Musk’s brain-computer interface company announced Thursday a video of a 9-year-old macaque named “Pager” playing a video game of table tennis. Pug has a Neuralink device implanted in his brain. In the video, Pug uses its brain activity to learn to control a computer to “taste a delicious banana milkshake provided through a metal straw.” When the Neuralink personnel disconnected the joystick, Pug was still playing the game. Musk said in a tweet on Thursday that Neuralink’s first product will allow paralyzed people to use their minds to play with their phones faster than those using their thumbs. However, this future seems far away.

As a new technology like “black technology”, brain-computer interface has emerged in recent years some new research and applications. For example, studies have shown that patients with cerebral palsy can drink beer with a robotic arm through a brain-computer interface. And Neuralink is undoubtedly a leader in the field of brain-computer interface. As early as July 2019, Musk said: “Monkeys can already be implanted with a brain-computer interface, and only use the brain to control the computer.” Previously, Neuralink had publicly demonstrated the experiment of implanting a brain-computer interface in pigs. Musk said: “A new goal in the future is to transmit signals from the Neuralink device implanted in the brain to the Neuralink device in the main nerve clusters of other body parts, so that paralyzed patients can walk again.” Musk also revealed that the device can Wireless charging, so the implanter looks no different from other people, and there will be no abnormal feelings.

I understand that many neuroscientists don’t like Musk because of his violent talk, but it is better to discuss things as much as possible. At the same time, public opinion is easily driven away by Musk’s propaganda, thinking that he has really achieved some breakthrough, and neuroscience is really like a long night before he injects capital. This is not good either. The following is my rough view: Mind playing games, monkeys playing ping-pong, these are not breakthroughs. Brain-computer interface has been a technology for more than ten years, and my colleague of ping-pong game has also been doing it for more than two years. Pay attention to the right hand of the monkey holding the straw, you will find that it actually controls the racket by controlling the right hand. The principle is also very simple. The neural signal used to control the racket is collected from the motor cortex and the area related to the hand. So that a simple single-layer neural network decoding is sufficient. Source: Neuralink However, Neuralink contributes. First of all, its equipment is very good. The main features include 1. No wires can be seen. I don’t know how it is powered, there may be hidden wires connected to the battery? Maybe the implant itself has enough power? In any case, it is much lighter than the electrodes commonly used in academia. (Added by Ma Daha in the comment area: wireless charging) 2 More than two thousand signal channels, in principle, it is possible to collect information on the activity of a single neuron of this order of magnitude. The electrodes commonly used in academia now require wires to be connected to external devices, which are very fragile and do not have so many channels. 3 On-chip calculations. There are many steps from electrode collection to nerve signal: signal amplification, filtering and noise reduction, digitization, and action potential detection. Neuralink’s implanted chip completes all these operations, and the rate at which Bluetooth outputs action potentials. This is not a new concept. I have seen some concept machines, but combining 1 and 2 needs to solve the power problem, and it is not easy to do a good job. Since there is no public data to see the signal quality, I will put a question mark for the time being. 4 Surgical robot. This can save a lot of time, and more importantly, it can be more accurate and repeatable than manually implanting electrodes. In addition, the objective function of the private enterprise is very different from that of the scientific community: the academic community pays more attention to publishing articles, and the enterprise can aim to treat the paralyzed market. Therefore, I personally welcome companies like Neuralink to develop brain-computer interface technology.


By zhiwo

0 0 vote
Article Rating
Notify of
Most Voted
Newest Oldest
Inline Feedbacks
View all comments
7 months ago

In the past two years, whenever Musk’s Neuralink released a new demo, the neuroscientist community on Twitter was a lot of ridicule, and then everyone followed again to review the literature of the BCI field for nearly 20 to 30 years, and every time I could I found one or two pieces of remains that I hadn’t paid attention to before. BCI’s classic work, such as those from Schwartz Lab, Shenoy Lab, and BrainGate, etc., have been written into the undergraduate textbooks, so I won’t mention them anymore (in fact, I mentioned them on Zhihu six years ago) . This time, I learned about the neurosurgeon of WashU (WashU University in St. Louis) who used ECoG (cortical brain electricity) to make a brain-computer interface to play the Space Invaders (Space Invaders, aka “Little Bee”) project in 2006. , It is also the cursor that controls the one-dimensional movement of the mind, and it has one more shooting action than Neuralink’s monkey. For another example, Frank Guenther, a neurolinguistic tycoon, published a PLOS ONE in 2009 (yes, the PLOS ONE that was rejected by everyone) and proposed a wireless brain-computer interface system called Neuralynx (Is the name familiar?) [2], using implanted electrodes to read the speech motor cortex signals of stroke paralyzed patients, and wirelessly transmit the signals to the decoding system, and finally synthesize speech. Therefore, whether it is “wireless transmission”, or “motion control”, or “playing games”, these technologies themselves are indeed prototypes that have been made in the laboratory more than ten or twenty years ago. In neuroscientists It seems that to some extent it is really worthless. Now if anyone wants to write a proposal to NIH or DARPA that I want to make a wireless BCI system that allows people to play games through ideas, they will obviously be questioned by their peers about the innovation, so it is difficult to be funded through review. Therefore, these wonderful prototype systems exist as the scientifically important “first time”, and then they are forgotten in the ocean of paper, waiting to be turned over again and again to be used to refute certain attempts to reinvent the wheel. proposal. This is the difference between academia and industry. Academia is only responsible for creating and solving the “first time”. Once a concept or technology is developed, subsequent repetitive and improved work is often not affected by the academic world. Special attention. The industry is to continuously improve and optimize existing technologies to achieve the optimization of costs and profits, and continue to develop new application scenarios for existing technologies. The latter is precisely because electronic technology, computer science and Internet technology can flourish, and the EE of 20 years ago, the CS of today can become an important reason for the sweet pastry: the development of the industry has driven a large amount of scientific research funding from the industry. , And the investment of funds has spawned more new technologies, thus forming positive feedback. Utilitarianly speaking, as a neuroscientist, I hope to see more people and companies like Musk and Neuralink, so that the cake can be made bigger, so that everyone will not struggle with NIH year after year. % On the funding line. Of course, the premise of all this is that Mr. Ma can respect the neuroscientists who provide them with the core technology… However, it must be emphasized that at this stage our technical reserves may not be enough to support a large number of people like Neuralink. The company exists. I am ashamed to say that the research results I posted in the previous article are all the technology of more than ten or twenty years ago. Today, we are not much more advanced in technology than it was then, and our understanding of the nature of neural coding is not much better. There is no fundamental breakthrough than twenty years ago. So after finishing my tongue, I still have to continue to write the grant proposal, continue to modify the paper, continue to type the code, and then silently envy the former colleagues in the next laboratory who joined Neuralink and may now be about to realize the freedom of wealth. Finally, to summarize: 1) If we in the neuroscience community can learn 20% of Musk’s PR skills, it is also a huge improvement in the entire field; 2) Everyone continues to work hard to post articles, maybe after two years your wheels will be Musk. Or Niusk made it; 3) Wait until the day when everyone’s wheels are crazily copied by the industry, I hope you remember to come back and shout with me: The 21st century is the century of !

7 months ago

I think from the thinking of Musk, he will try to treat patients first, and then try to promote it on a large scale. All science and technology companies created by Musk are all innovative and open new tracks. At the beginning, the audience is niche, and they strive to scale, reduce costs and increase gross profit, and then invest in R&D and innovation to open up another new track. Tesla Motors, Falcon carrier rockets, Starlink Starlink, etc., of course, mind control equipment is a great boon for paralyzed, disabled, and vegetative patients, but if combined with mechanical peripherals, it is also extremely useful for improving production capacity. helpful. Musk’s science and technology empire has always had a background of integration: exoplanets survive, and every time he starts his business is not aimless, but instead focuses on research and development around this goal or derivative industries. For example, Starlink seems to be an interconnected communication between heaven and earth, but is essentially preemptive. One of the few profitable sub-industries in aerospace is used to capture profits to provide financial support for the development of the moon, Mars base and BFR rockets. However, I can’t think of the relevance and intersectionality between the brain-computer interface and the survival of outer planets. I attach a few conjectures: 1. Vacuum and exoplanet surface environment mechanical peripherals fine operation 2. Improve the production efficiency of non-artificial contact operation in space 3 .Musk and other customers’ awareness uploads to complete immortality (tips from friends in the comment area) Musk’s current technology industry concept: humans need cheap vehicles to go to outer space or large-scale launches of spacecraft, SpaceX’s Falcon rocket and Crew Dragon spacecraft, BFR rocket. Human beings need cheap and more powerful vehicles to travel to outer planets on a large scale. SpaceX is developing a methane rocket BFR that can be recovered in the first and second stages. Human space life energy can only come from light energy, Tesla’s solar project Solar Roof. The energy collected by human life needs to be stored, Tesla’s PowerWall, PowerPack, and MegaPack. Humans need to use vehicles on planets without oxygen, such as Tesla’s Model X, Model S, Model 3, CyberTruck, Semi, etc. Human beings live on the surface of a non-atmospheric planet with risks of falling celestial bodies and high-intensity radiation. Boring company specializes in civil excavation and tunnel construction.

7 months ago

In what age, we are still focusing on the foreign affairs of others. Don’t you know that our country is leading the world in the field of human brain-computer interface? In 2021, Musk is still doing experiments on monkeys and publishing videos to be complacent. We have already achieved important clinical results in humans in January 2020. The brain-computer interface technology was jointly researched by the team of Professor Zhang Jianmin from the Second Affiliated Hospital of Zhejiang University School of Medicine and Professor Zheng Xiaoxiang and Professor Wang Yueming of the Qiushi Advanced Institute of Zhejiang University on the basis of animal experiments in 2012 and preliminary human brain experiments in 2014 . On January 16, 2020, the team implanted Utah array electrodes in the brains of patients with high paraplegia, then used algorithms to analyze the brain nerve signals collected by the electrodes, and output command signals that the robotic arm can understand, finally realizing the patient’s “ideas” Manipulate”. Judging from the experimental results obtained in our country, does Musk need to “declare” the fact that people who are paralyzed in the future can use their minds to play with mobile phones? The future development prospect of brain-computer interface must be in China! Great, my country!

7 months ago

This kind of technology just assists the brain to work, it is completely fine, it is a kind of technological advancement. The problem is: they don’t even let the browsing history of your mobile phone go, they even secretly take pictures of your stay in front of the shelf, and they can even collect and push merchandise even what you say casually. None of these actions have been authorized by you, they still did. Do you think an interface that can directly read thinking data is really just an interface in their eyes? Compared with the so-called “big data” of those browsing records, in the eyes of capitalists, the human brain is the largest database and the largest treasury. Unauthorized reading also means unauthorized writing. If they can delete the accident record on your car from the background, they can also fake a record and upload it to you, there is no technical problem at all. Brain-computer interface technology will inevitably develop rapidly in the future. Wouldn’t it be possible to stop many crimes by reading the mind directly? This reason is enough that everyone will be forced to install in the future. Regardless of what he said, please be cautious.

7 months ago

When Musk was engaged in electric cars, I: Wow Musk is great! When Musk was engaged in rockets, I: Wow Musk is great! When Musk got nervous, I: Oh, cooperating with him is a big flick… At least on Neuralink, what he did was something that neuroscientists could do years ago. Every time Neuralink makes a big news, it gets mocked by neuroscientists on Twitter. About half a year ago, Musk tweeted to recruit people for Neuralink, saying that he does not need any neuroscience background, because now people know almost nothing about neuroscience. We will teach you our cutting-edge neuroscience research results. Neuroscientists: Oh (indifferent face). Now, neuroscientists: So now you take out what we did more than ten years ago to “teach” us?

7 months ago

That’s right, I’m here to smash the scene (disenchantment). I will question the video and the project from both ethical and scientific directions. If you don’t want to look at ethical questions, you can skip 1 and go to 2. 1. I am curious how this monkey was trained to sit and play the world’s first console game “Pong” so meekly. Let me talk about some conditions that I think have a high probability: their training methods are very powerful, which may involve some tricky operations that do not pass ethical audits under normal circumstances. They are very rich and can buy a lot of monkey experiments. They don’t care that the screen is broken or the wire is torn off. (Update: I took a closer look. The screen is fixed on the glass, and the wire is passed around behind the glass, which avoids the hidden danger of monkeys. There are glass used in many places in the entire experimental facility.) Anyway, It’s impossible not to spend a lot of energy behind sifting out such an obedient and willing monkey to work. (This does not rule out giving monkeys drugs to stabilize their emotions or make them slightly addictive ← I guess, after all, they are commercial companies, not academic groups. What tricky operations they do internally can more easily avoid ethics. Review.) 2. A more normal idea is that the brain area they record may be PMC or simply PFC (update: should be PPC, posterior parietal cortex), by recording the brain signals during hand manipulation, to find the brain The representation mode of the motion planning (using this part of the data can get a trained decoder), and then in the subsequent non-manual control stage, the decoder obtained before is used to decode the brain’s motion planning information, which is directly reflected in the game’s operable The counterpart-the control of the baffle. (But in fact, you can construct a closed-loop task to allow the monkey to actively learn to use the signals from your recorded brain area to manipulate its counterpart: For example, at the beginning, every time you imagine an apple, it corresponds to downwards, imagine a pear, Corresponding upwards… Over time you will adapt. Apple slowly migrated directly downwards and no longer need to transit through it. After all, the plasticity of the brain is very high.) But in this video, there is a big problem. Yes, why are the tasks for monkeys different before and after. Someone would think that this is not better? After all, the latter game “Pong” is obviously more difficult! Is it really so? This is not the case. The degree of freedom to operate the Pong game is very low. You only need to decide “when to start moving the baffle, whether to move up or down”, which is OK, it is essentially a real-time binary classification task, and the decoding accuracy The requirements are not so high. But their first task was done with a handle, with a higher degree of freedom, and the direction of movement became a two-dimensional direction coordinate. This vectorized direction information can be discretized by angles. For example, we divide it into 360 directions. In terms of the difficulty of decoding, this is much more difficult than the previous binary classification task. This is just the direction of a stage, it doesn’t count how many times the direction needs to be changed during the whole stage, and no parameters such as speed are added. So there is nothing to be surprised about this thing. Commercial operations, deceive investors to continue to take money obediently. update: I neglected the middle part of the video. The first task ended up being a version of unplugging the joystick. This shows that neoalink’s equipment can indeed achieve target tracking on a two-dimensional plane, but unfortunately, this task (called visuo-motor tracking task, visual motion tracking task) is a very classic task in the field of motion control. Someone has already done it with neural decoding. As long as it detects those neurons that release motor intentions, this task can be done, because the simultaneity requirements are not so strict (with some delays). The difficulty in the current field is the planning of the target point motion control. If you have a sufficient understanding of the coding method of the motion planning information, you can design better tools for faster decoding-there may be almost no delay. But it seems that this is not reflected in the video, they bypassed this problem. Because they used Pong-this two-category, actually a simpler task than the first task. After all, when decoding the motion signals sent by the brain, it is difficult to have both speed and accuracy. This video of neuralink also fully illustrates the bottleneck that everyone is currently facing: if you want to be fast, do simple tasks, if you want to be more accurate, then slow down. This also shows that their research has not made any substantial contribution to the field of motion control. Although I was smashing the scene, I also think this is a direction worth doing in the future (but big breakthroughs require technological changes brought by other fields + new insights into brain signal coding methods in the field of motion control). Finally, I intercepted my previous neurolink’s answer under the pig related research and the evaluation of their system: all the technologies used have no underlying innovation, but the system integration is done well, and we can see that the goal is indeed commercialization. In addition, in addition to the limitations of the hardware itself, regardless of the input of the signal, a relatively large bottleneck of the brain-computer interface is how to efficiently and accurately perform real-time neural decoding. Even if technology advances and the sampling of neuron signals is no longer so sparse, it does not mean that we can accurately decode the information represented by the brain through more neural signals, especially those involving many advanced cognitive functions. signal. Even for the relatively simple behavior we think of as exercise, the brain’s planning and prediction of subsequent exercises involved in the exercise preparation stage is very complicated. Therefore, the worries of the public and some media are still unfounded. At least scientists have not found a universal and efficient decoding tool with commercial prospects. But if you are more optimistic, as the difficulty of data collection decreases and the amount of data continues to increase, the emergence of related tools may only be a matter of time. In short, the scope of application of this thing is still very limited. If it is to be widely used, in addition to solving various technical problems on the hardware, the neural decoding tool on the software must be further improved. Progress in the field cannot be seen so quickly. Scientific research requires one step at a time to gather accidental discoveries into insights for change.

7 months ago

I made it last year, but Ma Chuan still doesn’t believe it: as long as he throws out some things that have been tired of MIT labs to fool investors from time to time, he can deceive wave after wave of advocates chanting “Iron Man”. , Do you know why the pig was used this time? Because the mouse was used last time, the next time I change to an advanced primate like monkey, and then change to a more accurate behavior analysis of fancy, I can continue this project. In our field, it is really not a new thing, even mind typing is not a new thing, there is no theoretical and technical difficulty, and what Madison is playing is poor information. Let me predict that next will be paralyzed patients, gradually freezing people or conscious vegetative mind typing, or mind doing something else. After all, it is a ready-made technology, tell a good story, change a bgm, and come with the tears of joy from the family. All the onlookers stood up and cheered, and the new wave of leeks grew vigorously. I’m thinking about what he should do next, because the hot spots for publicity are running out soon, I’m afraid that I will not use implanted electrodes, but use patch-type EEG electrodes, and then make it full of technology. His helmet or headgear looks like it is really pure flicker. In fact, someone has invested in this field for a long time. Obviously, there are still many difficulties in commercialization in the short term (because it is really not practical and not profitable in the short term, and it is often cold after a wave of heat). Only he can make the momentum so big and let all People think that this field will not work without him. So I have been saying that he is a pioneer, because only he can bring this field to the public, and actually put money in to promote the development of this field; it is indeed a big flicker, and it will take less than ten or eight years to mature commercialization. The thing, but it blows every three to five. In my opinion, it is a concave design + extending the life of Tesla stock. Every time you answer this kind of question, it will eventually become a team game.

7 months ago

It is reported that two Neuralink devices were implanted in the brain of this macaque. Musk once described this device as a Fitbit (a smart watch) worn on the skull. Through a direct neural interface, it can greatly increase the transmission bandwidth from the cerebral cortex to an external digital carrier, by at least 1,000 times. , 10,000 times is also possible. You can see from the video that this macaque named “Pager” is using its brain activity to learn to control a computer. At first, Pager used the joystick to “taste the delicious banana smoothie provided by the metal straw”, and then interacted with the computer. When Pager is playing the game, the Neuralink device in its brain reads its brain activity, and then the computer decodes it. When the Neuralink personnel disconnected the joystick, Pager was still playing the game. Neuralink explained that the device implanted in its brain allows it to play “MindPong” games. Musk said on Twitter that Neuralink’s first product will allow paralyzed people to use their minds to play mobile phones faster than (normal) people using thumbs. He also hopes that future versions can shunt the glial cell signals in the brain to the glial cells in the body’s motor/sensory neuron clusters, so that paraplegics can walk again. However, although the vision is good, this demonstration is still only a stepping stone for Neuralink, because it hopes to provide the same capabilities for humans. This will bring some challenges, because the paralyzed person will not be able to move the joystick to help model the brain activity. Therefore, the “future” mentioned by Musk seems to be far away. But there is no doubt that this presentation is another big step forward for this ambitious company. In the last demonstration, Neuralink implanted a computer chip into the pig’s brain, and showed live demonstrations of reading the pig’s brain activity signals and the previous experimental video of successfully predicting pig behavior trajectories

7 months ago

In the third season of Black Mirror “Black Mirror”, there is an episode where a game player becomes a volunteer of this kind of team, brings a machine and starts thinking about entering a horror game. He saw a lot of things he was afraid of, and he also saw his girlfriend in real life, but his girlfriend later became a monster, and his pain became more and more real…Slowly he couldn’t distinguish between reality and game, and then he couldn’t remember Who I am. Returning to the realistic point of view, the staff said that the man only persisted for 0.01 seconds before he was in shock.

7 months ago

You can learn about black rock, a company that makes neuroscience instruments and has its own brain-computer interface department. People have done this for many years, and it has already been used by paralyzed patients. When the neuroscience annual meeting was held the year before last, they took the patients directly to the venue, demonstrated on the spot, grasped things, and operated the cursor. I also talked to the paralyzed patient, and he said that he asked Black Rock to let him play Final Fantasy with a brain-computer interface. When I thought that it was not just a mouse problem, but also the mapping of so many keys on the keyboard, I asked him if it was done? He said that he had been playing for a long time, but the calibration was too troublesome and a little tired. There are not many neuroscientists concerned about such a technology. Because after all, it is the signal of the primary motor cortex that is recorded. In the primary motor cortex, the cells are directly connected to the muscles, which is not magical at all. It is a purely technical problem. So speaking of it, neuralink, the monkey mind pong, is nothing new in the scientific circle. In the industry, black rock and other companies play the rest. The only point is to prove that neuralink’s new technology-robot electrode insertion is really useful, not bluffing.

Would love your thoughts, please comment.x