Eight years ago, a patient suffered from Lou Gehrig's Disease, which causes progressive paralysis. Although she can still make sounds, her words are no longer understandable. She is now dependent on an iPad or writing board to communicate.
After volunteering to have a brain implant placed, the woman is now able to communicate phrases such as "I don’t own my house" and "It’s just difficult" with a speed that is comparable to normal speech.
This is the claim made in a paper by Stanford University researchers that was published on bioRxiv over the weekend. Other researchers have not reviewed the study. According to the scientists, the volunteer, named only as "subject 12", broke all previous records using the brain-reading device to communicate at a speed of 62 words per minute, three times faster than the previous record.
Philip Sabes, an independent researcher at the University of California San Francisco, called the results "big breakthroughs" and stated that the experimental brain-reading technology could soon be available for use in the real world.
Sabes says, "The performance in the paper is already at an level that many people who are unable to speak would like, if it were available." This is what people want.
People with speech impairments usually speak at 160 words per minute. Even with keyboards, thumb-typing and emojis, speech is still the fastest form human-to-human communication.
Stanford University conducted the new research. The preprint was published on January 21 and has been gaining more attention on Twitter and other social networks due to the death of Krishna Shenoy (pancreatic cancer) this week.
Shenoy spent his entire career working to improve the speed of communication via brain interfaces. He meticulously maintained a list on his laboratory website. Shenoy was able to use his thoughts to type 18 words per minute in 2019, a record performance for the time. This feat was documented in MIT Technology Review's special issue about computing.
Shenoy's brain-computer interfaces are based on a small, sharp electrode embedded in the motor cortex of a person. This is the brain area most involved with movement. Researchers can record activity from several dozen neurons simultaneously and identify patterns that correspond to the motions a person is thinking about, even if they are paralyzed.
Paralyzed volunteers were asked to visualize making hand movements in previous research. Implants "decode" the neural signals of paralyzed volunteers in real-time, allowing them to steer a cursor, select letters from a virtual keyboard, control a robot arm, and even play video games.
The Stanford researchers wanted to find out if motor cortex neurons contained information that could be used to detect speech movements. They wanted to know if they could detect whether "subject T12", a woman trying to speak, was moving her tongue, mouth and vocal chords.
These movements are subtle and small. Sabes says that a handful of neurons can be used to predict the words a patient is trying to say with high accuracy. Shenoy's team transmitted this information to a computer screen where the patient's words were displayed as they were spoken by it.
This new research builds upon previous work done by Edward Chang, University of California, San Francisco. Chang has previously written that speech is one of the most complex movements humans make. The air is pushed out, vibrated to make it audible and then formed into words using our lips, tongue, and mouth. You can make the sound "f" by placing your top teeth on your lower lips and pushing air out. This is just one of many mouth movements required to speak.
The path to success
Chang used electrodes on top of his brain in the past to allow a volunteer to communicate with a computer. However, the Stanford researchers claim their system is better and up to four times faster.
"Our results show that there is a way forward to restore communication to paralysis patients at conversational speeds," said Shenoy and Jaimie Henderson, both neurosurgeons.
David Moses, who is part of Chang's UCSF team, said that the current work reaches "impressively new performance benchmarks". However, he added, "it will be increasingly important to demonstrate reliable and stable performance over multi-year times scales." A commercial brain implant might have trouble getting past regulators. This is especially true if the implant degrades over time, or if accuracy falls off.
[TAG0]
Future developments will likely include more advanced implants and better integration with artificial intelligence.
The current system already employs a few types of machine-learning programs. The Stanford team used software to predict the next word in a sentence. This improved its accuracy. "I" is often followed by "am", even though the words sound similar, and could create similar patterns in someone’s brain.
The subject's ability to speak quickly without making mistakes was increased by adding the word prediction system.
Language models
Newer language models like GPT-3 can write entire essays and answer questions. These models can be connected to brain interfaces to enable users to speak faster because it will be able to guess what they want to say based on partial information. Sabes says that the success of large language models in recent years suggests that a speech prosthesis may be possible.
Shenoy's team is part of BrainGate, which has implanted electrodes in the brains of over a dozen volunteers. The Utah Array is a rigid square of metal with approximately 100 needle-like electrodes.
Some companies, such as Neuralink, Elon Musk's brain-interface company, and Paradromics, claim they have created more advanced interfaces that can record thousands, if not tens of millions, of neurons simultaneously.
Some skeptics wonder if measuring more neurons at once will make a difference. However, the new report indicates that it will, particularly if brain-reading complex movements like speech is the goal.
Stanford scientists discovered that understanding "T12" meant making fewer mistakes if they had access to more neurons at once.
"This is a huge deal because it suggests that efforts by companies such as Neuralink to place 1,000 electrodes in the brain will make an impact, if the task's sufficiently rich," said Sabes, who was previously a senior scientist at Neuralink.
————————————————————————————————————————————————————————————
By: Antonio Regalado
Title: An ALS patient set a record for communicating via a brain implant: 62 words per minute
Sourced From: www.technologyreview.com/2023/01/24/1067226/an-als-patient-set-a-record-for-communicating-via-a-brain-implant-62-words-per-minute/
Published Date: Tue, 24 Jan 2023 17:38:50 +0000
Leave a Reply