I have a doppelganger. (Laughter) Dr. Gero is a brilliant but slightly mad scientist in the "Dragonball Z: Android Saga." If you look very carefully, you see that his skull has been replaced with a transparent Plexiglas dome so that the workings of his brain can be observed and also controlled with light. That's exactly what I do -- optical mind control.
我有个翻版。 (笑声) Gero 博士是动画片<<七龙珠Z-机器人传奇>> 里一个非常聪明但有点疯疯颠颠 的科学家。 如果你仔细观察, 你会发现他的脑壳被换成了一个 透明的有机玻璃圆顶盖子, 你可以透过盖子观察他的大脑活动 而且可以用光线去控制它。 这就是我的工作—— 光学精神控制。
(Laughter)
(笑声)
But in contrast to my evil twin who lusts after world domination, my motives are not sinister. I control the brain in order to understand how it works. Now wait a minute, you may say, how can you go straight to controlling the brain without understanding it first? Isn't that putting the cart before the horse? Many neuroscientists agree with this view and think that understanding will come from more detailed observation and analysis. They say, "If we could record the activity of our neurons, we would understand the brain." But think for a moment what that means. Even if we could measure what every cell is doing at all times, we would still have to make sense of the recorded activity patterns, and that's so difficult, chances are we'll understand these patterns just as little as the brains that produce them.
但是我跟我那邪恶的孪生兄弟不同, 他企图控制全世界, 而我没有险恶的动机。 我控制大脑的原因是 想了解它是怎样运作的。 你也许会说,等一下, 你不了解大脑 怎么能控制大脑呢? 那不是本末倒置吗? 很多神经学科学家都同意这个观点 认为对大脑的了解需要通过 很多详细的观察和分析来完成。 他们说:“如果我们可以记录我们的神经元的活动 我们就能了解大脑了。” 但是想一想这意味着什么。 即便我们可以侧量 每个细胞每时每刻的运作, 我们还是需要弄明白 记录下来的那些活动到底有什么规律, 而这是很困难的。 结果很可能是我们只能了解这些规律的一小部分 因为我们对产生它们的大脑知之甚少。
Take a look at what brain activity might look like. In this simulation, each black dot is one nerve cell. The dot is visible whenever a cell fires an electrical impulse. There's 10,000 neurons here. So you're looking at roughly one percent of the brain of a cockroach. Your brains are about 100 million times more complicated. Somewhere, in a pattern like this, is you, your perceptions, your emotions, your memories, your plans for the future. But we don't know where, since we don't know how to read the pattern. We don't understand the code used by the brain. To make progress, we need to break the code. But how? An experienced code-breaker will tell you that in order to figure out what the symbols in a code mean, it's essential to be able to play with them, to rearrange them at will. So in this situation too, to decode the information contained in patterns like this, watching alone won't do. We need to rearrange the pattern. In other words, instead of recording the activity of neurons, we need to control it. It's not essential that we can control the activity of all neurons in the brain, just some. The more targeted our interventions, the better. And I'll show you in a moment how we can achieve the necessary precision.
请看这个大脑活动模拟图。 在这个模拟中,每个黑点 代表一个神经细胞。 每当细胞放射一条 电脉冲时, 对应的点就会显示出来。 这里有一万个神经细胞。 所以你现在看到的相当于 一只蟑螂大脑的百分之一。 你们的大脑比这复杂 一亿倍。 在这个图案中, 有一部分 代表你, 代表你的感知, 你的情绪, 记忆, 你对将来的计划。 但是我们不知道到底哪块地方代表了这些, 因为我们不懂如何解读这些图案。 我们不了解大脑所用的密码。 要有进展, 我们需要破解这个密码。 但怎样解码呢? 一个富有经验的密码破译员会告诉你 要解开密码中各个符号的真正意义, 我们必须能够使用它们, 以我们的意愿重新组合它们。 在这个情况下也是如此, 要取得蕴含在这样的图案中 的信息, 关靠观察是不行的; 我们需要重新组合这些图案。 也就是说, 仅仅记录神经细胞的活动并不可取, 我们需要控制它们。 我们不用控制 脑部中所有神经细胞的活动,一些就足够。 我们对那些神经细胞活动的干预愈具定向性,效果愈佳。 我稍后会给你们展示 我们是怎样达到所需的精确度的。
And since I'm realistic, rather than grandiose, I don't claim that the ability to control the function of the nervous system will at once unravel all its mysteries. But we'll certainly learn a lot. Now, I'm by no means the first person to realize how powerful a tool intervention is. The history of attempts to tinker with the function of the nervous system is long and illustrious. It dates back at least 200 years, to Galvani's famous experiments in the late 18th century and beyond. Galvani showed that a frog's legs twitched when he connected the lumbar nerve to a source of electrical current. This experiment revealed the first, and perhaps most fundamental, nugget of the neural code: that information is written in the form of electrical impulses. Galvani's approach of probing the nervous system with electrodes has remained state-of-the-art until today, despite a number of drawbacks. Sticking wires into the brain is obviously rather crude. It's hard to do in animals that run around, and there is a physical limit to the number of wires that can be inserted simultaneously.
因为我个人比较现实而不浮夸, 我不会宣称只要具备能够操控神经系统功能的能力 就会一次揭开所有关于神经细胞的奥秘。 但是我们一定会从中学到很多。 现在,我决不是历史上 第一个意识到工具的发明是 多么重要的人。 试图 去改变神经系统机能的历史 长久卓越。 这可以追溯到至少二百年前, 十八世纪初 伽伐尼的著名实验。 伽伐尼证明,只要他把青蛙的腰椎神经 接上电流 青蛙的脚就会抽搐。 这个实验揭示了第一个,也是最根本的一个, 关于神经编码的事实: 这些信息是以 电子脉冲的形式编写的。 伽伐尼的 以电极探究神经系统的方法 即使在今天,仍然是最先进的, 虽然它有一些很明显的缺陷。 将电线接到脑部显然是相当残忍的。 这样的设计也难于在移动的动物身上实践。 当中亦有物理上的限制 只有有限数量的电线 能够同时安插在脑部。
So around the turn of the last century, I started to think, "Wouldn't it be wonderful if one could take this logic and turn it upside down?" So instead of inserting a wire into one spot of the brain, re-engineer the brain itself so that some of its neural elements become responsive to diffusely broadcast signals such as a flash of light. Such an approach would literally, in a flash of light, overcome many of the obstacles to discovery. First, it's clearly a non-invasive, wireless form of communication. And second, just as in a radio broadcast, you can communicate with many receivers at once. You don't need to know where these receivers are, and it doesn't matter if these receivers move -- just think of the stereo in your car. It gets even better, for it turns out that we can fabricate the receivers out of materials that are encoded in DNA. So each nerve cell with the right genetic makeup will spontaneously produce a receiver that allows us to control its function. I hope you'll appreciate the beautiful simplicity of this concept. There's no high-tech gizmos here, just biology revealed through biology.
因此,在上世纪末, 我开始思考, 如果我可以把这个逻辑 翻个个儿,那该多好啊。 不把电线 接到大脑的某一点, 而是改造大脑本身 使得某些神经系统的组件 可以感应到漫射讯号, 比如说一道闪光。 这样的方式能够,在电光一闪之间, 跨越很多科学发现的障碍。 首先,这显然是个无创伤性的, 无线的交流形式。 其次,就像无线电广播一样, 你可以跟很多讯号接收器同时沟通。 你并不需要知道那些接收器所在的地点。 再者,讯号接收器移动也没关系—— 就好象你车里的收音机。 我们的情况更加理想, 因为我们可以用DNA中所编码的材料 来制造这些接收器。 每个具有合适的基因构造 的神经细胞 都会自然产生出接收器 让我们能控制它的功能。 我希望你们能够欣赏 这个概念中 美丽的简单朴素。 这里面没有安置高科技的小玩意儿 只有用生物学揭示出的生物学。
Now let's take a look at these miraculous receivers up close. As we zoom in on one of these purple neurons, we see that its outer membrane is studded with microscopic pores. Pores like these conduct electrical current and are responsible for all the communication in the nervous system. But these pores here are special. They are coupled to light receptors similar to the ones in your eyes. Whenever a flash of light hits the receptor, the pore opens, an electrical current is switched on, and the neuron fires electrical impulses. Because the light-activated pore is encoded in DNA, we can achieve incredible precision. This is because, although each cell in our bodies contains the same set of genes, different mixes of genes get turned on and off in different cells. You can exploit this to make sure that only some neurons contain our light-activated pore and others don't. So in this cartoon, the bluish white cell in the upper-left corner does not respond to light because it lacks the light-activated pore. The approach works so well that we can write purely artificial messages directly to the brain. In this example, each electrical impulse, each deflection on the trace, is caused by a brief pulse of light. And the approach, of course, also works in moving, behaving animals.
现在,让我们来近距离看一下这些不可思异的接收器。 当我们把镜头拉近到一个紫色的神经细胞, 我们可以看到它的细胞外膜 布满着微细的气孔。 这些气孔能够让电流通过 以及负责 为神经系统传递信息。 但是你看到的是一些特别的气孔。 它们结合了光感应受体 这些光感应受体跟你眼睛里的很相似。 只要一束闪光射在这些光感应受体上, 那些气孔便会打开,电流亦会随之而接通, 那神经细胞便会发出电子脉冲。 因为那些光驱动气孔是编写在DNA里的, 我们可以达到无比的精确性。 这是因为 尽管我们身体里的每一个细胞 都有同一组基因, 但是在不同的细胞里 不同组合的基因会被开启或关闭。 你可以利用这个原理来确保 只有部分的神经细胞 具有我们设计的光驱动气孔,而其他的没有。 在这个动画中,蓝白色的细胞 在左上角的那些 并不会对光有反应 因为它们缺少了光驱动气孔。 这个方法很管用 我们可以把完全人工的讯息 直接编写入大脑。 在这个例子中,每一个电子脉冲, 轨迹的每一次偏斜, 都是由一束短暂的光脉冲造成的。 同时,这个方法亦适用 于移动中,表现良好的动物身上。
This is the first ever such experiment, sort of the optical equivalent of Galvani's. It was done six or seven years ago by my then graduate student, Susana Lima. Susana had engineered the fruit fly on the left so that just two out of the 200,000 cells in its brain expressed the light-activated pore. You're familiar with these cells because they are the ones that frustrate you when you try to swat the fly. They trained the escape reflex that makes the fly jump into the air and fly away whenever you move your hand in position. And you can see here that the flash of light has exactly the same effect. The animal jumps, it spreads its wings, it vibrates them, but it can't actually take off because the fly is sandwiched between two glass plates. Now to make sure that this was no reaction of the fly to a flash it could see, Susana did a simple but brutally effective experiment. She cut the heads off of her flies. These headless bodies can live for about a day, but they don't do much. They just stand around and groom excessively. So it seems that the only trait that survives decapitation is vanity. (Laughter) Anyway, as you'll see in a moment, Susana was able to turn on the flight motor of what's the equivalent of the spinal cord of these flies and get some of the headless bodies to actually take off and fly away. They didn't get very far, obviously. Since we took these first steps, the field of optogenetics has exploded. And there are now hundreds of labs using these approaches.
这是同类实验中的第一个, 可以说是光学版本的伽伐尼实验。 这是在六七年前 由我的研究生苏珊娜.利马完成的。 苏珊娜改变了左边那只果蝇的基因 使它大脑中二十万神经细胞中的两个 出现光驱动气孔。 你应该相当熟悉这些神经细胞 因为它们就是那些在你尝试拍打果蝇的时候 让你倍感沮丧的神经细胞。 它们造就了果蝇的逃跑反射,使果蝇 在你的手要靠近的时候飞到空中逃走。 你可以看到,一阵闪光跟拍打的动作有着同样的效果。 果蝇跳起,伸展翅膀,震动它们, 但它们不能真的飞离地面, 因为这只果蝇被夹在两片玻璃板中间。 要保证果蝇如此反应 并非因为它看到了那道闪光, 苏珊娜做了一个简单 但直截了当而有效的实验。 她把果蝇的头切掉。 这些无头的身体可以生存约一天。 但它们并不会有太多活动。 它们会站着 并不停的替自己梳理。 如此看来,断头之后,好像只有虚荣心这一个人格特征能够保存下来。 (笑声) 无论如何,一会后你将会看到, 苏珊娜能够启动果蝇的逃走运动神经 这相等于果蝇的脊椎 并令一些无头的身体 离地飞走。 很明显,它们飞不了很远。 自从我们走出了第一步之后, 光基因技术这个领域百花齐放。 现在,已经有数以百计的实验室 在使用这些方法。
And we've come a long way since Galvani's and Susana's first successes in making animals twitch or jump. We can now actually interfere with their psychology in rather profound ways, as I'll show you in my last example, which is directed at a familiar question. Life is a string of choices creating a constant pressure to decide what to do next. We cope with this pressure by having brains, and within our brains, decision-making centers that I've called here the "Actor." The Actor implements a policy that takes into account the state of the environment and the context in which we operate. Our actions change the environment, or context, and these changes are then fed back into the decision loop.
从伽伐尼及苏珊娜的第一步 令动物抽搐或跳动的成功到现在 我们取得了长足的进步。 现在我们能够 彻底地干扰它们的心理。 在我最后的例子中,我将会为你展示, 一个老生常谈的问题。 生命就是一连串的决择 这就造成了一个常存的压力,迫使我们决定下一步的行动。 我们以大脑来应对这个压力。 我们的大脑是我们的决策中心 我把它叫做“执行者”。 这个“执行者”执行一个政策 这政策会考虑到周遭环境的因素 以及我们生活的背景。 我们的行为改变环境,或背景, 而这些改变会反馈到我们的决策回路中。
Now to put some neurobiological meat on this abstract model, we constructed a simple one-dimensional world for our favorite subject, fruit flies. Each chamber in these two vertical stacks contains one fly. The left and the right halves of the chamber are filled with two different odors, and a security camera watches as the flies pace up and down between them. Here's some such CCTV footage. Whenever a fly reaches the midpoint of the chamber where the two odor streams meet, it has to make a decision. It has to decide whether to turn around and stay in the same odor, or whether to cross the midline and try something new. These decisions are clearly a reflection of the Actor's policy. Now for an intelligent being like our fly, this policy is not written in stone but rather changes as the animal learns from experience. We can incorporate such an element of adaptive intelligence into our model by assuming that the fly's brain contains not only an Actor, but a different group of cells, a "Critic," that provides a running commentary on the Actor's choices. You can think of this nagging inner voice as sort of the brain's equivalent of the Catholic Church, if you're an Austrian like me, or the super-ego, if you're Freudian, or your mother, if you're Jewish.
现在要把一些神经生物学的东西 添加到这个模型中, 我们建造了一个简单的一维空间 把我们最爱的实验对象,果蝇,放进去。 两架玻璃管中的每一个小室 都放有一只果蝇。 小室的左右两边 弥漫着两种不同的气味, 闭路电视会一直观察着 看着果蝇来回踱步。 这里有一些闭路电视的片段。 当一只果蝇到达小室的中间 两种气味会交汇的地方时, 果蝇必须作出决定。 它要决定是回头 留在同一种气味中, 还是跨过中线 去尝试新事物。 这些决策显然是 根据“执行者”的政策来执行的。 对我们的果蝇,这样有智慧的生物来说, 这个政策并非刻在石头上一成不变的, 它会根据自身的经验而转变。 我们可以将这样一个 适应性智能的元素加进我们的模型里 我们要假设果蝇的大脑里 不只有一个“执行者”, 而是也拥有多组不同的细胞, 包括一个“批评家”,不断地 给“执行者”的决定提出意见。 你可以把这个唠叨的内在声音 当成是大脑里的 天主教教堂, 如果你跟我一样是奥地利人的话, 或者可以把它当成佛洛依德学说中所说的“超我”, 如果你是犹太人,或者可以把它当成你的“母亲大人”。
(Laughter)
(笑声)
Now obviously, the Critic is a key ingredient in what makes us intelligent. So we set out to identify the cells in the fly's brain that played the role of the Critic. And the logic of our experiment was simple. We thought if we could use our optical remote control to activate the cells of the Critic, we should be able, artificially, to nag the Actor into changing its policy. In other words, the fly should learn from mistakes that it thought it had made but, in reality, it had not made. So we bred flies whose brains were more or less randomly peppered with cells that were light addressable. And then we took these flies and allowed them to make choices. And whenever they made one of the two choices, chose one odor, in this case the blue one over the orange one, we switched on the lights. If the Critic was among the optically activated cells, the result of this intervention should be a change in policy. The fly should learn to avoid the optically reinforced odor.
现在,明显的 这个“批评家”在我们的智力系统中 是一个重要的组成部分 所以,我们希望可以认出 这些在果蝇的大脑中 扮演着“批评家”角色的细胞。 我们实验的逻辑很简单。 我们的假设是:如果我们能够用我们的光学遥控器 来驱动那些“批评家”细胞, 那么我们应该可以,人工地,不断烦扰“执行者” 使它改变它的政策。 换句话说, 这飞蝇应该能够从错误中学习 它们会认为自己犯了错误, 即使它们其实并没有。 我们培植了一些果蝇 它们的脑部被多多少少地随机安置了 一些可以用光驱动的细胞。 我们拿出这些果蝇 给予它们做决策的机会。 每当它们作出两选一的决定, 选到一种气味时, 在这个具体案例中,它选了蓝色而非橙色的那种, 我们就亮灯。 如果“批评家”在光驱动细胞中, 这种干扰的结果会使 果蝇改变它的政策。 果蝇就应该学会去避开 那种受光学加强的气味。
Here's what happened in two instances: We're comparing two strains of flies, each of them having about 100 light-addressable cells in their brains, shown here in green on the left and on the right. What's common among these groups of cells is that they all produce the neurotransmitter dopamine. But the identities of the individual dopamine-producing neurons are clearly largely different on the left and on the right. Optically activating these hundred or so cells into two strains of flies has dramatically different consequences. If you look first at the behavior of the fly on the right, you can see that whenever it reaches the midpoint of the chamber where the two odors meet, it marches straight through, as it did before. Its behavior is completely unchanged. But the behavior of the fly on the left is very different. Whenever it comes up to the midpoint, it pauses, it carefully scans the odor interface as if it was sniffing out its environment, and then it turns around. This means that the policy that the Actor implements now includes an instruction to avoid the odor that's in the right half of the chamber. This means that the Critic must have spoken in that animal, and that the Critic must be contained among the dopamine-producing neurons on the left, but not among the dopamine producing neurons on the right.
这里有两个不同案例中的情况。 我们在比较两个不同品种的果蝇, 它们的脑部分别有 大概一百个可光驱动的细胞, 在这里以绿色来显示。 两组细胞的相同之处 在于它们都会制造神经递质多巴胺。 但是每个制造多巴胺的神经细胞 的身份 在左右两边显然是不同的。 以光学来驱动 大概一百多个细胞 在两个不同品种的果蝇里, 会有极其不同的后果。 如果你先看看 右边果蝇的行为, 你可以看到,当它到达小室的中心点 两种气味的交汇的地方时, 它会一直走过去,就像从前一样。 它的行为完全没有受光学驱动细胞的影响。 但是左边的果蝇,情况就大不相同了 每当它来到中心点, 它会停下, 谨慎地检查气味的接合点, 就好像在打探周遭的环境, 然后它会回头。 这证明“执行者”实施的政策中 现在包括了躲开那种气味的指令 该种气味是从小室右边散发过来的。 这说明“批评家” 一定对果蝇提过意见了, 这说明在左边的果蝇中,“批评家” 是那些能够制造多巴胺的神经细胞中的一个, 但在右边的果蝇中,“批评家”并不在那些能够制造多巴胺的神经细胞中。
Through many such experiments, we were able to narrow down the identity of the Critic to just 12 cells. These 12 cells, as shown here in green, send the output to a brain structure called the "mushroom body," which is shown here in gray. We know from our formal model that the brain structure at the receiving end of the Critic's commentary is the Actor. So this anatomy suggests that the mushroom bodies have something to do with action choice. Based on everything we know about the mushroom bodies, this makes perfect sense. In fact, it makes so much sense that we can construct an electronic toy circuit that simulates the behavior of the fly. In this electronic toy circuit, the mushroom body neurons are symbolized by the vertical bank of blue LEDs in the center of the board. These LED's are wired to sensors that detect the presence of odorous molecules in the air. Each odor activates a different combination of sensors, which in turn activates a different odor detector in the mushroom body. So the pilot in the cockpit of the fly, the Actor, can tell which odor is present simply by looking at which of the blue LEDs lights up.
通过很多这样的实验, 我们成功地把 “批评家”的身分 缩小到了十二个细胞。 这十二个细胞,在这里以绿色显示, 它们对一个 叫做“蘑菇体”的脑结构发出讯号, 在这里以灰色呈现出来。 我们从抽象模型中知道 那个接收“评论员”批评的脑结构 就是“执行者”。 这个分析指出 蘑菇体和 行动的决策有着一定的关系。 基于我们对蘑菇体的了解, 这是完全能够想像的。 事实上,这非常合理, 我们能够建构一个电子玩具电路 用以模拟果蝇的行为。 这个电子玩具电路中, 蘑菇体神经细胞以 在电路中心的竖排蓝色发光二极管 来表示。 这些发光二极管上接有感应器 用以探测空气中的气味分子。 不同的气味都会激活不同组合的感应器, 而它们再驱动 蘑菇体中另一个气味检测器。 所以在果蝇驾驶舱中的飞行员, “执行者”, 要知道哪一种气味存在 只要看看哪颗蓝色发光二极管亮起来就行了。
What the Actor does with this information depends on its policy, which is stored in the strengths of the connection, between the odor detectors and the motors that power the fly's evasive actions. If the connection is weak, the motors will stay off and the fly will continue straight on its course. If the connection is strong, the motors will turn on and the fly will initiate a turn. Now consider a situation in which the motors stay off, the fly continues on its path and it suffers some painful consequence such as getting zapped. In a situation like this, we would expect the Critic to speak up and to tell the Actor to change its policy. We have created such a situation, artificially, by turning on the critic with a flash of light. That caused a strengthening of the connections between the currently active odor detector and the motors. So the next time the fly finds itself facing the same odor again, the connection is strong enough to turn on the motors and to trigger an evasive maneuver.
“执行者”得到这个讯息之后的行为 取决于它的政策, 这些政策都是根据 气味检测器 与运动神经之间关联的强度来储存的 这驱动了果蝇的逃亡行为。 如果关联性弱,运动神经会保持关上 那只果蝇会继续前进。 如果关联性强,运动神经就会启动 那只果蝇就会作一个转身。 现在试想这样一个情况 就是当运动神经保持关上时, 那只果蝇继续前行 它就会遭受一些痛苦的后果 例如遭到电击。 在这样的情况下, 我们可以预期“批评家”会发表意见 并告诉“执行者” 要它改变它的政策。 我们人工地制造了这样的一个情境 以一束光来启动“批评家”。 这样就可以加强 正在起作用的气味检测细胞 与运动神经之间的关联。 所以,下一次 当果蝇再次面对同样的气味时, 关联性将会有足够的强度去启动运动神经 并引发果蝇进行回避。
I don't know about you, but I find it exhilarating to see how vague psychological notions evaporate and give rise to a physical, mechanistic understanding of the mind, even if it's the mind of the fly. This is one piece of good news. The other piece of good news, for a scientist at least, is that much remains to be discovered. In the experiments I told you about, we have lifted the identity of the Critic, but we still have no idea how the Critic does its job. Come to think of it, knowing when you're wrong without a teacher, or your mother, telling you, is a very hard problem. There are some ideas in computer science and in artificial intelligence as to how this might be done, but we still haven't solved a single example of how intelligent behavior springs from the physical interactions in living matter. I think we'll get there in the not too distant future.
我不知道你的想法, 但我觉得这样的实验非常令人兴奋 虚无的心理学概念 不见了,却引出了 对思维在物理学上,机能上的理解, 虽然这只是果蝇的思维。 这是一个好消息。 另一个好消息, 至少对于一个科学家来说, 就是世界上还有很多尚待发掘的东西。 在我讲诉的这个实验里, 我们发掘出了“批评家”的真正身分, 但是我们还不知道 “批评家”是怎样完成它的工作的。 试想一下,在没有老师,或者你的母亲大人告诉你的情况下 要知道你自己犯了错, 是一件非常不容易的事情。 电脑科学领域 以及人工智能领域中的一些想法 就是试图去解释这件事情是如何发生的, 但是到现在我们还无法解释 任何一个实际例子中 智能行为是如何 在生命体的物理交互中 产生的。 我想在不久的将来, 我们会找到答案的。
Thank you.
谢谢。
(Applause)
(掌声)