In July 2016, Elon Musk, the founder of SpaceX and Tesla, launched a new high-tech venture. Elon Musk started a new high-tech venture? Neuralink, which develops implantable brain-computer interfaces, though the news didn't become public knowledge until March 2017.
Neuralink virtually disappeared from the public eye for the next two years, until mid-2019, when the? mysterious? company only gained widespread attention when it held a launch event about its brain-computer interface research and published an unpeer-reviewed paper[1] in the preprint journal bioRxiv. At that time, the company had raised about $158 million in funding ($100 million of which came from Musk himself) and had 90 employees including neuroscientists, neurosurgeons, chip designers, biocompatible material scientists, brain-computer interface specialists, microfabrication engineers, and other specialists in various fields. This article will cover the ins and outs of Neuralink's creation and talk about the author's perspective.
Musk showed the latest progress of Neuralink at the launch event on August 28 this year. On the spot, Neuralink showed the state of a piglet after transplanting a brain-computer interface device.
Musk is an oddity among entrepreneurs. The companies he founded all have a sci-fi style, very ambitious ultimate goal, trying to solve a major problem in the continuation of human society; and at the same time there is a near-term goal that can be embarked on, can solve the immediate problem, commercial operations, for the ultimate goal to make technical and financial preparations. For example, SpaceX's long-term goal is to realize human migration to Mars or other planets, while the near-term goal is to develop recoverable rocket technology, reuse rockets, and engage in space commerce.
This is the same model adopted by Neuralink, which was created because of Musk's concerns about the rapid development of artificial intelligence. Since 2014, Musk has repeatedly warned of the dangers of AI on multiple occasions, arguing that it is developing so fast that it could far surpass humans in the future.
August 2014: ? We need to be doubly careful with AI. It has the potential to be more dangerous than a nuclear bomb.?
Two months later: ? If I had to speculate on what the biggest existential threat to us is, it's probably AI.?
July 2017:? I've been sounding the alarm, but until we see robots taking to the streets and killing people, people don't know how to react because it sounds so nebulous.?
April 2018: ? [AI is] a very important topic. It's going to affect our lives in ways we can't even imagine right now.?
Musk also argued that the US government is losing the war to control AI. He said: ? The way regulation is put in place is slow and linear, and we are facing an exponential threat. If you only have one linear response to an exponential threat, then the exponential threat is very likely to win. That's where the problem lies...? [3]
In Musk's view, people are masters of the planet because of their high intelligence and so on. Once AIs surpass humans in this regard, then they will treat us like we treat our pets today. But it's impossible to completely limit AI development, so his prescription is to develop brain-computer interfaces that can be implanted in the brain, allowing the two to merge into one before AI fully overtakes humans.
Neuralink's ultimate goal is to create a "whole-brain interface" (whole-brain interface) that would allow virtually all neurons in the brain to communicate smoothly with the outside world. Such a device would be so fully integrated into the brain that the whole-brain interface would be "part of the body" in sensory terms . This interface would allow the brain to wirelessly connect to the cloud, and thus be able to connect to computers, and even to other brains that have the same interface. In this way, the exchange of information between the brain and the outside world becomes a breeze, and feels like the same kind of thinking that goes on in your own mind.
The goal of creating a whole-brain interface is to enable people to "live with AI***" and become superhuman. At that point, Musk believes, ? when a person dies, he already has his own computer extensions and online extensions, like an online ghost? that? You exist more in the? The cloud. inside, rather than inside your body which is what some people call digital immortality.
Truly every brain neuron has to be built on a? micro-level electrode-neuron interface? , not only do we have to take into account the huge number of neurons in the brain (80 billion or more), but also the limits of current technology? Only a few hundred electrodes can be placed in the brain, and each electrode can measure up to about 500 neurons at a time simultaneously. This makes it impossible to measure 80 billion neurons at the same time. So the number of neurons that can be recorded simultaneously (what Musk calls ? bandwidth?) becomes the bottleneck for a whole-brain interface.
The second difficulty encountered by the Neuralink program is how to implant the electrodes in the brain. All current technologies for noninvasively recording brain activity are either poor in terms of spatial resolution, far below the level of recording individual neuron activity, or extremely poor in terms of temporal resolution, which does not allow for real-time recording of rapidly changing brain activity (Figure 2). Needless to say, single neuron recording, even of a small number of neurons, current techniques require invasive craniotomies, which are only acceptable if the patient's life is in danger. But Neuralink's long-term goal is human-computer fusion, which requires healthy people to undergo open-heart surgery as well. In this way, safety is a major obstacle. In addition, intracranial implants are expensive and can only be done by highly skilled neurosurgeons, making them difficult to popularize. According to Musk's vision, this technology should be as convenient as laser correction of nearsightedness to work.
The temporal and spatial resolution of the main techniques for recording brain activity.
Even after overcoming these two major bottlenecks, Neuralink faces other serious obstacles: communication between brain and machine should be wireless; implants need to be biocompatible, not cause rejection, and can be used for a long period of time in the intracerebral environment; how to place a large number of electrodes in the limited intracranial space; and how to process massive amounts of data in real time and obtain useful information from it.
According to Musk's usual strategy for founding companies, Neuralink's near-term goal is to develop medically implantable brain-computer interfaces that can have practical applications in the immediate future, while also preparing the technology and raising funds for long-term goals.
The devices could be used in epilepsy patients, giving early warning before a seizure occurs and reminding the patient to take his or her medication. Or to help quadriplegic patients use brain signals to control robotic hands or computer screen cursors.Neuralink plans to bring such devices to market around 2023.
Musk's timeline
By the end of 2020
Implant brain-computer interfaces into the human brain, first for quadriplegic patients.
In 8-10 years
It's possible to implant brain-computer interface devices into normal human brains, although that will "depend a lot on the timing of regulatory approvals and how well our devices work on people with disabilities."
In 10 years
It is hoped that it will be possible between healthy people to ? telepathy? (telepathy), in which two people, both implanted with brain-computer interfaces, can communicate directly with each other using brain signals.
Within 25 years
it is hoped that whole-brain interfaces will be developed, in which all of a person's neurons will be able to link up with an AI vector and use the AI as an extension of their own brain activity. In addition, the brains of healthy people with whole-brain interfaces could constitute giant brains that could communicate directly with each other, and what new phenomena such giant brains would produce we cannot yet imagine.
Already accomplished
In a paper published in early August 2019, Musk and Neuralink described three major accomplishments they've made so far:[1]
1 Flexible multi-"filament" electrode array
The filaments are very thin and flexible, roughly 4-6?m wide, and contain gold electrodes overlaid with a multimeric Each electrode extends outside the filament in a small piece to receive the signal, and these small pieces are arranged in a string along the filament. Compared with the electrodes commonly used in current brain-computer interfaces, these electrodes are very soft and can move with the tiny activities of the brain, thus causing less damage to the brain. At the same time, compared with the number of electrodes in the multi-electrode arrays typically used today, the number of electrodes in the multi-"filament" electrode arrays has been increased by an order of magnitude, with as many as 3,072 electrodes in the 96 strands of filaments in each array.
(a) Brain connected to AI, we can? not die or die?
(b) Electrodes? Filaments? Each filament includes 32 electrodes. (a) Appearance of the electrode filament, with the electrodes sticking out of the filament in a row of small pieces that are centered at a distance of 50 ?m. (b) Enlarged electrode filament. [1]
2 Surgical robots for implanting electrodes
Electrode wires are both thin and soft, and there are so many of them that they need to be implanted accurately into the brain in a short period of time, which is clearly impossible to do by hand. Musk's team has developed a neurosurgical robot that looks like a combination of a microscope and a sewing machine. The robot has an automatic implantation mode and can implant up to six filaments (192 electrodes) per minute. Each strand of wire can be implanted individually into the brain with micrometer precision and is able to avoid surface blood vessels and target specific brain regions.
While the entire implantation process can be automated, the surgeon retains complete control and can manually fine-tune the position of the filaments before each implantation into the cortex, if desired. Utilizing this system, the team achieved an implantation success rate of 87.1?12.6% (mean?s.d.) over 19 procedures.
3 Implantable chip for brain-computer interfaces
Neuralink's third achievement was the development of a chip that can be implanted in a brain-computer interface. The electrode array is packaged in a small implantable device with a low-power custom chip that amplifies and digitizes the signals recorded at the 3072 electrodes. The entire 3072 channel package takes up less than (23?18.5?2) mm3 of volume. A single USB-C cable transmits the data recorded on all channels simultaneously.
Musk's team has already installed the system in a rat brain, reading information from 3,072 electrodes, which is at least an order of magnitude higher than current brain-computer interfaces buried in the human brain. They later applied a similar system to monkeys, realizing that monkey brains control computers.
Brain-computer interface device permanently implanted in a rat. [1]
Next goal
The prototypes that have been successfully developed to transmit brain signals can only do so using USB ports housed on the animal's head, which is quite inconvenient. Therefore, the next goal is to realize wireless transmission.
Neuralink calls the proposed wireless sensors the ?N1 sensors? They intend to implant four N1 sensors, three in the motor area and one in the somatosensory area, so that they can not only control external devices with brain signals, but also receive sensory feedback and enable two-way communication between the brain and external devices. The sensors will be wirelessly connected to an external device mounted behind the ear, which can be controlled via a cell phone app.
The company believes this could help humans solve a range of medical problems relatively quickly and hopes to test it on five paralyzed patients by the end of 2020 to see if the technology can help patients use their brains to move a mouse cursor and type. The new crown epidemic is currently unknown whether the company's license application to the U.S. Food and Drug Administration (FDA, U.S. Food and Drug Administration) will be approved.
Currently, when placing the system, surgeons must drill holes in the skull to implant the wires, which can cause discomfort for patients. The company hopes that in the future it will be possible to use a laser beam to make a series of tiny holes in the skull and that it will be as automatic, painless and convenient as current laser surgery for myopia.
Schematic of the proposed wireless brain-computer interface implanted in the human body. [2]
Scientific skepticism
Despite Neuralink's impressive pre-existing accomplishments, many scientists are y skeptical of its proposed endgame.In July 2019, an article[5] published on the podcast Verdict questioned Musk's ambitions. In the article, Noel Sharkey, Professor Emeritus of Artificial Intelligence and Robotics, argued that Musk's rampant meddling with the brain in order to keep up with AI is ludicrous, and that there is no research or evidence to support Musk's claim that ? Artificial Intelligence will rise up and kill us? s argument that AI is just a tool and it is people who decide how to use it.
Toby Walsh, a professor of artificial intelligence and data at the University of New South Wales? Toby Walsh argued that the idea that humans need to merge with AI to be saved is dubious - we can't match the speed and memory of computers, but the emotional and social intelligence, creativity and adaptability we possess is what will keep humans ahead of the machines. He also said Musk is known for not being able to deliver on his promises on time, and that neural connections for healthy people might not be available until decades from now.
On May 13, 2020, Jerome Pesenti, head of Facebook's artificial intelligence division and a computer scientist at Carnegie Mellon University, posted a tweet criticizing Musk for not knowing what he was talking about when it came to AI and arguing that such a thing as an artificial general intelligence doesn't even exist yet, and that AI is far from being developed to the point of human intelligence .
It's clear that there's still a great deal of disagreement about Musk's idea of human-machine integration. What is right and what is wrong remains to be tested.
In the case of SpaceX and Tesla, Musk's approach to combining near-term and long-term goals in founding his companies is instructive and promising.
In the author's view, Neuralink's near-term goals also promise to be a boon to people with disabilities, though not necessarily on a timeline as optimistic as Musk envisions. But its long-range goals are quite questionable in principle and practical feasibility, and are very likely to become castles in the air.
SpaceX's and Tesla's long-term goals, while ambitious, are essentially pure engineering, backed by a solid theoretical foundation. In contrast, Neuralink's long-range goals are ? Establish whole-brain interfaces and brain-computer fusion while our understanding of the human brain is still very superficial, so far there is no theoretical framework about the mechanism of brain function, and it is difficult to find out in the foreseeable future. Therefore, the long-term goal of Neuralink is not purely an engineering problem, but also involves scientific issues. And scientific critical issues are often difficult to realize on schedule solely by virtue of massive investment of human and material resources, and sometimes rely on chance and luck, and on a flash of genius.
What is the problem with Neuralink's long-term goals?
First, there are 86 billion neurons in the brain, and measuring the activity of such a huge number of neurons at the same time doesn't seem possible at this point.
Taking a step back, the Neuralink team has set a long-term goal of recording a million neurons simultaneously (I don't know how this number was estimated; in terms of its sampling ratio, it's the equivalent of claiming that if you do a survey of 80,000 people, you'll have a handle on the entire world's population, a dubious presupposition.) While ? Stevenson's Law? suggests that the number of neurons we have been able to record simultaneously so far seems to double every 7.4 years. [4] But this is an empirical law, and it is questionable whether it will work forever; even if it were to continue at this rate, it would take until the end of the century to reach the number of one million, and until 2225 to record every neuron in the brain. However, any process that grows exponentially is bound to slow down significantly or even flatten out at some point due to other constraints.
In addition, it would be impossible to implant 86 billion electrodes in the skull, even with electrodes thinner than electrode wires, because of the limited volume inside the skull. Of course proponents of whole-brain interfaces could argue that entirely different new types of electrodes could be developed in the future, such as using the technology developed around 2011 for neural dust, a one-hundred-micrometer-sized silicon sensor that can be sprinkled into the cerebral cortex and communicate with the neural dust via ultrasound using a three-millimeter-sized device above the nearby soft meninges. But the neural dust itself is already the size of a large neuron, so it's still unlikely that the same amount of neural dust could be dispersed in the brain. New methods like optogenetics or the use of carbon nanotubes have also been suggested, but so far they are just ideas.
Even if it were possible to record the electrical activity of all the neurons at once, there would be a big problem of how to deal with such a huge amount of data, and just recording the electrical signals from the neurons might not reflect the whole picture of the brain. Because the brain is essentially an electro-chemical machine, in addition to electrical activity, chemicals like neurotransmitters and neuromodulators are also extremely important in brain activity, which are not taken into account at all in a whole-brain interface. Another overlooked factor is that the role of glial cells, which outnumber neurons in the brain, is unclear. Regarding these issues and the so-called ? mental uploading? and ? digital immortality? are just myths, the author has been discussing them with Dr. Karl Schlagenhofer. This has been analyzed in detail in a discussion with Dr. Karl Schlagenhofer [6] and will not be repeated here.
Proponents of whole-brain interfaces contend that ? There will always be some universal technology coming in the future that people in the past had trouble believing in? One always underestimates the power of human groups.? While there is some truth in this statement, we cannot absolutize possibility into inevitability. Although human group wisdom is unparalleled, and the development of technology is often beyond the general public's expectations, but this is not the same as saying that no matter what kind of vision will certainly be realized, especially in a limited time.
The author has observed an interesting phenomenon, that is, in terms of unraveling the mysteries of the brain, neuroscientists tend to be very cautious, while technologists tend to be very ambitious and aggressive. This may be due to the fact that technologists do not know enough about the brain, ? the ignorant are fearless? In fact, Musk himself has said, ? Without a full understanding of technology, I think it's hard to make the right decisions.? [4] He also admits that he is the least knowledgeable about neuroscience on the team, so is it right for him to be the one making decisions about the entire project?
In the early days of Neuralink, Tim Urban, a blogger at the well-known tech blog Wait But Why, was invited by Musk to spend some time with the company and talk to most of the founding team. Based on this experience, Urban wrote a lengthy article. [4]
In the blog post, he said:?AI will run on its own because conversations with people are too slow. The faster the communication, the more bonded you are to the AI? The slower the communication, the less integrated you are. The less integrated we are with the AI? the more independent the AI is? The more independent the AI is? The more likely it is to betray us. If the AI is completely independent and has far more intelligence than we do, how can you be sure that their optimal functioning won't work against the interests of humans? If we achieve a close ****ing birth with the AI, the AI won't become independent? It will become you and will have a relationship with your cerebral cortex similar to the relationship between the cerebral cortex and the limbic system.? To minimize the existential threat from AI, his (Musk's) strategy is essentially to make AI power ? owned, governed, and enjoyed by the people???
The above may summarize Musk's core idea in founding Neuralink, though in the author's opinion, the words themselves do not hold water. First of all, AI is not a subject, it does not have subjectivity or self-view, neither consciousness nor will. And all of these qualities have so far been unique to the brain. However, why the brain possesses these qualities is unknown to scientists at this point, let alone for an artifact to possess them. Therefore, in the foreseeable future, artificial intelligence can only be a tool, and its good or evil depends only on the people who use it. The author agrees with some AI experts that self-aware AI should not be developed at all, not to mention that there is no possibility of its development.
Taking a step back, even if the AI has a self and will, Musk's prescription will not solve the problem. His wishful thinking is that as long as people and AI are fused together, AI, even if it is strong, is still a part of people and is controlled by people. In fact, if the two are united, one is strong and the other is weak, it is highly likely that the strong one will dominate. Since Musk is worried about AI surpassing humans, the result of the fusion of the two may not fail to be that AI becomes the soul of the union, and people become puppets and tools of AI. Others who did not merge with the AI were reduced to slaves or pets of such a combine, a fate not much different from becoming slaves of people. As for Musk's claim that this danger can be avoided by allowing everyone to merge with the AI, it is also purely a pipe dream. It's impossible to achieve AI fusion for everyone at the same time. Considering that implanting whole-brain interfaces at the outset is bound to be expensive, and must be affordable to a few rich people like Musk, the dominant group that would be spawned would have the ability to rule over the masses, a prospect just as frightening as the AI tyranny that Musk fears.
The author argues that the dangers Musk describes don't exist in the foreseeable future, and his prescriptions don't solve any problems for the general population. As to whether Musk is a prophet saving humanity on this issue, or just another alarmist alchemist, readers are left to judge for themselves!