马斯克发布 Neuralink 进展（改进版）

0:00Welcome. I hope this is working. Welcome to Neuralink Live Update. We're going to tell you about

0:04the progress of the first patient with Neuralink, Berlin, and the recap of the progress there.

0:11Then talk about what changes we were making for the second patient, which we're hoping to do

0:16and implant in the next week or so. And this is for our first product, which is called

0:20Levathy, which enables you to control a computer or a phone just by thinking. So let's, in fact,

0:25So we'll start off with just some introductions.

0:27DJ, you want to start?

0:28Hi everyone.

0:29My name is DJ Sal.

0:30I'm an electrical engineer and a chip designer by training.

0:33I led the design of several generations

0:36of the Neuralink implant.

0:37Currently, I was on the founding team

0:39and currently a president.

0:40I'm Matthew McDougal.

0:41I'm a practicing neurosurgeon and I had a neurosurgery

0:44at Neuralink.

0:45Yeah, go ahead.

0:46Yeah.

0:46Head of Brain Interfaces Applications.

0:48And I'm Melissa.

0:48I'm a software engineer at Neuralink, trying to figure out

0:50how to turn Brain Interactivity into full stuff in the world.

0:52All right.

0:53Thank you.

0:53So we'll just get going into the presentation.

0:56Our first product is sort of like we call

0:58FLEP-T, which enables the person

1:01within your only game plan to control their phone

1:03or computer just by thinking.

1:04And once you can control your phone

1:05and computer, you can essentially control almost anything.

1:08Just literally just by thinking.

1:09So there's no eye tracking or anything.

1:11It is purely your thoughts.

1:13This is really quite a profound device

1:15that can help a lot of people who have lost

1:17the connection between their brain and body.

1:19Imagine people like Stephen Hawking,

1:21who imagine if he could communicate

1:23at the same speed as someone who still had the connection

1:25to their brain and body.

1:26It's really something that can help millions of people

1:29around the world.

1:30And it's part of our overall goal of enabling

1:33very high bandwidth connection between the brain

1:36and the rest of the world and your computers.

1:38The long-term goal to which sounds so esoteric

1:40is to mitigate the risk of the civil society risk of AI

1:44by having a closer biosis between human intelligence

1:48and digital intelligence.

1:50But that'll take many years.

1:52Along the way, we're going to help solve a lot of brain injury or spinal injury issues.

1:57And without first part, Clopathy, that's going to be really quite profound.

2:01There is also potential long-term for bridging the gaps of the damaged or severed neurons,

2:07able to span the gap between the brain's motor cortex to the spine to enable someone to use

2:12their body again.

2:13I think that would be very exciting.

2:15And that is something that is possible in that long-term.

2:19And then our second product, which we've demonstrated to work with monkeys, is Blindside, which would

2:24enable someone who is completely blind or lost both eyes or completely lost their object

2:29nerve to be able to see.

2:30So that's something that we hope to demonstrate in the future.

2:33So this just gives you a sense of what the device is.

2:35A way to think about the neuralic device is like a Fitbit or an Apple Watch with tiny

2:39wires or electrodes.

2:41Those tiny wires are implanted in the brain and they read and write electrical signals.

2:46A lot of people think the brain is an incredibly mysterious thing.

2:50It is mysterious in a lot of ways, but it does operate with electrical signals.

2:54So if you can read and write those electrical signals, you can interface with the brain.

2:58And the device is sized so that it is the same size as the piece of skull that is removed.

3:02So if it's like a few centimeters diameter of skull that's removed, we replace that with

3:06the device after implanting the tiny wires with the surgical robot.

3:11And that enables a read-write capability to the neurons.

3:14Yes, exactly.

3:15It's completely wireless.

3:16I could have a neural link right now, you wouldn't know. And it charges inductively, so you could just basically have an

3:21electromagnetic ad that you charge the device with. So yeah, it's like an Apple Watch, exactly.

3:26Except that it's actually a much harder technical challenge to solve,

3:30given that there's limit as to how much you eat the brain tissue, whereas in four phones and whatever.

3:35You don't have a hot watch, you don't have to be really care. I'm just sitting on a table.

3:38Sure. Yeah, it's got to go through scan and stuff as well, and okay. So it is a tougher

3:42challenged to charge and to have eye band with communications, given that it's got to go through

3:47skin and hair and stuff. We have something. We have something. Yeah. So our first step with telepathy

3:52is basically to unlock digital independence for people with policies and to allow them to control

3:58the computer just with their mind without moving the body. And our goal is to provide them the same

4:03level of control, functionality and reliability that I have when I'm using a computer. Even better

4:09of the level of control I have.

4:11It's not a high bar for you.

4:12It's going to be clear.

4:13This guy, he's controlling this with his brain.

4:14So he's not, you can't see his hands in this video, but he's not using a mouse and keyboard.

4:18Just thinking about how to move the cursor and playing civilization sets.

4:21No eye tracker.

4:22Right.

4:23There's no eye tracker.

4:24He's like, you can go watch this on Twitter.

4:25Just thinking.

4:26Just thinking.

4:27It's like the couple days.

4:28The cursor moved here.

4:29Yeah, it's like the last time or two nights ago or something.

4:31Yes.

4:32I think the way he also described it is he's using the...

4:34Yeah.

4:35There's many more videos on his own, the platform.

4:37He can is string that live and also can talk and move his head with him.

4:40Yeah, you guys, if you joined this livestream, you can ask him questions.

4:43He'll tell you all about what it's like to move.

4:45Also, I think I haven't played Civil Education myself, but I think this is actually not easy.

4:49But this is Emperor Mode.

4:51If you have played Sith, Emperor Mode is the highest of people to go.

4:54Just the point is, this is a kind of a weak demanding task while livestreaming,

4:57playing the hardest mode of game, and he's able to do that while moving.

5:00Talking, engaging with the audience while playing.

5:02One of the other games he likes to play a lot is chess.

5:05I think it gets lost sometimes that he's actually playing speed chess.

5:08Against speed, yeah.

5:09Which requires an incredibly high fidelity degree of control and speed of control in order to be able to win.

5:16Also, another cool stuff about our device is that it's anywhere, anytime.

5:21Also on a plane during a flight while trading really cool means of care.

5:25Also, our device unlocks things that previously were impossible for our participant.

5:30For example, that we were able to connect him to his aiming console switch.

5:34They, uh, mal-cut with friends and family and it was lovely to see them playing together after

5:39years and couldn't do it since he's injured.

5:41Imagine if you're sitting on a row over from the sky, I can't play.

5:45Look over here and get a cat meme, no hands, no movement.

5:48Even a were over.

5:49Yeah, it's a strange time.

5:50Yeah, and he loves using the device and using independently daily to watch videos, read, play games using the browser.

5:57And the key metrics that we care is to make sure our device is actually useful,

6:01is to basically the amount of hours you use the device daily and weekly.

6:05We track it since the surgery.

6:07And on weeks that he's not too busy and not traveling,

6:10he can even reach 70 hours of using the device a week.

6:14This is amazing.

6:15And he would of course, left to use it more, but need to run.

6:18And he needs to sleep sometimes.

6:20And also, of course, to charge the device once in a while.

6:22Hopefully, we'll improve that over time.

6:24Maybe not obvious to people who are watching this.

6:26It's a normal math book he's controlling.

6:27This isn't like some limited vision thing,

6:29but there's only a few options.

6:30you can just do anything that you can do on a MacBook Pro,

6:32same one I have on my desk actually, so you can see exactly.

6:34And maybe another interesting point is that

6:36on the first day he used BCI,

6:39was able to break the previous world record

6:42for cursor control, just by using the brain.

6:44And recently he even doubled it

6:46and was able to outperform about 10% of our engineered neural ink.

6:51And you can be sure that we are very good in this game

6:53and very quick.

6:54And if you wanna check out how well,

6:55how you can do it, you can do it on our website.

6:58And it's very addictive games.

6:59It's a very simple game. You just have to click on the square, but it actually even sounds silly.

7:04It's quite a yeah, it can be quite it sounds like you'd be quite addictive and it's

7:08Especially if you get a low score and you think let's know. I gotta

7:11It's anyone who wants to try this

7:13I recommend going to the neural link.com where it's website and seeing seeing if you can beat Nolan's record and it's you will find

7:19That's actually quite difficult to do

7:20This is really with version one of the device and with only a small percentage of the electrodes that are working

7:26So this is really just the beginning, but even the beginning is twice as good as the world record.

7:31This is important to emphasize.

7:33The media has a habit of saying that the glass is 10% empty, but actually it's 90% full.

7:39I think it's really quite an accomplishment of the Neuralink team to have achieved with the first patient,

7:43the first device, twice the world record for the range of computer bandwidth.

7:47That's really an astonishing and amazingly great outcome, and it's only going to get better from here.

7:53So the potential is to ultimately get, I think, to megabit level.

7:57So that's part of the long-term goal of improving the bandwidth of the brain-compute interface.

8:02If you think about how low the bandwidth normally is between a human and a device,

8:06the average bandwidth is extremely low. It's less than one bit per second over the course of a day.

8:12If there are 86,400 seconds in a day, you're outputting less than that number of bits to any

8:18given device, except in very rare circumstances. This is actually quite important for

8:22for AI, risk for human-AI semiosis,

8:27is just being able to communicate at a speed

8:29that I can follow.

8:30Yeah, just to emphasize again,

8:32he's performing at this extremely high level

8:34with about 15% of his channels.

8:36And so we wanna mitigate any of the problems

8:39that led to that situation.

8:40The brain is a fascinating organ.

8:42Share with you some of the secrets about the brain.

8:45During any typical brain surgery,

8:47a small amount of air is introduced into the skull.

8:50That's because neurosurgeons like to have as much room

8:52possible around the brain. And so there's this little known control mechanism of allowing

8:58the concentration in the blood to rise a bit, which allows the brain to either expand or

9:04contract depending on where you target that CO2. But typically neurosurgeons will have

9:08the brain shrink by lowering CO2. What we're going to do in our future surgeries is keep

9:14the CO2 concentration actually quite normal, maybe even slightly elevated. And that'll

9:18allow the brain to stay its normal size and shape during surgery. That should eliminate

9:22this air pocket that we saw in the first participant. That air pocket we think may have contributed

9:27to eating up some of the thread slack as the air bubble migrated to be under the implant,

9:33push the brain away from the implant. And so that's easy enough to fix. Another consideration

9:38that we want to focus on for our upcoming participants is that the brain, think of it like a really

9:45complex folded onion, its layer upon layer of sheets of neurons all over the surface of the brain

9:51folded into this odd looking shape. The folds of the brain travel down deep into the brain and along

9:58with it go those onion layers of neurons. And if we insert very close to one of the folds where

10:04there may be very useful information encoded in neurons, we may end up traveling with our threads

10:10parallel to some of the layers of neurons that we're most interested in,

10:14avoiding them entirely. To avoid that possibility, we're going to insert in our future participants

10:19more close to the middle of the apex of the folds, ensuring that we're crossing

10:24the layers of interest, layer five of the cortex. I also think that it's important to

10:29highlight here those tiny wires that Elon mentioned. They're fraction of a human hair.

10:33They're very flexible. Intentionally so, because brain is constantly moving and you want

10:37the electrodes to be moving with the brain causing less of the scarring. And it's actually

10:41impossible for a human to have a talented matt do is to actually maneuver them. But we have a surgical

10:46robot that we built that can actually precisely target them in any three-dimensional space,

10:51x, y as well as c with my current level precision while avoiding basketure so that you don't disrupt

10:56and cause immune response from happening. So we actually have the technology to be able to place

10:59them exactly where we want. Yeah, it was truly amazing to see the surface of the brain after the

11:04the robot had inserted all the electrodes on the first participant without a drop of blood

11:09in sight.

11:10It's really quite an achievement.

11:11Yes, so something that probably most people don't realize is that the brain appears to

11:15be somewhat undifferentiated.

11:17So if you look at the cortex, it looks like a whole bunch of folds that maybe, like it's

11:22not obvious just looking at a picture of the brain that the brain is highly differentiated.

11:26That there's, you pretty much know exactly where the part of the brain is that controls

11:31your right hand and your left hand and your leg and that kind of thing or vision.

11:36It's actually quite precisely located.

11:38It's not.

11:39Some people I think look the brain like Oak could be anywhere, but actually it's brain

11:42is highly differentiated even though it doesn't look it.

11:45It's yeah.

11:46Do you want to describe how we actually where like how we identify were to drill the queen?

11:50Yeah.

11:51So we can put a patient that is considering this implant into an MRI, so functional magnetic

11:56resonance imaging machine and ask them to imagine hand movements that because of the spinal

12:01cord injury don't happen, but just imagining those hand movements causes these areas of the

12:05brain to light up in the FMRI scanner.

12:08So we have a pretty good idea based in fact for each individual participant, which part

12:14of their brain is going to respond to imagined movements of the hand.

12:18So we can map those imagined movements, much as we all do when moving a mouse to controlling

12:24a cursor on a screen even without the use of a mouse.

12:26Yeah, but anyway, I think it's an important point that it's not like you're the party of

12:30brain that controls your hand might be anywhere in the cortex.

12:33This is not the case.

12:34It's going to be in a very specific region and it's going to be

12:36precision is key to the left-handed right-handed.

12:39My mind too, if you're right-handed, you want the device on the left side,

12:43like the left-handed to the dominant.

12:45Yeah.

12:45The left side of your brain controls right side.

12:46You've got another risk mitigation we're looking at in the future is that the

12:51implant has a certain size, the depth of the bottom of the implant, actually

12:55inner than the average human skull.

12:57And so what we want to be able to do is control the size of the gap under the implant,

13:03give the threads that travel from the implant into the brain as much slack as possible.

13:08We didn't do this in the first participant because we didn't want to manipulate any of

13:13their tissue that we didn't absolutely have to. In upcoming implants, our plan is to sculpt the

13:18surface of the skull very intentionally to minimize the gap under the implant, such that the bottom of

13:24of the entravels perfectly flush with the normal contour

13:28of the inner side of the skull.

13:30That will put the implant closer to the brain.

13:32It'll eliminate some of the tension on the threads.

13:35And we think it will reduce some of the tendency

13:37of threads to retract.

13:38And we actually built the tool to do that.

13:40Yeah, this is actually, this is a very important detail.

13:43You really want the inner contour of the skull to be flush

13:47so that the implant, there's no,

13:49the brain doesn't want to puck her up into the gap.

13:52That's really quite a big deal.

13:53minimizing air pocket and the implant being flushed with the inside contour of the skull

13:58are too very important.

13:59The additional benefit here is that we do see some amount of stick up, what we call stick

14:04up, so you might have bumped in the head, but this actually eliminates that even further.

14:08Yes, our goal is that if you run your hand over the top of the skull, you don't feel any

14:13mom, you don't feel any device, and that even if someone was bald, you wouldn't really

14:16even notice it.

14:17And then from the inner contour of the skull, the brain or physical standpoint doesn't

14:22really notice that there's a divot in the skull because there's no divot.

14:25Okay. Another aspect of the human brain that obviously differs from any of the animals that

14:30we tested in is that the human brain is a lot bigger. And so you may not realize that means

14:34that the human brain moves quite a bit more than any of these other smaller-brained creatures.

14:41So when we open the skull, we see the brain travel toward and away from the robot about

14:46three millimeters in total as the heart beats and the breathing takes place. And so that

14:51movement, it adds a small challenge for the robot in precisely choosing a depth to insert

14:56each thread. It's not an enormous challenge and we've already upgraded the robots' capabilities

15:01to be able to even more precisely target depth in even a very rapidly moving brain with a

15:07high amplitude of movement. You may think the most obvious mitigation for threads that pulled

15:11out of the brain is to insert them deeper. We think so too. And so we're going to broaden

15:16the range of depths at which we insert threads. For the very first participant, we had an enormous

15:22amount of data from our animal work and we had very highly optimized our insertion depth to maximize

15:28the crossing of layers of interest in the cortex with the electrodes that we're recording from.

15:34Now that we know retraction is a possibility, we're going to insert at a variety of depths that

15:39even in several cases of differing amounts of retracting threads, we're going to have electrodes

15:44at the proper depth and with the deepest threads, be able to track how much retraction has occurred

15:51across the surface of the brain from each thread. And so we're going to both have more threads in

15:55the right layer and have better data on how much retraction has occurred. If you're a BCI nerd,

16:00you might know that being able to control individual who's the depth or thread is not something that

16:05most neural interface devices offer. Most neural interface devices are static, fixed, rigid array

16:10that you push in and probably ruptures are on depth. To be able to do this is actually pretty novel

16:14part of the robot. Yeah, the historical approach is to actually pound in a sort of bed of nails

16:19with an air hammer into the brain. It looked crazy. Yeah, just out of camera. This is,

16:24it sounds like somewhat barbaric. This is not what we do, but this is what's been done before.

16:28It's literally just hammering in what looks like a bed of nails with the brain, which actually

16:32works. It's astonishing that it actually looks like. That's what people are doing. Like DBS probes,

16:35you're just sticking in by hand. Research is just guiding in it. There's several orders of magnitude,

16:40more volume of brain tissue that you're destroying compared to what we're doing.

16:43But that deep brain stimulation stuff does actually work.

16:46It actually helps people a lot.

16:47Yeah.

16:47Yeah.

16:48That's a great part of you.

16:49But I think we'll be able to do a much more

16:51finessed version of that down the road.

16:53It's really difficult.

16:53The neural device is something that really absolutely

16:56minimizes damage to the brain, absolutely

16:58minimizes the load on the patient.

17:00And the goal is to allow someone to live a completely

17:02normal life.

17:03You won't even notice that someone even has the device.

17:06So like I said, we're the ability to control your

17:09computer and phone.

17:10That's our flabothy.

17:11And then next device being able to allow people to see

17:13that could not see before.

17:15In fact, you could allow people to see Jordi LaForge

17:17and Saw Trek and any, but whatever.

17:19Infrared.

17:20Yeah, infrared, ultraviolet, radar.

17:22So I think another way of saying is that

17:25we want to give people superpowers.

17:27So it's not just that we're restoring

17:29your prior brain functionality,

17:31but that you actually have functionality

17:32far greater than a normal human.

17:34That's a super big deal.

17:35And I also think oftentimes the questions

17:37that we get a lot is why do you have to actually go

17:39into the brain or if you place it on the surface

17:42or outside the school?

17:43Basically, the long story short, the physics of how it works, you really need to get the sensors, which are these facing in the brain, next to the source, which is as close to it as possible.

17:51Otherwise, what you get is you get a population response and not be able to do the level of controls that we believe.

17:57Yeah, maybe a good sort of analogy would be like, if you're trying to understand what goes on in a factory, you need to go into the factory.

18:03You can't just put a stethoscope on the wall and try to figure out what's going on.

18:06Like, anything on the outside of the trying to read things from the outside is like putting a stethoscope on the wall of a factory trying to understand what's going on in the factory.

18:12It's not gonna be effective.

18:13It could be, threads are gonna be in there.

18:15But I just wanna emphasize again,

18:17like the goal is to get people superpowers,

18:20not just to restore prior functionality.

18:22It's very exciting, and I think that should give hope

18:25to a lot of people in the world

18:26that the future is gonna be exciting and inspiring.

18:29And the technology is gonna give them superpowers.

18:31That's amazing.

18:33Yeah, yeah.

18:33It could can you multitask, but yeah, in fact,

18:36you look at Nolan's streaming,

18:37and you can just check out Nolan's streams on the X platform.

18:40He's multitasking all the time,

18:41So he's while talking and listening to podcasts.

18:44Yeah, exactly.

18:45It's really just if you're using your hands

18:47and you're playing a video game while talking.

18:50Don't take that word for it.

18:50They just go watch.

18:51He's out there on the internet doing his thing.

18:53Yeah, exactly.

18:54So can you do keyboard shortcuts or is it just a mouse?

18:56Yeah, that's what you were working on right now.

18:58Sure.

18:59So currently he's working in the mouse,

19:00but we are also exploring and recording more dimensions

19:04from the neural activity.

19:05Multiple clicks, so to the shortcuts or just able to do

19:08more games, draw a book controller.

19:11that also in the future we plan to expand to decode the text,

19:14not just the mouse control, but also allow participants to...

19:17Yeah, actually, so maybe going back to the discussion

19:19of third-order traction, one of the very exciting parts

19:21to me about the story is that we're able to do so much

19:24of 15% of channels.

19:25You have more channels, what that actually offers you

19:27is not just faster mouse control,

19:29as in the motor cortex neurons don't all represent

19:31the same thing, so you're trying to understand

19:33what an individual finger is trying to do.

19:34You might or might not have an electro next to it,

19:36and the more channels you have in the brain,

19:37the higher likelihood you have representation

19:39or decodability of all fingers on hand.

19:42So if you're trying to do something

19:43like output text at a fast rate,

19:45there's something that matters a lot

19:46for people who are completely locked in,

19:47who cannot speak at all, who are trying to just,

19:49I love you to a loved one in their family

19:51or ask for a glass of water or a scratch or whatever.

19:53Being able to type at a fast rate is extremely important.

19:56And the more fingers you have access to,

19:57higher probability you can do that efficiently.

19:59And yeah, I'm super excited about how concealing is

20:02you can not get to as we solve this down your track to two.

20:04Yeah, we're currently at approximately 10, 10,

20:06best for a second, be great.

20:07But ultimately we want to get to megabit and I think ultimately whole brain interface.

20:12I think many years from now I think gigabit level is possible.

20:15So that's pretty astonishing.

20:17This is still version one about device.

20:19As we mentioned, it's version one with only 15% of the threads working.

20:23The current device has 64 threads with 16 electrodes on each thread.

20:28Our next device has threads with eight electrodes per thread.

20:31Because as we get more confident about how we're exactly to place the thread,

20:36you need fewer electrodes per thread.

20:38So we can essentially with the current device,

20:40without substantial changes,

20:41to actually double the bandwidth

20:43if we are accurate with the placement of the threads.

20:47And then our next generation device will have maybe

20:49You need even more channels.

20:50Yeah.

20:51So.

20:51A thousand in the field.

20:53Yeah, so next device for every four, yeah, 3,000 channels.

20:56This will just keep getting better and better.

20:58Really moving up, I think, and of course magnitude

21:00in batches of 10, basically, in what kind of bandwidth.

21:03I think one of the people that long before someone with a neural link device can communicate

21:08faster than someone who has a fully functional body and I think faster than the fastest speed

21:13type of or auctioneer.

21:14The esports tournament is going to be like you want to be able to speak faster than someone

21:19can communicate with a neural link to the device.

21:22It may be a very interesting part of this.

21:25We can't connect standard inputs to the computer through apps and keyboards.

21:28very soon as we will have a much harder bandwidth, we need to think about new ways to actually

21:34build the interface for devices. Is this something that we've said?

21:37Yeah, no, that's a good point because the current input devices are centered around human hands.

21:43Yeah.

21:43So it's like we've got these little meat sticks that we move and this is a certain rate of which

21:49you can move your fingers. And so we've got like the mouse and the keyboard and

21:53well the joystick control or something like that. But you really don't need that. You can actually,

21:58you're at need since you're not going to use your hands, you actually don't need those

22:02conventional control mechanisms. And so ultimately, I think you'll be able to do conceptual

22:08telepathy, like where you can communicate entire concepts to someone else within your link or to

22:14the computer. Even today, we have some problems here, right? If you don't feel the mouse clicking

22:18under your finger, how do you know it actually happened? Because you're seeing it on the screen,

22:21but you don't actually feel the mouse clip. You don't have the appropriate septic feedback of

22:24the keys under your fingertips or the trackpad under it. There's all sorts of interesting UX

22:27out to actually give the user some sense of their decoder is actually doing their,

22:32but the neural thing is actually doing how many are trying to do.

22:34Wireless. Yeah, it's Bluetooth. Just a Bluetooth connection, just like how your normal Apple

22:39mouse or like Apple Magic keyboard connects to your computer, same exact name. In fact,

22:43in, yeah, we can basically have this exposed as an HID interface if we want to use just a name

22:46of the protocol for sending bits from a mouse into a computer. Yeah, I can plug into basically

22:51anything. Yeah. Yeah. I think we chose that interface because it's ubiquitous. Yeah.

22:55Basically any devices have Bluetooth capabilities.

22:59Our long-term goal is to actually have our own protocol that is safe and secure, but for

23:03now, we've chosen it for interoperability.

23:05The question is, can a Neuralink chip prepare the paralysis in the long-term?

23:09We can't do that right now.

23:10We have done preliminary work in planting a second Neuralink in the spinal cord, and

23:16we can restore naturalistic looking and leg movements in animal models.

23:22But this isn't something that is, don't hold your breath waiting for it.

23:25going to be a while, we've got a lot of work to do. Yes, there's no reason in theory that we can't

23:30repair paralysis. Yeah, essentially, there's no physics barrier to fully solving paralysis. That

23:36is perhaps a way to say that you've got signals coming from your motor cortex that if they are

23:42transferred past the point where nerves are damaged, essentially just it's basically a

23:47communications bridge. So you bridge the communications from the motor cortex past the point in the

23:52necrospline where the nerves are damaged and you should like it is physical it is possible from a

23:57physics standpoint to restore full body functionality. It's a very hard technical problem but there is

24:02nothing that prevents it happening from a physics standpoint. In terms of next phases of rollout we

24:07really want to make sure that we make as much progress as possible between each neural link

24:12patient. So this is we're only just moving now to our second neural link patient but we hope to have

24:17if things go well, high single digits this year,

24:21and I don't know, maybe this is somewhat

24:23dependent on regulatory approval

24:24and how much technical progress we make,

24:26but within a few years, hopefully thousands.

24:29Yeah, and I think one thing that is important to highlight

24:31is that it's not that we built only one device

24:34and one surgery, we've done hundreds of surgery,

24:36we built thousands and thousands of devices,

24:39even for just the ability to earth

24:41any sort of low frequency failure mode.

24:43We have already been investing very heavily

24:46infrastructure to be able to scale this thing on the device manufacturing side, as well as

24:50on the surgery side of things. We want to be able to help as many people as quickly as we go

24:54through, obviously, the appropriate hurdles that are regulatory challenges and proving out the device

24:58with... Yeah, the device implantation really needs to become almost entirely, if not entirely,

25:03automatic. In the same way that, say, Lasic eye surgery is done, you don't have an ophthalmologist

25:08with a laser cutter by hand that would be crazy, but the ophthalmologist oversees the

25:13a laser machine and make sure that the settings are correct and then the machine does everything

25:17and restores your eyesight. It's really remarkable how many people have had their eyesight restored

25:22with laser, and I think there's another one called smile. They keep making it better. We need to

25:27have something similar for a neural ink implantation so that you basically sit down and whatever the

25:32whatever kind of upgrades or rain fixes are needed. That's reviewed by medical expert. So we

25:37want to make sure that is reviewed correctly, but it really needs to be automatic. So you sit down

25:42and within 10 minutes you have a neural device installed very fast. It's very cyberpunk,

25:47Deus Ex if you played those games. When we nearly start to interface with other devices like

25:52Wilcher, it's a great question. We're currently focusing on controlling computers and unlock

25:56independence in the virtual world. Of course our plan is as you mentioned earlier, Robotic Arm and

26:01Wilcher, I like independence in the physical world. Of course, at the additional risk if you make

26:05mistakes with your computer there. I'm sure it's to that but we are working with the FDA to allow us to...

26:11Well, it seems like if the wheelchair has a...

26:13The wheelchair just seems to have an air face.

26:15So if the wheelchair has a Bluetooth air face,

26:17you could just Bluetooth air face to the wheelchair.

26:20And that's probably something we should do

26:22pretty soon.

26:23It's really not a matter of paperwork.

26:24You don't want to try it off a cliff.

26:25I think we can limit the speed.

26:27Yeah, yeah.

26:27So it doesn't go careening or off into disaster.

26:29But so just we could go slowly at first.

26:32But yeah, so being able to...

26:33Really, the neural device just should work generally

26:36for anything that's got a Bluetooth air face.

26:38Including potentially an Alt in the system.

26:39Yeah, yes, you could communicate with optimists.

26:42Yep, absolutely.

26:43Optimists will also be able to talk to optimists,

26:45but like.

26:46It's not just be management.

26:46But you could just, yeah, instead of talking,

26:48you could just be in it directly.

26:49Or if someone has lost the use of speech,

26:51then they can still communicate to an optimist.

26:53They can communicate telepathically to optimists

26:56or by Bluetooth.

26:57And so even if someone has completely lost the ability

27:00to speak, they could still control out

27:02with their computer or phone.

27:03Also, if you have an optimist and you have a neuro,

27:04you can just directly map the brain signal

27:06to control the physical arm of the robot.

27:08And that's a very meaningful thing.

27:09If you're folks at High Spawning Court Injury,

27:11one of the biggest requests is to be able to scratch yourself.

27:13It's something that quite annoying, actually.

27:15And if you have a scratch on your face,

27:17you can't fall asleep until you scratch it.

27:19It was very convenient to be able to move something

27:20physically towards you to scratch.

27:22Similar things like eating food.

27:23If you need somebody to feed you,

27:24very hard to have dinner with friends in a way

27:26that is a normal social experience.

27:29And so if you can feed yourself, pick up a fork,

27:31and actually eat a pizza chicken on your own.

27:33That's a big deal.

27:34It prevents and saves a lot of interactions with caretakers

27:37and other people in your life

27:38that you rely on to take care of.

27:39It really increases your.

27:40I think an exciting possibility long-term also

27:42is to say if you take parts of the Optimus humanoid robot

27:46and you combine that with a neural link,

27:48let's say somebody has lost their arms or legs,

27:50we could actually attach an Optimus arm or Optimus legs

27:54and do a neural link implant.

27:55So that the motor commands from your brain that would go

27:59to your biological arms now go to your robot arms

28:01or robot legs.

28:02And again, you'd have basically cybernetic superpowers.

28:06Actually, so the latency from the neural link to your hand would probably be slightly

28:09faster than it is just to go to your physical hand.

28:11So you can imagine if you're a piano player or anything that requires extremely fast

28:15hand movements that you could actually have a pretty imbalanced right hand robotic arm

28:18control versus left hand physical hand control.

28:20Yeah, I like to say, it's just a cyberpunk, Deus Ex, in a future where you have cybernetic

28:25upgrades that are actually better than your biological.

28:29And certainly, as particularly as we expand to a large number of customers or patients

28:35for neurolink, the understanding of the brain

28:37will improve dramatically because really,

28:39there isn't a very fine-grained understanding

28:42of the brain today because the sensors aren't good enough.

28:45You've got FMRI, which is pretty good,

28:47but it's still not as good as actually having

28:49high bandwidth electrodes in the brain.

28:50Yeah, I think this is underappreciated as a research tool

28:53to move that whole effort forward of really knowing

28:56what the physical substance of human thought is.

28:59We don't know to the degree that we need to.

29:02So neurolink is actually a very powerful research tool.

29:05Yeah, I think we can ultimately understand and fix quite severe psychosis or if somebody's got,

29:10like the, if somebody's got like a delusion that they have a chip in their brain.

29:13I was wondering if you're going to mention that one.

29:15We just want to be clear, there's only one person within your link chip in their brain.

29:18So for people out there who think we've put a chip in their brain, we'd like to assure you

29:22or what it's worth, you probably won't believe us, but we did not put each of in your brain, okay?

29:27So this actually, or a remarkable number of people who think we have put a chip in their brain,

29:31but we have not.

29:32But in the future, if you would like us to put a particular brain, which will perhaps

29:36help with the issue of thinking that you have a chip in your brain, then we will be able

29:40to do so.

29:41There are people that have severe schizophrenia, they've got basically things that their brain

29:45is malfunctioning in some way.

29:47And this is actually due to really physical circuitry issues.

29:50You can think of the brain as really, it's a biological computer.

29:54And if some of the circuits are crossed, it's going to crash or it's going to have issues

29:59that, but with an early device, we can fix those issues and give someone who think has

30:04a safe, sophisticated brain or psychosis of some kind, allow them to live a normal life.

30:09I think that is one of the likely things in the future.

30:11Yeah.

30:12You can certainly imagine, I'm sure people have parents, grandparents who have memory that's

30:16not working as well as it used to be.

30:18Sometimes they forget who their grandchildren are or what date is and this is something that

30:23a early device could help fix.

30:25That's actually one of the personal reasons.

30:26in many way forms of you're literally losing part of your identity.

30:30Yeah.

30:30Which is a great thing.

30:31Yeah.

30:32It's really just a glitch in the biological computer that is a fixable glitch.

30:35It's a short circuit essentially.

30:37How does the device charge and how long does it charge to last?

30:39Yeah, so the current version that Nolan has lasts for the four to five hours on his charge

30:44and it takes about minutes to charge.

30:46The thing we have learned from Nolan is that's actually one of the main limiters for him using

30:49it more.

30:50It's actually pretty hard to use a product more than like 70 hours a week.

30:53But that's about what he has used it for in some weeks?

30:56Yeah, 70 hours a week.

30:57Just for context, you sleep roughly eight hours a night.

30:59So that's, we're doing better than the bed.

31:01The bed is 56 hours a week if you use roughly.

31:04And so 70 hours a week of use is,

31:06I challenge you to think about products

31:07that you've actually used for that duration.

31:08That's the, okay,

31:09in some of these points we're worth emphasizing again,

31:11like that Nolan Nuffus and your link recipient

31:14has used the your link device for 70 hours in a week,

31:17which is incredible.

31:18Probably won't enjoy that.

31:19I'm sharing his computer use publicly.

31:20But I assure you it's for productive things only.

31:23But actually, so one of the things we learned

31:24is that in the next version of the device,

31:25really until double or increase that battery life. And so I think the next version is going

31:29to be double, actually double, without, without increasing the charge. And charge your time

31:33to double the battery life, meaning you should get roughly an hour to use. And the goal is

31:36to actually get to all they use. Yeah. So you can just charge maybe in your sleeve,

31:39or sleep in pillow. Exactly. As soon as you've got 16 hours of usage, then you basically

31:44have 24 hours of usage because it can charge while you're sleeping.

31:47One of the things that's important, I think to call up here is if you're paralyzed, you

31:49can't, you know, put the charger over your head yourself. And so it's important to think

31:52about. It's not just a duration of better use, but also can you recharge it yourself independently.

31:57We spend a lot of time thinking about how to make that feasible because then that means that you can

32:00this is what no one does. You can use the device, charge it, use the device without needing anybody

32:06to come in and help you with that. Which is a big deal if you're trying to play a set of until

32:105 a.m. a night when your family's in. The way in which he does that is that there is a charge of

32:14coil that's a big one about this big and we actually put it in the sleeve of a bat. It's a beany,

32:20Or a beanie and then he wears it and then says with the voice command charge charge energize

32:25That's the one you're writing work. Yeah, the current device that Nolan has is reading

32:29So it's trying to read his essentially horse movement from one one hand

32:33That's also worth pointing out in the future

32:35They were pretty cool to Nolan the second implant that would allow the other hand to be used and also have higher

32:41Obviously higher active electrode counts

32:42So then you could play to essentially play games to add it because that's not only how you play games

32:46But then with writing, it's really just, it's electrical impulse.

32:50Instead of reading electrical impulses from the neurons, you issue an electrical impulse,

32:54which is obviously critical for vision.

32:56So vision is writing, which is just trading an electrical impulse in the vision part of

33:00the brain, and that activates a pixel.

33:02So we actually do have this working in monkeys, and we have a threat working with monkeys

33:06for a while now, where you can flash a pixel and then you watch where the monkey, obviously

33:10the monkey is like, what?

33:11The monkey is a little surprised to see, hey, there's a flash here and a flash here, but

33:15It's used to it after a while, but it just you can see that the pixel is in the right location because the monkeys eyes will dark to that location

33:21It's not on the screen. There's no pixel on the screen. There's no pixel on the brain

33:24Yeah, I think just verify that that you're triggering pixel in the right part of the brain

33:28So the initial resolution for vision will be relatively low sort of Atari graphics type of thing

33:34But over time it could potentially be better than normal vision order locations as bliss mentioned

33:39There is feedback. It was appropriate septive feedback. There's a cocktail feedback

33:42Especially on robot arm if you're trying to grasp a call.

33:44Right.

33:45You need to know if you got it.

33:46One-on-one leg.

33:47Yeah.

33:48Yeah.

33:49It's a very much a delicate balance of not just initiating the movement but getting the

33:53feedback and controlling it accordingly.

33:55There is a somatosensory cortex that's right adjacent to motor cortex.

33:59That could be better.

34:01Motor movements.

34:02Any changes in neural growth after the device is inserted.

34:04We don't take any signs of neural damage but I guess we have seen some rebound on some

34:11of the electrodes, right?

34:12Correct. And then also a brain is very plastic.

34:14It's not that plastic. It does diminish quite a bit after each

34:1820. Throughout childhood, actually when you get to about 25

34:21brain really done cooking. There are there is a little bit of damage done with each insertion,

34:26but it's a minuscule amount compared to anything else up there. And it's an easy amount of

34:31damage to recover from and it's really only detectable on cutting pieces of the brain after

34:37the animals no longer alive and looking at them under a microscope. You can't really tell during

34:41life that there's been any.

34:43Another way to interpret this question

34:44have the changes in neural growth that

34:45the device has inserted.

34:46One way to interpret that is the user learning

34:48how to use the device.

34:49I think on that side of things, there's

34:51been tremendous progress.

34:52He's put in hundreds of hours trying

34:54to figure out the best way to use this device,

34:55because he really thinks that.

34:56He can figure this out, and he helps share this knowledge.

34:58He's like, I'm Friday night at 8 p.m.

34:59And he's starting a session of figuring out himself

35:01how to push his own performance to the next level.

35:03And that's really a unique learning process,

35:05because there's not many people in the world

35:07that have had the experience of moving something.

35:09And so there's a lot of nuance to, OK,

35:10how exactly should I imagine our attempt to move my wrist to get the thing to him.

35:14He's really dialed that into it.

35:15So just a sheer number of hours that he's using even in the past six months, right?

35:19In many ways, he's using it in his travel and his plane ride.

35:23Effectively, PCI has left a lot.

35:25One of the questions is how close we're converting thoughts into text.

35:28Right now, it's more about moving a curse from the screen on a virtual keyboard,

35:31but long-term, you should be able to really estimate entire words

35:36faster than any more compulsory type.

35:37able to type a low world today. But we're still in the early days making that a positive experience.

35:41The other things that we're looking at is signaling. At the end of the day, it is a movement of

35:45into... Yeah, that's true. ...is the brain trying to naturally push the threads out. This is a universal

35:50feature of any implant in the body. The body tries to reject it and the goal of the surgeons and the

35:55technology team is to fight that. And so with artificial hips and with screws in the spine,

36:00we've done a really good job of finding biocompatible materials and techniques to fix those implants in

36:06in the body, past a certain age, it's getting hard to find someone without some kind of implant.

36:11Me kind of screws in their spine.

36:13So we've got this problem pretty well solved to answer your question.

36:17Yes, the body is trying to get rid of any implant, but we can ensure that basically can't.

36:22It's also worth highlighting that the threads have not actually moved in the past five months.

36:28There's still minor movements in terms of some maybe getting pushed in a little bit, pushed

36:31out a little bit, but it's more or less very stable and stable for them.

36:35The reason for that is, you know, when you do a brain surgery,

36:38it takes some time for the tissues to come in,

36:40and then, you know, the heart tissue,

36:42or the neomembrane to actually come in

36:43and then anchor the threads in place.

36:45And once that happens, it has been stable

36:47and it's seen much movement.

36:48That's where the world record performance starts to come in.

36:51Yeah, it was a couple weeks ago.

36:52Yeah, the threads, it is important that the threads

36:54be extremely tiny.

36:55If they're extremely tiny, then the brain does not,

36:57the smaller they are, the less likely the brain is to react

37:00to them.

37:01That's why you want the threads to be extremely tiny

37:03and also to minimize any damage to neurons.

37:05On that note, we do plan to actually share some of the tissue response in detail in some of the later upcoming updates.

37:12Yeah. It is quite a challenging, it's challenging on many fronts to do something like this because you're trying to read and write electrical signals

37:20that you need to have threads themselves need to be electrically isolated and not subject to corrosion in the body.

37:26So like the just metal by itself is somewhat such a corrosion or being attacked.

37:31So it's like in terms of the perverse coatings and things to actually make this electrode work

37:35while not actually eroding its performance over time is very difficult.

37:39Human body is a very harsh environment.

37:41Very harsh environment.

37:41It's a bag of salt water with bath sensors that's elevated temperature that is well regulated.

37:47I'm sure people have experience dropping their electronic devices in a seawater and in an instance.

37:52So we've got to wrap this up soon. There's like a few last questions.

37:56I guess so a good question. So what about upgrades?

37:59So yeah, we do think it's going to be important to be able to upgrade the device over time.

38:03Just like you wouldn't want like an iPhone one stuck in your brain forever.

38:06If you've got an iPhone 15, you probably want the iPhone 15, not the iPhone one.

38:10So I think people over time will be able to upgrade their new link.

38:13So you're able to take your new link device out and put a new one in.

38:16And we have done this with some of our animals and they're actually in one case we did it with,

38:21we upgraded device three times and with a pig.

38:24We did it with a monkey as well.

38:26Yeah, yeah.

38:27Yes, and my team's doing fine.

38:29Yeah, Pedro has a lot of latest employees.

38:30I actually hit his record though last year with an upgrade.

38:34It's still beat him though.

38:35Still beat him? Yes, this is true.

38:36Humans are top of species lead award right now.

38:38Pages like what, like eight or something?

38:39Pages like 8.5 BPS.

38:41Okay.

38:42Well, that's a very high score.

38:43Yeah.

38:43I'm not trying to put Pedro down.

38:44And also to train a monkey to do that,

38:46it's like a whole challenge on it.

38:47So we have the best animal care team.

38:49Yeah, I just do want to emphasize we do our absolute best to take care of the animals.

38:53And when we had our USDA inspector come through,

38:55She said that this was the nicest animal facility she's ever seen in her entire life

38:59So the monkey orders room service

39:03Yeah, we have monkey room service, which is rare. We're the only ones who offer monkey room service

39:07We really do everything we can to maximize welfare of the animals. All right with that. Thank you everyone for tuning in. Hope you found this interesting