Decodificadores cerebrales semánticos 101: lo que la decodificación cerebral significa para su privacidad

Transcrito automáticamente de la transmisión en vivo (11,806 palabras). Las marcas de tiempo abren el video en ese momento. Es posible que haya errores menores de transcripción.

andx.com. Let's go live on Tik Tok. Um, hello, hello, hello, Tik Tok. Welcome to the evolutional semantic brain recorders. Everybody, welcome. We're here. If you have any questions, don't hesitate to just uh put them down here so we can uh address your questions. And basically uh we're going to talk about uh the evolution of the semantic brain decoders. Um, I've been used to to utilize this technology against my will. And this is one of the reasons why I'm just uh creating this series of uh videos so everybody knows what's going on in this country. Okay.

Again, this has been done without my consent. Uh, the evolution of semantic brain decoders. uh and basically I'm going to talk about the foundation of brain decoding and neuro signaling and the introduction to brain decoding the concept and historic background that's the first part that I'm going to cover uh so let's go good morning good afternoon uh about half an hour in Miami talking about semantic brain decorder so welcome to today's uh class here on small seminar on semanticing brain decorders. Today we're going to dive in one of the most fascinating areas of modern neuros science and artificial intelligence AI, the decoding of human thoughts and intentions. And to me this is a very very very um important topic because we today currently we have uh people that they are decoding your thoughts and your intentions without your consent and nobody's doing nothing because there's no laws. this is new technology and some and most of the time um the way that the politics system United States and around the world is like we tend to vote for um old people.

Okay. And old people are behind on technology. So these old folks that are in the Congress uh know about some information but they have not put the laws to protect you. Okay. So, so if you get get this information, talk with your uh representatives, talk with your congressmen and let them know what's going on. Okay? Uh again, uh we have uh people in the law enforcement utilizing the coding human thoughts and your intentions. Um as I going to continue to talking about this I am going to uh disclose some information based on my current experience with this technology.

Okay. [snorts] Uh so um I've been with

this technology already for five five years uh to my knowledge. Uh so let's go ahead and start working uh and giving you the concepts of history so you understand the technology and then we're going to move forward and we're going to disclose everything like where how and where because in porquet okay so the brain decoding refers to techniques that interpret brain signals [snorts] and convert them into meaningful outputs so such words actions or images uh the interd disciplinary field has seen remarkable progress particularly with the advent of machine learning and functional neurom imagining technologies like fMRI. I am going to talk about what is a fMRI in a little bit. Okay.

the journey into brain computer interfaces. Learn about this acronym because throughout the next two hours uh or hour and a half we're going to be talking about brain computer interfaces and there that's called BCIS. Okay, BCIS began uh around a decade ago. Okay, that's more than 10 years. Okay. Uh with the goal of assisting people with sever motor impairments, allowing them to control prosthetics or even communicate with decoding motors intentions. But over the last 10 years, we have shifted from decoding simple motor actions to interpreting more complex cognitive content such as language, emotions, and abstract thoughts.

The uh the first wave of BCI's brain computer interfaces focus on direct brain interfaces to control external devices like robotics arms or neuroprothetics using electrodes implanted into the brain. For example, neuro decoding of motor intentions was a key focus allowing people with paralysis to operate technology with their minds. This work extensively published in neuroscience journals like the journal of neuroscyence of 2021. Okay. Uh the way for the next evolution semantic decoding. So if you are uh want to find more information okay go to the neuros the journal of neuroscience okay which you're going to find there a lot of technologies and a lot of u truthful documentation about semantic brain decoders and anything related with your brain. Okay.

Uh the right the the rise of semantic brain decoders. Okay. So, so before I I jump to the second uh part of this, so you understand basically

uh let's say that you this this technology was created for this purpose around 17 to 20 years a 20 20 years ago. Okay. Let's say that you got in a car accident and your brain half of your brain is not working properly and you lose your ability to move your arm. Okay? or you don't even have an arm. Okay. So, they put a prosthetic and a cable used to go to your brain and if you wanted to move your arm, you could able to to grab a cup and and something, you know, or or or grab something. It was something like that, you know what I mean? Like 20 years ago. Now you fast forward uh with technology, artificial intelligence, neural networks and all these different things, 6G technology etc etc etc. Uh now we have a whole different thing okay uh before everything was like you used to put a electrode let's say that you were not you got a stroke okay normally and this happens today to you get a stroke and you lose your ability to talk. So back in the days 20 years ago this technology you put electro a cap with a bunch of sensors. Okay. And you will like to your thoughts you will talk and say I want water and there was a printer it will come out a little strip that says I want water. And they will bring you water and and and that's how you used to communicate.

Now there's no electrodes. Now it's like Wi-Fi operation that you just connect your brain connects to your phone. Okay. Just like that. Okay. And [snorts] we will talk about this is the technologies that like we're going to be able have the ability to we will have the ability to dis uh turn off and turn off light systems on smart homes. We're going to be able to do coffee uh do all different kind of things inside of a smart home. Okay. Connected through Wi-Fi. But we'll talk about that later on in a different end. But we're going to talk about the basics of everything. Okay. Uh the rise of semantic brain decoding. Let's now move into the real semantic decoding which builds upon this earlier technique but goes much farther. Semantic decoding refers to interpreting the meaning behind the brain activity rather than simply identifying words or motor movement. In this area recent breakthroughs have been nothing short of revolutionary. In 2023, researchers from the University of Texas Austin including Jerry Chan and Alexander Hood made a major advance in this field by developing a non-invasive system that can decode continuous narrative from brain activity. The model uh these researchers that we're talking about are the guys the pioneers that of

this technology. These are the the the researchers um uh the professor Alex Hoth and Jerry Tang from the University of Texas like 17 or 18 years ago. But on 2023 there are opening up and and and and talking about major advancement in this field by developing a non-invasive system that they can decode continuous narrative from a brain activity. So it means uh they have the ability through BCIS to brain decoders uh to know what are you're thinking constantly as a noninvasive they don't have to open your brain they don't no no they just get that information because we're an antenna and your brain is always giving out signals okay uh their model doesn't just interpret isolated work it can understand complex following semanticing content This technology was able to reconstruct stories directly from your brain activity patterns recorded via fMRI which is a breakthrough in terms of decoding continuous thoughts in case of isolated words. A key paper published in nature neuroscience on 2021 discusses how transformer-based AI models were integrated with brain activity data to predict not only words but also the meaning and context behind those words.

This opened up a new possibility for applications from assisting people with speech impairment to potentially enable new forms of human computer interactions and national security and people now have the ability to abuse your constitutional rights. Okay, that's that's a different translations of what I am reading but just very being being very honest about uh what is currently happening. Let me see if I can find uh so uh Some people might be upset because I am doing this. And you know what? Here's what I have to say. Uh, you should ask before you tap on someone else's brain.

Okay? Um, I was not born to allow a federal uh government to tap into my brain without my consent.

[A passage about a private individual has been omitted from this transcript.]

people while you asleep. Okay. So that's the reason with a lot of people don't know about uh how is this been done because they come at you at you asleep. A good way to understand how this happens is like um when you go to sleep you you you go on different uh stages. Okay. But if you get wake up immediately you going to see what's happening. Okay. So I suggest you to to to ask for someone to wake you up when you asleep that you're like two hours asleep that they wake you up and there you're going to like your brain you're going to catch what is happening in your brain. Okay.

[snorts] Um let's see. So basically uh a new artificial intelligence called semantic decoder can translate person brain activity while listening to a story or silently imagining telling the story into continuous streams of text. The system developed by researchers at the University of Texas at Austin uh might help people who are mentally conscious yet unable to physically speak such as deerly by strugg to communicate intelligently again. The study published in the journal nature neuroscience was led by Jerry Tang a doctoral student in computer science and Alex Huth an assistant professor of neuroscience and computer science at Alton Austin Austin Texas. The work relies in part by transformer model similar to the ones that powers open AI chat GPT and Google bar. Uh now uh bar is Gemini. Okay. Um I am going to get more information because uh based on my research that I have conduct on this for the past five years ago. I know there's more students that are part of the crew uh the initial crew that they're working with Jerry Tang and Mr. Alex Hoth. So I am going to give them credit because I think that what they have done is a great job uh discovering and and coming on with this technology. Okay. Uh my hats off to Dr.

Jerry Tang and Mr. Alex Hoth and their assistance and everything that they have uh done. Uh I don't have no problems with them at all. I do have problems

with law enforcement federal agents utilizing these technologies against your will and and without your consent and without you having a lawyer next to you. Okay, that's what I have a problem with my constitutional rights being violated by federal employees. Okay, not the professors and the the researchers that have conducted this technology and there other people doing other types of technology. I just want to be very very clear with this, okay? I do not have any problems or any um I'm a researchers myself. I'm an inventor myself as well and I am not uh in any way, shape or form upset with the people from uh University of Texas, Austin or MIT, Harvard and other places that this has been happening in DARPA. No, I don't have no problem with none of these people. I do have problems with the the any user of the technology that is utilizing the technology against my will, against your will and is utilizing this technology uh without your consent and is violating your constitutional rights on daily basis, weekly basis, monthly basis, year after year. Okay, I do have a problem with that. So, let's continue talking about this. Okay. Or likely other language decoding systems in development. This system does not require subjects to have uh surgical implants making the process non-invasive. Participant also do not need to use any only word from prescribed list. Brain activity is measured using an fMRI scanner after extensive training of the decoder in which the individual listening to hours of podcast in the scanner later provided that the participant is open to having their thoughts decoded.

They're listening to a new story or imagine telling story allow the machine to generate corresponding text from a brain activity alone. So basically uh we know that our brain is always sending a signal and the best way that I can describe that is um every person have their own frequency. Okay. And I'm going to tell you also the frequencies that are being used because that is your frequency of your brain. So we are all different. We all think different but we all have different frequencies. Okay. So, so that's how they are tapping to your frequency. Um, back in the day, you like purchase a scanner, you know, uh, when the cell phones were were new, the technology of cell phones were new, the scanners were able to pick up on everybody's conversation. You just have to have a scanner and I think you used to put them on 700 800 uh, signal or Yeah. Yeah. 7 7 or 800. I can remember that uh megahertz

and you were able to get all the channels of all the different um phones talking around your antenna on your scanner and you will like listen there everything. Okay. Um just like the same technology now you know there are scanners getting into your brain. How other scanners if you drive by your city you will see antennas everywhere. Okay. And those antennas are being used. Uh if you are in the Miami date area, you go to Biscane Boulevard, you go to the to the Bayront Park, you will see that uh Crown Castle were installing antennas.

And those antennas are spread away around neighborhoods. Those antennas are in front of schools. Those antennas are in front of uh banks information. Uh those antennas are everywhere. Okay? and and they're utilizing those antennas that they're needed for your phone. They're needed for um your um your EVs, your vehicles to have more information and and XYZ um different kind of things. Okay. So, uh for a non-invasive method, this is a real leap forward compared to what has been done before, which is typically single words or short sentences. Mr. Hoth say we are going to we are getting the model to decode continuous language or extended period of time with complicated ideas. So this was in 2021 we are in 2026. Okay imagine just need to think uh how artificial intelligence work is constantly evolving. So, so this now um you are thinking about your password of your bitcoins.

Okay, if you're thinking about your password of your bitcoins, someone else is getting that password. Okay, your password of your email. Okay. Um, if you're thinking a good idea, if you're thinking that you're going to go out and you're going to go out and said, "Hey, hey, I am thinking of uh going tonight to this nightclub and I am thinking of calling so so and so and I'm like someone else is getting that." Okay. So, so, so just so you will understand how this is working. Okay. The result is not word for word transcript. In state, researcher designed to capture the the gifts of what is being said or thought.

Uh albe imperfectly about half of the time when the coder has been trained to monitor a participant brain activity, the machine produces text that closely and sometimes precisely matches the intended meaning of the original words. This was in 2021. Today is more precise.

You know what I mean? But hey, there's way to uh play the the the artificial intelligence. There also ways that I have uh been a successful um develop to evade the semantic brainers activity in my brain. Okay. Uh but I'm not part of the research. This is my information. uh for example in experiments and a participant listening to a speak speaker say I don't have my driver's license yet had their thought translated as [snorts] she has not even started to learn to drive yet.

Um listening to the words um I didn't know whether to scream, cry, run away instead of say leave me alone. Uh I didn't know whether to scream, cry or run away. Instead, I say, "Leave me alone." What's the code says? Started to scream and cry and then she just say, "I'm told you to leave me alone." So, let me see if I I am able. I am going to be okay.

Right. Let's see if I am able to at here Okay. Um, I'm able to show uh people right now what I am reading. Uh, So basically uh let's see if I can what I am sharing can be seen. Yes. Yes. Yes. And yes. Uh so basically let me just show you uh more. This is an uh on the left side uh it says actual Yeah. No. Yeah. Yeah. Yeah. Everything okay? [laughter] Uh so actual the the left side on the screen um is the actual stimulus. Okay.

And this is your thought. This is exactly what your thought said. And then on the right side of the screen you will see the decoded stimulus of how the artificial intelligence is decoding your brain thoughts. Okay. So just I want for you to like look at the screen if you are in YouTube and uh x.com and Twitch. I am not able to screen uh share screen my my uh my screen share my uh my uh my laptop because I still don't have uh stream keys from Tik Tok and this is a process that I have to go through. Uh once that I get those stream keys I am going to be able to be more uh cool with everybody at Tik Tok. So, uh, for for now, um, the actual symbol says, I go up from the air mattress and press my face against the glass of the bedroom window, expecting to see eyes starting back at me, but instead finding only darkness.

The decoded stimulus says, "I just continued to walk up to the window and open the glass. I stood on my toes and peer out. I didn't see anything and look up again. I saw nothing. I'm sorry. Um, this image shows decoder predictions from brain recordings collected while user listened to four stories only. Okay. So if you know how models work and artificial intelligence work basically with more time and more training and more listening and more the systems are able to predict and are able to provide better data. Okay. So uh from for from brain recordings collected by the user listening to four stories example segments were match manually selected and annotated to demonstrate typically decoder behaviors. Uh the decoder exactly reproduce some words and phrases and capture g of many more. Credit to the university of te Austin Texas. Um, so actual stimulus was um I didn't know whether to scream, cry, or run away. Instead, I say, "Leave me alone. I don't need your help." Adam disappear and I clean up alone crying.

The decoded uh prediction said started to scream and cry and then she just said, "I told you to leave me alone. you can hurt me anymore. I'm sorry. And then storm off. I thought he had left. I started to cry. And you see there's is it was six years ago was was kind of off. But

with training and training and training, all these models get more robust uh and the servers get faster and more precise. And beginning with an early version of the paper that appear as a preprint online, the researchers address questions about potential misuse of the technology. The paper described how decoding work only with cooperative participant who has participated willingly in the training of decoder. Result for individuals who on whom the decoder had not been trained were unintangible and a participant on whom the decoder has been training later put up resistance for example by thinking other thoughts result were similarly inible. We take very seriously the concern that it could be used for bad purposes.

[A passage about a private individual has been omitted from this transcript.]

residents of Miami Day County and a lot of residents in other counties, but I am just talking where I'm at right now. This was currently also happening while I was living in Palm Beach County, okay? uh and when I was living in West Palm Beach. Okay. So, um I can say that while I was in West Palm Beach and when I've been in Miami Day County, it's been happening. And while I am also in in Brocker County as well or anywhere that I've been traveling in the state of Florida since I moved back in here.

Um we take very seriously the concern that it could be used for bad purposes and have worked to avoid that. Tank say we want to make sure people only use these type of technologies when they want to and they and that it helps them. Okay, that is not helping me at all. In addition to having participants listening or thinking about stories, the researchers asked subject to watch for short silent videos. While in the scanner, the semantic decoder was able to use their brain activity to accurately describe certain events from the videos. The system currently is not practical for use outside of the laboratory because it's reliance on the time needed on FM fmr machine. But the researcher thinks this work could transfer to others. More portable brain imaging systems such as functional near infrared stro spectroscopy fnirs. Let me repeat that again. Functional near infrared spectroscopy. Okay. uh fn

fnirs uh measures where there's more or less blood flow in the brain at different points in time which it turns out is exactly the same kind of signal that the fMRI is measuring. H say so our exact kind of approach should translate to fnirs also he noted that resolution with f frris will be lower. Okay. And here uh this is uh Dr. uh uh Mr. Jerry Tang. Okay. Right in here.

This is Jerry Tang. Uh PhD student Jerry Tam. See, prepare to collect um prepares to collect brain activity data in the biomedical imaging center at the University of SA Austin. The researchers are training their semantic decor on dozen of hours of brain activity data for participants collected in an MRI scanner. So let's uh review uh this. Okay, this is a MRI scanner. Okay. So if you if you being in um MRI scanner, okay, the MRI scanner is going to scan whatever the doctor prescribe, but the way that these scanners work, they scan everything, you know, they just only send what the doctor request, right? So other sources get your frequency. So if you've been to an MRI scanner, okay, and I'm going to I'm going to be very honest with you.

If you ever go anywhere else and you step into a machine that behaves like a MRI scanner, know that that machine is also taking your brain signal as well. Okay, I am not going to I am not saying what machine is, but if you're smart and if you've been into MRI scanner, you can just search what is what it does in my scanner. An MRI scanner, what it does is have a magnet going in circles and you when you're inside there, you can hear them.

So if you are anywhere around the world and someone tells you get in here please and you see a machine again going in circle but you're not laying down.

Hello. and they're they're getting your your your brain activity um and ID just like getting the ID of a cell phone or just getting an ID of anything. Okay, I think that that's very uh very simple now to to understand. This work was supported by the White House Foundation, the Alford Peace Loan Foundation and the Bur Burroughs Welcome Fund. uh you can find more information about those uh foundations and funds that are supporting this uh research in the University of Austin, Texas. There's other foundations and there also other type of funds in other university uh in uh United States of America. they're currently doing in advancing what uh Mr. Jerry Tank, the PhD student that he's a doctor by now after all this remarkable work working uh alongside um with uh Alex Hut and professors and other people that I am going to uh talk about later on. Okay.

All right. So, um now that we got that away, uh you will see that I am not um making things up. You know, I'm reading it from the horsemouth. This was uh University of Texas uh paper and that was posted online and um I have I had so much information about you know you want to connect yourself. Um I have everything repos I have all everything that you need to do to connect and have your own system as well for research. Um, Oh,

So, um hello, hello everybody on Tik Tok. Welcome, welcome, welcome. Uh so, basically, uh copyright and disclaimer under section 107 of the copyright act of 1976. Uh this video is for educational and information purposes only. It may contain copyrighted material the the use of which has not always been specifically authorized by the copyright owner. This content is made available in accordance with sections 107 of the copyright act of 1976 which allow for fair use of copyrighted material for criticism, comments, news, reporting, teaching, scholarship or research. I am a researcher and I am conducting research. So anything that you see on my screens uh this is my copyright disclaimer uh because I am utilizing uh this information for research and I am doing the research live uh not just behind the scenes. All copyright uh content use in this video remains a property of respective owners. So I just show you a piece and um uh uh blog post from the University of Austin, Texas. And due to that I am giving this copyright disclaimer under section 107 of the copyright act of 1976. Okay. Uh this vis this video is for educational and information purposes only and I am conducting my research on semantic brain decor for approximately already four and a half or five years already. Okay. And I am just sharing uh this information with all of you in YouTube, Twitch, uh Tik Tok X, everybody around the world.

Thank you very much for being here. Thank you, thank you, thank you for Hi, hi, hi, hi everybody on Tik Tok. Hello, hello, hello, hello. Thank you, thank you, thank you. Tick, tick, tick, tick, tick, tick, tick.

So, so basically uh now now that I have to to do [laughter] so everybody that is adding me to do um adding me to to try to do a co-host uh I am I am I am doing this once that I finish with this I am going to go ahead and do um adding call holds and invite people to guess in so we can talk about semantic brain recorders. Uh so basically uh let's go ahead and Okay. Uh decoding the brain activity methods and techniques. Okay. We're going to talk about that. Uh let's now talk uh take a closer look at at how semantic decoding actually works. Uh both from the technical and practical perspective. The core technology behind most brain decoders is fMRI a noninvasive imagine technique that detects changes in blood flow in brain in the brain. the critical because increased blood flow indicates that certain brain regions are active and processing information. The fMRI scanner capture temporal and spatial pattern of neural activity as subject listening uh listening to stories visualize concept or even images action. So the semantic decoders are uh meta also uh is doing currently doing research uh uh Mark Zuckerberg team um >> uh let's See, so basically let's see here.

uh meta is utilizing um is like is utilizing using AI to decay language from the brain and advance our understanding of human communication. Let me just go ahead and transfer this to everyone uh so you're able to see it. Um

so you're able to see this uh using AI to decode language from the brain. Uh this is at ai.mmea.comblogbrain AI research human communication. Okay. And this was done um uh today. Okay. Today we have this information. Uh over the the last uh 10 years the meta fundamental artificial intelligence uh research better called as fair uh lab in Paris has been at the forefront of advancing scientific research. Uh we lead breakthroughs in medicine, climate science and conservation and have kept our commitment to open and reproducible science as we look to the next decade.

Our focus is on achieving advanced machine intelligence AMI and using it to power products and innovation for the benefit of everyone. So as you see this technology from 20 years ago now is being developed here in America and also in Europe on Paris France. Okay. From our with partners Mr. Mcronone up there he's a cool dude. Okay. So uh today uh in collaboration with the Basque Center of Cognition Brain and Language a leading interdisciplinary research center in San Sebastian Spain. We are excited to share two breakthrough that show how AI can help advance our understanding of human intelligence uh leading us closer to AMI.

Uh bye-bye. See you soon. Okay. Uh building on our previous work toward decoding the perception of images and speech from brain activity. We are sharing research that successfully decodes the production of sentences from non-invasive brain recordings, accurately decoding up to 80% uh of characters and does often reconstructing full sentences solely from brain signals. In a second study, we're detailing how AI can also help us understand these brain signals and clarify how the brain effectively transform thoughts into sequences of words. This important research uh wouldn't be possible without the close collaboration we have fostering in the neuroscience community. Today Meta is announcing a 2.2 million donation to the Rothschild Foundation Hospital again to the Rush Rush High Rothchild Foundation Hospital in support of this work. This continues our track record of

working closely with some of the leading research institution in Europe including Neuros Neurospin Ca ANS PSL and CNRS. These partnerships will continue to be important to to us as we work together to explore how this breakthrough can make a difference in the world and ultimately improve people's lives. So we are doing research in United States of America, Spain and France. All it's just we're in the same team. Okay. Thank God.

Right. Uh but also this technology is being developed also as well in China and Russia and everywhere that there's like uh artificial intelligence is being used. Okay. Using artificial intelligence to decode language from non-invasing recordings of the brain. Every year million of people suffer from brain lessons that can prevent from communication communicating. Current approaches show that communication can be restored without neuroprothesis feeding command signal to a to an AI decorder. However, invasive brain recording techniques like stereostatic electronics pography. Uh let me repeat that word again. Hello hello hello hello hello hello everybody. Um uh estereatic electron patholraphic and electrocarticoraphy require neuros neurosurgical interventions and are difficult to escape until now using non-invasive approaches has typically been limited by the noise complicity of signals they record. For our first study, we use both MEG and EEG, non-invasive devices that measures the magnetic and electric field elected by neuro neural activity. the the record 35 healthy volunteers at BCBL while their type sentences. We then train a new AI model to reconstruct the sentences subtly from the brain signals on on new sentences. Our AI model decode up to 80% of the characters typed by the participants recorded with MEG [snorts] at least twice better than what can be obtained with the classic EEG system.

Okay. Uh and writing there um this is some of the research that is being conducted on on on around the world. Uh how's the brain turn thoughts into words? So there this is a machine that they have been utilizing. This is like the coders and the B basque center in Spain. Okay. Uh this is how they're conducting. These are researchers when I am showing you live in the screen right now. Um but this is just uh some

technology that other people are doing. Uh we're more advanced than than than what you're currently seeing in here. Okay. So this is how what is being decoded. The EMG scan uh has 80 uh and like uh these are the different uh technologies they've been using and and the research they're currently conducting. you can see with your own eyes on my screen. If you're on YouTube X or Twitch, you're able to see uh this uh right now uh live as I place this video. So, you can see the technology and um you cannot say I'm invested in this, okay? Because you're watching it with me right now, okay? I am not I am not I'm not crazy or nothing like that.

I just I just know the the what's going on, okay? So uh this video was produced by Meta and is publicly out there for you to find out and see what are currently do the research uh is conducted on on different parts around the world. Okay. This research could create a new avenue for non-invasive brain computer interfaces again BCIS. Okay. You're talking about the same thing that we've been discussing for a few years now. Okay. brain computer interfaces to help restore communication for those who have lost the ability to speak. But several important challenges remain before this approach can be applied in clinical settings. The first is performance related. The coding performance is still imperfect. The second is more practical because MEG requires subjects to be in magnetically shielded room and remain still finally.

Uh while this research was done with healthy volunteers, future work will need to be done to explore how it could benefit people suffering from brain injuries using artificial intelligent to understand how the brain forms language. We also sharing a breakthrough toward understanding the neural mechanisms that coordinate language production and the human brain. Studying the brain during speech has always proved extremely challenging for neuroscience in part because of simple technical problem. Moving the mouth and tongue heavily hurts neuromaging signals. Uh again this is another video of the meg scan. Uh it has uh they're reporting 80% accuracy. I know based on my research has more than 80% accuracy. Okay. to explore how the brain uh this is just basically uh you know we are 80% we need more funding okay so uh let's send more funding to these guys to explore how the brain transform thoughts into integrate sequence se sequences of motor actions

we use AI to help interpret the meg signals where participants type sentence uh by taking 1000 snapshots of the brain every second we can pinpoint the precise moment where thoughts are turned uh into words uh syllables and even individual letters. Our study shows that the brain generates a sequence of representations that starts the most abstract level of representations the meaning of a sentence and progressively transform them into a marate of actions such as the actual finger movement of the keyboard. Importantly, the study also reveal how the brain coatly and stimul represents successive words and actions. Our result shows that the brain uses dynamic neural code, a special neural mechanism that chains successive representation while maintaining each of them over the long time period. The brain generates hierarchy of linguistic representation before typing each word.

uh phrases uh GPT2 we're using here uh GPT2 llamas all the different AI models the AI has and available uh for the research um so word spacey fact text syllable and the letter o e uh these are how um the brain signal travels and is being uh from the different representation of each word in the hierarchy of linguistic phrase, word, syllable and letter. And and you can see here in the chart on analytically how each of the segments of of the word phrase, word syllable and the letter are it takes on seconds in milliseconds as well. Okay, so we're talking about this happens so fast that it's not even a second. Okay, this is like negative1ative7.5 negative50 negative. 255 zero is the onset and then it move after the the it happens. So there's like a a mechanism in your brain um that that you build uh when you're talking um it goes like your thought goes through a process all the way to assembly and then assembly construct and put all the vocabulary that you have in your brain constructs the sentence of the paragraph or or whatever script you want to come out of your mouth and assembly takes all the letters and put it to put it all together. and comes out of your mouth. You know, you you thought you before you think, before you talk, before you everything. This happens so fast. That's when all this happens when you're a kid, when you're a baby, you start learning the the vocabulary, start learning to talk and eventually all this happens. We have our brain is the most

um efficient machine and computer ever created, okay? By the universe. uh cracking the neuro code of language remains one of the major challenges of AI and neuroscience. The capacity for language which is specifi specific to humans has endowed our species with an ability to reason, learn and accumulate knowledge like no other animal on the planet. Understanding it neural architecture and its comput computational principles is thus an important path to developing AMI. So we are developing multiple technologies.

Okay. Um artificial intelligence this AMI uh sh we'll talk about this another time. Okay. Um enabling health breakthrough with opensource AI. Okay. At Meta, we are a unique position to help solve some of the world's biggest challenges using AI. Our commitment to open source has enabled the artificial intelligence community to build on our models to achieve their own breakthroughs. Last month, we share how brightart a company based in France is using Dino V2 as part of its AI software to help clinicians identify or rule out signs suggestive of congenital heart defects. in fetal heart ultrasounds uses our on babies. Uh Dinov2 is a model uh from Meta as well.

Okay. Uh last year uh Brear achieved FDA uh 500 510K clearance for its software uh which they attribute in part to meta open source contribution. We also share how Virgo, a company based in the United States, is using Dinov2 to analyze endoscopy video, achieving state-of-the-art performance in wide range of artificial intelligent benchmarks for endoscopy such as anatomical landmark classification, disease severity score for ulceratives, qualities and poly segmentations. Um basically um we are also with all these breakthroughs that we're going to see in the next five years 10 years to come we will be able to uh around the world all these researchers are going to be able to to all these models from Meta and all the models available from other companies like uh Chat GPT um and and and others um we're going to be able to to diagnose cancer. We're going to be able to uh fight uh a lot of um illnesses that until today we have not

been able to tap and this is a new industry of healthc care that are being developed right now. So uh if you're looking for invest these are the good things to invest because it's going to be in the future. Okay. Uh as we look towards the next 10 years, it is exciting to think about how the breakthroughs we share today could benefit the great good. We look forward to continuing the important conversation we're having with the community as we move forward together to tackle some of the society greatest challenge. Uh you can read the paper for brain to text read the paper from thought to auction and let's see let's see we're here back again uh this is the basically we're we're we're going about and talking about the evolution of semantic brain quarters where are the uh BCIs and where everything is standing okay Um, let's See?

So uh some of the the the models that they that they're available to conduct some research uh I have conduct research on Yama uh on the on the beginnings of Yama Yama 2, Yama 3. I I I I played with those uh models a few years ago. Uh also uh Meta has blenderbot. Meta's conversational AI train for chatbots and dialog systems. uh Robert a robustly optimized bird retraining approach a variant of bird uh bird it's um it used to be Google um birectional encoder representation from transformers focus on improving training techniques uh PyTorch uh library while not model itself.

PyTorch is Meta's deep learning library that widely used for developing AI models. Uh OPT open penetrated transformer meta language model family design for efficiency and ethical use on LLP task. Uh Dino and Dino you know is being used right now for semantic brain decorders is a self-supervised learning model for visualized representation learning. So uh uh and the acronym of Dino is distilled knowledge from noisy labels. Uh you have an open AI the open AI have uh GPT4 GPT 3.5 GPT3 codeex uh DAL uh 2 clip whisper and chat GPT.

Okay. Uh uh GPT4 generative pre transformer 4. The most recent model on GPT series known for its advanced language capabilities including improvement performance in test requiring reasoning translations and more. GPT3.5 is a privilege to GPT4 used to early interaction with chat GPT famous for conversational abilities and Y applications. GPT3 generative pre-trained transformer 3 a milestone model known for generative human like text and for being foundational for many AI applications in text generation codeex a derivative of uh GPT3 specifically fine-tuned to generate and understanding code powers GitHub copilot uh D2 an image generational model based on text from text to image AI I clip convasive uh language imagine uh pre-training an AI model designed to understand images and text together you use in conjunction with other models like Dale E uh Whisper opens AI speech recognition system that can transcribe and translate speech into multiple language and chat GPT a conversional AI built on GP3 uh 3 and GPT4 and we are at chat GPT 9.2 and if you are getting the beta uh we are also already on G chat GPT uh 6.

Okay. Um then these are the two uh these are two uh independent um AI corporations that are doing research but they're talking and they're sharing the data as well which is benefits the whole community of researchers and the technology uh size of it. Okay. Um

See, >> so uh basically uh cloud-based AI platforms we have Google AI uh TensorFlow, Palm and Bard. Okay, TensorFlow Google's open-source machine learning library and Palm is pathway language model, Google's advanced language model focus on multitasking and higher efficiency and Bard is the Google chatbot AI model designed for conversations. You have Microsoft AI with Azour AI and GPT4 via Open AI API. Uh, Assur AI, a suite of cloud-based artificial service offering everything from machine learning to cognitive service. Auror cognitive service include tools for vision, language understanding and speech recognition. GPT4 available through Microsoft integration with open AI through a API access. Uh we have IBM Watson uh known for AIdriven applications across industries including healthcare, finance and customer service. IBM Watson is also being used inside of meta and chat dpt. There's like everybody's using all technologies because this is part of their research. You see what works and what doesn't work. We also have Jeff with Amazon web services AWI AI uh which in in my opinion they have a lot of years of experience. Okay. Uh, SageMaker AWA's machine learning services for building, training, and deploying models. Lex is a conversational AI service for building chatbots and recognition images and video analysis of AI.

So if you know uh J Besso started with Alexa you know we know Alexa it's been for a while and now uh Rufus is an AI AI power uh shopping assistant is integrated into Amazon shopping app. So if you if you haven't see uh Rufus, you can go to download the uh Amazon uh new app in your phone and you will see that they have uh Rufus there and say you know and and you will can interact directly with Rufus as well.

uh Amazon Q, a generative AI assistant specifically designed for business enterprise and developers and project ALIA uh an AI assistant designed to help third party sellers on the platform with an example of improving product listing. So everything is getting used with artificial intelligence. Okay. uh in Amazon in the AW and the AW side of AI we have S maker which is AWS machine learning service for building training and deploying models uh lex

that we already discover a conversional AI service for building chatbots and recognition image and video analysis AI Alibaba cloud AI now in China okay uh offers severity of AI services including image recognition, voice recognition, machine learning platforms. So if you go to Alibaba cloud AI, you can tap into China's technology for artificial intelligence and do your own research as well. Uh we also have Oracle cloud AI uh provides tools for enterprise AI uh applications including chatbook machine learning models and LPbased systems. Okay, so let's see.

Everybody now is getting all the information of um So what does this mean for the real world? Uh real world applications of semantic brain decoding. Okay. Uh we have medical applications, culary references and psychological and cognitive research. Uh what does this mean for the real world? Well, the implications are wide reaching especially in the fields of healthcare communications and assisted technologies. Uh one of the things that Meta is trying to do is like you have this they have this uh Ray-B bands um artificial intelligence uh sunglasses that they can communicate with your brain and those little sunglasses communicate and you can like turn off video, you can turn on and turn off lights. This is some technologies that are going to continue to de develop in the next few years and we will discuss them here in in my channel with all of you uh learning and getting ahead of the curve on artificial intelligence because not a lot of people are discussing artificial intelligence like this. Okay. Um medical applications one of the most immediate application is for individual with severe speech and motor impairments. Uh for example, people that with lock lucket in syndromes where person completely aware but unable to move or speak. Okay, just like people with u to get you're totally healthy person, you're a father, you have kids, you have family and suddenly boom, you can a car accident or you get a stroke. and um the medical applications. This is a good way for this technology of artificial intelligence help um families and people

around the world. Okay. So we are going to move um and also prosthetics as well. Prosthetics before was just opening and closing. Now they you can type, you can move your hands more. um symmetrical like a like a real hand. Okay. Um uh a fasia a fascia is a condition that impairs language ability could potentially use brain decoding to communicate directly uh through thoughts. uh Alzheimer patients, uh people that are in homes, people with your loved ones, your grandma, your granddad, uh people that their brains are not behaving correctly and they are um now uh in the end of life technically. and and we are going to be able to to have our folks and our families uh with us uh being able to communicate and and be able to be with them more than what we currently had in the past uh 2,000 years. Okay. So this is a this is like really uh great uh technology coming up.

Uh two 2021 study in the e transactions of neural systems and rehabilitation engineering highlighted the potential of neuroprothetics that translate brain signals into speech or text giving these individuals a voice again. Again people are going to have voice again. Okay. And this is very very very um advanced, okay? That you lose your ability to to to communicate and suddenly you can purchase and then have uh a system that you can communicate with your family and and continue with your life. Okay. Uh psychological and cognitive research. Imagine being able to decode connective states or emotions in real time. Semantic brain recorders could help psychologists study mental health by allowing them to observe pattern associated with conditions like depression or PTSDs by analyzing brain activity related to specific emotion. Researchers can develop more accurate diagnosis tools. Uh ethically uh ethical societal and future consideration the ethical consideration in brain decoding uh while the technology potential is vast. There's a serious technical uh ethical consideration surrounding semantic brain decoding that we need to address.

Privacy concerns. I I have a problem with that too. Okay. Uh privacy concern ability of reading someone thoughts raises questions about mental privacy. uh brain activity is incredibly personal and without strict ethical safeguard there's a real risk okay uh that this technology could be misused either for

surveillance or extract private information without your consent. Uh and that's right in there. Um, one chicken. So basically uh fourth amendment and fifth amendment of the US constitution here how this amendment relate. Fourth amendment protection against unreasonable searches and seizures. The fourth amendment protects citizens from unreasonable searches and seizures by the government. If private information is extracted from an individual or their property without consent or a valid warrant, it constitutes a violation of this amendment. The government generally needs probable cause and a warrant to obtain personal data unless certain exceptions apply such as a consent from the individual. So for example, warrant surveillance or unjust to fate data collation without the individual consent could be a violation of the fourth amendment protection against reasonable searches. So I do have a problem with this technology.

[A passage about a private individual has been omitted from this transcript.]

problem with that. I also the fifth amendment protections against self-inccrimination. The fifth amendment provides individuals the right not to incriminate themselves if private or sensitive information is extracted in a way that could lead to self-inccrimination or violation of one personal privacy. It might infringe upon this right. Additionally is if forced or coercing private information may also violate a person due process right under the fifth amendment. So um basically these are the where we at right now um with with what what this technology is currently um um is being currently used uh and and you know also people are enriched in themselves with people's bitcoins because uh in in a in a wallet you need your password. You think about your password.

Uh you can get at night when you're sleep and you're sleeping [snorts] with this technology coming to your brain and start talking with you while you're asleep. Um remember you're you are asleep but your your brain is always working. Okay. So this technology comes when you're asleep and extract information. U I'm going to give you a a simple uh uh a simple how I see things. Uh, so uh for example, you have a bitcoin.

What is the bitcoin value? Uh, Bitcoin is is it was way way high uh maybe a year ago went all the way to 116,000 123,000 and now is half of that. Okay. So basically a lot of big fish uh made money cash out and and and you know that's that's the pattern going up and going down and some people make money some people lose uh they lose their ass on on on Bitcoin. So a lot of people pulling out of Bitcoin is that okay with the technology uh let's discuss about Bitcoin and their cryptocurrency that not that are not insured because they are fundamentally different from traditional financial assets. Here are the few the few reasons why decentralization. operates on a decentralized network, meaning there is no central authority like a bank or insurance company that can oversee, protect or ensure traditional insurance is typically provided by centralized entities that can manage and regulate the risk involved in the case of Bitcoin. No single entity or authority can guarantee protection. Okay. uh lack of regulatory framework. Cryptocurrencies uh included Bitcoin fails outside of the traditional financial regulatory framework. In many countries, insurance company typically require clear regulatory guidelines to assess risk and underrite policies which is challenging when dealing with the largely unregulatory digital asset. Uh volatility. Bitcoin is highly volatile with prices of different fluctuating dramatically within short periods. This is create a challenge for insurers uh who will need to account for the extreme price swings and uncertainty and votal volity of the market. Uh insuring

something that could lose significant value in matter of hours or days without extremely difficult uh custody risk. Bitcoin is stored in digital wallets and the person or entity holding that private keys to those wallets control the bitcoin. If the private keys are lodged, stolen or compromised, the bitcoin is gone forever and there is no way to recover it. Traditional insurance policy typically cover physical assets or assets with a third party example bank or financial institution that can be held accountable which is not the case with Bitcoin. Basically, the way that I describe this, you know, uh do you do you have uh $20 in your pocket, you you can open your wallet, you can pull out your $20, right? You you can see it. You can touch it. Uh now, if you have Bitcoin, show me that. You cannot do it. It's well, it's it's in the computer. It's a bunch of numbers. It's it's is there. You have to trust me, and I have to trust you. You see what I mean? like but you know the way that we have always uh done business uh traditionally uh from many centuries is I give you something and you give me something back you could uh with BCoin it's just a bunch of numbers that you translate and not everybody are good with this technology uh and and the security concern while Bitcoin's transactions are secure on the blockchain Okay. Okay. They are completely secure on the blockchain. User are responsible for securing their own wallets and private keys. If they fail to do so, uh they bear the risk of theft of loss. Traditional insurance often cover loss due to theft. But in cases of Bitcoin, if someone's wallet is hacked, that's that's the question, right? That's the that's that's the the what you need to to to think about is is if someone wallet is hacked or they lose their private keys there is no recourse. Okay. No centralized guarantee unlikely bank deposits which are insured by government entities like the uh FDIC. In the US US Bitcoin does not have a central backing or guarantee from a government or institution. This make it difficult to offer insurance in the traditional sense. Okay. So, banks are going to fight bitcoins. Okay. But at the same time, they're going to show that they're supporting bitcoin. Okay.

Uh and also the blockchain the ledger okay is a good technology because on the ledger you can see the transactions okay so the the the blockchain is not the problem Bitcoin is the problem so how your keys and your password are compromised okay let's say that you

decided to put your password uh so you can remember your password. Your password is uh 20 or 30 or 50 characters and you have memorized those 50 characters because remember if you lose your password you cannot go in into your p into your into your wallet. So once that you have all your um your password when you're asleep this new technology of semantic brain recorders come [snorts] into your sleep ask you about your password and you say your password while you're at sleep a hacker or uh so federal employee with ability to use This technology goes in, extracts those bitcoins, and now when you wake up, you're not going to look at your wallet. You didn't use your wallet. Right now, six months later, you decide to do a transaction with your Bitcoin wallet. And when you open your wallet, everything was gone six months ago. You don't even remember what happened.

And then who are you going to call if something gets stolen like that? Your first thing is you're going to call your your police department. And your police department is going to tell you that that is fraud, that they are not they're unable to help you out. And they're going to direct to direct you to IC3, which is the FBI, Federal Bureau of Investigation. And when you go to the Federal Bureau of Investigation, they're going to tell you, well, you can fill this paperwork. We will try to locate your bitcoins, but uh they're not insured with the FDIC and we're very limited on what we can do. And you lose in the game of trying to get rich with the technology and this is how I know this is the future of of stealing all the bitcoins around the world. Okay, they're not insured by no one. There's no one behind them. So, uh there's uh a lot of powerful people that they want those bitcoins gone. Okay.

And and then basically that writing there is giving you uh what is currently happening. So let's say that you are a federal employee working for X Y three letters agency. Okay. and you went to a school and you have a badge and a shiny badge, but you have access to this semantic brain decor. But it also happens that you are part of the

department that do fraud investigations. You have the perfect crime because no there's no oversight on these three words letters of agency. their employees are very quiet about the whole situation. And you know who's going to catch the guy that works on the threeletter word organization inside of the federal government? Who is going to catch them if they are the one they investigate fraud? Nobody. You know, like do you think these guys are going to put themselves in their handcuffs and going to go in front of the judge and say, "Oh, we still we've been stealing bitcoins for the past 10 years." You know, so um so you know when the bitcoins came into my life, I had the opportunity to get bitcoins for a dollar, $2, $3, $5. Um I was a l ex-lic banker and I knew how the world of business work. Um, I said this is going to be just like a roller coaster ride going up and going down eventually. Went all the way to 120. It's coming down to 60,000. Eventually, some more people are going to come in. Um, and and basically, they're going to just um to continue with the roller coaster and getting more people and getting their bitcoins. And nobody's going to stop that. So if you have bitcoins, you know, uh my advice that I have always give to everyone, just put it on something you can hold, you know, gold, silver, uh cash, CDs, you know, something that you can just put in your home [snorts] under your bed, your pillow. um you open that pillow, you open your hands and there's the $20 again, you know, and you can like go out and buy food and whatever. Uh so that's how I how I see it. Uh but I as always, you can see it differently and that's I'm pretty cool with that too as well. You know what I mean? That's the cool thing about living in United States of America. You can see things your way and I can see things my way too as well. So, uh let's see here. Um psychological and cognitive research. We talk about that. Um and then we're talking about the ethical considerations in brain decoding. Uh privacy concerns the ability to read someone's thoughts raise a question of mental privacy.

Uh the journal of law and bioscience of

2021 outlines the legal and ethical challenge of BCIS uh emphasizing the need of informed consent and robust data protection measures to safeguard individual thoughts and intentions. Autonomy and consent ethical guidelines must ensure that individuals retain control over their brain data. Informed consents is paramount. This means individuals should fully understand how the brain data will be used and have the ability to withdraw consent at any time. Without this, the technology could be used to infringe on personal autonomy.

The future of semantic brain decoding uh look forward. The future of semantic brain decoding is incredibly promising but still face many challenges. Real time portable decoding. One major area of research is making brain decoding technologies more portable and capable of providing realtime data. Currently fMRI machines are large, expensive and require stationary subject. But researcher are investing more affordable portable solutions like EEGs and near infrared spectroscopic and IRS and all that technologies right here in your phone. Okay, it's been developed in applications that can be used in your phones. Okay, because what is that we always carry? We carry a phone where applications get into your phone without your consent in the back end. And those um those are the technologies that are being developed. Okay. Um uh 2021 study in the e transactions of biomedical engineering explored the challenges of real time egs uh decoding which could provide a more immediate and accessible alternative to fMRI. Uh integration with artificial intelligence model. Future advancement with C AI models become better at predicting thoughts and intentions with less training data, enabling us to understand cognitive states and emotions more quickly and accordly. Researchers are exploring how reinforcement learning can be integrated into brain decoding models to improve predictions and accuracy. uh scholars uh reference an artificial intelligence in medicine on 2020. Uh discuss how transformers and recurrent neural networks are andn are being used to enhance decoding accuracy and speed and how this can be applied to practical brain computer interfaces. Uh basically I'm going to just cut it in here. This is the closing rem remarks and open discussions. Uh to conclude the field of semantic brain decoding is stills uh in infancy. Okay.

This is still on baby steps with artificial intelligence behind it. So baby steps on artificial intelligence is not baby steps. Okay. Uh but but still is the this is new technology um that is still in infasy but holds enormal potential from healthc care application for those with a speech and motor impairment possible changes in how we communicate with machines. This technology is poised to change the future of human interaction and AI integration. Uh and basically also um this technology is going to be used for national security.

Um something that I have seen in the news recently here in United States of America. Uh in 2001 we saw uh how uh buildings in America fall and we went after the terrorists. Okay. Uh here in America there's no terrorists anymore in my opinion because of this technology. Okay. All these tech, all these terrorists, they're being round up and they're either on jail in Cuba or they're in jails around the world. Okay?

Or they are in their countries. um the next wave they're coming they're coming after American people in my opinion and this is what I have been able to see in the past year and a half or two years uh consecutively you see American citizens being the target of the department of homeland security and due to this technology and due to the funding and how the congress work and how the laws are out there. Uh the Department of Homeland Security has the ability to abuse you. Uh I don't know who is going to intervene. Either the FBI, the CIA, the White House, Congress. Uh but today, the Department of Homeland Security in my opinion is being very irresponsible utilizing these systems. Um I don't think that the Federal Bureau of Investigation um or the CIA are utilizing the systems um the way that the Department of Homeland Security are using them because the Department of Homeland Security is a organization within our government that is still in baby steps as well. It's in infancy uh compared to the Federal Bureau of Investigation. uh and and the three words that you know

that you can see DHS, Department of Homeland Security, those are also three words. Uh three words also are CIA, NSA, uh FBI, uh there are multiple agencies that have three words. Okay. But all these different uh organizations that I have used and in my conducting my research and I talking about them. Number one, I have respect for them. But seeing a new organization within our government with all this power and and how this organization get their employees and how some of these employees got their positions. you know, uh I know for a fact that I I know people that are working for the Department of Homeland Security that they have a bachelor's degree from the University of Phoenix, but they never they did not completed the University of Phoenix. Someone else did their homeworks, someone else did all their work and they receive a bachelor's degree and those are the guys that today are using this technology. I do have a problem with that. And if we continue with not oversight and we not and we don't have uh if we don't get the bull by the horn and we don't control these organizations of three letters, [snorts] uh there are they don't have oversight because there's no laws or nothing is being created. Uh the direct people that are going to get hurt are the American citizens or anyone inside of our country or anywhere around the world. Uh again, thank you for listening to me. Uh I'm going to shut down this transmission and I am going to start coming later on to do uh we'll talk more about artificial intelligence and we're going to talk about all these new technologies that as we continue to learn I going to get more information so you know uh how these technologies are being used and what is the new uh advancement and achievements that we have in the artificial intelligence realm. Okay, my name is Victor Rosario.

Thank you. Thank you for listening to everything. And again, I am the next mayor of Miami Day County. Uh vote for me on November 7, 2028. You can go to my website at victormiami.com so you can learn more information that I am posting on my blog uh for Miami Day County. See, if you're from Miami Day County or the state of Florida, thank you for listening. I love you. Have a good one. Bye-bye.