Resurrecting Speech: AI and Brain Implant Aid Paralyzed Woman's Communication

0:00The first thing I want to talk about here, and I guess the really big headline is that essentially this is a very remarkable advancement in the field of neuro technology. So researchers from the University of California, San Francisco, UCSF, and the University of California, Berkeley, along with a team from Stanford University, have developed separate yet transformative systems that allow individuals with speech impairment conditions to communicate. So these developments were all published in the scientific journal Nature. And I really think that these bring on a whole new era in the human machine interface that could very drastically improve the quality of life for many, many people that are suffering from paralysis or conditions like Lou Ginger's disease, ALS, and a lot of other ailments.

0:47And I think that AI is able to do some really incredible things here. So kind of going on the backstory about this one woman in particular, back in 2005, Ann Johnson was left severely paralyzed and speechless after she had a stroke. Now, this is a condition that can happen after strokes. So despite the inability to vocalize, her brain continues to obviously generate language signals. signals. So UCSF and Berkeley researchers implanted a device on the surface of her brain, specifically in the regions associated with speech and language. So the implant featured 253 electrodes that kind of just captured all of the brain signals, which were then transferred to a computer bank via a port and a cable installed on her head.

1:34AI algorithms then translate these signals into sentences articulated by a digital avatar, right? So they created like a digital avatar of her speaking and just based off of her, the brain signals that they could pull and the AI decoding it. Now they have this digital avatar of herself, which is talking absolutely incredible. So Dr. Edward Chang, who's the chair of neurological surgery at UCSF and is the author of the study, in particular, when it got published, expressed his super, super, he was super happy with the results. And he said, quote, there's nothing that can convey how satisfied it is to see something like this actually work in real time.

2:15I mean, honestly, like, think about this, it had been 18 years since this stroke happened. So this is absolutely incredible. The technology essentially translated Johnson's brain signals into speech at a rate of nearly 80 words per minute, with a median accuracy of around 75 when using a vocabulary of 1024 words. So to give the avatar a more kind of personalized touch, the system actually used a recording of her voice from her wedding and incorporated the facial movements and emotional expressions. So it was actually able to use her voice, right? So using, there's all sorts of technology, there's 11 labs and all sorts of others that you can train an AI clone of your voice.

2:58And while a lot of people talk about like Like some of the downsides are like, oh no, well, you know, what if hackers or bad actors train AI model on your voice and try to trick people that you know into over the phone to do things, right? Like there's all sorts of like bad use cases. This is an incredible use case. She was able to clone her voice and is able to speak again thanks to this technology. So in parallel, a Stanford University study showcased some really similar technology that helped Pat Bennett. that's a woman with ALS, and Bennett's system involved implanting two small sensors on her brain, so enabling a software program to decode signals and convert them into words on a computer screen.

3:37This technology achieved a rate of 62 words per minute and was about 91 % accurate when Bennett used a 50-word vocabulary. So I will say the difference here is, of course, this one is a little bit slower, a little bit more accurate. But also, I think it's probably the accuracy is just due to the fact that there's only 50 words in the vocabulary of this accurate one, whereas the other one had over 1000 words. So I mean, if it was me, I think going with the 75 % accuracy on 1000 words, in my opinion, would have been would have been better than 90 % 91 % accuracy on just 50 words. 50 words seems like not a lot, but maybe I just have a lot of words to express what I'm saying.

4:19So, I mean, it's up to everyone. But anyways, both of these are incredible technologies. Both systems, I think, mark a really significant advancement over previous attempts at similar technologies. So they had Francis Willett was the lead author of the Stanford study. And Francis said, it is now possible to imagine a future where we could restore fluent conversation to someone with paralysis. So I will say, however, that these technologies aren't without limitations. editorial commentary alongside the studies pointed out that the systems currently demand the expertise of very highly skilled researchers for operation and maintenance and moreover both johnson and bennett have some residual facial movement leaving questions about kind of the effectiveness of this technology for individuals without any muscle control um so dr jamie henderson who is a neurosurgery professor at stanford acknowledged that the constraint essentially just acknowledge those constraints there, but also kind of emphasize the potential for ongoing improvements in here, right?

5:22I think we can see a lot of room for this to improve and grow. I think what's really cool here is seeing a lot of different research teams at different places working on the same technology and coming up with results, right? Albeit they're slightly different, right? Some of them have higher accuracy rates or higher word counts or all sorts of things. But I think like honestly, developing these things all kind of independent, finding out really great use cases for this eventually the you know the best models will all converge um and we're going to build something really incredible but this is a major problem and this is incredible the way they're trying to solve this so i think as the technology continues to evolve it really paves the way for more adaptable efficient and possibly even wireless communication solutions for these uh and for people that are unable to speak right because currently this is like i'm implanting your brain if you look at the pictures there's a wire coming straight out of her head into the system.

6:13And so being able to do this wireless, I think will be a big deal. I think probably what Neuralink and others are doing is going to also be able to help in a lot of these cases. So for Johnson, I think this technology is going to make her aspirations of becoming a counselor a reality. And for Bennett, the chance to stay connected with the world becomes more attainable. I think despite the hurdles that obviously are going to lie ahead, I think these groundbreaking studies point to a future where barriers to human communication can be significantly lowered if not entirely eliminated and this is something that is very very exciting for me as I'm following this field.