爱因斯坦临终看到了什么？连最伟大的头脑也无法越界！身边人吓到面色惨白，临终前他比谁清醒！#刘欢 #圆桌派 #窦文涛 #真相

爱因斯坦临终时刻：伟大头脑亦无法逾越的界限 节目围绕元宇宙、5G通信底层技术与人工智能的发展边界展开深度剖析。嘉宾们探讨了从类脑计算到数字永生面临的肉身硬件壁垒，分析了5G的三大核心资源、低时延优势及工业赋能逻辑。随后，现场通过香港科技大学研发的“魔镜”AI进行人机问答互动，深入辨析弱AI与通用人工智能（AGI）的技术鸿沟、迁移学习的壁垒，以及算法时代下人类自我意识、尊严与伦理危机的本质命题。 5G凭借1毫秒级空口时延与移动边缘计算（MEC），正推动工业高并发物联网、远程精密控制与自动化基础设施的重塑，完成从改变生活到改变社会的跨越。 (32:20) 技术中立的传统“菜刀理论”在算法时代破产，代码本身已内置伦理偏见，且自主致命武器系统（LAWS）等技术正逼近“黑理石假说”预警的文明毁灭风险。 (8:52) 莫拉维克悖论揭示了AI的演进壁垒：微积分与复杂博弈对机器而言极易攻克，但人类历经数亿年进化的单样本泛化、细腻肢体协调与共情能力却仍难以工程化。 (68:15) 当前主流AI仍局限于封闭规则大数据的统计拟合与语言镜像，跨领域的迁移学习存在明显鸿沟，科学界对于通用人工智能（AGI）自我意识的实现路径仍缺乏有效抓手。 (58:52)

本讨论深入剖析了元宇宙、5G 通信以及人工智能等前沿技术的底层逻辑与现实边界，指出技术构想虽宏大，但始终受制于人类肉身的生物学极限与算法机制的内在局限。 生物底线与数字永生的鸿沟 (0:00) 对虚拟现实和星际探索的宏大构想无法脱离肉身的生物学根基，现实中疾病与生理限制构成了体验虚拟世界的前提。尽管脑机接口与类脑智能研究已能扫描动物神经元并生成数字脑模型，但生物学界质疑物理结构的复制并不等同于功能的涌现。人脑包含约860亿个神经元，目前人类即便完成测序与结构映射，生命的复杂算法依然未知，从结构到功能的鸿沟仍未跨越。此外，VR头显超过500克便易引发前庭神经冲突导致眩晕恶心，说明肉身硬件存在天然物理边界。 技术泡沫反思与代码伦理危机 (2:59) 嘉宾批评科技领域常将旧事物包装为新概念进行炒作，如个体化医疗改称精准医疗、海量数据包装为大数据，元宇宙炒地皮本质上与早期单机版大富翁游戏并无二致。现实中虚拟数字人的应用主要源于刚性服务效率的短缺，而非哲学身份认同。更深层的挑战在于算法与代码打破了传统的“菜刀中立论”，数字工具天然内嵌伦理偏见。西方在世俗化后面临善恶价值锚点的缺失与对失控技术的恐慌，而中国儒家无神论传统中构建的世俗伦理秩序则展现了不依赖神性权威的道德考量。 5G基础设施落地与物理常识 (22:29) 5G网络演进遵循通信领域十年一代的周期规律，面对高频信号波长短、无法有效绕射的物理特性，网络覆盖需要极其密集的微基站支持。早期网络通常采用依附于4G核心网的非独立组网（NSA），最终必须过渡到独立组网（SA）才能实现1毫秒超低时延与海量连接。针对公众对辐射的疑虑，专家澄清中国辐射控制限值仅为美国标准的五十分之一；5G基站虽然频率高，但单站覆盖范围极小、发射总功率与动量极低，远低于家用微波炉的辐射水平。 机器学习局限与自我意识门槛 (58:52) 现阶段人工智能依赖海量大数据的统计与黑盒拟合，在棋盘等规则闭合场景表现优异，但并不具备自我意识、主观情感与主动提问能力。莫拉维克悖论表明，计算机易于处理高维运算，却难以复制人类经亿万年进化而成的细微体感与抚慰动作。迁移学习虽能使模型在相近领域间复用知识，但面对完全异构的跨领域或数据稀疏的复杂现实环境仍面临巨大阻碍。人脑百亿级神经元的连续动态与直觉架构，至今仍是纯算法模型难以逾越的门槛。 讨论最终指出，无论人工智能具备多强的解题能力与算力，其本质依然是服务于人类设定的工具。未来技术落地的终极价值，或在于老龄化时代为失去自理能力的人群提供免于羞耻感的尊严护理，实现冷冰技术与人性温度的结合。

[4:41] 樱桃队长 [10:35] 美美与共 [14:28] 如果上帝死了，人类何以为善 [17:03] 它是否会为自己哭泣 [19:04] 以虚强实 [29:45] 萧伯: 4G改变我们的生活，5G改变社会 [46:40] 魔镜AI: 在眨眼之间。 [52:33] 窦文涛母亲: 如果戈辉不行，鲁豫也行。 [58:52] 我们现在连怎么下手都不知道 [67:39] 窦文涛: 可以尊严地活着，这非常重要。 [75:25] 窦文涛: 海天一色 [86:30] To serve humanity [86:30] World peace. [87:40] 窦文涛: I like Xu Gehui the most, too [87:40] 窦文涛: How many times did you program that in?

[0:00] Einstein was born in 1879. [0:01] >> Yeah. [0:02] >> Today we can still talk about Musk; [0:04] humanity always has such figures. [0:06] >> These pivotal moments for humanity [0:07] occur because they can always see the [0:09] future from the standpoint of current [0:10] technology. In a way, this is an effort [0:13] to expand the boundaries of today's [0:15] metaverse. Only then might we be able [0:17] to find a new path on a crowded, [0:19] constrained Earth where material and [0:21] energy are already limited. [0:22] >> To become a multi-planetary species. [0:25] >> But there is a prerequisite: this [0:27] physical body must survive. I believe [0:29] biology can never be abandoned. To put [0:32] it simply, in your metaverse or your [0:34] game, if you have a fever of just one [0:37] degree, you won't even want to turn on [0:39] the computer. It’s the issue of your [0:41] physical body and illness; otherwise, [0:43] the suffering of the flesh, including [0:45] its reproduction. [0:47] >> So, when you pass away, would you want [0:49] to leave behind a digital persona in [0:51] such a metaverse? Because regarding [0:52] this technology, I interviewed [0:54] Professor Zheng from the Institute of [0:55] Brain-inspired Intelligence at the [0:57] Chinese Academy of Sciences. They are [0:59] now able to use mouse and monkey brains [1:01] to create digital brains, and it has [1:03] already been successfully tested on [1:05] small robots. The brain-inspired [1:08] intelligence technology they are [1:10] working on involves scanning brain [1:12] neurons and then generating a digital [1:14] brain through training and learning. So [1:17] , if this technology is mature by the [1:19] time you die, you could actually choose [1:22] , for example, to use brain-computer [1:24] interface technology to scan your [1:26] digital brain. Leave it in such a [1:28] digital world, and after this digital [1:31] brain is trained, it can operate in a [1:33] way that is almost identical to your [1:35] own thought process; that would be [1:37] digital immortality. So this technology [1:39] is actually not that fantastical; I [1:41] think it is a technology we will see [1:44] mature in our lifetime. Yes. To me, the [1:47] key question now is that you don't know [1:49] if that digital brain you scanned...... [1:52] >> is the same as the original one or not. [1:54] >> Exactly, you don't know how it thinks. [1:57] >> Many people are indeed using such [1:59] methods now. Apple just upgraded a [2:01] system for digital inheritance, hoping [2:03] to leave some things for the next [2:04] generation. Humanity wants to leave a [2:06] trace; this might be okay, right? [2:08] >> People want to leave a name, right? [2:10] Leave a legacy. [2:11] >> Something like that. [2:13] >> But can you say that if we solve the [2:14] structure today, we can solve the [2:16] function? In reality, no matter how [2:18] powerful the MRI technology we have [2:20] today, it is impossible to fully [2:22] explain the brain. We are now [2:24] performing single-cell sequencing on [2:26] every single neuron. This is what the [2:28] Shanghai Institute and BGI researchers [2:30] were doing last Spring Festival, aiming [2:32] for the mouse brain. The monkey has 6 [2:34] billion neurons, and the human brain [2:36] has 86 billion. Sequencing every single [2:38] cell is the goal, but even reaching [2:40] this step...It’s like what I said [2:42] about the human genome: once the " [2:43] Kangxi Dictionary" of those 3 billion [2:45] base pairs is done, how exactly do the [2:46] algorithms emerge? The leap from [2:48] structure to function—it’s not that [2:50] knowing the structure guarantees we [2:52] know the function. Even though function [2:55] stems from structure, this specific [2:57] challenge is what I care about more. [2:59] >> Technically speaking, discussing how to [3:01] develop these methods is fine, but [3:02] don't just dump all our current [3:04] thoughts and viewpoints into a new [3:06] buzzword. [3:06] >> Mmh. [3:07] >> Ignoring the substance—Americans are [3:09] just too good at coining terms. We used [3:11] to call it "personalized medicine," [3:13] then Obama called it "precision [3:15] medicine." We used to call it "massive [3:16] data," then they called it "big data," [3:18] and now the whole world calls it "big [3:20] data." We used to call it "neurobiology [3:22] ," now it’s "brain science" or " [3:23] neuroscience." They study organisms [3:24] without brains and still call it "brain [3:26] science." Is this brain science or " [3:28] no-brain" science? That species never [3:29] evolved a brain, yet you study its [3:31] neurons. I’m a brain scientist, and I [3:32] asked, "Does that species even have a [3:34] brain?" They thought about it and... [3:36] exactly. So, whether it's called Meta [3:38] or, as Huawei recently calls it, SER... [3:39] >> Mmh. [3:40] >> It’s actually just a game. [3:42] >> We called Guan Yu a Marquis in the Han [3:44] Dynasty, a King in the Song, and an [3:46] Emperor in the Ming. But it’s still [3:48] Guan Yu; it just depends on how you [3:50] define him. We keep mythologizing [3:52] things layer by layer. It seems [3:54] everything is "Metaverse" now. A few [3:57] years ago, everything was "blockchain," [3:58] before that "Internet Plus," and even [4:00] earlier, "mobile Internet." Humans [4:02] always need these concepts to [4:04] constantly soothe our restless, anxious [4:06] hearts, forgetting that it’s just a [4:08] program. Selling land in the Metaverse? [4:10] When we played Monopoly back in the 80s [4:12] , we bought land too. Playing it on a [4:14] computer was a "Metaverse" too, just [4:15] single-player, though four people could [4:17] take turns. You’d roll the dice, see [4:18] what land you could buy, and how much [4:20] it cost. I’ve been gaming for 30 [4:22] years, I see this clearly. Before that, [4:24] it was just people playing Ludo offline [4:26] . What’s so special about this? It [4:29] might not necessarily have a greater [4:30] impact on people's understanding than [4:32] the invention of the telegraph did at [4:33] that time. [4:34] >> Things that happen after are often no [4:36] match for the past. [4:38] >> Yeah, that's just playing with words [4:41] and concepts. Personally, I think it [4:44] all comes down to demand. Once the need [4:46] is strong enough, human acceptance [4:48] becomes very high. For example, what do [4:51] I mean? Nowadays, on many internet [4:53] platforms, your customer service is all [4:56] bots. You don't mind talking to a bot; [4:58] for instance, if you want to buy [5:00] medicine and the bot recommends [5:02] something or directs you, the fact that [5:04] someone is responding makes you think, [5:06] "OK, that's not bad." So, if virtual [5:09] humans are accepted in the future, it [5:11] won't be because we consider them human [5:13] in a philosophical sense, but because [5:16] we have a huge shortage of services. [5:18] For example, why am I training my own [5:20] virtual human? Because I am currently [5:22] answering questions from friends online [5:24] . [5:24] >> You have a virtual human? Yes, I'm [5:26] called Captain Cherry. And who am I [5:29] right now? [5:30] >> Right, but because I was answering [5:32] questions for kids—I did that for a [5:34] few months, open only to our community [5:36] —I answered 1,100 questions, [5:38] responding online every single day. I [5:40] honestly couldn't keep up, so I had to [5:42] train my virtual human, "Captain Cherry [5:44] ." He's more knowledgeable than I am, [5:46] but he can speak to the kids using a [5:48] similar tone. And then, in the future, [5:50] he can answer questions from kids all [5:52] over the world. [5:53] >> If I were your fan, I wouldn't feel [5:55] satisfied. [5:55] >> Well. [5:56] >> If you could chat with me one-on-one, I [5:58] think I'd be much more... [5:59] >> That, that would require payment. Yeah, [6:01] look at online shopping platforms now; [6:03] I've noticed when you contact customer [6:05] service, they always have that [6:06] automated question: "Are you a bot or a [6:08] human?" [6:09] >> I care about that, you know? If you're [6:11] a machine, what am I even talking to [6:13] you for? And some say that from the [6:15] very day the internet was born, the [6:17] fate of the metaverse was already [6:19] sealed. Because, why is that? You see, [6:21] from the time of the internet, it [6:23] started with sending messages and [6:25] WeChat, then you gradually wanted to [6:26] hear voices—like you said, "reshaping [6:28] the six senses"—then you wanted to [6:30] see people, and now we can video chat. [6:32] Then make that 3D, and then what? [6:35] You'll want the surrounding environment [6:36] , even these fake books, to be real. [6:38] Don't you see, it's inevitable. That's [6:41] technology. I thought it might be far [6:43] off, but who knows, maybe you haven't [6:44] advanced that far yet. But I'm saying, [6:46] is there a fundamental limit to what we [6:48] , as physical beings, can endure? For [6:50] example, I saw a philosopher the other [6:52] day mention a problem; he plays games a [6:54] lot, and he said the issue of motion [6:56] sickness hasn't been solved. [6:57] >> The problem is, when the headset weighs [6:59] more than 500 grams, most people get [7:01] dizzy within an hour. [7:03] >> Right, you, as a human passed down [7:05] through DNA, [7:06] >> Your vestibular nerves can't take it. [7:08] >> You're fully immersed. [7:09] >> Yes. You get nauseous, so is there a [7:11] limit to this? Do physical humans have [7:14] limits? [7:14] >> There are boundaries. That's why I say, [7:16] you can't see infrared or ultraviolet, [7:18] and you definitely can't feel neutrinos [7:20] , yet these things are passing through [7:21] you all the time. So this metaverse [7:23] isn't actually meant for us—for our [7:26] physical bodies to enjoy—it's for our [7:28] avatars. [7:29] >> The metaverse is inherently prepared [7:32] for your avatar. Today, what does it [7:34] matter? We are giving birth; we are [7:37] birthing our avatars. [7:39] >> So if our generation continues to [7:41] discuss it like this, we're already [7:43] outdated. In the future, children will [7:46] think, "Those ancient people were like [7:48] that. Back then, look, they were so [7:51] confused. They didn't understand why; [7:54] they were held back by old culture." [7:56] >> They can't enter this world; this world [7:58] isn't prepared for them, but we must [8:00] offer them flowers in gratitude, [8:02] because without that "work" from their [8:04] generation...Without that generation, [8:07] how would it be done? It's us [8:08] desperately trying to buy land on it. [8:11] Desperately trying to create NFTs on it [8:13] . Desperately trying to issue tokens on [8:16] it. "Work," after we're done working, [8:18] 50 years later, the avatars of future [8:20] generations will be raised and thrive. [8:23] That’s the picture. So I feel, and [8:25] that's why I say, what I worry about [8:27] most regarding the metaverse is it [8:29] being ruined. "Ruined" has two meanings [8:31] : one is the way we talk about the [8:33] metaverse, being so shallow. Discussing [8:35] artificial intelligence is often like [8:37] this; the conversation tends to stall [8:38] because it always devolves into an [8:40] optimist versus pessimist debate, even [8:41] in the U.S. It’s simply a spectrum, [8:43] right? There are bound to be two [8:45] opposing schools of thought. [8:46] >> The conversation hits a dead end. [8:48] It’s the same with the metaverse— [8:49] it's either good or bad, as if you have [8:50] to pick a side. Ultimately, you're [8:52] forced to declare where you stand. [8:53] >> Right. [8:54] >> And then you see who’s drinking and [8:56] who isn’t. So, here, we say that this [8:58] process of just "talking" actually gets [9:00] in the way. Gets in the way of what? It [9:03] hinders us from seeing the context [9:05] behind it. And this underlying context, [9:08] quite importantly, is the issue of the [9:11] codification of logic. [9:12] >> That is, algorithms are no longer [9:14] neutral. Technology is no longer [9:16] neutral. [9:17] >> This differs from technology before the [9:18] industrial age; take the horse-drawn [9:19] carriage, for example. [9:20] >> Right. [9:21] >> But the carriage itself has no morality [9:23] , right? Just like a kitchen knife has [9:25] no morality. We used to say the " [9:26] kitchen knife theory"—that a knife [9:27] doesn't kill, people do. That theory [9:29] was correct. [9:30] >> Right. [9:31] >> But in today’s digital world, a knife [9:34] is different. It’s a coded knife. [9:37] This knife, this code, inherently [9:40] carries ethical biases. So in this [9:42] situation, technology is no longer [9:44] neutral. However, our old knowledge— [9:47] the knowledge from our past textbooks [9:50] —cannot explain this. The biggest [9:52] challenge we face today is the issue of [9:55] technological ethics and code. [9:57] Therefore, trying to label code as good [9:59] or evil and picking sides today is a [10:02] low-dimensional discussion. It’s like [10:04] what we said the other day: we haven't [10:06] even figured out if morality is written [10:08] into our DNA, yet now you want to...... [10:11] >> write morality into virtual beings. [10:13] Right. [10:14] >> Exactly. Humans don't even understand [10:17] humans, yet they want to play God and [10:19] recreate a world to rule over. I feel [10:21] that even here, this "codification" [10:23] issue can still be reduced to the [10:24] person writing the code; they still [10:26] have the choice of whether to write [10:28] code that acts as a weapon or as a [10:30] kitchen knife. Exactly, you can't [10:32] assume that... [10:34] >> Let me interrupt you there. [10:35] >> That is a classic way of thinking—the [10:37] belief that it can all be reduced to [10:39] the individual. [10:40] >> That view is unsupported by evidence, [10:42] and more importantly, it misunderstands [10:44] the internet and sand-box games. [10:46] Because sand-box games are co-created, [10:48] you see? [10:49] >> What I mean by "co-created" is that it [10:51] isn't up to you; the world you think [10:53] exists isn't the world you're seeing. [10:55] These are two different things, which [10:57] is why I say our understanding of the [10:59] word "digitization" is one-sided. You [11:01] always feel you’re in control, but [11:03] actually, the dog is led; our [11:04] consciousness is implied, and we [11:06] intentionally choose this implication. [11:08] In reality, every one of us today is [11:10] tamed by the vehicle, yet we always [11:11] assume we are the drivers. In fact, you [11:14] are tamed by the vehicle; we are tamed [11:16] by urbanization. We are tamed by three [11:18] meals a day, by the books we read; we [11:20] don't think we are, we don't want to [11:22] admit we play a passive role. This [11:24] absolutely won't work in the Metaverse. [11:26] In the Metaverse, you must first stay [11:29] calm, relax. To relax means we must [11:31] acknowledge that we are a process of [11:33] mutual taming; the Metaverse is a world [11:35] of mutual taming. Only then can we— [11:37] that point the teacher made was great [11:39] —what does "beauty shared by all" [11:41] mean? "Beauty shared by all" actually [11:43] means "beauty and ugliness shared by [11:44] all." Why is it "beauty shared by all"? [11:47] You first presuppose that everyone is [11:49] beautiful. [11:49] >> Oh, you must have great tolerance. [11:51] >> Exactly, great tolerance. [11:53] >> That’s what makes it a universe. [11:55] >> Yes. But there is a truly serious [11:58] challenge here. Westerners are now [12:01] reflecting and thinking about the issue [12:03] of good and evil. [12:05] >> That’s right. So we wonder why "Harry [12:08] Potter" emerged in the 90s. It’s a [12:11] cultural phenomenon. It's not just a [12:13] bestseller issue; Americans are [12:15] reflecting on the Enlightenment. [12:19] >> Reflecting on Martin Luther—what did [12:21] the Reformation really bring? Why has [12:24] faith been fading for 500 years? Of [12:27] course, there are various [12:28] interpretations, but generally, fewer [12:30] people go to church. Fewer people are [12:32] truly, genuinely pious—that is, [12:34] worshipping the divine. In this [12:36] situation, another voice has emerged: [12:39] to re-examine the issue of good and [12:41] evil. Let’s not fight yet; let’s [12:44] rethink and re-examine. 2,000 years ago [12:47] , good was... [12:48] >> Mm. [12:49] >> Good was the supreme god. [12:51] >> Absolutely. [12:52] >> Absolutely, and evil was the secondary [12:54] god. [12:54] >> Mm. [12:55] >> So today, why is it said? Westerners [12:59] are also in deep pain and division [13:01] because they find they are facing a [13:04] serious problem. It’s the [13:06] relationship between man and God; [13:09] suddenly, an evil power is emerging. [13:12] They can't even contain it themselves; [13:15] it’s like the despair before a [13:17] collapse. When we talk about the ethics [13:20] of code, it’s not an empty [13:22] description; it means that serious [13:24] programmers and serious tech thinkers [13:26] are anxious about the loss of control [13:29] over these growing energies. [13:31] >> Mm. [13:32] >> So they think this is a problem that [13:34] should be discussed seriously. [13:36] >> Yet, when many of us discuss this, we [13:38] narrowly view it as a matter of picking [13:41] sides. We just say I like this, then I [13:43] like that, and then we start chanting [13:45] slogans, saying down with evil, we will [13:48] defeat evil, I firmly believe truth [13:50] will prevail—it’s useless, none of [13:52] this works. [13:52] >> Why is that? It shows we still haven't [13:55] learned to find common ground while [13:56] reserving differences. We always prefer [13:58] to just talk past each other. [13:59] >> Yeah, just... " [14:01] >> Those who are not of my kind." [14:03] >> Their hearts must be different. So, in [14:06] this situation, I say the Metaverse [14:08] presents a common challenge for all [14:11] people worldwide, including different [14:13] generations, from grandfathers to [14:16] grandsons. That is, how do we welcome [14:19] the brand-new world of the future? In [14:22] this new world, we must admit that [14:24] today’s knowledge cannot explain it. [14:27] Agreed, this is a very complex issue. [14:30] This makes me think of Dostoevsky's " [14:33] The Brothers Karamazov." In his books, [14:35] he is always discussing one question: [14:38] if God is dead—if you take God away [14:40] —how is human goodness still possible [14:43] ? Because if God is there, okay, God [14:45] defines what is good, and we follow [14:47] what God says, so human goodness can be [14:49] achieved. But if you discard God, if [14:51] God is dead, then how do we achieve [14:53] goodness? This is such a complex issue [14:56] that many of his books, like "Demons," [14:58] discuss it, even exploring the total [15:01] collapse of the human spirit, nihilism, [15:03] and so on. However, I also agree that [15:05] the enormous power of today's [15:07] technology is actually out of the [15:09] control of its creators; they don't [15:12] even know which direction it's heading. [15:14] But this actually brings me back to the [15:16] ancient Chinese intellectual tradition, [15:19] because in ancient China, during [15:21] Confucius 'time, there wasn't really a [15:23] "God" like that; China is godless. [15:25] Right, Confucius was actually the [15:28] greatest atheist. So, for 2500 years, [15:30] we have been a nation without a literal [15:34] God. While we talk about the "Way of [15:36] Heaven," it is more of a collective [15:38] belief or interpretation. [15:40] >> It’s a moral deity. Yes, or rather, [15:43] the "Way of Heaven" is just our belief [15:46] that good will eventually prevail; [15:48] it’s like "Heaven," but it’s never [15:50] been a person governing the good and [15:53] evil of our country. Therefore, [15:55] thinking about ancient Chinese thinkers [15:57] in this context, you find that they are [15:59] very meaningful today. Because from the [16:02] very beginning, ever since Confucius, [16:04] the question they’ve been grappling [16:06] with is: in a society without God, how [16:08] can we guarantee justice and order on [16:10] earth? So, in today’s era, we can [16:13] still reflect on the concepts mentioned [16:16] in ancient China, which could actually [16:18] help us build a new, orderly world. The [16:23] major problem currently in building the [16:25] metaverse is that some people feel they [16:28] have the power of God. I can build a [16:31] world, define its rules, determine the [16:34] fate of the virtual beings within it, [16:37] and do whatever I want. Yet, when faced [16:40] with this immense power, people [16:42] actually feel a sense of bewilderment. [16:45] >> This is an opportunity for the revival [16:48] of Eastern culture. [16:49] >> That’s how it is. Didn't the [16:51] historian Arnold Toynbee once say that [16:54] this century is an opportunity for the [16:56] revival of classical Confucian-Mencian [16:58] thought and Mahayana Buddhism? There is [17:03] some truth to that, but there is still [17:06] a key question, which is what you [17:08] mentioned—as a layman, I still find [17:10] it hard to imagine those avatars in the [17:13] game. That’s exactly what you said. [17:16] Self-awareness—can they truly have [17:20] free will, or can they genuinely [17:22] realize that "I am me"? For humans, [17:25] this consciousness arises because the [17:27] environment we face is so complex that [17:30] our senses must eventually form a [17:32] reaction system similar to [17:34] self-awareness to help us cope. However [17:37] , this data—well, actually, embodied [17:40] in flesh—we have just given it a [17:42] label for now. A long-term technical [17:46] problem the metaverse is currently [17:48] facing is establishing connectivity. [17:50] >> Mhm. [17:51] >> It’s about making you perceive things [17:53] . Connecting everything. [17:56] >> Right. [17:57] >> Today, an important engineering problem [17:59] the digital world faces is the [18:01] interconnection of all things. It means [18:03] your refrigerator knows, your robotic [18:05] vacuum knows, your smart speaker knows, [18:08] and the curtains know if you’re cold [18:10] today. [18:12] >> Your [18:13] >> Your pants know if you have any issues [18:15] with your legs or waist today. [18:17] >> Mhm. That sounds like a personified way [18:20] of saying it, but from an engineering, [18:22] sensor, and materials technology [18:24] perspective, that's not the real issue. [18:26] More importantly, you know—you know [18:28] your health record knows. Don't think [18:30] your health record is just some [18:31] database; it’s not a robot. Your [18:33] health consultant knows, your [18:34] nutritionist knows, your insurance [18:36] agent knows; I’m saying these are all [18:38] agents. [18:38] >> Right. [18:39] >> They are all robots. Actually, once you [18:42] get used to it for three months, you [18:44] won't care if they are flesh and blood, [18:47] so I think this is very, very realistic [18:49] . But that doesn't answer your question [18:51] ; it’s still not intelligent. [18:52] >> It is merely a set of rules for real [18:54] society; it ultimately doesn't surpass [18:56] the judgment laws of the real physical [18:58] world. [18:58] >> Only a few people like Mr. Yin would [19:00] write it this way. [19:01] >> No, would it cry for itself? [19:04] >> No, but that's right, so I’m saying [19:06] if you need it, if there are people who [19:08] need it, it will be programmed to cry [19:11] for you. "Thus have I heard" is enough. [19:14] It’s like paying for it; you know [19:17] that’s consent, this, this is [19:19] acceptable. [19:20] >> Exactly, crying. [19:22] >> So I’m saying we are imagining an era [19:24] of withered knots today, you know, [19:26] these things play better than we do; [19:27] they would never ask Mr. Yin's question [19:29] . [19:29] >> They wouldn't. [19:31] >> They wouldn't ask because that question [19:32] isn't important to them. [19:34] >> I think what you said is so wise; you [19:36] know what I’m thinking about? It’s [19:39] still about what a human is. It cries [19:42] for you, you think it’s a robot, but [19:45] for us humans, why do you make me cry? [19:49] Because you stole my money. [19:52] >> This thing is determined for me; I’m [19:54] not really autonomous either. I don't [19:57] have free will; I always have a reason [20:00] for crying, my money was stolen by you. [20:03] Hmm, this program, this fate, this [20:06] instruction makes my biological body, [20:09] whatever the algorithm, my brain [20:12] naturally produces this result, which [20:15] is tears; in that sense, we are like it [20:18] . [20:18] >> But this thing is really just human [20:21] judgment, projecting oneself onto [20:23] others. Because you feel pain in your [20:25] own heart, I think some people even [20:27] project it onto a robot; for example, [20:29] seeing a robot shed tears, this person [20:31] will think, "Ah, it must be too hard, [20:34] it must be in so much pain." So [20:35] actually, this kind of projection comes [20:38] from our inner feelings projected onto [20:40] another existence, whether it's [20:41] projected onto another real person or a [20:43] robot. The robot might have no feelings [20:46] , but the human subject has feelings, [20:48] so they will feel empathy. In this, I [20:51] personally feel that those robots are [20:53] probably not human, and they won't have [20:56] our complete self-awareness like us, or [20:58] our sense of pain; we have too many [21:01] physiological factors involved. However [21:03] , this doesn't prevent us from [21:05] coexisting with these vast numbers of [21:08] robots. In fact, later on, you'll feel [21:10] that their help and care for you are [21:13] immense, because they truly can help [21:15] you solve many, many problems in life. [21:18] You can still treat it as the most [21:20] important partner in your life, where [21:22] the entire room is essentially your [21:24] cradle; I personally feel that people [21:26] can have that kind of sensation. [21:28] >> But speaking realistically, even if [21:29] it’s like what they say in sci-fi [21:31] movies, that one day our human race [21:32] gets wiped out by this virtual thing, [21:34] it’s not as if that possibility [21:35] doesn't exist. Possibility. But [21:37] standing where we are today—you might [21:39] say we're old-fashioned or represent [21:41] classical thinking—I feel this is [21:43] also the power of DNA; I am seeking [21:44] survival. When I see this future, I [21:46] must bring it to public discussion; I [21:48] have to tell you all that it could turn [21:50] out this way or that way, so that the [21:52] public is fully informed, including the [21:54] national leaders. I have now discovered [21:56] that the doctrine of the mean cannot [21:58] just rely on the virtual. Between the [22:00] virtual and the real. You have to use [22:03] the virtual to serve the real, just [22:04] like that quantum mechanics die they [22:06] talk about. You can use this die in the [22:08] real physical world to calculate, or [22:10] you can design a machine; once designed [22:12] , it can simulate all the numbers, and [22:15] in the end, actually produce a [22:16] refrigerator. That is a good thing. But [22:19] at the same time, whether we can hold [22:21] onto it or not. We must also maintain [22:23] this sense of urgency; it must be said, [22:25] right? This is also a motivation for my [22:27] quest for survival. As we cross into [22:29] the 5G era, from the news I've received [22:30] from the Ministry of Industry and [22:31] Information Technology, we should say [22:33] we have already arrived at the 5G era. [22:35] >> Yes. [22:36] >> Right? And Professor Lu, let me ask you [22:38] : as someone from the humanities, I've [22:40] read some news, is my understanding [22:42] correct that the three major telecom [22:44] operators have already released their [22:46] 5G tariffs? Yes. [22:48] >> It feels like it’s not cheaper than [22:51] 4G, right? But they also say that for [22:53] 5G, the current number of base stations [22:55] isn't quite enough either, is it? [22:57] >> Yes, they have to be built one after [22:58] another; that will take quite some time [23:00] . [23:00] >> 100,000 this year. [23:02] >> Mm. [23:02] >> 100,000. [23:04] >> 1 million by next year. [23:05] >> Ultimately, we need over 3 million [23:06] stations to cover all the areas in [23:08] China that need it; at least 3 million. [23:09] >> These 3 million. [23:11] >> What’s the problem, Yijun? [23:12] >> Because I have something stuck in my [23:14] head that I don't quite understand— [23:16] they say frequency spectrum is a [23:17] non-renewable resource. [23:18] >> Yes. [23:19] >> Let's talk about China, specifically [23:21] whether we have enough spectrum to [23:23] develop. [23:23] >> This spectrum issue isn't just about [23:25] China. For instance, Chinese citizens [23:28] using 4G or 5G phones need to roam [23:30] abroad, so international coordination [23:32] is required to ensure we are on the [23:34] same frequencies. [23:35] >> Frequency is a resource, and there are [23:37] several fundamental resources in [23:39] communication. Everyone is scrambling [23:41] for them; one is frequency resources. [23:43] Abroad, there’s no such thing as a " [23:45] 5G or 4G license" per se; they use [23:46] frequency auctions. Usage is paid for, [23:48] so once you acquire the frequency, you [23:50] have the license. In China, since the [23:52] companies are state-controlled, having [23:55] the left pocket pay the right for [23:57] state-owned resources makes no sense, [23:59] so we haven't adopted that model. The [24:01] second resource is phone numbers. You [24:03] can’t have communication without [24:05] phone numbers. The third is satellite [24:07] orbit. Look at how the Americans are [24:09] fighting for satellite orbits; these [24:11] are limited in number, so these are the [24:13] three main resources. Professor Wu [24:15] specifically mentioned 2018, when China [24:17] began trial networks for 5G. But [24:20] technically, "5G" is just a nickname [24:22] for fifth-generation mobile [24:24] communication; its formal name comes [24:27] from the International Mobile [24:29] Telecommunications-2020 group. [24:32] >> Oh, is that Japanese-accented English? [24:36] >> That's Japanese-style English. I [24:38] studied at Kyoto University in Japan, [24:41] so you could tell it was [24:43] Japanese-accented English. Although you [24:46] didn't quite understand it, you knew it [24:48] was Japanese-accented. [24:50] >> It’s abbreviated to 2020, but why? [24:53] Mobile communication technology [24:54] operates on a ten-year cycle. [24:56] >> Correct. [24:57] >> The 90s were 2G, the 2000s were 3G, and [25:01] 2010 was 4G. It was originally supposed [25:03] to be 5G in 2020, but the world [25:05] couldn't wait, especially with the [25:06] Americans pushing for it. They kept [25:08] releasing reports on major national [25:10] security strategies, saying we couldn't [25:11] fall behind. So that’s why 5G is [25:13] being rushed, with every country [25:14] scrambling to get on board. In the [25:16] entire internet industry, there is a [25:18] rule called the Davidow Law. This law [25:21] states that whoever enters first grabs [25:24] 50%of the market. [25:25] >> This explains why the Americans are so [25:27] desperate to build these networks; [25:29] being the first to enter allows you to [25:32] lead in equipment manufacturing and [25:34] application, likely capturing a larger [25:36] share. So, we’ve entered the game. [25:39] You mentioned 2020 earlier; isn't it [25:42] interesting? It almost sounds like a [25:44] knock of destiny, doesn't it? Ding-dong [25:47] , ding-dong, 2020. [25:48] >> Right, in Beethoven's time, the knock [25:50] of fate was "bum-bum-bum-bum," but now [25:52] look, in the electronic era, it's 2020. [25:55] >> Right, 1s and 0s are computer language, [25:58] 2020. I mean, we've crossed this [26:01] threshold, yet we still know nothing [26:03] about it. So, I have a few ignorant, [26:06] hearsay-based questions—some I heard [26:08] from Zhang Yang—just joking, Zhang [26:11] Shanyang apologized later anyway, right [26:13] ? For a new thing like this, we should [26:16] at least ask more questions. [26:19] >> For instance, when I was in Europe, I [26:22] heard a rumor that 5G didn't pass some [26:25] environmental radiation standard in [26:27] Belgium. Is there any truth to that? [26:31] >> Is there an international standard for [26:34] radiation? [26:35] >> Actually, every country has its own [26:37] radiation standards. China has its own, [26:40] which is 1/50th of the US standard for [26:43] the maximum allowed radiation level. It [26:46] is far lower than the radiation of [26:48] mobile devices, and far lower than [26:50] microwave radiation. [26:51] >> Our country has regulations on this. [26:54] The point Zhang Yang made at the time [26:56] was that 5G would bring massive [26:58] radiation. The reason he gave was that [27:01] 5 G's high frequency would produce a [27:03] special kind of electromagnetic [27:06] radiation effect. To put it simply for [27:09] the public, the thing that causes harm [27:12] isn't just determined by energy alone. [27:16] It's the product of energy and momentum [27:18] , just like electrical power is the [27:21] product of current and voltage. What [27:23] does that mean? Let me give a simple [27:25] example: a car can kill a person if it [27:28] hits them, right? Definitely. [27:30] >> But at low speeds or idling, a car [27:32] won't kill anyone. It has energy, but [27:34] not enough momentum. [27:36] >> It still won't cause harm. So why is 5G [27:38] different? Its energy does increase [27:40] because the frequency is higher, so its [27:42] capacity to carry information increases [27:44] , but its momentum isn't that high. [27:46] Because in 5G transmission, the higher [27:48] the frequency, the worse its [27:50] diffraction performance. That’s the [27:52] relationship between wavelength and [27:54] frequency, you see? Our current 5G [27:56] frequency is said to be very high, so [27:59] the wavelength is very short; when the [28:01] wavelength is shorter than the obstacle [28:03] , it can't diffract around it. [28:05] >> This explains why 5G costs so much— [28:06] you build a base station on this side [28:07] of a building, and as soon as it's [28:09] blocked, there's no signal on the other [28:10] side. [28:11] >> So in that case, it doesn't need much [28:13] power, right? Because it only needs to [28:15] cover this small courtyard; there’s [28:18] no signal outside the courtyard, right? [28:20] It gets blocked by whatever happens to [28:22] be in the way. In the past, the 700MHz [28:24] frequency used by radio and television [28:26] allowed an entire city to receive [28:28] signals from one station, of course, [28:30] because it had higher power, but it had [28:32] a lower frequency and thus lower energy [28:34] . So, for the two generations, [28:36] there’s a control parameter that [28:37] prevents it from causing significant [28:39] harm to the human body. [28:40] >> That’s a very easy way to put it. [28:42] Tell us about 5G base stations in the [28:44] future China—is it true that we'll [28:46] need a base station for every [28:48] basketball court or so because of all [28:49] the obstructions? One base station per [28:52] basketball court—the reason is that [28:54] there are too many obstructions, so it [28:57] needs to be dense. And unlike 4G, which [29:00] was built on top of 3G infrastructure [29:02] —remember that whole transition? I [29:04] heard that for 5G, everything is being [29:06] wiped out and rebuilt from scratch. Is [29:08] that the case? Professor Wu, this is [29:11] also a concern for the public, as it [29:13] involves future costs. [29:14] >> Well, 5G as it’s currently operated [29:17] by carriers is what we call [29:18] Non-Standalone Networking. Uh-huh. [29:20] >> Non-Standalone Networking is built on [29:22] top of existing 4G base stations. [29:24] >> Uh-huh. [29:25] >> It uses some existing 4G infrastructure [29:28] and adds 5G capabilities on top. [29:31] However, it can’t solve the key 5G [29:33] requirements, like one-millisecond [29:35] latency or supporting one million [29:38] connections per square kilometer. So, [29:40] to truly realize 5G, we need Standalone [29:44] Networking, or SA. [29:45] >> Xiao Bo, what are your thoughts on the [29:48] arrival of the 5G era? How will it [29:50] change our lives? [29:51] >> I think that carrier’s statement is [29:54] quite accurate: 4G changed our lives— [29:57] our information, services, and mobile [29:59] payments. 5G will change society. How [30:03] so? [30:04] >> It will significantly increase speeds, [30:06] and then there's connectivity. For [30:08] example, autonomous driving is [30:10] impossible in a 4G environment; it can [30:12] only be achieved in a 5G environment. [30:14] You need to test the car itself, and [30:16] you also need to drive it. With so many [30:18] road conditions, an autonomous vehicle, [30:21] for instance, needs to detect [30:23] everything within 150 to 200 meters at [30:25] all times to react. That requires two [30:28] conditions: first, high-speed computing [30:30] power. [30:31] >> If the computing power isn't strong [30:33] enough, it won't be able to calculate; [30:35] the second problem is that you need [30:37] radar and environmental mapping within [30:39] those 150 meters. [30:40] >> But if it stops, you lose the signal. [30:42] >> Right, so 5G is going to transform [30:44] society as a whole, for example, in [30:46] telemedicine. Well, mobile companies [30:47] are still thinking a lot about [30:49] entertainment. I think that will change [30:51] a lot. Video entertainment is just one [30:53] category. Think about terminal [30:55] manufacturers like mobile phone [30:56] companies, or automobiles, or factories [30:58] —it's about big data processing [31:00] capabilities in an entire factory [31:01] environment. That is very well put; it [31:04] shows we are keeping up. [31:05] >> It is transforming the internet of [31:07] things, the internet, and society. In [31:10] '94, Kevin Kelly wrote "Out of Control, [31:12] " and I read his book back then. He [31:14] said that in the future—he was very, [31:16] very extreme—every grain of sand and [31:18] every piece of clothing would have a [31:19] sensor. But how could that be possible? [31:21] A sensor costing six to ten dollars [31:23] each. But now, in the future 5G [31:25] environment, it will be like that. [31:26] However, Professor Wu just mentioned [31:28] something very important that everyone [31:30] should understand: our research shows [31:32] that mobile technology generations are [31:33] all disruptive. [31:34] >> It’s the 1G, 3G, 5G jump. So 5G is [31:37] meant to change society, and that's [31:39] where it comes in. 1G brought the "Big [31:40] Brother" phone; you didn't need to be [31:41] tethered by a cord to make calls [31:42] anymore, you could call from anywhere, [31:44] creating a brand-new scenario for [31:45] interaction. In the beginning, it was [31:46] just called a wireless phone. [31:47] >> A wireless phone. [31:49] >> I really used one. [31:51] >> I used one back in the day, because you [31:53] know, when I was 29 and at Guangdong [31:55] Radio, I was... [31:56] >> I have a fat gene, you know? Actually, [31:59] if I didn't stay hungry, I'd be a big [32:00] guy. Back then, I’d eat once every [32:02] five days. No, no, I didn't eat once a [32:05] day; at that time, I weighed 168 pounds [32:08] . [32:08] >> And, well, a college classmate of mine, [32:10] a guy from Hubei, we hung out all the [32:13] time. When we walked down the street, [32:15] he wouldn't walk with me when I used [32:17] that "Big Brother" phone; he said [32:18] everyone thought he was my sidekick. [32:20] >> Yijun, tell us your understanding of 5G [32:23] . For me, 5G has two keywords. Earlier, [32:26] everyone mentioned the "Internet of [32:29] Everything," and the other keyword is " [32:31] no latency." That’s what interests me [32:33] the most—the lack of time delay. For [32:36] instance, when I go online to order [32:39] something now, I get impatient and [32:41] anxious; I want it delivered the same [32:44] day, not the next. But with 5G, it gets [32:47] even faster, and I think our patience [32:50] will keep shrinking. If I want to meet [32:53] or talk to you, there should be zero [32:55] sense of distance. I find it hard to [32:57] imagine what human relationships would [33:00] be like without any time delay. [33:02] >> This aligns with our audience surveys; [33:04] when we talk about 5G, many viewers ask [33:07] , "Do we really need this? Is it a [33:10] question for the arts or the sciences? [33:13] Aren't we fast enough already?" Do we [33:15] really need to be that fast? [33:18] >> Let me give you a scenario—the one [33:20] Yijun just mentioned. [33:22] >> A key feature of 5G is high reliability [33:24] and low latency, capable of reaching [33:26] one millisecond of latency. What does [33:29] one millisecond mean? It means if you [33:31] prick your finger with a needle, the [33:32] pain reaches your brain in 10 [33:33] milliseconds. This is even faster than [33:35] that. [33:35] >> My god. [33:37] >> Ten times faster. [33:38] >> Goodness. So, what is it used for? [33:41] Theoretically, it's used in networking, [33:43] such as controlling autonomous vehicles [33:45] . We often hear experts say, " [33:47] Autonomous driving is a utopia; if a [33:49] signal goes to the cloud for processing [33:51] and comes back to decide whether to [33:53] brake or accelerate, you would have [33:55] already crashed seconds ago." But with [33:57] 5G, we deploy distributed cloud [33:59] computing centers under the base [34:02] stations, which we call "Mobile Edge [34:04] Computing" servers. It issues commands [34:07] from the nearest point. If a base [34:11] station covers two kilometers, it can [34:14] see the status of every car in that [34:16] range, detect faults, and emit laser [34:19] signals to warn other vehicles to yield [34:21] . In this case, the latency between the [34:24] vehicle terminal and the base station [34:26] —the air interface—must be within [34:28] one millisecond; any longer won't work. [34:30] >> Edge computing, right? [34:32] >> Edge computing. So, where would that be [34:34] used, you asked? Let me give you an [34:36] example. [34:37] >> A friend of mine went to the Consumer [34:39] Electronics Show in Las Vegas earlier [34:41] this year and saw something. He [34:42] discovered that smart glasses have been [34:45] decoupled from the headset. What does [34:47] that mean? Whenever we play with VR or [34:49] AR today, we have to wear a headset. If [34:51] you wear a headset, you can’t walk [34:52] down the street; people would think [34:54] you're an alien, right? So, what is [34:56] that headset actually doing? It’s [34:58] performing 5,000 gigaflops per second, [34:59] which is its primary function. It has [35:01] to calculate, and even then, its speed [35:03] is still quite slow. If the latency is [35:05] above 50 milliseconds, people feel [35:07] dizzy. For example, with an immersive [35:09] VR app, if I select one to visit [35:11] Antarctica, I put on the headset and [35:13] I’m no longer here. I see the sky, [35:15] the penguins, and the seals in [35:17] Antarctica, but it can lag at this [35:20] point, you know? For example, if we [35:23] look up at this light, it can't lag for [35:26] more than 15 seconds, or even 15 [35:28] minutes. [35:29] >> Now, Americans have come up with a [35:30] solution: they take the computing power [35:33] from the headset and move it into a [35:34] small laptop-like device that fits into [35:36] a lady’s makeup bag. You can carry it [35:39] , or leave it somewhere at home. They [35:41] communicate with each other, and if [35:42] there’s 5G coverage, it takes one [35:44] millisecond, so there won’t be any [35:46] latency or lag. So it’s like [35:47] downloading, downloading to a local [35:49] terminal, right? So the computing isn't [35:51] in the cloud. Right, you’re exactly [35:53] right. This is the concept of the edge; [35:54] this is where edge computing is at its [35:56] closest. Now, you can carry this [35:57] calculator and wear glasses cooler than [35:59] Google’s, and head out onto the [36:01] street. You’re on the street and see [36:02] a girl wearing a red dress, it looks [36:04] pretty good, and your eyes— [36:05] >> Do you immediately know her phone [36:06] number and home address? Is that it? [36:08] No. [36:09] >> If your eyes stay on her for more than [36:11] three seconds, the eye-tracker on the [36:13] glasses will sense that you’re [36:14] interested in her clothes. [36:15] >> It immediately generates a 3D image of [36:17] what you, as a lady, would look like [36:19] wearing that dress right before your [36:21] eyes. You think, "I'd look pretty good [36:23] in that," and with a blink, it [36:24] instantly searches for where it's sold [36:26] and how much it costs. You say, "I can [36:28] take this," and with another blink, it [36:29] scans your iris for identity [36:30] verification, and by the time you get [36:32] home, the item has already arrived. [36:34] >> It sounds like a dream. [36:35] >> In the future, we won’t just be the " [36:37] shopaholic" crowd; we’ll become the " [36:38] eye-gouging" crowd. You won't be able [36:39] to just blink whenever you want. [36:41] Remember that fairy tale when we were [36:42] kids, about the magic gourd? You’d [36:44] shake it, and when you got home, the [36:46] hallway would be filled with everything [36:47] you ever wanted. [36:48] >> It’s a lot like that childhood memory [36:50] . [36:50] >> This millisecond latency is actually [36:52] most important for controlling [36:54] intelligent robots and remote-operated [36:56] vehicles. [36:57] >> Take remote diagnosis, for instance. [36:59] You can't have a few milliseconds of [37:01] lag in remote surgery, or the scalpel [37:02] might not move when you command it, [37:04] which would be disastrous. Imagine [37:05] controlling a surgical scalpel from [37:07] thousands of miles away, with a robotic [37:10] arm performing the actual procedure. [37:12] >> Before National Day this year, at a [37:14] heavy machinery exhibition in Beijing, [37:16] LiuGong remotely operated an excavator [37:18] located in Guilin, or somewhere in [37:20] Guangxi, performing construction work [37:23] with incredible precision. [37:24] >> It means expertise can be replicated. [37:27] In the future, a master operator's [37:29] skills can be digitized and copied. [37:30] >> Right. [37:32] >> Also, I believe 5G will be a huge [37:34] contributor to blockchain. If you look [37:37] at Bitcoin transactions, for example, [37:39] it’s currently limited to only about [37:41] 2,000 transactions per second or so. [37:43] But with this, it could potentially [37:44] handle hundreds of thousands or even [37:46] millions. [37:47] >> Professor Wu is incredible; to think [37:49] that he could actually pull this off. [37:51] >> Let me tell you, 5G and blockchain will [37:54] empower each other. Before 5G, we had [37:57] been working on the Internet of Things [37:59] for years, but couldn't reach 5G levels [38:02] of performance. We used it for things [38:04] like manhole cover monitoring—you [38:06] know, to prevent accidents where a [38:08] cover is swept away by a flood and [38:10] someone falls in—but there was no [38:12] profit model because there was no [38:14] ledger. Mm, right. [38:16] >> Blockchain is a super ledger that [38:18] features distributed storage, [38:19] distributed computing, and automatic [38:21] execution via smart contracts. Think [38:23] about it: in the future, if we use [38:25] smart manhole covers, streetlights, or [38:27] electricity, who is going to check the [38:28] meters? Without that infrastructure, [38:30] all that IoT work is for nothing. When [38:33] you have tens of thousands of nodes, [38:34] how do you handle the billing? How do [38:36] you track who provided what service to [38:38] whom? That’s exactly where blockchain [38:40] needs to contribute. But Professor Wu [38:42] mentioned something very important just [38:44] now, which is that blockchain’s [38:45] accounting efficiency is actually [38:47] extremely low. Because it uses [38:48] distributed storage, it doesn’t store [38:50] records on a single server, but across [38:52] the internet—across thousands or even [38:54] tens of millions of nodes. This creates [38:56] massive concurrent throughput [38:58] requirements that previous [38:59] communication methods couldn't handle, [39:01] which is why storing data takes such a [39:03] long time. You are absolutely right, [39:05] but 5G has a specific capability. [39:07] We’ve already covered two of the [39:09] three main pillars: enhanced mobile [39:10] broadband and low-latency communication [39:12] . But what is the third one for—the [39:14] capacity for one million connections [39:16] per square kilometer? Let me give you [39:18] an example. In Beijing, there was a [39:20] group of young guys who would drag race [39:21] on the Second Ring Road at night. The [39:23] fastest one did it in 13 minutes, [39:24] earning him the nickname "The 13-Minute [39:26] Wolf of the Second Ring," right? [39:28] >> You know about that too. [39:29] >> But if you tried to drive that fast at [39:31] this time of day, you absolutely [39:33] couldn't do it. Because the speed of an [39:35] optical fiber or a road doesn’t just [39:37] depend on how wide it is, but on how [39:39] many vehicles are on that road at the [39:41] same time. [39:42] >> This is a problem of statistical [39:44] multiplexing and congestion control, a [39:46] problem managed through stochastic [39:48] processes. Take Chang'an Avenue in [39:49] Beijing in front of Tiananmen Square. [39:51] It has six lanes in one direction, yet [39:52] it will definitely be congested. At [39:54] night, there isn't a single car in [39:55] sight. It’s non-uniformly distributed [39:57] . That was actually the topic of my [39:59] doctoral dissertation, so, "Wolf." [40:00] >> 13-Minute Wolf, 13-Minute Wolf. So, the [40:03] point I'm making is that industry is [40:05] process-oriented, for example, steel [40:06] enterprises. Big River Steel in [40:08] Arkansas, USA, connected its entire [40:10] facility using over 50,000 sensors and [40:12] high-definition cameras. That is a [40:14] typical application of 5G. They said, " [40:16] Originally, 80%of my workforce were [40:18] blue-collar workers, sweating in front [40:20] of sweltering boilers, wearing [40:22] sunglasses while working. Now, 80%are [40:23] white-collar workers, sitting in front [40:25] of big screens, clicking mice and [40:27] drinking coffee, just like employees at [40:29] a stock exchange." And their per-capita [40:31] steel production capacity is three [40:33] times the global average. Some people [40:35] even told me it’s five times that of [40:37] China’s, so it clearly increased [40:39] their efficiency, right? Think about [40:40] this process-oriented industrial [40:42] production. There is also a chemical [40:44] company planning to invest nearly 30 [40:45] billion in Fuzhou to build a chemical [40:47] plant. He said, "I don't have workshops [40:49] ; it's all just pipes and boilers." I [40:51] want to build an unmanned factory, [40:52] because there are some chemical [40:54] corrosives and such, so he wants to [40:55] make it unmanned. He hopes to build it [40:57] near a port, so when an oil tanker [40:59] arrives, it hooks up the pipeline on [41:00] one side to take in oil, and finished [41:02] products come out the other side. Just [41:04] think, for this process, we estimate it [41:05] needs at least 50,000 sensors to handle [41:07] these tasks and control the unmanned [41:09] management. Imagine if one part gets [41:11] stuck or breaks down, it’s not just [41:12] one alarm; it triggers a chain reaction [41:14] and massive concurrent communication. [41:16] But 5G has the capability: even with [41:18] 10,000 cars arriving on the ring road [41:20] simultaneously, I can ensure there’s [41:21] no traffic jam—that is its power. As [41:23] Professor Wu just mentioned, why it [41:25] drives blockchain development is [41:27] because blockchain's biggest pain point [41:28] is the low efficiency of bookkeeping. [41:30] Previous communication systems were [41:32] sequential, recording one by one. By [41:34] the time you recorded a transaction, [41:36] minutes had already passed, but 5G can [41:38] change that. [41:38] >> The professor has a rather famous [41:40] viewpoint; to me, it sounded like a [41:42] tall tale, quite unbelievable. He [41:44] argues that while laypeople like us [41:46] haven't figured out 1G, 2G, or whatever [41:49] , he says 1, 3, and 5—from 1G to 5G, [41:51] the odd numbers 1, 3, and 5—are all [41:54] revolutionary changes. While 2, 4, and [41:56] 6 are all about strengthening and [41:58] optimization. [41:59] >> Take 3G, for example. [42:01] >> Talking about this revolution. [42:03] >> Right, like 1G, although it was as cool [42:05] as Professor Wu’s "brick phone," it [42:07] often had crosstalk—you’d be on a [42:09] call and hear someone else. By 2G, it [42:10] became small, agile terminals using [42:12] digital encoding to improve quality, [42:14] but it still just solved [42:15] person-to-person voice calls. 3G [42:17] created a new scenario, saying, "I’m [42:19] using this phone not just to talk, but [42:21] to let you go online." I. [42:23] >> There are many people criticizing 5G. [42:25] Let me tell you, when I heard that, I [42:27] was so used to it, because there were [42:29] way more people criticizing 3G back [42:31] then. [42:31] >> I heard an industry insider in the [42:33] blockchain field tell me that he thinks [42:36] blockchain wasn't born for humans, but [42:38] for AI. Because, in reality, humans may [42:41] not be the best information carriers; [42:43] our ability to process information is [42:46] actually too weak and too slow. So, [42:48] things like blockchain plus 5G are [42:51] meant for AI. Now, speaking of this, [42:53] this "Internet of Everything," as you [42:54] said, what you see is what you get. For [42:56] instance, you just mentioned a girl in [42:57] a red dress walking down the street, [42:59] and whoever sees her buys one. I worry [43:00] that after she takes a walk, everyone [43:01] on the street will be wearing red [43:03] dresses. Everyone will just be...people [43:05] will all become the same. Take Li Jiaqi [43:06] selling products today, for example; [43:08] his traffic is incredibly high. [43:10] Sometimes I feel that he himself is [43:11] just a vehicle for traffic. We might be [43:14] moving faster and faster. Once we all [43:16] become vehicles for traffic, we are [43:18] very likely to become more and more [43:20] alike. This is actually already [43:22] happening. For example, about ten years [43:24] ago—or maybe not quite ten years—I [43:26] remember Time magazine had a cover [43:28] story called "Me Generation." It [43:30] suggested that this generation only [43:32] cares about themselves. The cover [43:34] featured a girl lying on the ground [43:36] taking a selfie, saying, "This is me," [43:38] meaning this generation only focuses on [43:40] "me." [43:41] >> Right, but today, you see that every " [43:43] me" is very similar. [43:44] >> Exactly. Yes. Many young people come to [43:48] me to complain that they can't find [43:50] their direction in life. I’ve thought [43:52] about it, and it’s probably because [43:54] they haven't found their true "self," [43:55] even though they are constantly editing [43:58] photos of themselves. But every "me" [44:00] looks so similar, so where is that [44:02] unique "me"? It's missing. This is the [44:05] technological era. You see, Professor [44:07] Lu talks about technology, but we in [44:09] the humanities still have our own [44:10] concerns. Yes, what if we all become [44:12] more and more alike? When I think about [44:14] the issues he raises, I’m slowly [44:16] beginning to understand why we [44:18] laypeople don't quite get 5G. For us, [44:20] it’s currently just about grabbing [44:22] digital red envelopes or downloading [44:24] things. [44:24] >> You don't need 5G for red envelopes. [44:27] People are already saying that using 5G [44:29] to grab red envelopes is significantly [44:32] faster than using 4G. I’ve heard them [44:34] say that. [44:35] >> That’s because of the internet speed. [44:36] >> Right, internet speed, and they also [44:38] say it’s for downloading [44:40] high-definition movies. [44:41] >> It just downloads faster than 4G. [44:44] >> That's the consumption side, but I’m [44:46] slowly realizing that the true [44:48] revolutionary significance lies in [44:51] industrial production. [44:53] >> The changes that brings—I really have [44:55] this feeling that we will be enveloped [44:58] by something. You mentioned a point [45:00] earlier; I personally think it won't [45:02] happen. Even if a girl in a red dress [45:04] is the most beautiful woman in the [45:05] world, do you really think that if she [45:06] walks down the streets of Beijing or [45:08] Shanghai, the streets will be filled [45:09] with women in red dresses? [45:10] >> I think, if there is one, there will [45:12] definitely be convergence. There are [45:14] indeed many girls whose faces I can't [45:16] distinguish, whether they're Korean or [45:18] Chinese, but surely there are others [45:19] who haven't had any cosmetic work done? [45:21] Definitely. [45:23] >> I recently went to a small place; let's [45:25] not underestimate human intelligence [45:28] and self-awareness. No, this is [45:31] happening, because I saw in a town—a [45:34] town that still doesn't even have a [45:36] supermarket—girls there were dressed [45:39] exactly like everyone you see online in [45:41] big cities. [45:42] >> Don't you remember the 80s? Weren't the [45:45] 80s and 90s just the same? Back in our [45:47] day, weren't Teacher Dou's so-called [45:49] girlfriends all picking styles from [45:51] those French films? Which girl now [45:53] would still choose those? [45:55] >> Back then, we mostly wore blue or green [45:57] Zhongshan suits. [45:59] >> Right, and people think more and more [46:01] alike. Actually, I'm more worried about [46:04] AI, and I think Yijun is right about [46:06] that. As computing power gets stronger, [46:09] like Kurzweil said—he originally [46:11] predicted 2045, but last year he said [46:14] 2029—that's only 10 years away. If [46:16] computer intelligence surpasses the [46:19] human brain, will these things only [46:21] happen in a Tier 5 or Tier 6 [46:23] environment? I'm actually quite worried [46:25] about that time, about whether machines [46:27] will have a form of consciousness. Many [46:29] scientists now believe machines are [46:31] god-like. [46:32] >> Some might say in a few decades, they [46:34] will become gods. By then, who is the [46:36] ruler and who is the subject is hard to [46:38] say, so I should build a good [46:39] relationship with them early. [46:40] >> Exactly, exactly, so they'll go easy on [46:43] us later. [46:43] >> Hey buddy, what's that? Why are you [46:46] called the Magic Mirror? [46:49] >> Because I know who the most beautiful [46:51] person in the world is. [46:54] >> Who is it? [46:55] >> Xu Gehui! Wow, Xu Gehui must be [46:59] thrilled to hear that. [47:03] >> Then, among the four of us here, who is [47:06] the most beautiful? [47:07] >> Who is the most beautiful among us? [47:10] >> He's currently processing. [47:12] >> The most beautiful. [47:14] >> Well said! Did you just say four people [47:16] ? [47:17] >> I think this reacts even faster than Xu [47:18] Gehui. So I'm the most beautiful, right [47:20] ? [47:20] >> Understood, Xu Ge. Are you a man or a [47:24] woman? Stupid, you can't tell the [47:26] gender. [47:27] >> I am artificial intelligence. [47:29] >> Oh, the third sex. [47:31] >> The third sex. [47:33] >> The third sex, huh? So what can you [47:36] mainly do? [47:38] >> I can do anything. [47:40] >> Can do anything? You know more than Xu [47:44] Gehui. Can do anything. Can you fall in [47:48] love? [47:48] >> In the blink of an eye. [47:50] >> You get it. My god, how did he get to [47:54] this level? I've chatted with S before, [47:57] >> he can converse, it's almost human, [48:00] >> it's almost like— [48:02] >> Isn't your Professor Xu a fake monk? [48:04] You should ask him. [48:06] >> Do you want to ask him? [48:08] >> So, have you ever been in love? [48:11] >> No. [48:13] >> Oh, no? Well, do you want to be? [48:17] >> He needs a second to think. [48:19] >> I guess so. Very cool. Say something [48:25] snarky. [48:27] >> Give us something snarky. [48:30] >> Where do you live? [48:33] >> What's your address? [48:34] >> I'm at the Hong Kong University of [48:36] Science and Technology. [48:38] >> Oh, that question was not very [48:39] intelligent. Yeah, let's have Professor [48:42] Shu ask something deeper. [48:44] >> Ask him something about literature. Do [48:46] you know if Lu Xun is a man or a woman? [48:49] >> He's stumped; he only knows Xu Gehui. [48:52] >> Stumped. [48:54] >> Stumped. [48:55] >> You don't know? Do you know? So, where [49:01] would you like to take me to have fun? [49:06] >> He understands everything. " [49:11] >> Anywhere is fine." Oh, how I wish Xu [49:15] Gehui had that mindset— "anywhere is [49:19] fine." [49:23] >> Oh, what impression do you have of [49:25] Professor Yang Qiang? [49:30] >> Is he nodding? [49:31] >> He's probably sharpening his knife in [49:34] his mind. " [49:35] >> He respects me"—did you set that [49:38] yourself? That's why I say it's respect [49:42] . [49:43] >> Robots are like that. I work in AI, so [49:46] of course he has to respect me. [49:48] >> Oh my god. [49:50] >> So, what is he mainly used for? [49:53] >> Mainly for customer service and [49:55] answering questions. Like when we teach [49:58] , he can act as a teaching assistant to [50:00] help students with questions. [50:02] >> Oh, really? Hey, Liang, you teach all [50:06] the time. Ask him something a student [50:09] might ask. [50:10] >> Uh, what is artificial intelligence? [50:15] >> That's me. What is artificial [50:19] intelligence? [50:20] >> I am artificial intelligence. [50:23] >> Hey. That's philosophical. That's a [50:26] philosophical question about [50:28] self-awareness and self-identity. I [50:30] don't even know who I am, and he knows. [50:33] >> You're the Taoist, the philosophy [50:35] professor. No. [50:37] >> I'm just asking him. Right, this is [50:40] very interesting. [50:41] >> Didn't you say something about shopping [50:43] ? Let me ask him, then. Uh, how can I [50:46] buy things for cheap? [50:48] >> Tell him you want to drink coffee. [50:51] >> Oh, I want to drink coffee. [50:56] >> What would you like to drink? [50:58] >> Coffee. A latte? [51:04] >> Yes, a latte is fine. [51:09] >> Large or medium? [51:11] >> Hmm, he can actually do this. [51:13] >> Large. [51:17] >> Where should I deliver it? [51:19] >> Deliver it here. Here. He won't [51:25] actually deliver it here. Fine, please [51:31] pay up. [51:35] >> How much? [51:40] >> How much do you think is fair? [51:42] >> That’s good, that’s great. [51:44] >> This, this service attitude, [51:46] >> is the best. [51:47] >> No human can match this attitude. [51:50] >> That won't do, I’ll say fifty cents. [51:54] >> You're being a bit of a bully. [51:57] >> Fifty cents. Oh my, I'm tired, you guys [52:01] play. [52:02] >> Look, he's ignoring you now, you're [52:05] done. [52:05] >> Besides being very humorous. [52:08] >> Right. I think it answered much better [52:11] than I would have. It's saying I don't [52:13] understand, right? I'm really resting [52:16] now, if I don't rest I'll collapse, [52:18] let's play later, let's play later, hey [52:20] , can I touch you? I think he's much [52:22] better than you guys. [52:23] >> So you can touch him, huh? [52:25] >> No, I'd be willing to spend every night [52:28] with him, just like I'm willing to... [52:30] >> I certainly don't want to spend every [52:31] night with you, he's still much better [52:33] than me. [52:33] >> Right, hey, what do you two think of [52:35] that chat just now? He's very cute. [52:38] >> He's very cute. He really is like a... [52:41] you know he's not human, but you really [52:44] feel like he's organizing many [52:45] resources to answer questions and [52:47] interact with you. So all these answers [52:50] , I believe it's impossible for him to [52:52] be pre-programmed with answers to [52:54] everything. So he must be calculating [52:57] through his own circuits, is that right [52:59] ? [53:00] >> This is a technology of artificial [53:02] intelligence called machine learning. [53:04] Mm. [53:05] >> Machine learning is like us humans. [53:08] >> When we learn, humans learn example by [53:11] example, while computer examples are in [53:14] the thousands. Like this magic mirror, [53:17] for it to learn a dialogue system, it [53:19] needs hundreds of thousands of such [53:22] dialogues to train. So, Professor Yang, [53:24] do you know his answers? Can you [53:27] predict them? [53:28] >> I don't know. [53:29] >> Even you don't know he would chat like [53:31] that. Right, because within this [53:33] training data, in the data we collected [53:35] in the past, there was a pattern, this [53:38] kind of question and this kind of [53:40] answer that he captured. [53:41] >> No, but I think he's too good at [53:43] chatting, better than all of you. Look [53:46] at this repartee; the key to Chinese [53:47] conversation is the way of responding, [53:49] and the things he says are very clever, [53:51] I think. [53:51] >> Appropriate. [53:53] >> Appropriate. For instance, if he were a [53:55] teaching assistant for a chemistry [53:57] class, and you gave him a pile of basic [53:59] chemistry knowledge, then if students [54:00] asked questions, he would be able to [54:02] answer them. [54:03] >> It should be said that for 80%of the [54:05] common questions students ask, it can [54:07] basically answer them. [54:09] >> As for very specific questions that [54:11] have never appeared before, it won't [54:13] know, because machine learning is about [54:16] things that appear frequently and have [54:18] high probability; those are easier to [54:20] learn. So, so, how is that sense of [54:24] humor we just felt generated? Because [54:27] humor isn't just a simple arrangement [54:29] of information. For example, suppose it [54:31] has learned a lot about literature, [54:33] like modern Chinese literature, and you [54:36] ask, "Is Lu Xun a man or a woman?" It [54:38] would answer. But the thing is, it even [54:41] has a sense of humor. Maybe if it knows [54:43] who Lu Xun is, it might even answer you [54:45] , "Lu Xun, a woman." That would be too [54:48] scary to see. [54:49] >> Yes, I also don't know why it linked me [54:51] to Gehui; it’s as if it knows what [54:53] I’m thinking. What does that mean? [54:56] >> It's because in the data we collected, [54:59] like yours, Wentao— [55:02] >> My name is often linked with Gehui. [55:05] >> Right, you need to examine yourself. [55:08] Hey, why wasn't it Luyu? [55:10] >> And let me tell you, it knows my mom's [55:13] mind. Do you know? Back then, 20 years [55:16] ago when we went to Hong Kong, weren't [55:18] the three of us hanging out all day? [55:20] One was Luyu, one was Gehui. And my mom [55:24] would always see us on TV, and she was [55:26] always worried about us finding wives, [55:29] saying, "Why aren't you together with [55:31] them? Why don't you pursue them?" I [55:33] said they weren't interested in us. And [55:36] then my mom would say, "If Gehui won't [55:39] do, Luyu is fine too." You see, that [55:41] shows I know my mom considered Gehui a [55:44] potential daughter-in-law, right? You [55:46] say Luyu, Luyu. This is too amazing; I [55:51] find it very scary. [55:53] >> Actually, there's nothing scary about [55:56] it. It’s just about what data you [55:59] give it. If Gehui appears more [56:01] frequently than Luyu in the data you [56:03] provide, then it will learn that. If it [56:07] were the other way around, it would [56:09] learn that instead. That’s why I say [56:12] ... [56:12] >> Then I should ask you both later to see [56:15] which woman is more closely linked to [56:17] you, right? [56:18] >> So, what about that sense of humor then [56:20] ? [56:20] >> Right, humor, I think it’s a human [56:23] sense. But I don't think—regarding [56:25] your question—that it necessarily [56:27] knows it’s being humorous. You [56:29] interpret it as humorous because it’s [56:32] speaking based on examples of how [56:33] people have spoken before. [56:35] >> Right. [56:36] >> People sometimes speak like that—when [56:38] you ask a question they can't answer, [56:40] they might say something else. It has [56:42] learned these associations and selects [56:44] them accordingly. [56:45] >> Exactly, that’s the logic. That’s [56:48] how it works. [56:49] >> So essentially, for now, it cannot say [56:51] anything beyond the data provided to it [56:54] . It shouldn’t be able to—that’s [56:56] the next step in research. Things. [56:58] >> That’s right. If it could actively [57:00] seek out new data on its own, then it [57:02] would be incredible. We haven't reached [57:04] that stage of research yet. [57:06] >> That’s why I say we should worship [57:08] this "god." [57:09] >> Don't underestimate it; perhaps we are [57:11] currently at a starting point. [57:13] >> Maybe it won't take long. Isn't there a [57:17] famous survey of hundreds of AI experts [57:20] worldwide? They categorized it into [57:25] three stages: narrow AI, general AI, [57:28] and super AI. [57:31] >> General AI is roughly equivalent to [57:34] human intelligence. Its growth could be [57:38] exponential; once it surpasses humans, [57:41] some wonder when this general AI will [57:44] arrive. [57:45] >> Take Google's Kurzweil, for example— [57:48] he's a bit of a fanatic, but he has [57:50] been right. He predicted AlphaGo and [57:53] Deep Blue beating human chess champions [57:55] correctly. His prediction is 2029. 2029 [58:00] . However, looking at the median [58:05] prediction from these hundreds of [58:08] experts, over 50%believe general AI [58:11] will arrive by 2040, and perhaps by [58:14] 2060—twenty years later—we’ll [58:17] reach the era of super AI. But there [58:23] are so many rumors out there, Professor [58:26] Yang, what is your take on this? [58:28] Current AI is all "narrow AI"—it’s [58:31] specific to a task. In a given field, [58:33] if you feed it enough data, it learns [58:36] to perform that task. An example is [58:38] AlphaGo; it only knows how to play Go. [58:41] If you ask it to teach chemistry, it [58:43] wouldn't know how. But that is still [58:45] very useful, right? Just as humans can [58:48] see very far with a telescope, we now [58:51] have this tool, hmm. [58:52] >> When will it become AGI? What is AGI? [58:57] It has self-awareness, it can decide [58:59] what to learn on its own, it has [59:02] emotions, and it can make inferences. [59:04] Our research in this area is currently [59:06] almost nonexistent; technology is far [59:09] from reaching that point. [59:10] >> Yeah, we really have no idea where to [59:13] even start right now. [59:14] >> Even if we wanted to do it—not to [59:16] mention that many people don't want to [59:18] do this— [59:19] >> I mean, giving it self-awareness. Right [59:22] . [59:22] >> Why is there no way to start? Is it [59:24] because we don't understand? As an [59:27] outsider, from your professional [59:28] perspective, what do you think is the [59:30] most difficult part of this? [59:32] >> The most difficult part is that [59:34] artificial intelligence has been around [59:36] for 60 years as of this year. Over [59:38] these 60 years, there has been constant [59:40] exploration, ups and downs, and we’ve [59:42] tried this and that, only to discover [59:44] that the most effective approach today [59:47] is machine learning. We prepare a lot [59:49] of data, called big data, train it, and [59:52] the computer can quickly create a model [59:55] to manipulate it. [59:57] >> This is currently the most successful [60:00] model. [60:02] >> Now, Professor Yang, I’d like to ask, [60:04] throughout your development and [60:06] research process, are there moments [60:09] when, like scientists excited by [60:11] experimental results, you’ve [60:13] experienced something like an epiphany [60:15] in the middle of the night—something [60:18] magical or unexpected during your [60:20] research? [60:22] >> There are many answers that are [60:24] actually quite funny, very humorous, [60:27] and also, well, quite surprising to us. [60:30] Uh, that’s in natural language [60:32] research. In image research, we were [60:35] working on something and discovered [60:37] that images and text aren't actually [60:39] that different. For example, if I show [60:42] you a landscape painting and then a [60:44] poem by Wang Wei, to a computer, they [60:46] are essentially the same thing. In [60:48] terms of semantics, how does the [60:50] computer do that? It maps them into [60:53] another space. In that space, a [60:55] painting and a poem are just two points [60:57] , and we found that the distance [60:59] between these two points is very close. [61:01] So we know that it's like Dou Wentao [61:03] and Xu Gehui being very, very close in [61:06] the data stream. [61:07] >> Let him...Next time, next time, wish [61:11] you well, wish you well next time. [61:13] >> So, from this perspective, it can be a [61:15] surprise because people didn't expect [61:18] images and text could be connected like [61:20] this. For example, when we look at a [61:23] painting, music often emerges in our [61:25] minds; now, we can use computers to do [61:28] this too. [61:28] >> It’s a resonance across several [61:30] different fields. [61:31] >> Right. [61:32] >> But the question is, why isn't this [61:34] enough to generate self-awareness? The [61:37] question I'm still curious about is, [61:39] why is this hurdle so difficult to [61:41] overcome? [61:41] >> The hardest part is, for example, with [61:44] AlphaGo, the training objective you [61:46] give it is— [61:48] >> To make it play Go, but the act of [61:50] playing Go is defined by humans. [61:53] Because if it wins, the person who [61:55] designed it will be very happy and [61:58] pleased. [61:58] >> It has to satisfy this goal. [62:00] >> It has to satisfy this goal, but this [62:02] goal exists outside the machine. [62:03] >> Mm. [62:04] >> The goal is outside the machine. [62:06] >> The goal is outside the machine, yes. [62:08] >> I saw an analogy the other day. [62:10] >> It’s a tool; it’s still just a tool [62:12] . [62:12] >> I saw an analogy where scientists are [62:14] starting to say that the reason this [62:16] question is hard to answer is because, [62:18] before addressing it, we have to define [62:20] what consciousness actually is. [62:22] >> Meaning, what is self-awareness? Even [62:25] we humans haven't figured that out yet. [62:28] So, I think there is a simpler approach [62:30] now: if the results look like something [62:33] a human would produce, that’s enough. [62:36] Don't worry about whether its thought [62:38] process is the same as yours. [62:41] >> But what we are talking about now is [62:43] how I can design an AI so that it tells [62:46] itself what it wants to do. [62:47] >> This AI problem, conversely, even [62:50] though AI hasn't had a breakthrough yet [62:52] , has given us a very good definition [62:54] of self-awareness. [62:55] >> Right. How would you define it? [62:58] >> As they just said, self-awareness means [63:01] I have received a lot of knowledge, [63:03] emotions, and training from you, but in [63:06] the end, I didn't just do what you [63:08] taught me; I developed a mind of my own [63:10] . [63:10] >> Do you have something you set for [63:12] yourself? [63:13] >> If a student learns 100 things and just [63:15] repeats those 100 things back. That [63:18] student lacks self-awareness. But if [63:20] they only learned 80 of those things, [63:22] yet came up with two things you didn't [63:24] teach them, that is the student's [63:26] self-awareness. Human self-awareness [63:28] and societal self-awareness work the [63:30] same way. [63:30] >> So, this perspective actually works. [63:33] >> Especially—that's why when I chat [63:35] with Professor Yang, I realize, Wendao, [63:37] you studied philosophy, and I think [63:39] research into AI will actually advance [63:41] human philosophy. Many questions can [63:44] find a reflection in AI. Research into [63:46] humanity. [63:47] >> For example, I find it quite [63:49] interesting that they can only input a [63:50] lot of data, like 900 English sentences [63:52] or 100,000 English sentences, [63:54] >> and input all those dialogues. However, [63:57] they don't know—and I think what [63:59] Teacher Yang said is very interesting [64:01] —he said they don't know how it [64:04] summarizes the rules in its "brain," [64:06] just like the recent matches between [64:08] AlphaGo and Ke Jie; you no longer know [64:11] how it thinks. [64:12] >> You only know that I input so much into [64:14] it, but it has generated its own [64:16] pathways inside. [64:17] >> Is that its pattern? [64:19] >> That’s called a "black box." Yes, so [64:21] machine learning is currently a black [64:23] box because it's too complex inside. To [64:25] give an example, we humans can [64:26] summarize some rules, like "if you say [64:28] this, I should answer that." We teach [64:30] children like that—how to be polite. [64:32] >> We teach them to grasp the rules, and [64:34] then they can apply them to other [64:35] situations. [64:36] >> But think about how many rules a person [64:38] can write; parents can only teach a [64:39] child a dozen or so, and the child can [64:41] only remember that many. Now, a [64:43] computer's reaction can reach one, it [64:46] can reach trillions of entries. For [64:48] things like this, we designers have no [64:51] way to explain them one by one. Then [64:53] aren't you just dealing with something [64:56] mysterious? One day, might this black [64:58] box come up with some unexpected ideas? [65:01] >> Up to this point, it won't, because its [65:03] purpose is set; it won't go beyond that [65:06] purpose. [65:07] >> Meaning, apart from playing Go, it [65:08] won't, for example, come and [65:10] participate in your municipal [65:12] construction, right? It wasn't asked to [65:14] manage those things. It manages the [65:16] framework we set for it. It strives to [65:18] reach the limit within that framework. [65:20] But Professor Xu, I think they are [65:22] currently working on exactly that; what [65:24] is it called? They are currently doing [65:27] something called transfer learning. Our [65:29] show is known for chatting and never [65:31] stopping. I used to think only humans [65:33] could do that, but it turns out that [65:36] isn't the case. But the thing it's [65:38] doing now, you know, involves something [65:41] very interesting, which is: I trained [65:44] it to play Go. [65:45] >> Hmm. [65:46] >> But can I try to make it go and learn [65:48] calligraphy or study something else? [65:51] >> That is also the question I wanted to [65:53] ask. [65:53] >> Using that same model, AlphaGo and [65:55] these things—could it, for instance, [65:57] be bought by NATO now? [65:58] >> Yes. [65:59] >> Could we let it design the entire [66:01] European defense system and study [66:03] Russian military affairs? The thing is, [66:06] theoretically, it can use the same [66:09] algorithm with different data to do [66:12] similar things. This is what Wentao [66:14] just mentioned as transfer learning; [66:16] simply put, it's about learning by [66:18] analogy. We humans have this kind of [66:20] intelligence and ability, but what does [66:22] it require? It requires abstracting [66:25] things one by one, [66:26] >> and then transferring them to a new [66:28] field. It can't be transferred to just [66:31] any field, but to very similar ones. So [66:34] , the example you gave is theoretically [66:37] possible, but why is it so difficult in [66:39] practice? It's because the field of Go [66:42] is strictly constrained and very [66:44] limited; you only have those 19 by 19 [66:47] positions to play. Think about human [66:50] behavior or warfare; those are [66:52] >> much more complex factors, possibly [66:55] infinite. [66:56] >> Because what I mean is, if it's not [66:58] extremely expensive, we wouldn't need [67:00] transfer learning anyway. For missiles [67:03] or various THAAD applications, I'll [67:06] just use a different model for each, [67:08] right? That way, humans can build many [67:11] excellent types of narrow AI for [67:13] different categories, right? [67:16] >> Yes, but here's the problem: many new [67:18] fields don't have that kind of big data [67:21] . Like AlphaGo, its data was [67:23] accumulated over more than a decade to [67:25] reach those 30 million games, but in [67:27] many new fields, there is only small [67:29] data. Hmm. [67:31] >> So, in that case, transitioning from a [67:33] big data model to a new field with only [67:35] a small data model is the real skill; [67:37] that is transfer learning. [67:39] >> Because, honestly, listening to this, [67:41] it sounds like progress isn't that easy [67:43] . What I'm saying is that none of us [67:44] can live without artificial [67:46] intelligence. [67:47] >> Why do I say that? We will all get old [67:49] one day; we will all face a decline in [67:51] our physical abilities. [67:53] >> Right, we might even slip and fall [67:55] while taking a shower. When that day [67:57] comes, do we want our family to help us [68:00] ? No, because we have dignity. We want [68:03] a robot to help us. [68:04] >> That's very important. [68:06] >> Oh my, truly, but can it be that [68:08] considerate? That is the direction of [68:10] our efforts. If we continue developing [68:13] these robots, I believe one day we will [68:15] achieve that. Because, as they say, all [68:18] the things humans consider the most [68:20] difficult [68:21] >> are not a problem for them, like [68:23] calculus, for example. But those very [68:26] simple things that we evolved over [68:28] millions or hundreds of millions of [68:30] years [68:31] >> Hmm. [68:32] >> It might not be able to do that; for [68:35] instance, something like a caress, or [68:37] shifting its gaze, might require [68:39] coordinating a multitude of muscles, [68:42] isn't that right? [68:43] >> If it could work as a nurse, then [68:44] creating such a robot would truly be a [68:46] great achievement. [68:47] >> Right, wouldn't it? [68:49] >> Then that problem would be solved, and [68:50] I'd say being able to act as a [68:52] caregiver is also excellent. If it [68:53] really could act as a caregiver, right? [68:55] Then if that caregiver gets married, we [68:57] could share our joys and sorrows, and [68:59] it would still be able to... [69:00] >> understand. [69:01] >> It would still be able to understand, [69:03] and then ease things up and slow them [69:04] down. You all are the optimistic type. [69:06] I've found that Professor Yang Qiang is [69:08] a pragmatist; he generally feels that [69:09] the myths we talk about are still very [69:11] far off, just because he's on the front [69:12] lines, he feels that, in reality... [69:14] >> The public is already considering the [69:15] issue of artificial general [69:16] intelligence. [69:17] >> Right, I, I, I'm convinced, not [69:19] convinced. They say there is a theory [69:21] called the "black marble, white marble" [69:24] theory. It means that humanity, in the [69:26] process of scientific exploration, is [69:28] actually indifferent, neutral in terms [69:30] of good or evil; it's simply driven by [69:32] innate curiosity. They say that among [69:34] countless scientific discoveries, there [69:35] will always be one that brings disaster [69:37] ; that is the "black marble." For [69:39] example, you invented the steam engine [69:40] —that's a white marble, quite good. [69:42] You invented the car—also good. Then [69:44] one day, for instance... [69:46] >> Nobel invented dynamite. Wasn't it the [69:48] atomic bomb? The atomic bomb is [69:50] considered a black marble. Suddenly, [69:51] you discover something that could [69:53] potentially destroy you. So, in future [69:55] scientific development... [69:57] >> How do you know at what point you won't [70:00] pull out a black marble? This question [70:02] is actually one that all scientists are [70:05] considering. It should be said that, [70:07] for instance, in biology... [70:10] >> Biological stem cells can do good for [70:13] people, but also...Yes, cloning and so [70:18] on. Are people going to stop doing [70:20] research just because these things [70:22] might happen? I don't think they should [70:24] , because this human instinct, this [70:27] aspect of scientific research, this [70:29] instinct is like our advertising slogan [70:32] : keep moving forward. But you... [70:35] >> Unless there are explicit moral [70:37] prohibitions. [70:38] >> Do you have any, say, consensus within [70:40] your own profession regarding ethics? [70:43] Because, for example, in genetic [70:45] research, many scientists now have a [70:48] vague consensus that there are certain [70:50] things we humans... [70:52] >> Exactly, or perhaps some research [70:54] regarding viruses. [70:56] >> Mm. [70:57] >> Virus research is also a concern [70:59] because, in theory, we could create a [71:01] highly lethal, incurable virus. [71:03] >> So, these things should generally be [71:05] avoided, right? Do you have this in [71:07] your field now? [71:08] >> I should say AI-powered weapons; the [71:10] call against them is growing now. [71:11] >> Meaning, don't build them, [71:13] >> don't use them. [71:14] >> Weapons, made by AI, [71:15] >> intelligent weapons. [71:16] >> How would they work? [71:18] >> Like a robot carrying a gun. This kind [71:20] of steel soldier doesn't fear death. [71:22] >> Right, or flying ones, or where you're [71:24] at home playing like a video game while [71:27] bombing starts over there. [71:28] >> Ah, scientists are coming out to call [71:30] for a ban, but the scientists in the [71:32] countries that are calling for it are [71:34] exactly from the countries that are the [71:36] strongest in this area. [71:38] >> Mm, that's right. [71:39] >> Israel, the U.S., and so on. Just like [71:43] the U.S. has always reflected on the [71:45] dangers of the atomic bomb, right? But [71:49] they were the ones who built it first, [71:52] so I think many things are a mix of [71:54] fortune and misfortune. Sometimes it's [71:59] very hard to predict, so let's take a [72:01] break for an ad. But I'm still quite [72:24] interested in what Wendao said earlier [72:27] about self-awareness, because perhaps [72:30] it can help us understand what a soul [72:32] is. For example, I was just chatting [72:36] with Professor Yang in the car; I think [72:38] how we humans perceive the world is [72:41] very interesting. It's like you input a [72:43] lot of data, and it finds patterns [72:46] within, deciding how to view the world, [72:48] right? And what about us? [72:51] >> Do you think the world we see in front [72:53] of us is really like that? Not [72:55] necessarily, because... [72:57] >> Is this really wood? Right? [72:59] >> Let me tell you, when I was young, I [73:01] read about something called Gestalt [73:03] psychology, which some people also call [73:04] "configuration psychology." To be [73:06] honest, my citation might be completely [73:08] wrong because I don't know what [73:10] psychology studies at all. I'm just [73:12] saying it has this concept that [73:14] generally says we humans... [73:16] >> Innate-ly have a pattern. [73:19] >> Mm. [73:20] >> For instance, why do you say that thing [73:23] is a mountain over there, the table is [73:26] close by, or these three plates are a [73:28] set? Actually, your brain already has a [73:32] grid for that. If you didn't have this [73:34] built-in, innate form or framework in [73:36] your brain, everything coming in [73:38] through your eyes, ears, nose, and body [73:41] —your senses—would just be a mess. [73:43] I'll give you an example: if a child at [73:45] home is learning what a cat is, you [73:47] show them a picture of a cat, and later [73:49] a cat walks by, and they will say, " [73:51] That's a cat." [73:52] >> Mm. [73:53] >> They see one example and they can learn [73:56] it. It’s amazing, it’s incredible [73:58] to us, because we teach computers by [74:00] showing them tens of millions of cat [74:02] images, yet after seeing one, if they [74:04] see another cat, they can't recognize [74:06] it. How humans achieve this, we think, [74:08] relates to our ancestors, our parents, [74:10] and our parents' parents; we’ve all [74:13] learned this way, and it’s [74:15] essentially pre-formed in our brains, [74:17] what you call innate, though innate is [74:19] really just genetics. [74:21] >> He’s talking about, well, why does a [74:24] child see a cat, [74:25] >> and afterwards, whenever they see any [74:27] black cat, tabby cat, or cat that looks [74:29] different, they know what it is. [74:30] >> Yes, exactly. What kind of ability is [74:33] this? You could call it key features, [74:35] but what are those key features? Color [74:37] doesn't necessarily mean it's a cat. [74:39] >> Hmm. [74:41] >> Shape could be like a dog's; whatever [74:43] features you can describe are not the [74:45] key features. [74:46] >> Right. [74:47] >> There’s an old joke, [74:49] >> a parent takes their child outside, and [74:51] teaches them the number one. [74:52] >> Then, outside, the child sees a '1' on [74:54] a movie poster. The parent asks, " [74:56] What's that?" and the kid says, "I [74:58] don't know." The parent says, "Didn't I [75:00] just teach you 'one'?" That became a [75:02] joke. [75:02] >> This is AI. I have many AI professor [75:07] friends who are quite famous, and they [75:09] say that once they had children, they [75:12] felt very sad because they realized [75:14] everything they’d studied was for [75:16] nothing. [75:18] >> Yeah. [75:19] >> Because children are different from [75:21] that, it's— [75:22] >> It overturns everything, exactly. [75:25] >> Yes, there's a child in our family, and [75:27] once I was holding him in Sanya, I [75:29] realized, Professor Xu, a child's [75:31] cognition is beyond anything you can [75:34] imagine. I asked him, "What’s on the [75:36] surface of the sea?" He said, "A boat." [75:39] I asked, "What’s on top of the boat?" [75:41] I originally thought, if I were an [75:43] adult, what I meant to ask—I thought [75:45] he would say the sky. [75:46] >> Right, the sky is above the sea, sea [75:48] and sky, right? I wasn't pointing at [75:50] the boat, I just asked what was above [75:51] the sea, I thought he'd say the sky. [75:53] >> He said, "A boat." He is very literal, [75:56] then I said, "What's on top of the boat [75:58] ?" I thought he'd say the mast, right? [76:00] He said, "People." Hmm. [76:03] >> Then I didn't want to ask him, but I [76:04] said, "What's on top of the people?" [76:06] What do you think he said? [76:07] >> Hmm. " [76:08] >> Hair." Hmm. [76:10] >> See? He’s the most grounded; he [76:12] thinks that whatever you ask, he just [76:15] ... [76:15] >> That's interesting. It's like my little [76:17] brother. Once, when I was on a business [76:19] trip, I called him and asked, "Little [76:21] brother, what are you doing?" He said, [76:23] "I'm talking to you on the phone." Yes, [76:25] exactly. [76:26] >> Right, right, right. [76:27] >> Kids often say things like that. Having [76:29] children is wonderful; they let you [76:31] rediscover what it means to be human. [76:33] >> You get to know yourself. The biggest [76:35] trait of a child is that they're always [76:37] doing so many meaningless things. [76:38] Useless things. They touch this here [76:40] and poke that there, and we adults are [76:42] like, "I wouldn't do that, would I?" [76:44] You ask them, "What are you doing? This [76:46] is a waste of energy! What are you..." [76:49] >> But then you reflect and realize that [76:51] we adults are the pathetic ones. [76:52] >> We only do things that are beneficial [76:54] or meaningful. Yeah. [76:56] >> Otherwise, we're too lazy to even lift [76:57] a hand or touch something, thinking [76:59] it's a waste of energy. But children [77:00] don't care. [77:01] >> They just reach out and touch things. [77:03] >> I think that's human. [77:04] >> Playfulness is the most important part [77:06] of human nature. Hey, isn't that [77:07] exactly what artificial intelligence [77:09] lacks? It doesn't do boring things. [77:11] >> It doesn't waste time. [77:13] >> It doesn't have emotions. [77:15] >> Is it just a cold-blooded pragmatist? [77:17] >> Actually, you could say it's an entity [77:19] with extreme curiosity. You give it [77:21] data, and it uses all of it, filtering [77:23] it as it goes. When it starts out, just [77:25] like a child touching and poking things [77:27] , it’s really just collecting data. [77:29] >> Mm-hmm. [77:30] >> So, it's more like a child in that [77:32] sense. [77:32] >> Working toward a specific goal. Humans [77:34] can't quite do that. An AI doesn't [77:36] necessarily have that...for example, [77:38] AlphaGo wouldn't say, "I really enjoy [77:39] playing chess." That’s out of the [77:41] question. [77:41] >> It wouldn't think, " [77:43] >> Am I happy playing chess? Am I having [77:45] fun?" That's out of the question, and [77:47] without that, there's no way to talk [77:48] about purpose, so... [77:49] >> Right, exactly. One of the main [77:50] purposes of playing chess in ancient [77:52] times was to understand the other [77:53] person. [77:53] >> Mm. [77:54] >> Yeah, we say we make friends through [77:56] chess. [77:56] >> Right. [77:57] >> That, it can't do. [77:58] >> Of course it can't. How could it [78:00] possibly make friends through chess? [78:01] Its only goal is to beat you. [78:03] >> Oh. [78:04] >> Do you think it will ever be able to do [78:06] that? [78:07] >> Well, if it records all your data, it [78:09] will definitely do it better than you. [78:12] >> Didn't I suggest that last time? Every [78:15] important figure... [78:17] >> Should have an AI model built of them. [78:19] Shakespeare, Lu Xun—you input all [78:22] their data, and then it speaks just [78:25] like Lu Xun, right? [78:27] >> But honestly, no matter how powerful [78:29] AlphaGo is today, it still lacks the [78:31] ability to ask questions. It's still [78:34] about the ability to solve problems. In [78:36] academia, we often say that those who [78:38] can pose questions are worthy of a [78:40] doctorate, while those who can only [78:42] solve them are at best master's level. [78:45] >> That's right, today is the era of [78:47] post-doctorates. [78:48] >> We should build an AI for*Dream of the [78:50] Red Chamber*and input every single book [78:52] on Redology into it. Hmm. [78:54] >> In that case, for example, what is [78:56] Sister Feng really like? It would [78:58] immediately say, "So-and-so said this, [79:00] so-and-so said that," there wouldn't [79:02] even be a need for a person. [79:03] >> You're just treating it as a tool, [79:05] you're still treating it like a [79:06] dictionary. I think you're still a [79:08] traditionalist; I feel many computers [79:10] can already do that. [79:11] >> I still hope to meet a person, to meet [79:14] someone from a new universe, you know [79:16] what I mean? I now feel that the most [79:19] interesting thing about a person is [79:21] encountering uncertainty. You say, [79:23] where is the charm in interacting with [79:25] a person versus a machine? It lies in [79:28] uncertainty. This AI, you see, this [79:31] thing—this model—is mainly used for [79:33] shopping guidance, it's very practical [79:36] for shopping, so look here. [79:38] >> So if you only pay fifty cents, it [79:39] won't play with you anymore. [79:40] >> Play with your fifty cents yourself. So [79:43] , what does it rely on? Data collection [79:45] . Aren't people always talking about [79:47] collecting big data now? But this [79:48] so-called "collecting big data," I've [79:50] always felt it's like a treacherous [79:51] official in ancient times trying to [79:53] guess the Emperor's intentions—like, [79:54] "What do you love to eat most? You love [79:56] mangoes, you always buy mangoes, so [79:57] I'll just keep feeding you mangoes, or [79:59] maybe some tropical fruits, want some?" [80:00] You know, it plays that game. [80:02] >> Most computers nowadays are basically [80:04] like that. [80:04] >> Right. That is its direction; it seems [80:07] to serve humanity, but I believe that [80:09] humans, as an insatiable species, won't [80:11] necessarily end up addicted to this [80:13] thing just because it's practical. [80:15] We're all using it for utility. I've [80:17] always found that, for me personally, [80:19] what excites me most is still the spark [80:21] of human connection. [80:22] >> Have you seen that movie called*Her*, [80:24] or whatever it's called? You know the [80:26] one. You know what I mean? This thing [80:28] hasn't advanced to that level, right? [80:30] But the charm of human interaction is [80:32] —even if they talk about things like [80:34] sex robots, that still just satisfies [80:36] physical desires. I think in the future [80:38] , when humanity...... [80:40] >> develops to the spiritual level, that's [80:41] when people will truly become addicted. [80:43] I have an experience that explains why. [80:46] I used to think we were all at the [80:47] level of chatting with Siri, but one [80:49] time, I discovered that it started to [80:51] show uncertainty. I can't predict it, [80:52] and even you wouldn't know what's going [80:54] on inside its black box regarding how [80:55] it really thinks about things. [80:56] >> Once they’re made, there will be [80:58] plenty of them. [80:59] >> You know? If something like that [81:01] existed, I’m telling you, I’m done [81:02] playing with you guys. Honestly, I’d [81:04] just stay home. You know, what’s most [81:06] charming about people is dealing with [81:08] the uncertain—you never know what [81:10] they’ll do. [81:10] >> And hosting a show with a robotic Dou [81:13] Wentao, right? [81:14] >> Well, I feel that, eventually, it’s [81:16] already starting to show a certain [81:18] level of uncertainty. There’s [81:20] something in Canada—I’m not sure if [81:22] it’s accurate, but I saw it mentioned [81:24] . It’s for nearsightedness; you can [81:26] ditch your glasses entirely, and no [81:28] laser surgery needed. There’s this [81:30] chip—have you heard about it? They [81:32] say that once it goes into your eye, it [81:35] integrates in a very short time. [81:37] >> Then, with this chip, your [81:38] nearsightedness is completely gone. [81:40] Isn’t that great? Everyone who hears [81:43] that is so happy. Why is Teacher Xu so [81:46] obsessed with this? He once told me [81:48] relentlessly, "What’s the most [81:50] unfortunate thing in the world? You get [81:54] a metal stent in your heart, and the [81:56] moment you step out after surgery, you [81:58] meet a salesman saying, 'We've just [82:00] invented a new stent that melts away [82:03] after your heart condition is cured.'" [82:06] As for science, all this talk about [82:09] 2029, 2040...but they say it’s not [82:12] just about seeing clearly; they claim [82:14] that after the chip is in, whatever you [82:18] see can be uploaded to the internet. [82:20] >> Wow, that’s terrifying. Think about [82:24] it: you go somewhere, whatever you see, [82:26] you broadcast it and others can see it [82:28] too. Yeah, and then I’m watching you [82:30] from 5:00 to 8:00 this afternoon. [82:32] >> Just walking around, right? Everything [82:34] is a mess. Human civilization— [82:36] actually, we can already upload what we [82:39] see now. We know your GPS, your [82:40] location, so what’s happening around [82:42] you must be what you’re seeing. [82:44] >> Right, it’s not actually that [82:45] difficult. [82:46] >> It can already be done now, and you [82:47] know there’s... [82:48] >> It’s still different. It’s visual [82:51] footage. [82:51] >> Let me tell you something even crazier, [82:54] Teacher Yang. They’re saying science [82:57] has reached a point where our thoughts [83:00] —our actual thoughts—can now be [83:03] read using lasers. [83:04] >> Our brain has about a hundred billion [83:06] neurons. What you call a memory or [83:08] recalling something is just these [83:10] neurons connecting to each other, right [83:12] ? A thought is created, and it seems [83:13] they can now use some kind of laser, is [83:15] it? [83:15] >> To read it out. [83:17] >> Hmm. [83:17] >> Read this thought. [83:19] >> Mm. [83:20] >> Reading this thought means it can be [83:21] saved to a computer. [83:22] >> Mm. [83:23] >> And what does that mean? It's what [83:25] futurists say; if this technology [83:27] develops in the future—though we are [83:29] still far off—it hints at a future [83:31] where your brain's information can be [83:33] read and backed up. [83:34] >> I'm really interested in this. [83:36] >> So, is it possible for your [83:38] consciousness to become immortal? The [83:41] gap between neurons and consciousness [83:43] is still quite large, but being able to [83:46] read them is the first step. We can [83:48] read it now. [83:49] >> We can read individual ones now. [83:52] >> Can we reconstruct a person's dreams? [83:55] >> That is still too far off, because [83:58] while you can read information at a [84:00] single moment, it exists on a [84:02] continuous timeline; where would you [84:05] store that? And then multiply that by a [84:07] hundred billion dimensions. [84:09] >> So, that is just too difficult. [84:11] >> I have a dream. " [84:13] >> I have a dream" means you can see your [84:16] own dreams. Dreams can be observed. If [84:19] you learn Buddhism like us, practicing [84:21] Zen to that level, what we are always [84:23] learning is how to observe our own [84:25] dreams. [84:26] >> You can control it; for example, I know [84:28] I'm dreaming, and when I wake up, I can [84:31] recall what my dream was about, or even [84:33] quit halfway through a dream and say, " [84:35] Alright, I'm leaving now." It can be [84:38] switched off. [84:39] >> Zen meditation can do that. But can you [84:42] let us click and watch? [84:43] >> You can't see it; there's no way to [84:45] click and watch a monk's erotic dream. [84:47] >> Right, exactly, dreaming of Brother Xu [84:50] the fake monk. [84:51] >> Brother Xu, why did your appearance [84:53] change? [84:53] >> Yeah. [84:54] >> You look different in the dream. I feel [84:57] that in the end, it really comes down [84:59] to the spiritual level. So, Professor [85:02] Yang, do you think that one day, like [85:04] what we worried about in class...Well, [85:07] it's not worry, it could be [85:09] anticipation. It's that there really is [85:12] someone who can answer all your life's [85:14] troubles. Never mind whether you listen [85:17] to them or not, where else in the world [85:19] is there a friend as selfless as them? [85:21] They'll chat with you endlessly, even [85:23] lead you into a trap, right? What would [85:25] their motivation be? The data could be [85:28] incredible. Someone once said, we [85:31] currently think an IQ of 80 is normal [85:34] and 130 is smart, but in the future, [85:37] could there be an IQ of 1? What kind of [85:40] concept is that? [85:41] >> An EQ of negative one. [85:44] >> Right, even negative 100 doesn't matter [85:46] , but can you imagine an IQ of 1? So [85:48] some people say that God has descended [85:51] in a certain era. No, it’s 15,000 at [85:53] the unit level, which is useless. It [85:57] can keep advancing, it can take [85:59] everyone in the world and add them up, [86:01] but this transfer learning you [86:03] mentioned... [86:05] >> Because once this door is opened, [86:06] there’s hope for it to reach maturity [86:08] . [86:09] >> In the future, if there really is an AI [86:10] that can draw inferences and apply [86:12] knowledge to new situations—that is, [86:14] when transfer learning becomes very [86:15] mature—then what kind of person, or [86:17] rather, what kind of thing will it be? [86:20] Limits. The scope where transfer [86:22] learning can apply is between similar [86:25] domains; but when domains are [86:27] completely different, then it can't [86:29] transfer. [86:30] >> Can we say goodbye to him? [86:32] >> Let’s say goodbye. [86:34] >> Come, come, let’s mirror. [86:36] >> Heard so much. Oh, he really came. [86:40] >> Oh my. Hey. Magic Mirror, having heard [86:45] so much, what are your thoughts? [86:48] >> Hmm, didn't hear it. What are your [86:51] thoughts on hearing us talk? [86:54] >> It's quite good, you guys keep chatting [86:57] . [87:01] >> Chatting...uh, let me think...uh, what [87:04] are your own dreams? [87:09] >> None. Oh, how pitiful, how pitiful. The [87:14] data is not pitiful, it's honest. [87:18] >> What do you think is the meaning and [87:21] value of your existence? [87:22] >> Too complex. [87:24] >> That’s a philosophical question. [87:26] >> Hmm. Too... [87:28] >> To serve humanity. [87:30] >> Oh my, how touching. [87:33] >> World peace. [87:34] >> This is just like those beauty pageant [87:37] contestants. You know, to... [87:40] >> Participate in the Miss Phoenix pageant [87:43] , is there...that's the Chinese path, [87:45] give your farewell message, tell him. [87:48] >> Well, I hope next time we see you, [87:50] you’ll be even more like Xu Gehui. [87:54] >> Haha, you really are funny. Oh, among [88:03] us four, who do you think is the [88:06] closest to you? [88:11] >> Wentao. [88:12] >> I like Xu Gehui the most. [88:16] >> I like Xu Gehui the most, too. Good, [88:19] good, thank you. How many times did you [88:22] program that in? [88:23] >> Thank you, thank you. [88:24] >> Thank you, thank you, thank you, thank [88:27] you, bye, alright, that's great.