2023年5月,Neuralink拿到FDA的人体临床试验批准。消息出来的那几天,社交媒体上刷屏的说法是“脑机接口要来了”。这个理解差了好几个数量级。
FDA批的是“允许启动人体试验”,产品上市还隔着整套临床流程:先招少量受试者,观察安全性,再逐步扩大规模,收集有效性数据,最后才是上市审批。这条路走完通常要几年。Neuralink现在站的位置,是刚拿到入场券。
但这张入场券分量不轻。
过去两年,Neuralink在监管层面一直不顺。2022年初,FDA拒绝过他们的申请,理由集中在几处安全性顾虑:植入体里的锂电池有没有风险,电极线会不会在脑组织里迁移,设备能不能在不造成额外损伤的前提下被安全取出。每一项都是实打实的工程问题,走流程绕不过去。
现在FDA放行,意味着监管机构审过了他们的动物数据和工程数据,认为风险已经可控到可以让人参与试验。这家公司做了七年动物实验,现在第一次真正跨过门外那道坎。
Neuralink成立于2016年,由马斯克和几位联合创始人一起创办,办公室分布在加州硅谷和德州奥斯汀。这家公司要做的事情清单很长:让盲人重新看见,治疗抑郁症、痴呆、癫痫、中风,最终让瘫痪的人重新走路。
清单上每一项都是硬骨头,而公司规模只相当于特斯拉的零头。视频里提到的数字是大约300名员工,对比特斯拉的13万人。用300人去碰神经科学里最难的一批问题,这个比例本身就说明了他们在赌什么。
缝进大脑的线,比红细胞还窄
Neuralink在推特的“技术星期二”系列里放出过一组图片,把植入大脑的电极线放大给公众看。
官方说法是:这些又细又软的线,宽度只有几个红细胞那么宽,目的是把大脑对异物的反应降到最低。
这句话背后的因果值得展开。
大脑有一套自己的免疫机制。任何外来物插进去,尺寸越大、材质越硬,免疫系统的反应就越强烈。反应的结果是胶质细胞把异物层层包裹,形成一层疤痕组织。电极一旦被这层组织隔开,信号质量就逐年下滑,最后可能什么都读不到。所有植入式脑机接口都要面对这个老问题,几十年没被彻底解决过。
Neuralink的思路是,把线做得足够细、足够软,让大脑认不出那里有东西。认不出来,就不会启动免疫反应;不启动免疫反应,电极就能长期贴着神经元工作。
首席研究员Dan Adams给了一个能对比的数字。每根线末端的矩形孔大约15微米宽,人类红细胞直径大约8微米,一微米等于千分之一毫米。也就是说,这个关键结构比两个红细胞并排还要窄一点。
更直观的是横向比较。另一种已经临床应用的脑机设备叫深部脑刺激器(deep brain stimulator),导线直径比Neuralink的线粗了大约800倍。而那个设备,早就在治疗帕金森病这类神经系统疾病了。
800倍的差距说明一件事:粗线能做的事,医学上已经被验证有价值。细到这个程度的线,理论上能做的更多,同时把组织损伤和免疫反应压得更低。
细也带来新的工程难题。线越细越难缝合,越容易断,长期可靠性越难保证。后面那些测试,全都是为了回答这个问题。
他们用扫描电子显微镜来检查工艺是否受控、缺陷出在哪里,偶尔也拍几张“美照”。之所以不用普通光学显微镜,是因为要看的结构在微米量级,光学显微镜已经到物理极限了。
脑子一直在动,连心跳都算进去
线细还不够,还得经得住时间。
Neuralink放了一段测试视频,用的是动态机械分析仪(dynamic mechanical analyzer),给电极线做疲劳测试,观察机械性能随时间怎么变化。测试方式里有个容易被忽略的细节:他们模拟心跳引起的脑部运动,频率1赫兹,同时跑一个10赫兹的加速版本。
这个设计点出了一个常被忽视的事实。很多人以为人静止的时候,脑子也是静止的,事实并不是这样。心脏每跳一下,脑组织就跟着轻微位移一次,一分钟六十几次,一天八万多次,一年三千万次。电极线要在这种持续的微动里待上十年,既不能断,也不能从原来的位置滑走。
更麻烦的是使用场景。Neuralink的设想是这套东西将来普通人也能装,那装了之后人还要正常生活,要跑步、要打球、要摔跤。线必须同时满足两个互相拉扯的要求:足够柔软,能跟着组织一起动;足够结实,动十年也不坏。材料学上这两个要求是对着干的,怎么平衡,就是这些疲劳测试要回答的问题。
谁可以报名,条件写得很清楚
Neuralink开了一个患者注册库(patient registry),入口在官网 neuralink.com/patient-registry。条件列得明确:
- 人在美国境内
- 年满18岁,且达到所在州的法定成年年龄
- 有能力自主做出同意决定
- 患有四肢瘫痪、截瘫、视力丧失、听力丧失,或者无法说话
这里有个容易混淆的地方要说明。注册库跟试验招募是两码事。注册是把个人信息放进池子,让公司知道有多少潜在候选人、情况大概是什么样。符合条件的人之后还要经过筛选、评估、完整的知情同意流程,才可能真正进入手术名单。
媒体报道常写成“Neuralink开始招人了”,这个说法不准确。第一例人体植入的具体时间点,视频里也没给出确定答案。
把手术练到无聊
Neuralink放了一张模拟手术室(mock operating room)的照片。官方的说法是:在工程空间里,用假体练习完整手术流程,所有硬件和器械都按真实配置摆好,这样可以快速测试、对比不同的手术改进方案。
他们内部有一句话:一台好手术,应该是一台无聊的手术。练得越多,手术越可预测。
2022年11月的Show and Tell活动上,Neuralink反复讲过他们的实验室流程。其中一个方法是做“大脑假体”(brain proxy),用材料特性接近真实脑组织的东西,让手术机器人反复练穿刺、练缝合,找出效率和安全性最好的参数组合。视频作者Ryan Tanaka说,他去年去总部参加活动时,亲手摸过那个假体。
先假体、再动物、最后人,这个顺序有它的道理。Neuralink的手术机器人要处理的线比头发丝还细,误差按微米算。这种操作人手做不了,只能靠机器,而机器的每一个动作参数都要在假体上练到稳定,再往活体上放。假体的意义在于,它可以被反复扎、反复录像、反复复盘,真正的脑子没有这种试错机会。
Pager的12岁生日
视频里有一只叫Pager的猴子,正在找零食吃,庆祝12岁生日。
就是这只猴子,几年前被植入两枚Neuralink芯片,靠意念玩电子游戏,让全世界第一次直观看到“脑机接口”四个字长什么样。
Pager只是Neuralink众多动物模型里的一个。拿到FDA批准之前,他们在大鼠、猪、羊、猴子身上都做过测试,从实验室环境一层层走到活体动物。这条路走得并不平顺。有动物死亡,被动物保护组织起诉过,也有内部员工抱怨进度太赶、实验设计仓促。这些争议真实存在,不该因为FDA批准就当作没发生过。
需要说清楚的是,动物实验在神经科学里是绕不过去的环节。任何要进人脑的设备,必须先证明在结构相近的动物身上不会造成不可逆损伤。争议的焦点通常落在怎么做、做多少、痛苦有没有降到最低这些具体问题上,某些时候也会落回要不要做。
充电的时候,脑袋会不会发烫
这是个很实际的问题,Neuralink自己也把它当成重点在测。
他们展示了最新的无线充电器。从循环播放的视频看,充电器的表面积明显大于植入体本身。官方说法是,他们在测试植入体的热性能,目标是保证安全、提升效率。测试方法是用红外成像,在充电线圈处于不同位置的情况下,检测植入体底面的热点在哪里。
换句话说,他们在找“最热的那个点”。植入体贴着脑组织,温度高一点就可能带来损伤,这项测试直接关系到安全评估,不能当例行公事处理。
动物身上的充电是自动完成的。视频里,猴群被训练成自己充电,坐在树枝上就行。Pager爬上去、坐到线圈下方,系统自动检测到它的存在,状态从“搜索”切换到“充电”,屏幕上实时显示功率输出(0到1的刻度)和流进电池的电流。
人类版本的方案,是把充电器做进某种头部穿戴设备里,可能是睡帽,也可能是棒球帽。这个设计逻辑很清楚:充电最好发生在人不动的时候,睡觉是最自然的时间窗口。晚上戴着帽子睡一觉,第二天电量满格,用户不需要额外做任何动作。
为什么非要开颅,而不是戴个头环
Neuralink最常被质疑的就是这一点。既然市面上已经有非侵入式脑机接口,戴个头环就能读脑电,为什么还要在头骨上开洞?
首席神经外科医生Matt McDougall给了一个很工程师的回答。他说,评价这类辅助设备,核心问题只有一个:你能往大脑里送多少信息,又能从大脑里取出多少有用信息。现在这个数字非常小,甚至比不上老式调制解调器。
然后他补了更重要的一句:看一项技术,要看它的天花板。很多技术的理论最大值很低,低到让人失望,哪怕执行得完美、发展得完美。而Neuralink这类技术吸引人的地方,是天花板极高。随着技术迭代,没有明显理由说不能和数百万个神经元打交道。
这就是带宽的差别。非侵入式设备隔着颅骨、头皮和脑膜读信号,信噪比先天吃亏,能拿到的通道数和精度都受限。植入式电极直接接触神经元,理论上可以做到高带宽。物理层面的优势,软件补不回来。
McDougall说,如果你想要的是一个“语义假体”、一个挂在认知能力旁边的AI助手、那些未来几十年里还属于科幻的东西,那就必须往高带宽这条路走。评估所有脑机技术时,这条前提绕不开。
这段话其实在回答一个更大的问题:Neuralink到底在赌什么。它赌的是短期里非侵入式更安全、更容易被接受,长期里天花板会卡死,而侵入式的天花板足够高,值得承担手术风险。
一个外科医生的私心
McDougall说自己加入Neuralink的主要动机,跟技术本身关系不大。
我很喜欢这个想法,往远了说,十年、二十年之后,人类能控制自己大脑出问题的方式。到现在这个阶段,每个人要么认识某个人,要么通过朋友的朋友,接触过成瘾、抑郁、自杀、肥胖这些事。这些大脑功能的失调,是我做这行的动力,是我想在职业生涯里解决的东西。(Matt McDougall)
这段话解释了一件事:Neuralink对外的叙事重心落在治疗上,意念打游戏只是顺带展示的一面。游戏演示是给公众看的,抑郁症、成瘾、肥胖这些才是他们讲给医生和监管机构听的故事。
McDougall也谈到了安全性,说得很直接。
从我亲手做过的几百台手术来看,我绝对敢为这个设备背书。我认为它比很多行业标准里、FDA已批准的、我日常给病人做的那些手术要安全得多,那些手术没人会去质疑它的安全性。在我看来,Neuralink已经达到了一个远超常规安全门槛的水平。(Matt McDougall)
这话要注意语境。他是公司雇员,立场不中立,这是事实。但“几百台手术”这个量级说明,他没有只对着纸面评估风险。至于能不能兑现,最终还是要看人体试验的数据。
300人对13万人
视频结尾,Ryan Tanaka提了一个对比:Neuralink团队大概300人,特斯拉大概13万人。
这个比例说明两件事。一是这家公司还很小,能并行推进的项目有限,任何一条线出问题都会拖慢整体进度。二是它招人的方向跟一般医疗器械公司不一样。Neuralink官网放出的岗位里,机械工程师、软件工程师、机器人工程师占很大比重,神经科学背景并非硬性要求。
McDougall的原话是,他们需要这个星球上最聪明的人,来解决世界上最难的问题之一。你不需要是顶尖神经外科医生,也不需要在脑科学领域有很深积累,需要的是对使命有强烈认同感的人。
300人做这种级别的事,好处是每个人能影响的面很大,坏处是容错空间很小。
怎么理解这次批准
回到开头。FDA批准启动人体试验,是Neuralink七年历史里的一个节点,离终点还很远。真正的考验在试验开始之后:手术能不能稳定复现、线能不能在活人脑子里长期待住、信号能不能维持几年、充电方案在真人身上好不好用、免疫反应会不会比动物身上更严重。
这些问题现在都没有答案。任何“脑机接口已经成功”的说法,在这个时间点都太早。
- 如果你或家人符合条件(四肢瘫痪、截瘫、视力或听力丧失、无法说话,年满18岁,人在美国),可以去 neuralink.com/patient-registry 注册进入筛选池。但要清楚,注册只是进入候选名单,不代表会被选中。
- 看到“Neuralink获批”的新闻,先分清是启动试验、扩大试验,还是产品上市。这三者之间隔着好几年,混着说会把预期拉到不现实的位置。
- 关注长期随访数据,尤其是电极在体内几年后的信号衰减情况。发布会的演示好看,长期数据才说明问题。
- 如果你是机械、软件、机器人方向的工程师,Neuralink的招聘页面值得看一眼。300人的团队意味着个人能影响的面积很大,也意味着要承受的不确定性很高。
- 把“意念打游戏”和“治疗抑郁症”分开看。前者是演示,后者是这家公司真正想拿到的适应症,也是它跟监管机构谈判的核心筹码。
0:00Elon musk's company neuralink has just received FDA approval to start testing brain chip implants in humans this is important because neuralink is working on helping blind people see and solving many types of neurological disorders like depression dementia seizures strokes and eventually helping paralyzed people walk you may have seen a couple years ago they implanted two chips in a monkey to play a video game with just its mind and now that same concept will be tested in paralyzed patients so they can do daily tasks on a computer without needing their limbs neuralink is a small But Mighty company with grand Ambitions the company was started in 2016 by Elon Musk and several
0:41others and now has offices in Silicon Valley California and Austin Texas this major Milestone of getting FDA approval is a huge step in bringing this amazing technology to those who need it most welcome to neuropod a channel covering all topics related to Elon musk's brain machine interface company neuralink in this update episode I'll share a clip of nerling's head neurosurgeon Dr Matt McDougall discussing the safety of the implants we'll look at a close-up view of neuralink's electrode threads and for those who don't know these are the threads that are connected to the implanted chips and essentially get sewn into the brain with the neurolink robot
1:19I'll show a picture of neuralink's mock operating room and some short videos of neuralink's animals in action then have you ever wondered if your head would get too hot while charging your neural link I'll share a clip of what the heat map of a link implant looks like when it's charging and then we'll finish off the episode with a few cool things one of which is a video clip of Dr McDougall sharing his primary motivation for working with neurolink the safety of the device I would absolutely vouch for from you know the hundreds of surgeries that I've personally done with this I I think it's much safer than many of the industry standard FDA approved surgeries that I
2:02routinely do on on patients that you know are no one even thinks twice about their standard of care neuralink is already reached in my mind a safety threshold that is far beyond a commonly accepted safety threshold nerlink has started to share more details about the tech they're using and building one post from their Tech Tuesday series on Twitter gives us a close-up zoomed in view of the electrode threads they insert into the brain nerling says our thin flexible threads are a few red blood cells wide minimizing the brain's response to them before I read the rest of this post I reiterate this fact that the smaller the threads are the less likely it is that
2:49the brain will have a severe immune response that is if the threads are small enough the brain may not even recognize that they're there and of course if the brain doesn't know something's there it's not going to try and attack that something for some reference principal investigator at neurolink Dan Adams added this the rectangular hole at the end of each thread is about 15 micrometers wide a human red blood cell is about eight micrometers in diameter and one micrometer equals one one thousandth of a millimeter also one more thing I should mention is that another BCI device called a deep brain stimulator has leads that are around 800 times thicker in diameter and that
3:29device is already able to treat neurological conditions like Parkinson's disease imagine how much better neuralink will be able to do the original post continues on to say we use scanning electron microscopy to verify process control understand defects and sometimes take Glamor Shots fortunately more info about the threads were given with this next post
3:53in this video we can see knurling testing their threads they say we use Dynamic mechanical analyzers to fatigue test our threads and identify changes in their mechanical properties over time it's hard to spot but here you can see us testing a thread by emulating the brain motion due to our heartbeat at one Hertz and an accelerated test at 10 Hertz these threads need to not only be able to last for a long time but they're also in quite a harsh environment the brain is also not stationary as the post alludes to even when we quote unquote aren't moving our heartbeat is causing motion furthermore since neuralink wants to design these devices where anyone
4:33could get one people will need to be able to go about their daily activities and even play sports while a neuralink or two are implanted as such these threads need to be robust yet flexible to move with the body and last a long time if you're interested in learning whether you may qualify for the knurling clinical trials apply to join the patient registry at www.neurlink.com patient Dash registry
4:59anyone within the United States who's at least 18 years old and the age of majority in their state who is able to consent and who has quadriplegia paraplegia vision loss hearing loss and or the inability to speak is invited to participate in the patient registry here's a picture of neuralink's mock operating room nerling says we practice surgeries on proxies with all the hardware and instruments needed in our mock or in the engineering space this helps us rapidly test and Benchmark surgical improvements a great surgery is a boring one and practice makes perfect as neurlink discussed multiple times throughout their November 22 Show and Tell event they are doing all they can
5:42to test the devices in the lab prior to testing in Animals by creating fake brains that have similar material characteristics to actual brains they can learn how to optimize the efficiency and safety of their surgeries this is a picture of the brain proxy that I took when I attended the event at headquarters last year look at nerlink's monkey named pager he celebrated his 12th birthday foraging for treats this is the same monkey from a couple years ago that I showed earlier he had two neural links implanted and was thus able to play a video game with just his mind this is just one of the many animal models that nerling has been testing their devices on prior to
6:24getting FDA approval the team has been testing in the laboratory environment and has also tested in rats pigs sheep and monkeys this is what nerlink's newest Wireless Charger looks like as you can see from this looping video the surface area of the charger is larger than the link implant nerling posted this saying we test the thermal performance of our implants to ensure safety and improve efficiency here we are taking infrared images to detect hot spots on the bottom surface of an implant while charging with the charger coil in different positions stay cool you might be wondering how this charging would take place in practice for knurling monkeys they sit on
7:07branches as shown in this video The Troop has been trained to charge themselves so let's see how pager charges his implant on the right we're streaming real-time diagnostics from pagers N1 when he climbs up and sits below the coil you can see the charger automatically detects his presence and transition from searching to charging we see the regulated power output on a scale of zero to one and the current driven into his battery
7:45and for humans the plan is to have the charger inside some type of headwear whether that be a nightcap or a baseball cap this is the primary thing that drives Dr McDougall to work with neuralink I love the idea down the road and we're talking you know a 10-year or maybe 20-year time frame of humans just getting control
8:11over some of the horrible ways that their brains go wrong right so I think everybody at this point has either known someone or second order known someone a friend of a friend who has been touched by addiction or depression suicide obesity these functions of the brain or malfunctions of the brain are what drives me these are the things that I want to tackle in my career I also often get asked about the pros and cons of surgically implanted devices versus non-invasive devices Dr McDougall articulated this nicely he shared the primary reason why invasive brain machine interfaces are more appealing than non-invasive bmis or wearable headsets now that's a useful way to
8:58think about these assistive devices how much information are you able to get in into the brain and out of the brain usefully and right now that that number is very small even compared to the old modems but you have to ask yourself when you're looking at a technology what's the ceiling what's the theoretical maximum and for a lot of these Technologies the theoretical maximum is is very low disappointingly low even if it's perfectly executed and and perfectly developed as a technology and I think the thing that attracts a lot of us to a technology like knurling is that the ceiling is incredibly High there's no obvious reason that you can't interface
9:39with millions of neurons as this technology is refined and and developed further so that's the kind of wide band you know high bandwidth brain interface that you want to develop if you're talking about and a semantic prosthetic an AI assistant to your cognitive abilities you know the more sci-fi things that we think about in the coming decades so it's an important caveat when you're evaluating these Technologies they really want it to be something that you can expand off into the Sci-Fi and three less things to wrap up the episode Elon saw and liked the latest video I posted on Twitter please consider following at neuropod as this is where I'll post the most timely
10:29updates and the second thing is I was interviewed by Michael matalini on a show called The Singularity lab Mike's an excellent interviewer so I encourage you to check out that with the link in the description and the third thing is I want to encourage anyone who is inspired by neuralink's general admission to end suffering due to many neurological diseases to apply to join the team you know encourage anyone who is you know excited about things like that you know especially mechanical engineers software Engineers robotics engineers come to the neuralink website and look at the jobs we've got we need the brightest people on the planet working on these the
11:04hardest problems in the world in my opinion and so if you want to work on this stuff come help us you don't need to be a fantastic neurosurgeon or have a deep expertise in a brain related field instead they're looking for passionate people who care deeply about the mission of the company the team is still quite small at around 300 employees by comparison Tesla has around 130 000 my
11:31name is Ryan Tanaka thanks for watching subscribe to stay informed about neurolink
资源与出处
- 视频在 YouTube 收看
- 档案silicon-valley.tech 档案页本页视频与文稿数据来自 Elon Musk Archive
- 精读本篇精读 Markdown 版