In 2017, Tim Urban, a blogger on the well-known tech blog Wait But Why, was invited by Elon Musk to a lengthy visit to Neuralink, the neural connection company he founded, and had in-depth discussions with Musk and most of his founding team at meetings or in private. After the visit, Urban published his summary in a blog post quoting Musk[1]:
I can imagine a bouquet of flowers and have a very clear picture in my mind. But if you want to describe it in words, you need to use a lot of language and words, and you can only describe a general appearance.
You have a lot of ideas in your head, and you have to compress them into words or typing, which is extremely slow to transmit data. That's language. Your brain runs a compression algorithm on the transmission of ideas and concepts. In addition, you have to listen, you have to unzip the information you hear. Data loss during this process is also very serious. So, when you are decompressing and trying to understand, you are also trying to reconstruct the other person's state of mind to understand its source, and to reorganize in your own mind the concepts that the other person is trying to convey to you. ...... If both of you have brain interfaces, you can communicate without compression directly with the other person.
This conceptual communication, musk calls "some kind of non-linguistic consent consensual conceptual telepathy." [2]
Musk's dream is not new. Science fiction novels and science fiction movies such as "The Spokesperson for the Dead", "The Destroyed Man", and "Avatar" have long described scenes where the hearts are directly connected, and in many other thematic science fiction novels, the description of the ideas of other people and even other creatures can be communicated without language, and the brain can directly receive the ideas of other people and even other creatures, and is often associated with human progress, ultimate goals (such as crossing wormholes, all human consciousness is uploaded to the network to meet together) and so on. One of Musk's original intentions in founding Neuralink was also to allow us to communicate directly with "real thoughts" that were not verbally coded.
Of course, Musk is not the first person to propose a "brain-brain interface". In fact, as early as 1994, nobel laureate in physics Murray Gell-Mann wrote in his book Quark and the Jaguar: "For better or worse, one day a person can connect directly to an advanced computer (not through a colloquial language or an interface like a console) and connect with one or more other humans through that computer." Thoughts and feelings will be completely shared, unlike language that may be selective or deceptive... I'm not sure if I'm going to recommend doing it (though if all goes well, it might alleviate some of the toughest problems we humans have ever faced). But it will certainly create a new form of complex adaptation system that is a true synthesis for many. ”[3]
Quark and the Jaguar
The first person to actually put brain-brain interfaces into practice was Miguel Nicolelis, professor of neuroscience at Duke University, pioneer of brain-computer interfaces, and senior expert. In 2011, in his book Beyond Boundaries: The New Neuroscience of Connecting Brains with Machines and — How It Will Change Our Lives," he reported that they had two rats implanted in a brain-brain interface to accomplish a preset task together. At the 2014 World Cup in Brazil, Juliano Pinto, a 29-year-old paraplegic patient, used his brain to control mechanical bones made in NicoLelis's lab and successfully tee off.
On August 28, 2020, Neuralink held its second press conference, and Musk introduced the latest progress and physical demonstrations. In the past few years, they have made significant advances mainly in technologies such as chip miniaturization, surgical robots, wireless transmission, etc., and have taken a big step forward in the transformation from principle to practicality. In 2019, they successfully recorded the neural activity in the pig's brain and predicted the pig's movements. The press conference won enough eyeballs all over the world and achieved a great public relations victory. But there is no innovation in terms of ideological principles, and Musk's advocacy of making people superhuman through the fusion of human brain and artificial intelligence is at least a myth for the foreseeable future.
On April 8, 2021, Neuralink posted a 5-minute video online (see below) showing a 9-year-old macaque named Pager playing a ping-pong game on his computer. In the video, Peggy doesn't need a gamepad (joystick) and can move the ping-pong racket just by "thought" (which is actually just a brain signal), and plays quite well. This is another big progress report on neural computer interfaces since Neuralink's second conference in August 2020, although in 2008 NicoLelis successfully trained monkeys to control robots far away in Japan to walk synchronously through "ideas". For a time, the whole network boiled, and many people were very optimistic that Neuralink had successfully achieved "mind manipulation" and was not far from "transmitting hearts" from person to person.
Neuralink
The researchers trained Peggy to use "mind" to manipulate the ping-pong racket
Nicolelis, however, is unequivocally opposed to Musk's philosophy. At the WeWe Conference of Tencent Scientists in November 2020, Nico Lelis bluntly said that Musk's idea control, memory upload and even immortality of the brain-computer interface is just a marketing strategy, which is of no benefit to the scientific development of the brain-computer interface field. "I don't agree with a word of what he said." [5]
Why does Nico Lelis say that?
To answer this question, we need to see where people have come so far in working on brain-brain interfaces.
First, look at a concrete example of a "brain-brain interface" published by nicoLelis Labs in 2013 [6].
In the experiment, behaviorally trained mice that knew how to decompress the lever according to the indicator light were divided into encoder and decoder groups, locked in two rooms with the same setting, and could not see each other. Both groups of mice were implanted with microelectrodes in the motor cortex of the brain, and the electrode cables were connected by an artificial signal acquisition conversion device. When the coder mouse presses the A lever correctly according to the instructions, the microelectrode accurately collects the corresponding neuronal dense firing, which is artificially processed into a string of high-frequency pulse signals (A signal); when the B lever is correctly pressed, the neuronal firing pattern collected by the microelectrode is processed as a single pulse (B signal). At the same time, different modes of pulse signals are sent to the microelectrodes in the decoding mouse brain, slightly stimulating the cerebral cortex, called intracortical microstimulation (ICMS). When the ICMS is a string of high-frequency pulses (A signal), the A lever is pressed; when the ICMS is a single pulse (B signal), the B lever is pressed.
In this way, it is possible to encode which lever the rat presses, and which lever the decoder mouse also presses. The researchers believe that "the encoding mouse and the decoding rat make the decoding rat rely entirely on the neural patterns of the encoding mouse to reproduce the behavioral choices of the encoding mouse through the brain-brain interface." [5] In this way, "heart-to-heart transmission" is realized.
So, how does the decoding mouse read the "neural pattern" of the encoding mouse? In other words, how does the decoder know that a high-frequency pulse means pressing the A lever, and the pulse alone pressing the B lever? Could it really be connected to the coding rat's mind?
The answer is: the researchers told it.
The study is divided into two parts, and before the experimental part, there are important training phases. The researchers used a conditioned behavioral training method (recalling a Barplov's dog) to let the decoding rat learn to associate different ICMS with different levers. In this way, in the experiment, the cortical discharge mode of the encoding mouse is artificially converted into different pulse signals, and the decoding mouse presses the appropriate lever according to the rules that have been learned long ago.
In other words, the decoding mouse was able to "reproduce the behavioral choices of the encoding mouse" because it learned to respond to the ICMS accordingly during the training phase. The authors did not say whether the decoding rat could do so if it had not been trained. My guess is no. If so, then the decoding mouse does not actually know the choice of the encoding mouse, but the experimenter converts the selection of the encoding mouse into a suitable stimulus that can cause the corresponding action of the decoding mouse, so this is actually just a reflection.
Invasive brain-brain interface
In 2020, the Lab of Luo Minmin, Beijing Institute of Life Sciences/Beijing Brain Science and Brain-Like Research Center, developed an optical brain-brain interface that can transmit information about motion speed from one mouse to another, and control the movement speed of the latter in precise and real time [7].
In the brainstem, there is a nucleus called nucleus incertus (NI), which contains a class of neurons that express neuromedin B (NMB). Luo Minmin's group has long found that the activity of such neurons can accurately predict and control the movement speed of animals. They had two mice (a coder mouse, a decoder mouse) with their heads fixed, but the body could run freely on the treadmill, record the changes in the calcium ion signal of a group of neurons in the undetermined nucleus of the encoding mouse, and convert it into light pulse stimulation of different frequencies through machine learning, which was applied to the same type of neuron population within the undetermined nucleus of the decoding mouse, allowing the movement speeds of the two mice to be highly synchronized.
This work of Luo Minmin's group is certainly a big step forward from the earlier work of Nico Lelis and others, and the activity of the decoded rats controlled is no longer a simple task of "choosing one or the other", but a continuously variable quantity - the speed of motion.
But they also do not use the primitive brain signal encoding the mouse to directly control the activity of the decoding mouse, but to artificially convert the primitive brain signal into a light stimulation pulse sequence, and then use the light pulse to stimulate the decoding mouse. Does this count as "heart-to-heart"?
By inserting microelectrodes directly into the brain, although it can achieve high resolution and signal-to-noise ratio, it is difficult for healthy subjects to accept. Not long ago, the U.S. animal protection organization PCRM (Physicians Committee for Responsible Medicine) complained to the USDA about Neuralink's collaborative research with the University of California,Davis between 2017 and 2020, and PCRM considered it cruel to implant chips into the skulls of macaques.[8]
As a result, many laboratories are also studying non-invasive brain-brain interfaces.
Non-invasive brain-brain interface
The Lab of Rajesh P. N. Rao[9] at the University of Washington is one of the international centers for non-invasive brain-brain interfaces. Since publishing their first article on human brain interfaces in 2013, they have done a series of related work. This article only describes two of these representatives.
Experiment 1[10]
Experimental mission: Two subjects work together to complete a game: fly a missile or an airliner on the screen of the "sender", ask the "sender" to manipulate the "receiver's" hand through the brain interface, and pull down the missile with a button. Two subjects' brain-brain interface devices consisting of electrical brain (EEG) transcranial magnetic stimulation (TMS) are connected to each other.
Mission training: Collect the sender's electroencephal electrical (EEG) signal, train him to see the missile flying on the screen, move the one-dimensional cursor by imagining wrist movements; for the recipient, find out in advance which piece of cerebral cortex is responsible for controlling the wrist artration muscle (the muscle of the wrist extension), and place a transcranial magnetic stimulation coil above this cortex, so that the magnetic pulse emitted by TMS can cause the hand to move upwards and pull the button.
In the experiment, the two subjects were in two different buildings, one mile apart, and it was impossible to hear or see each other. The sender imagines that he or she moves his wrist and induces an EEG signal, which is wirelessly transmitted to the recipient's TMS device after detection, and the control coil sends a corresponding magnetic pulse, causing the subject's wrist to move and pull the button. This allows the two subjects to work together to complete the game simply through the brain-brain interface.
*Note
Transcranial Magnetic Stimulation (TMS) is a non-invasive, painless, non-destructive brain stimulation. TMS technology uses a pulsed magnetic field to act on the cerebral cortex, changing the membrane potential of cortical nerve cells to produce an induced current, affecting brain metabolism and nerve electrical activity, thereby triggering physiological and biochemical reactions (such as causing a simple action).
Experiment 2[11]
In this experiment, three subjects—two senders and one receiver—sit in separate rooms and work together to complete a game of Tetris, with the rules of the game as shown in the following figure:
The sender and receiver need to cooperate: the sender decides whether the falling block needs to be rotated, and "tells" his decision to the receiver through the brain interface, and the recipient manipulates the placement of the block and eliminates the bottom row of blocks.
On both sides of the sender's screen, one side displays the word "Yes" to indicate that the building blocks need to be rotated, and there is a LED under it that flashes 17 times per second; on the other side, the word "No" means that there is no need to rotate, and there is a LED under it that blinks 15 times per second. Different flicker frequencies can induce different frequencies of EEG components.
When the sender makes a judgment and looks at a word, the control device determines whether the TMS coil at the back of the recipient's head emits a magnetic pulse based on the EEG frequency collected by the head. Magnetic impulses stimulate the occipital cortex (responsible for visual information processing) in the back of the recipient's head, allowing the recipient to see the flash of light, which means that the sender means "rotating blocks" according to prior agreement.
Schematic diagram of the brain-brain interface (i.e., the control device in the figure) flow. The control device converts the sender's EEG signal into a pulse signal that stimulates the receiver.
—
Mark Stone/University of Washington[12]
The whole game requires three people to communicate and work together to complete. The recipient does not decide whether to spin the bricks until they receive instructions from the two senders (visual signals). The recipient's EEG signal is also collected and communicated to the sender in a similar way, letting the sender know the recipient's decision and giving feedback again. Communicate back and forth like this, and in the end, the outcome of the game will be communicated to all three people at the same time.
Cross-species mixed brain-brain interface
In experiments using human brains to control animals, researchers generally use hybrid brain-brain interfaces to collect brain signals in a non-invasive way for humans, while microelectrodes are implanted for animals to control animal movements. Relatively speaking, the effect of intrusive interfaces is more refined and accurate.
Zhang Shaomin et al. of Zhejiang University [13] have developed a brain-brain interface from the human brain to the rat brain, the experimental participants imagine themselves waving their left or right arms, the associated EEG signal is converted into a left or right turn control signal, wirelessly transmitted to the microelectrode installed in the rat motor cortex, discharge stimulation mouse brain.
The controller can see the rats in the maze on the screen. The researchers used a complex three-dimensional maze (below) in which the rats needed to go uphill, down ladders, avoid obstacles, detour, cross aisles, and more. In the experiment, the controller can imagine having the rats smoothly circumnavigate the complex maze in a predetermined route within a specified time.
Figure 3 Schematic diagram of the complex maze used by Zhang Shaomin et al. of Zhejiang University. [13]
From the above representative studies, it can be seen that the vast majority of brain-brain interface research so far is still difficult to say that it is the true meaning of "heart-to-heart". This may explain why Nico Lelis, a senior expert on brain-computer interfaces, believes that Musk's "heart-to-heart" is a "marketing strategy."
Indeed, if viewed only from the superficial phenomenon of behavior, these experiments can show that the receiver can do the actions that the experimenter wants to do only according to the commands imagined in the sender's mind, rather than according to the instructions of language. If we confuse the sender's brain signals with "thoughts" or "thoughts" and interpret the receiver's actions as accepting the sender's "thoughts" or "thoughts," then we assert that this is "teleportation." But brain signals are not the same as "thoughts" or "ideas," and Benjamin Libet's classic experiments have long taught us that the associated "readiness potential" can already be recorded in our brains before we realize we want to turn our wrists. So prepare the potential (brain signal) before the "mind" we perceive. We can use an EEG device to detect the preparation potential, and of course, we can also process this potential to stimulate another person's brain to make some kind of action. In my opinion, this cannot be regarded as a heart-to-heart transmission. Because from the receiver's point of view, the experimenter already knows in advance what kind of stimulation to stimulate, and which part of the recipient's brain will cause the recipient to cause the recipient's kind of experimenter to see the action, which is actually just a reflex.
Libett's experiment
In the early 1980s, the American neuropsychologist Libett conducted an experiment in which subjects decided when to move their wrists and record their muscle and brain electricity.
It has long been known that when muscles are exercised, the muscle electricity can be recorded in the corresponding parts as the beginning of the exercise; in addition, muscle movement is controlled by the primary motor cortex of the brain, and before that, some brain regions have long made exercise plans and reached the primary motor cortex, and then the primary motor cortex issues commands to control muscle movement. Activities related to the earlier "exercise program" can be recorded in the EEG and are an EEG component called the "preparation potential". The readiness potential appears 1 second or more before the actual movement begins. The average person thinks that we first generate the idea (decision) of "I want to exercise", and then make a plan for the cortex responsible for planning the movement, issuing commands through the primary motor cortex to control the movement of the wrist muscles (measuring the emulsion). Libette asked subjects to rotate their wrists while staring at a point of light on the screen that was constantly rotating along a clock face (pictured). He asked the subject to report afterwards that he had made up his mind to turn his wrist when the point of light had turned. As a result, Libett found that the subjects appeared before making a decision, more than half a second ahead. This result shows that the preparation potential precedes one's own awareness of the movement of the wrist (mind), and that the EEG signal is not equivalent to the mind itself. In fact, it's still unclear what the neural matrix of the mind is, or in which brain region it occurs.
Figure: Schematic diagram of a libert experiment.
Blackmore, 2005
The brain-brain interface experiments described in this article so far can be divided into two parts, one of which is to measure a certain brain signal associated with the sender while he is thinking (only correlation, not cause and effect!). We don't know what this "thinking" neuroma is). The other part, which examines the pattern of stimulation and which part of the recipient's brain is given to enable the recipient to perform the actions the researchers want, is actually a kind of stimulus-reflex, not "understanding." Finally, through machine learning, the recorded sender's brain signals are converted into the stimulation patterns required by the experimenter. In this way, the first two parts are joined into one, which gives people the impression of "passing on the heart". This is even more evident when the action is one-of-a-kind (all the experiments in this article except the optical brain-brain interface experiment in Luo Minmin's laboratory).
It is worth pointing out that the second half of all experiments is to control the recipient's movement, because the experimenter knows where the brain area that drives this movement is, and the neural coding of motor control is a population code, without looking for any special neuron, so that the experimenter can know in advance which part of the brain should be stimulated in what way. The foundation of this work is based on the principle of population coding that has been well understood in brain research on motor control. If what the recipient is asked to accomplish is not a motor task, but a mental activity, then it is impossible to achieve, because now we have no idea what the neural mechanism of this activity is, and what kind of stimulation must be given to cause the corresponding mental activity. That's what Greg Horwitz, a neuroscientist at the University of Washington, put it: "If you want me to move my arm, I know where to put the electrodes," and "Even if you can insert electrodes anywhere in my head, if you want me to vote for Biden or Trump, I don't know where you should stimulate to achieve it, or in what mode." ”[14]
- Mouse Bullhorst -
Since Nico Lelis formally proposed the brain-brain interface, more than a decade has passed, and although progress has been made in all aspects, there has been no breakthrough in the problems mentioned above. This is not something that can be solved by improving the technology of brain implants alone, and Musk proposed that 8 to 10 years have passed since the realization of telepathy, which is probably a promise that cannot be fulfilled.
Of course, science should not exclude bold ideas that inspire scientists to shock the unknown, and may be realized at some point in the future, but they should not be confused with reality. We can look at what other ideas there are about brain-brain interfaces.
Nicolelis et al. have proposed: "Finally, it must be emphasized that the topology of the brain-brain interface does not have to be limited to only one sender and one receiver. Instead, we pointed out that, theoretically, if a grid of many interconnected brains were used instead of only two brains, it would be possible to improve the accuracy of the channel. This computational structure may set a precedent for an 'organic computer' that can solve heuristic problems that ordinary Turing machines cannot compute. ”[15]
If many brains are interconnected to communicate directly with each other, it may constitute a "giant brain", just as neurons are interconnected into a much more powerful brain than it is. It is hard to imagine what new phenomena such a giant brain would produce.
Rao et al. argue that "a great deal of information in our brains cannot enter consciousness from introspection, and therefore cannot be expressed at will in the form of language" [8]. This is where it is difficult for surgical experts and music masters to pass on their knowledge and expertise to novices. They were unable to tell students exactly "how to position and move their fingers while performing critical operations" [8]. They hope that brain-brain interfaces may eliminate such inherent problems in language communication.
Of course, some people have begun to worry about the negative effects of brain-brain interfaces. [16]: Could the brain-brain interface cause the sender to have some coercive effect on the receiver, causing the latter to lose some sense of autonomy? Does extracting information from the sender's brain records violate their right to privacy? There is often more important content in what people don't say than what they say, and privacy in the human brain is the core of individual autonomy. Developing brain-brain interfaces can be more than worth the loss... Of course, judging from the current development of brain-brain interfaces, it is impossible for some experts to imagine in the foreseeable future, so these concerns are still too early, but the alarm bells may be of significance to how to carry out research in this area healthily.
bibliography
[1] Tim Urban (2017) Neuralink and the Brain’s Magical Future. Wait But Why 2017年4月20日（https://waitbutwhy.com/2017/04/neuralink.html）
[2] https://www.wired.com/story/elon-musk-neuralink-brain-implant-v2-demo/?bxid=5cec254afc942d3ada0b6b70&cndid=48167859&esrc=&source=EDT_WIR_NEWSLETTER_0_SCIENCE_ZZ&utm_brand=wired&utm_campaign=aud-dev&utm_mailing=WIR_Daily_082820_Science&utm_medium=email&utm_source=nl&utm_term=list1_p2
[3] Murray Gell-Mann (1994) Quark and the Jaguar. W. H. Freeman and Company.
Chinese translation: Gell-Mann, Yang Jianye et al. (2002) Quark and Jaguar, Hunan Science and Technology Publishing House
[4] Miguel Nicolelis（2011）Beyond Boundaries: The New Neuroscience of Connecting Brains with Machines---and How It Will Change Our Lives. Times Books。
[5] https://news.sciencenet.cn/sbhtmlnews/2020/11/358719.shtm
[6] Miguel Pais-Vieira, et al. (2013) A Brain-to-Brain Interface for Real-Time Sharing of Sensorimotor Information. Nature SCIENTIFIC REPORTS, 3 : 1319 | DOI: 10.1038/srep01319
[7] Lu, L., Wang, R., and Luo, M. (2020). An optical brain-to-brain interface supports rapid information transmission for precise locomotion control. Sci China Life Sci 63(6):875-885, https://doi.org/10.1007/s11427-020-1675-x
[8] https://www.theguardian.com/world/2022/feb/15/elon-musk-neuralink-animal-cruelty-allegations
[9] https://en.wikipedia.org/wiki/Rajesh_P._N._Rao
[10] Rao, Rajesh et al. “A Direct Brain-to-Brain Interface in Humans.” PLOS ONE 2014: 1-12.
[11] Linxing Jiang, Andrea Stocco, Darby M. Losey, Justin A. Abernethy, Chantel S. Prat, Rajesh P. N. Rao. (2019) BrainNet: A Multi-Person Brain-to-Brain Interface for Direct Collaboration between Brains. Scientific Reports, 9 (1) DOI: 10.1038/s41598-019-41895-7
[12] Anthony Cuthbertson (2019) First brain-to-brain interface to communicate using only your mind successfully tested, researchers claim. (https://www.independent.co.uk/life-style/gadgets-and-tech/news/computer-brain-interface-university-washington-neuralink-a8984201.html)
[13] Shaomin Zhang et al. (2019) Human Mind Control of Rat Cyborg’s ontinuous Locomotion with Wireless Brain-to-Brain Interface. Scientific Reports, 9:1321 ( https://doi.org/10.1038/s41598-018-36885-0)
[14] Adam Rogers (2020) Neuralink Is Impressive Tech, Wrapped in Musk Hype. Wired 04/09/2020 (https://www.wired.com/story/neuralink-is-impressive-tech-wrapped-in-musk-hype/?bxid=5cec254afc942d3ada0b6b70&cndid=48167859&esrc=desktopInterstitial&source=EDT_WIR_NEWSLETTER_0_DAILY_ZZ&utm_brand=wired&utm_campaign=aud-dev&utm_content=Final&utm_mailing=WIR_Daily_090620&utm_medium=email&utm_source=nl&utm_term=list2_p5)
[15] Pais-Vieira, Miguel; Mikhail Lebedev; Carolina Kunicki; Jing Wang; Miguel A. L. Nicolelis (2013). A Brain-to-Brain Interface for Real-Time Sharing of Sensorimotor Information. Scientific Reports. Nature Publishing Group. 3: 1319. doi:10.1038/srep01319 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3584574).
[16] Martone, Robert. (2020) Scientists Demonstrate Direct Brain-to-Brain Communication in Humans. Scientific American Mind. 31 (1):7-10
Author: Gu Fanji | Cover: Maus Bullhorst|Type: Light and Shadow
The original text is reproduced from "Return to Simplicity":
https://mp.weixin.qq.com/s/Qrbz1VyTTPB-3Qe4u3BNxw