write ends with a new published result.
Project Steno is a multi-year collaboration between Facebook and Chang Lab at the University of California, San Francisco. The goal is to create a system that converts brain activity into words.
A new research paper published in the New England Journal of Medicine shows that the technology can be applied to people with speech disabilities.
But over the course of the research, Facebook has made it clear that it is abandoning the idea of commercial head-mounted brain-readers and instead building a wearable interface on the wrist.
The new research has no clear applicability to mass market technology products. In a press release, Facebook said it's turning its attention to head-mounted brain-computer interfaces.
"Facebook is not interested in developing products that require implanted electrodes," the press release said. She said she still believed in the long-term possibilities of head-mounted optical computer interface technology, but decided to focus on a different neural interface method with a short-term market path.
Chang Lab's ongoing research involves using implanted brain-computer interfaces to restore people's language skills.
This new article focuses on a participant who lost the ability to speak after suffering a stroke more than 16 years ago. The lab provided the man with implanted electrodes that can detect brain activity. Then the man trained a system to recognize certain patterns for 22 hours.
In this exercise, try to read individual words from a vocabulary of 50 words. In another chapter, try to use these words to form complete sentences, including basic verbs and pronouns, as well as specific nouns and useful commands.
This training helped create a language model that is able to respond when a person wants to utter certain words even when they can't.
Facebook is abandoning the idea of mental writing
The researcher modified the model to predict which of the 50 words he meant. It includes a typical English probability system similar to the predictions of smartphone keyboards.
The researchers reported that in the latest experiment, the system could decode an average of 15.2 wpm (allowing for errors) or it could properly use decoded words to decode 12.5 wpm.
Previous research by Chang Lab on the Steno project, published in 2019 and 2020, showed that electrode arrays and predictive models can create relatively fast and complex intelligent writing systems.
Most of the previous input options involved using brain implants to mentally move the cursor on an on-screen keyboard. Although other researchers have experimented with visual handwriting and other methods.
Previous laboratory research has involved decoding the brain activity of people who speak normally. This article shows that it can work even if people aren't talking out loud.
Research today is invaluable for those not served by existing consoles and other interfaces. Having a limited vocabulary can help them communicate more easily.
But it fell short of Facebook's ambitious goals in 2017. A non-invasive brain-computer interface system that enables people to type 100 words per minute. This is similar to the maximum speed you can get with a traditional keyboard.
And Facebook will not abandon its entire head-mounted brain interface system. But it plans to open up the software and share hardware samples with outside researchers while closing its own research.