Compile | Chen Junda
Edit | Panken
Zhidong reported on August 6 that Musk revealed for the first time in an in-depth podcast interview released on August 2 that Neuralink successfully implanted a brain-computer chip for the second trial patient, and there are currently 400 electrodes in the second patient's brain (about 40% of the total number of implants) working, much higher than the 10%-15% required for normal work. If regulatory conditions allow, Musk plans to complete eight more brain-computer chip implantation surgeries by the end of this year.
Musk called the progress of the second patient "exceptionally good", and he expects huge improvements in the transmission of brain-computer interfaces as the number of implants increases, including a significant increase in the number of electrodes and an increase in signal processing capabilities.
In the future, brain-computer interfaces may be transmitted at megabits per second, more efficiently than humans can type or speak. In Musk's view, this means that BCI users can communicate with other BCI users 10 times or even 100 times faster.
Ultimately, Musk hopes to increase the communication bandwidth of humans through brain-computer interfaces, helping humans communicate with AI faster and better, thereby improving the symbiotic relationship between humans and AI.
Neuralink's second brain-computer chip implant was greenlit by the United States Food and Drug Administration in May this year, and was originally scheduled to be officially carried out in June this year, but was postponed to July due to personal health problems of patients. The details of this clinical trial have not yet been disclosed in detail on the official website of Neuralink, and we can only get a glimpse of the relevant information from Musk's interview.
Neuralink's executives and first patients also participated in this podcast interview. Neuralink's COO Dongjin "DJ" Seo revealed that they are already working on new products such as multi-implant technology and visual stimulation implants, and are also communicating with regulators about the technology related to the idea of manipulating robotic arms.
Noland ·Arbaugh, the first implanter, was able to manipulate a computer with his mind and play complex games after receiving a brain-computer chip implant, but after a period of time, the implanted electrodes fell off, resulting in reduced control efficiency, which made him very frustrated. However, Neuralink quickly fixed the problem with improvements in the recognition algorithm. Nolan has amassed more than 200 pages of clinical reports for Neuralink, which has helped Neuralink significantly improve product performance.
1. The implantation effect of the second case has been greatly improved, and 8 more implants may be completed by the end of the year
Information about patients who have undergone clinical testing of the Neuralink brain-computer interface has not yet been disclosed. Musk mentioned in the podcast that the patient's spinal cord injury was similar to the first patient, and so far the second implant of Neuralink is going well, with about 400 electrodes providing signals. According to Neuralink's official blog, their implants have a total of 1,024 electrodes, which means that in the second implant, about 40% of the electrodes were able to provide signals.
▲Neuralink Implantation Body 实拍图 (图源: Neuralink)
The 400 electrodes is more than the number of electrodes needed for the proper functioning of this system. In the first implant case, only 10%-15% (about 100-150) of the electrodes of the brain-computer interface were able to successfully send signals. However, by modifying the recording algorithm, the Neuralink team improved the sensitivity of the brain-computer chip to neural population signals, and at the same time improved the technology to convert these signals into cursor movement information.
These adjustments allow the 100-150 electrodes in the brain of the first patient to achieve a communication bandwidth of up to 8 bytes per second (the amount of information equivalent to 4 Chinese characters or 8 English letters), breaking the world record for human use of brain-computer interface to manipulate the communication bandwidth of the cursor.
Musk expects that as the number of implantable cases increases, the effect of brain-computer chips will make huge progress, which includes a significant increase in the number of electrodes and an increase in signal processing capabilities. In the next few years, Neuralink will dramatically break this record, reaching 100 or even 1,000 bytes per second of communication bandwidth, and perhaps even megabytes per second in five years, faster than anyone can communicate by typing or talking.
After the second implant procedure, Neuralink will accelerate other clinical trials. They are already planning to recruit new patients in Canada and the United Kingdom, and have received applications from more than 1,000 quadriplegic people worldwide, of whom more than 100 are eligible. Musk plans to push Neuralink to complete eight more implant cases by the end of this year, but whether this goal can be achieved depends on whether the regulatory authorities approve it.
Second, brain-computer interface or human-computer communication fast channel, similar technology will be used to help blind people regain their sight in the future
In fact, the information density (i.e., the amount of data transmitted per second) when humans communicate in natural language is very low. Musk said that the average amount of data transferred by humans per second may be less than 1 byte. In addition, when expressing and receiving information in natural language, people must actively compress and decompress information, a process that inevitably brings loss to information transmission.
In Musk's eyes, brain-computer interfaces may become a new channel for efficient communication between humans. If both parties are equipped with brain-computer interfaces, they may be able to communicate efficiently and clearly at 10 or even 100 times the normal speed, just like watching a video or listening to a podcast at 2x the speed. He believes that the information transmitted by the brain-computer interface is not just a simple text, but an overall piece of information, including rich text, images and other content. It's like a meme we send in a chat, where images, actions, and words combine to convey extremely complex meanings. Musk believes that billions of people could use brain-computer interfaces in a few decades.
▲马斯克接受采访（图源：The Lex Fridman Podcast）
Brain-computer interfaces also have the potential to become a new way for humans to interact with AI. At present, the communication speed of AI is about several megabytes per second, while humans only have 1-2 bytes per second, and this huge gap greatly reduces the communication efficiency of AI and humans. Musk said that in the long run, he hopes to further increase the communication bandwidth of brain-computer interfaces, and although this may also lead to redundancy of information, the increase in communication bandwidth will eventually improve the symbiotic relationship between AI and humans.
Neuralink's second product is Blindsight, a visual restoration technology that Musk believes is the next node that Neuralink can unlock in its development tree. This product is designed to restore sight to visually impaired people who are completely blind by directly stimulating neurons in the visual cortex.
Musk believes that Neuralink's product is actually a general-purpose input/output device. All human sensations and manipulation of limbs are transmitted in the form of electrical signals, and if the transmission "cable" is broken, it will not be able to complete these functions normally. The brain-computer chip can capture all these signals and transmit them to a designated receiving object.
Essentially, brain-computer chips are repairing neuronal damage, so Musk also said that similar technologies can be used to solve problems such as schizophrenia and epilepsy. He also cited the example of the late well-known theoretical physicist Stephen Hawking, who was paralyzed, saying that similar techniques would be of great significance if they could be applied to people like him.
3. The first implant patient came forward to say that Neuralink executives said that they had experimented with multiple implants on monkey brains
Noland Arbaugh, the first brain-computer chip implant, also participated in the podcast interview · this podcast. After a diving accident, Nolan was completely paralyzed from the neck down. He was initially hesitant to participate in the first clinical trial of Neuralink, as the implant would likely continue to improve in the future.
Looking back on how he felt on the day of surgery, Nolan said he was still under anesthesia after the surgery, but quickly regained consciousness. When he didn't get out of bed after surgery, the brain-computer interface was able to record his nerve signals. He found that when he gave the idea of moving his finger, a specific waveform appeared on the tablet that was responsible for displaying the signal. He eventually helped Neuralink's researchers discover more than 100 different waveforms, each of which corresponds to a command that controls a certain part of the body.
A week after surgery, Nolan was able to move the mouse with his mind, but he wasn't thrilled by it, knowing that it was just an electrical signal from the machine learning algorithm to recognize that he wanted to move the mouse. This action is called "Attempted Movement", which is to move the mouse by concretely imagining the trajectory of the mouse.
However, after a period of training, Nolan suddenly discovered that he could achieve "Imagined Movement". He only had to stare at the end of the line where he wanted to move the mouse, and the mouse would fly over quickly, without having to actually imagine the trajectory of the mouse, which surprised Nolan.
▲诺兰通过想象鼠标移动至蓝点实现快速控制（图源：The Lex Fridman Podcast）
Nolan's control of electronic devices during daily use is achieved through an app. This program "translates" signals from the brain into control instructions for electronic devices. Nolan has amassed more than 200 pages of clinical reports in clinical trials, citing the problems he encountered in his use and the features he hopes to add, which has greatly helped Neuralink improve the performance of its brain-computer interface applications.
Neuralink's COO, Dongjin "DJ" Seo, is the only one of Neuralink's seven founding members who hasn't left, and he is now serving as CTO of Platform X. In this interview, DJ Seo revealed that they have experimented with implanting 2 brain-computer interfaces in monkeys, 1 on the left and right brains, and they have also tried to implant dedicated brain-computer interfaces in different functional areas of the cerebral cortex.
DJ SEO shared that in fact, the original implant N1 can already obtain all the signals in the human brain, and now they only focus on the conversion of motion signals. When researching visual implants, they will focus more on giving brain stimulation to present the picture, but the rationale behind it and the product design are similar to the original implants.
At present, brain-computer interfaces can only help users move 2D cursors, and DJ Seo believes that similar operations can be achieved in robotic arms and wheelchairs. However, there is still a risk of using ideas to influence the operation of the physical world, and they are still communicating with regulators.
The potential of brain-computer interfaces does not stop there, human consciousness and thoughts are also a series of electrical signals. In the future, DJ Seo hopes that Neuralink will use implants to record these signals to better understand how humans work behind their brains and consciousness.
Conclusion: With the acceleration of the development of Neuralink technology, security has become a concern
According to Musk's assumption, if approved by regulators, Neuralink will perform a total of 10 brain-computer chip implantation surgeries in one year this year and conduct relevant clinical tests. However, many in the academic community have questioned the lack of transparency in the way Neuralink is advancing its research. Neuralink's blog post is limited to a handful and does not share the details of the procedure in detail. ·United States Sameer Sheth, a neurosurgeon who specializes in implanted neurotechnology, said in an interview with the academic journal Nature: "Neuralink only shares what they want us to know, and researchers have a lot of concerns about it."
According to the email information disclosed by the United States FDA on its official website, Neuralink has been in close communication with the FDA since 2021, but it took more than 2 years to obtain the first clinical test approval. Currently, Neuralink's studies in Canada and the United Kingdom are still in the stage of accepting potential patients, and there is no news of local government approval. The hesitation of regulators may also reflect the risks behind this technology.
来源：The Lex Fridman Podcast