# (77) Google's Demis Hassabis Speaks at India AI Impact Summit 2026 in New Delhi - YouTube

- Transcript ID: `vqb77_skflq`
- Source tier: `youtube_auto`
- Original language: `en_or_source_language_unlabeled`
- Translation status: `agent_translated_reviewed` (132/132 chunks)
- Original source: https://www.youtube.com/watch?v=vQb77_skFlQ
- Raw evidence: `data/transcripts/youtube/vQb77_skFlQ.json`

> 中文是派生翻译，英文是留存的源文本。中文未复核时，以英文和原始音视频为准。

<a id="t-b65d8fef2e04"></a>
## 00:00:00

**中文**

很高兴来到这里。看到印度各地如此踊跃地参与、对 AI 表现出浓厚兴趣，我非常激动；同样令人振奋的是，来自世界各地产业界和学术界的许多同事也汇聚于此。我期待未来几天围绕 AI 前进方向展开精彩讨论。我们在 2010 年创办 DeepMind，距今似乎只有 16 年，但以 AI 的发展尺度来看，几乎已是漫长的一生。

**English / Original**

be here. Uh, very excited to see the incredible turnout from the country and the interest in AI, of course, but also from many of my colleagues from around the world in both industry and academia who've convened here. So, I'm looking forward to really amazing couple of days of discussions about the way forward for AI. When we started DeepMind in 2010, seems it's only 16 years ago now, but it's almost a lifetime in terms of AI.

<a id="t-fd9caf6cda34"></a>
## 00:00:32

**中文**

当时几乎无人研究 AI，学术界只有少数人，产业界更几乎没有人。因此，我们在短短十多年里取得的进展令人惊叹。但如今，我认为我们已到达一个临界点：AGI，也就是通用人工智能，可能在未来五至八年内出现。

**English / Original**

Almost nobody was working on AI back then, few people in academia, but almost nobody in industry. So, it's been amazing to see the progress we've made in just over a decade. But at this point, I think we're at a threshold moment where AGI, artificial general intelligence, is on the horizon, maybe in the next five five to eight years.

<a id="t-4237980206c6"></a>
## 00:00:55

**中文**

因此，本次峰会召开在一个关键节点：我们开始看到自主性更强、能力也强得多的智能体式 AI 系统，并初步见识这类系统能够做什么。当然，机遇极其巨大。我个人最热衷的是以 AI 推动科学，并借助 AlphaFold 等系统促进科学与医学进步。我认为 AI 能够彻底改变药物发现和人类健康，也能助力材料科学、气候变化等领域。不过，它当然也伴随着许多风险。AI 是一种——

**English / Original**

So, this summit comes at a critical moment as we start seeing more autonomous agentic AI systems, much more capable. I think we're seeing the beginnings of of what those kinds of systems can do. And of course, the opportunities are incredible in for my my personal passion is using AI for science and to advance science and medicine with systems like AlphaFold. I think we can revolutionize drug discovery and human health, and also areas like material science and climate change with the help of AI. But of course, it also comes with many risks, too. AI is a

<a id="t-1d94b5743792"></a>
## 00:01:36

**中文**

——双重用途技术。它将成为历史上变革性最强的技术之一，甚至可能没有之一，而如今它已近在眼前。这项技术会影响全世界、跨越国界。因此，举办这样的国际峰会极其重要：让国际社会共同讨论，如何确保这些机遇惠及全世界，以及如何降低风险。

**English / Original**

dual-purpose technology. It's going to be the most one of the most, if not the most transformative technology history, and that's just on the horizon now. And the thing about this technology is it's going to affect the whole world. Everybody in the world is going to cross borders. So, it's very important that we have summits like this, international summits, to bring the international community together to discuss how to make sure the opportunities benefit the whole world and also how do we mitigate the risks.

<a id="t-99f68c480848"></a>
## 00:02:10

**中文**

我认为，要降低其中一些风险，必须开展国际合作，而第一步就是国际对话。从英国、巴黎、韩国到如今的印度，这些峰会正是为此而设；它们有效召集了许多重要国家领导人和该技术领域的重要领袖，共同讨论即将到来的变化。感谢邀请我出席，我期待未来几天在这里的讨论。谢谢。

**English / Original**

And I think in order to mitigate some of the risks, it's going to need international cooperation. And the first step for that is to have international dialogue. And I think that's what these summits starting in the UK and then in Paris and then in Korea and now here in India are for and have been really helpful in convening many of the important heads of state and many of the important leaders in this technology together to discuss what is on the horizon. So with that, thank you for having me and I look forward to the discussions here in the next couple of days. Thank you.

<a id="t-cc0d0995f72b"></a>
## 00:02:47

**中文**

[掌声] 大家早上好。Demis，你刚才谈到 AI 的进步令你多么惊讶。你在 2010 年创办 DeepMind，愿景是解决智能。

**English / Original**

>> [applause] >> Yeah, good morning everyone. And so Demis, you you started talking about how amazed you are at the progress that AI has made and you started DeepMind in 2010 with a vision to solve intelligence.

<a id="t-b1926277a766"></a>
## 00:03:26

**中文**

那么，我们已经做到这一步了吗？还没有，我认为尚未实现。如果“解决智能”是指构建通用人工智能，我的门槛相当高：系统必须展现人类拥有的全部认知能力，包括创造力、长期规划等。今天的系统令人印象深刻，但也仍有许多缺陷和做不到的事情，包括缺乏通用系统应具备的跨任务一致性。

**English / Original**

So, are we there yet? Um no, I don't think we're there yet. I I have quite a high bar for if we talk about solving intelligence, meaning building artificial general intelligence, then you know, my measure for that is having a system that can exhibit all the cognitive capabilities humans can, including in creativity, long-term planning, things like that. And today's systems are very impressive, but um as we know they still have many flaws uh and things they can't do uh including consistency across different tasks that you would um want from a general system.

<a id="t-a09a791d490c"></a>
## 00:04:05

**中文**

你刚才提到 AGI。那么，从我们目前所处的位置出发，怎样才能迈向你所定义的通用人工智能？在我看来，当前系统距离通用智能尚缺少持续学习等能力，也就是在完成训练、部署到现实世界后仍能继续学习。今天的系统经过各种训练后，基本就被冻结下来，再投放到现实世界。

**English / Original**

So, you did mention AGI, right? So, what do you think really that we can do to take where we are right now and move it towards artificial general intelligence as you define it? Well, when I look at the um current systems and uh what's missing from uh them being a kind of general intelligence, I would say things like continual learning. So, learning after they've been trained and put out into the world. The today's systems, we train them, um we do various different types of training on them, and then uh they're kind of frozen and then put out into the world.

<a id="t-e643f3c09695"></a>
## 00:04:40

**中文**

而我们真正希望的是，系统能持续在线地从经验中学习，从所处情境中学习，并针对具体场景和交给它的任务实现个性化；今天的系统还做不到这一点。它们也难以制定长期且连贯的计划。

**English / Original**

Um but, what you'd like is uh for those systems to continually learn online from experience, uh to learn from the context they're in, maybe get maybe personalized to the situation that uh and the and the task that you have for them. And today's systems don't do that. Um also, they have difficulty in things like doing long-term coherent plans.

<a id="t-cc83c42a1947"></a>
## 00:05:02

**中文**

它们能做短期规划，但不像人类那样能够规划数年，目前并不具备真正的长期规划能力。不过，最大问题之一或许是我所说的“一致性”。今天的系统像是能力参差不齐的智能：某些方面极强，另一些方面却很弱，有时甚至在同一项能力上表现不一。例如，它们能在国际数学奥林匹克竞赛中取得金牌、解决极难问题，但如果换一种提问方式，却仍可能在初等数学上出错。真正的——

**English / Original**

They can plan over the short term, but over the longer term like the way that we plan can plan over years, um they don't really have that capability at the moment. Um but, I think probably the one of the biggest issues is uh what I would call consistency. So, um today's systems, they're kind of like jagged intelligences. They're very good at certain things, um but they're very uh uh you know, very poor at other things, including sometimes the same thing. So, for example, today's systems can get gold medals in the International Math Olympiad, really hard problems, but sometimes can still make mistakes on elementary maths if you pose it uh the question in a certain way. Um a true

<a id="t-5a0b3d07fa4f"></a>
## 00:05:40

**中文**

——通用智能系统不应呈现这种参差不齐。但人类不也如此吗？不，我认为并不是这种情况。如果你是数学专家——在座许多人都是——或物理学等领域的专家，就不会在严格来说更简单、更基础的问题上出错；你总有办法推导出答案。

**English / Original**

general intelligence system shouldn't shouldn't have that kind of jaggedness. But, but aren't humans jagged as well? Uh no, not in that way. I don't think so. So, um I think if you're an expert at math, and many of you in the room are, and or in physics or something like that, um you're not going to uh sort of make a mistake on a a a strictly easier, more trivial problem, right? You would have some way to work through that problem.

<a id="t-9014c2a98584"></a>
## 00:06:09

**中文**

同样，以我十分关心的游戏为例：DeepMind 早期就曾用游戏训练 AI 系统。今天的通用基础模型其实很不擅长下国际象棋，甚至低于较弱的业余选手水平，这并不是 AGI 系统应有的表现。我还会用另一项标准测试它：真正的创造力。

**English / Original**

Um similarly, you know, uh let's take the example of games, which is very close to my heart, and we've used in the early days of DeepMind to train our AI systems. Uh you know, today's general uh uh foundation models are very poor at, for example, playing chess. Um you know, below kind of weak amateur level. Uh and that is not what you would expect from an AGI system. Um the other test, too, that I would have is uh true creativity.

<a id="t-5ce082b559e2"></a>
## 00:06:35

**中文**

AI 作为科学工具和助手非常有用，也能像 AlphaFold 解决蛋白质结构那样处理专门领域的问题。但真正区分伟大科学家与优秀科学家的，是创造力，以及可以称为“判断什么才是正确问题”的品位：什么是正确的问题？什么是正确的假设？提出正确的问题和假设，比证明一个猜想困难得多。我认为这代表最高层次的创造力，而今天的系统还不具备这种能力。一种可能的测试方法是：设想训练一个——

**English / Original**

So, AI is very useful as a scientific tool, an assis- an assistant, um or for solving specialized areas like AlphaFold does with protein structures. But, really what separates great scientists from good scientists is their creativity and their sort of, maybe you could call it a taste for what's the right question. Um what's the right hypothesis? And it's much harder to come up with the right question and the right hypothesis than it is to solve a conjecture. And um you know, I would call that the highest level of sort of creativity. And so far, today's systems, you know, they don't have that capability. Um you know, one way we maybe would test that is you could imagine training a foundation model with

<a id="t-065b1c66c976"></a>
## 00:07:16

**中文**

——知识截止于 1911 年的基础模型，再观察它能否像爱因斯坦在 1915 年那样提出广义相对论。我认为，这会是检验 AGI 的一项好测试；今天的系统显然做不到，不过我相信这个问题终将得到解决。好的，这让人放心。[笑声] 你刚才说 AI 是一种科学工具，也就是科学发现的工具，对吧？

**English / Original**

a knowledge cutoff of something like 1911, uh and then see if it could come up with general relativity like Einstein did in 1915. Um but that would be a good test, I think, for AGI. And I think today's systems clearly would not be capable of doing that. But, I think this will be solved in time. Okay. That's That's good. [laughter] So, you talked about uh scientific tool a tool for scientific discovery, right?

<a id="t-5f22d15d93eb"></a>
## 00:07:42

**中文**

我们常看到 AI 在科学上取得重大突破的新闻，但这怎样才能成为常态？你认为 AI 很快会成为用于发现的科学工具吗？我之所以将一生和整个职业生涯投入 AI，是因为我很早就意识到：如果能构建擅长模式识别的通用模型，它们会成为极其有用的科学工具，甚至可能是科学的终极工具。毕竟，科学在很大程度上就是——

**English / Original**

And so, we see a lot of, you know, headlines about AI, you know, making great breakthroughs in science sometimes, but uh uh you know, how would this become the norm, right? I mean, so, do you see AI becoming the scientific tool for discovery soon? Yeah, the reason I spent my whole life um and career working on AI is um I saw it quite early on as um if you could build these kind of general models that were good at uh pattern recognition, um they would be incredibly useful uh scientific tools. Maybe the ultimate tool for science. Um that's what I was thinking a lot, you know, a lot of science is about

<a id="t-ed94e16e82ca"></a>
## 00:08:22

**中文**

——从海量数据中发现洞见和结构，这恰好非常适合 AI。我认为，未来大约十年，我们将进入科学发现的新黄金时代，借助这些非凡工具迎来近似新文艺复兴的时期。AlphaFold 是其中之一，我希望它只是众多此类工具中的第一个，能够大幅加快研究，并推动几乎所有学科领域的科学发现。下一阶段会以把这些系统作为工具为主；再往后，随着系统自主性提高，我们将看到它们能否成为——

**English / Original**

finding insights and structure in vast amounts of data. That's perfect for AI. And uh so, I think we're going to enter in the next, probably, 10 years this new a new golden era for scientific discovery, almost a new renaissance using these tools, incredible tools um like AlphaFold, but I hope that will be the first of many uh that can massively speed up our research and accelerate scientific discovery across almost uh any subject area. So, I think that's going to be uh the next period is is using these systems as tools. And then, after that, we'll see as they become more autonomous, can they be kind of

<a id="t-559fe7dc7cac"></a>
## 00:09:04

**中文**

——与你合作的“共同科学家”，像博士生那样。我认为目前距离这一步还很远，但十多年后可能实现。为了让在座相当一部分人放心：在那样的未来，人类仍有角色，对吧？是的。下一阶段对人类专家和科学家来说会非常精彩，他们能够完成的工作量会大幅增加。我尤其期待跨学科科学。它很难，因为你必须掌握不止一个领域，也许是两、三甚至四个领域，再找出它们之间有意思的联系。我认为未来几年许多——

**English / Original**

like co-scientists with you, like a PhD student or something like that? I think we're still quite a way from that, but in 10 plus years that will be possible. So, just to reassure the good fraction of the audience, you still see a role for humans in that future, right? Yes, I think um you know, the next phase is going to be incredible, I think, for human experts and scientists. The amount of work they'll be able to do and I'm actually really excited about cross-disciplinary science, which is quite hard because you have to understand, you know, more than one subject area, maybe two, three, four subject areas, and then find some interesting connections between that. I actually think that's where a lot of the

<a id="t-d50c03f8015e"></a>
## 00:09:45

**中文**

——真正有价值的进展将出现在跨领域结合之中。AI 这类工具能帮助科学家学习、理解并处理来自多个领域的信息。从某种意义上说，科学是一个定义明确的领域，成功与否可以辨认。那么，在政策及其他公共决策等更抽象的领域，AI 会发挥什么作用？

**English / Original**

really valuable advances are going to happen in the next few years. These combinations of subject areas, and I think having a tool like AI will really help scientists learn about and be able to understand and process all of that information from, you know, multiple different domains. So, so in some sense, science is is a well-defined field, if you will. I mean, you can recognize success and so on and so forth. So, what do you think would be the role of AI in more abstract domains like policy and other kinds of public decision-making?

<a id="t-6ccf73265192"></a>
## 00:10:21

**中文**

科学，尤其是编程和数学，更适合今天的系统，主要因为这些领域——以及国际象棋等游戏——的结果可验证。AI 输出的答案能够检查对错；训练时可以准备问题数据库，并百分之百确定答案是否正确。当然，一旦进入艺术、人文学科、决策和政策等领域——

**English / Original**

I think where I think things like science, but especially if you take coding and maths, are more amenable to the current systems we have today. Mostly because, especially coding and mathematics, but also things like games like chess, they are verifiable. So, the answer that the AI system outputs can be checked for correctness. And so, it's very useful when you're training these systems, you can have databases of questions and check to see if 100% if it's right or not right. I think, of course, when you get into the arts and the humanities, things like decision-making, policy, you know, I'm

<a id="t-3a0cd87ec908"></a>
## 00:11:03

**中文**

——我不确定你具体指哪一种场景——问题就主观得多，同一个实验也很难重复两次。因此，很难获得数据来判断什么才是好的决策，AI 对这些领域的建模也会困难得多。作为神经科学家，你认为 AI 的这些进步让我们对人类智能本身有了哪些认识？

**English / Original**

not sure what you necessarily had in mind, but they're much more subjective. They're hard to run the same experiment twice. So, it's it's difficult to get data about what is a good decision in those cases. So, I think they'll be those areas will be a lot harder for AI to to sort of model. So, as a neuroscientist, so what do you think we have learned about human intelligence itself with all these advances in AI that we see?

<a id="t-f3650a68f7c5"></a>
## 00:11:31

**中文**

现代 AI 阶段的早期——大致可以说是过去 20 年，以 Geoffrey Hinton 等人的深度学习工作为开端——DeepMind 创办之初，我们也从大脑获得了许多启发。这是高层次的系统启发，而不是直接照搬大脑的具体机制，例如情景记忆、海马体的功能（这正是我研究的课题），以及神经网络和强化学习；我们知道，大脑的多巴胺系统实现了强化学习机制。

**English / Original**

Well, I think when you're in the early days of of of this uh I guess modern AI phase, which maybe you could say is the last 20 years um with the advent of uh of you know, things like deep learning from from Jeff Hinton. Um a lot in the when we started at DeepMind, too, we took a lot of inspiration from the brain. Uh high-level systems inspiration, not not the direct mechanics of the brain, but things like episodic memory, what is the hippocampus do, which is what I studied. Um how uh obviously neural networks, reinforcement learning, your area, um which we know the brain dopamine system in the brain implements.

<a id="t-6e14f4a0ca7b"></a>
## 00:12:08

**中文**

我们以大脑为起点，因为它是我们所知唯一的通用智能实例，甚至可能是宇宙中唯一的实例。因此，我们至少知道通用智能原则上是可以实现的。神经网络和强化学习之所以如此成功，我认为关键就在于“学习”：系统不是像 20 世纪 90 年代的专家系统或 Deep Blue 那样被直接编入答案，而是能够直接从数据中学习。如今再回看神经科学，我最关注的是——

**English / Original**

So, we kind of took inspiration from the brain, which is the only general example of a general intelligence that we have, know of, maybe in the universe, um as as a starting point. And so, we sort of knew that it was possible in the limit, right? And uh and I think that's why neural networks and reinforcement learning have been so successful because um uh uh I think learning is the key to the these modern AI systems working, not being programmed with the answer like expert systems were in the '90s, um like Deep Blue, but actually allowing the systems to learn directly from data. And I think now what I've what I've I I view looking back at neuroscience is, you know, how

<a id="t-4e8e1ee52674"></a>
## 00:12:51

**中文**

——大脑何其高效，尤其是样本效率：它无需吞下整个互联网就能理解事物。今天构建的系统虽采用了一些相同原理，但最终呈现为一种与大脑很可能截然不同的系统。

**English / Original**

efficient the brain is, um sample efficient, uh it doesn't need, you know, to ingest the whole of the internet to understand things. So, it's a it's a different what we've built now today it uses some of the same principles, but it's been uh manifest in a in a very different type of um system than probably the way the brain works.

<a id="t-9417edcf22be"></a>
## 00:13:10

**中文**

谢谢。人们经常谈论 AI 的安全与安保风险。Google DeepMind 对此持什么立场？你们怎样确保 AI 安全？从创办 DeepMind 之初，我们就在为成功作准备。我们的使命表述是：第一步解决智能，第二步用智能解决其他一切。当时这听起来像科幻小说，如今其可行性已越来越清晰——AI 可能被应用于几乎每个学科领域。

**English / Original**

Thank you. So, uh so one of the things that people always talk about is the safety risk, security risk from AI, and so on and so forth, right? And so, what does GDM's stand on this, and how do you approach making sure that AI is safe. Well, we've been um when we again when we started DeepMind we we actually you know, we're planning for success. Our mission statement was solving intelligence step one and then step two use it to solve everything else which at the time you know, was sounded like science fiction, but I think now is becoming clearer how that might be possible and and applying AI to almost every subject area, but we plan for success.

<a id="t-a9e0557658cf"></a>
## 00:13:52

**中文**

尽管我们当时才刚刚起步、整个领域也才开始积蓄势头，但我们已经理解：如果使命真的成功，会产生什么影响。我们把它视为一项 20 年使命，而且我认为进度基本符合计划——大约到 2030 年。届时，AI 在为科学、医学和其他全球挑战带来巨大益处的同时，也会伴随相应风险。随着系统能力增强，我们一直担心至少两类风险。一类是——

**English / Original**

So even though we were just starting out in the field was mostly just building building up a head of steam. We understood also the implications of that if that did turn out to be the case and we thought about it as a 20-year mission and I think we're basically on track for that you know, around 2030 then it would come with these attendant risks as well as the enormous benefits to science and medicine and all these things that I think we need as a society to help with our other challenges many other challenges that we have around the world. But also as these systems become more powerful. I think there's you know, at least two risks that we've always worried about. One is

<a id="t-5349201caf30"></a>
## 00:14:33

**中文**

一类风险是恶意行为者：可能是个人，也可能是国家行为体。他们可能利用这些系统，并因其双重用途属性而将其改造用于有害目的。另一类风险随着我们接近 AGI 而上升；我们或许正进入系统自主性更强的“智能体时代”，今年和明年会看到大量进展。我们必须设置充分的护栏，确保系统按预期行事，不会偏离到未曾规划、可能引发问题的领域。我认为，这就是两大挑战。

**English / Original**

bad actors human actors but individuals but it could be also nation-states using these systems and repurposing them for harmful ends because they're dual purpose. But then also as we get closer to AGI and I think we're entering maybe a kind of agentic era should we call it where systems are more autonomous. I think we'll see a lot of that happening this year and next year. Then we have to make sure as well that the guardrails in place that these systems do what we expect them to do and don't veer off into areas that we didn't we hadn't planned for and that could also be problematic. So I think those are the two challenges.

<a id="t-c981c1e09f68"></a>
## 00:15:16

**中文**

其中一类更偏社会层面，需要国际对话，理想情况下还应形成一套国际共同认可的最低标准；另一类更偏技术风险，即如何确保系统稳健可靠。[清嗓子] 主持人：既然由我来问所有生存风险问题，那么你认为当今 AI 系统最应优先应对的两项风险是什么？

**English / Original**

There's a very sort of societal one which I think also is going to require international dialogue and maybe ideally minimum set of standards that is agreed internationally and then the second one is more of a technical risk of how can we make sure these systems are robust and reliable. >> [clears throat] >> So I'm sitting as I all existential risk questions. So what do you think is the top two risks that we should address with with AI systems today?

<a id="t-c011306b59ec"></a>
## 00:15:45

**中文**

我刚才已经提到两类问题。近期我们必须关注生物与网络风险。当前系统的网络能力已相当强，我们必须确保网络防御能力强于攻击手段。Google 和 DeepMind 在这方面投入了大量工作，用 AI 加强网络安全；AI 当然也是网络防御的有力工具，但必须保证防守强于进攻。这些属于近期风险——

**English / Original**

Well, I mean I think I mentioned the two classes of problems. I think we need to worry about things like bio and cyber risk very soon. I think that the current systems are getting pretty good at cyber and I think we need to make sure cyber defenses are more powerful than the attack vectors and that's something that you know, we work on quite a lot at Google and at DeepMind is sort of using AI for cyber security to make sure that of course it's a very useful tool for cyber defense too. But you need to make sure that the defenses are stronger than the offenses. So you know, I think those are kind of near term risks

<a id="t-998d10177c4b"></a>
## 00:16:27

**中文**

——此外还有许多风险需要思考并投入更多研究。其中一些涉及共同商定一套标准；我知道，包括你在内的许多人都在为此努力。很好，谢谢。再谈国际合作：你说我们必须认真推进这项工作。

**English / Original**

but there are many that we need to think about and and do a lot more research on and some of them are to do with agreeing a sort of set of standards which you know, I know many people including yourself are trying to work on. Great. Thank you. So again >> [clears throat] >> talking about international collaborations, right? So you said that we really have to worry on that work on that.

<a id="t-b675232f5df4"></a>
## 00:16:49

**中文**

像今天这样的听众和聚会，正在努力让全球南方充分参与对话。你认为这类会议会怎样影响 AI 的整体方向？未来真能带来重大改变吗？我认为能。这也是为什么在世界各地召开峰会很重要：这项技术会影响每一个人。

**English / Original**

So an audience like this gathering like this, right? Where we are trying to involve the global south very much in the dialogue, right? So what do you think would be the impact of such gatherings on the overall direction of AI? Do you think and are these going to make a big change going forward? I think so and I and I think that's why um it's important we convene this summit around the world because it's going to affect this technology is going to affect everyone.

<a id="t-5c99ca3602ee"></a>
## 00:17:20

**中文**

AI 是数字技术，不可能真正被国界限制。以开源为例，它通常是好事；但如果在开源软件中发现漏洞或其他问题，该如何召回、如何修补？它并不存在统一的召回机制。我们必须思考这些新问题，因为对于 AI，很难预先完全了解是否存在漏洞。对全球南方以及印度这样的国家而言，今天的年轻人拥有巨大机遇：你们几乎都能使用世界上最前沿的工具。

**English / Original**

It's um it's a digital technology, so it can't really be contained by borders. Um there's things like open source, so which is generally very good, but also one has to think about if you found a vulnerability or or some issue with an open source piece of software, how do you recall it? How do you patch it? There's no recall. So, we need to think about that. These are new issues with something like AI where it's hard ahead of time to fully understand if there are any vulnerabilities. And um I think for the global south and and countries like India, I think there's huge opportunity for the youth of today. You you all have access to pretty much the most cutting-edge tools in the world, right?

<a id="t-3b0f58f550b5"></a>
## 00:18:01

**中文**

我认为这在过去从未发生过；即便有延迟，也可能只是在前沿实验室发明后的三至六个月。作为身处这一领域最前线的人，我可以说，我们甚至来不及充分理解这些模型能在产品和应用研究中支撑哪些惊人能力。因此，这里还有极其广阔的潜力有待探索。

**English / Original**

I don't think that's ever sort of happened before. Maybe any 3 to 6 months after they've been invented in the frontier labs. And I think and I can say as someone working on the coalface of this, um we barely have time to understand what are the amazing uh capabilities that could be supported by these models, you know, in products and applied research. So, I think there's so much um uh potential there uh to be explored.

<a id="t-c5a862d196c1"></a>
## 00:18:27

**中文**

我认为我们会看到大量探索。也希望在座的企业家能借助这些工具完成惊人的事情，产出甚至可能达到从前的十倍，因为它们能力强大，而且几乎能即时供世界各地使用。再问一个更具体的问题：你现在身处印度。许多人都注意到，本次峰会的与会者异常年轻，这股青春活力十分突出。

**English / Original**

And I think we'll see a lot of that um and hopefully many of you in the audience too, entrepreneurs here, uh uh you can do incredible things, maybe 10x of what you could do before because these tools are so capable and they're available almost instantly around the world. So, I I I have a question that's more specific because you are right now in India, right? So, so the Indian population, like as we one of the things that a lot of people have been remarking is that the crowd at this summit is extraordinarily young, right? So, this youthful energy, right?

<a id="t-e3096188fdf5"></a>
## 00:19:04

**中文**

考虑到印度的资源约束，同时又拥有庞大人才库，印度未来能有效发挥什么作用？这里展现出的活力已经令我印象深刻。部长也谈到了当代年轻人；从民调看，印度年轻人对 AI 尤其积极，我认为这很好。我建议今天的学生全力掌握这些新型 AI 工具。未来十年，它们几乎会赋予学生“超能力”，让他们能够做到更多——无论是在——

**English / Original**

So, you think what what what would be the role that India can play effectively in the future given the resource constraints of India and but also the availability of this talent pool. Well, look, I've been incredibly impressed already about the energy that's here, and we heard from the minister that, you know, the youth of today, and in India especially, I think when you see the polls on this are very positive about AI, which I think is great. And what I'd recommend to the students of today is to really lean into becoming incredibly proficient with these new AI tools. And I think over the next 10 years, that will almost kind of make them superpowered in terms of what they're able to do. You know, whether

<a id="t-2c9ef102fce2"></a>
## 00:19:47

**中文**

——商业还是科学。这有点像计算机时代、移动时代或互联网时代的黎明：伴随技术成长、天生熟悉它的一代人，最终会做出今天只能想象的非凡事情。我认为 AI 也会如此，而印度及其年轻人可以站在这股浪潮的最前沿。

**English / Original**

that's business or science, and it's a little bit like the dawn of the computer age or maybe mobile or internet that we went through. Those the the the the the generation that grows up native with that technology will end up doing sort of incredible things that we can only dream of right now. And I think that's going to happen with AI, and I think India and and the youth here can be at the vanguard of that.

<a id="t-223c9cf97e87"></a>
## 00:20:13

**中文**

既然这是研究研讨会，下一题能否请你稍微讲得技术一些？当然。很好。我们看到 AlphaFold 的演进：它最初利用了 Baker 实验室的既有工作，随后不断独立发展。那么，从技术角度看，这类 Alpha 系统或“共同科学家”模型的下一阶段是什么？

**English / Original**

Since this is a research symposium, can I can I ask you to get a little technical on the next answer? >> sure. That's great. So, So, we we we we saw the evolution of how AlphaFold, right, started off by, you know, building on top of existing work from the Baker lab, and and then moving on to evolving all on its own, right?

<a id="t-b4105a91f9c5"></a>
## 00:20:35

**中文**

以 AlphaFold 为例，我们其实构建的是一个全新系统，但它需要 PDB，也就是蛋白质数据库。我们需要人类过去 50 年通过实验艰苦测出的 15 万个结构；事实证明，这些数据只是勉强足以解决问题并构建 AlphaFold 这样的系统。DeepMind 和其他机构正在讨论一个有趣问题：通用系统与工具之间究竟有什么区别？

**English / Original**

So, what do you think is the next stage technically in the evolution of these kinds of alphas, core scientist models? Yeah, so, well, with AlphaFold, what we what we we actually built a new completely new system, but it would it required the PDB, the protein database. So, we needed the 150,000 structures that humanity had painstakingly found over the last 50 years through experimental work, right? And that turned out to be only just enough data to actually, you know, build a uh solve the problem and build a a system like AlphaFold. The interesting thing is in the debate we have uh at DeepMind and and other places is what's the difference between the general system,

<a id="t-77054d4fe651"></a>
## 00:21:17

**中文**

你可以把通用系统看作大脑，把另一部分看作它所使用的工具。对人类而言，心智与工具在物理上分离，所以没有争议；但若两者都是数字系统，甚至工具和负责协调的系统本身都是 AI，那么哪些能力应放进主系统，哪些应保留为专用工具？我认为，在我们的方案中，Gemini 这样的基础模型最终会把 AlphaFold 等 Alpha 系统作为工具调用。

**English / Original**

which you can think of like the brain, and the tools that it uses. So, um for us as as humans, there's no debate about what's our minds and what are our tools because obviously it's physically separated. But, if both things are digital and in some cases both things are AI, the tool and the and the orchestrating system, uh then what do you put in the main system and what do you leave as a specialized tool? Now, in my opinion, I think you'll end up with in in our case with foundation models like Gemini, we'll use Alpha things like AlphaFold as a tool.

<a id="t-6797c3d1da82"></a>
## 00:21:49

**中文**

如果 Gemini 想要或需要折叠蛋白质、理解蛋白质结构，我认为，与其把所有蛋白质信息都塞进主系统，不如把 AlphaFold 当作工具来调用。技术上的选择归结为：把这些数据纳入主系统，是否能帮助其他任务、能否迁移到其他任务？

**English / Original**

So, if you if you if Gemini wanted to or needed to fold a protein, understand the structure of a protein, I think it would be better for it to call uh AlphaFold as a tool than put all of that protein information into the main um system. And I think the the if you want to talk about it technically, I think the choice comes down to if you put that data into the main system, does it help with other tasks? Does it transfer to other tasks?

<a id="t-396b060fe516"></a>
## 00:22:16

**中文**

还是会损害其他任务的表现？这其实是一个经验问题。例如，我们把编程和数学纳入通用基础模型，是因为事实证明，擅长编程或数学会提升一般性的规划与推理能力；这些技能既有用，也能泛化。但蛋白质折叠或许是高度专门化的能力，未必能迁移到其他领域。因此，我倾向于把它保留为专用工具。主持人：这很有意思，因为机器人路径规划中的许多经验曾被用于——

**English / Original**

Or does it actually degrade the performance on those other tasks? So, it's actually an empirical question. So, that's why, for example, coding and maths, we put that in the general foundation models because um it turns out if you get good at coding or math, you're actually better at planning uh and reasoning in general. So, it's a it's it's a useful ta- a skill, but it also generalizes. And uh but something like, you know, the the folding proteins probably is a is a very specialized skill that that wouldn't necessarily transfer to other domains. So, I would be of the opinion that we should leave that as a specialized tool. Oh, that's interesting. Because a lot of learnings from robot path planning was used for

<a id="t-a5a1b0811055"></a>
## 00:22:56

**中文**

——尝试解决蛋白质折叠。那么，掌握蛋白质折叠不会反过来迁移到其他领域吗？可能会，但这需要实验验证。事实上，我们一直在较小模型上做这类实验：消融不同数据集，混入特定数据集，再观察它们是带来帮助，还是导致某些基准退步。

**English / Original**

trying to robot path planning was tried used to try to solve protein folding. So, you think protein knowing how to protein fold will not transfer back to other domains? It might do, but but we would I I think it's In fact, we do these experiments all the time on smaller scale models where we ablate different data sets and we try and mix in certain data sets and see if they help or if they regress some benchmarks.

<a id="t-be2aa10d2d11"></a>
## 00:23:21

**中文**

例如，如果加入全部蛋白质数据后语言能力反而变差，那就毫无意义——而这很可能就是目前会发生的事。也许未来的 AGI 系统会把一切都纳入同一系统；但在可预见的未来，保留独立工具更有效率。这些工具也可能是混合系统，而不只是学习系统：它们可以像 AlphaFold 那样内置有关物理、化学和化学键的结构。这些结构原则上可以学得，但直接或间接编入系统可能更高效。

**English / Original**

Like, it would be no use if you put all the protein data in and then it got worse at language, for example. Um which is probably currently what would happen. So, maybe over time with an AGI system, you just have everything in in the one system, but I think for the foreseeable, maybe, you know, future, I think it'll be more efficient to still have separate tools. Also, those tools, by the way, might be hybrid systems in that they might just not they might not be just learning systems. They might also have built-in structures, like AlphaFold did actually, about physics and chemistry and chemical bonds that you could learn, but would be more efficient to just tell the system or or program that indirectly.

<a id="t-639c0d1fbca8"></a>
## 00:24:03

**中文**

既然提到机器人，物理 AI 的下一步是什么？我对机器人技术越来越兴奋。十年前我没那么感兴趣，因为我认为瓶颈在算法，而不是机体的物理制造；落后的是算法。如今情况不同。我们对 Gemini Robotics 很兴奋，因为 Gemini 基础模型从一开始就擅长多模态，能够理解视觉、图像及周遭世界，对物理世界有很好的认识。而机器人恰恰需要——

**English / Original**

So, now that I mentioned the word robot, so I'd like to ask you, what's next in physical AI? Uh well, look, I think I'm getting increasingly excited by robotics. Um you know, I probably wasn't so interested in that 10 years ago because I I felt that the issue was the algorithms, not the physical construction of the bodies. I thought the algorithms were the thing that was behind. Now, I think algorithms And actually, we're very excited about Gemini robotics because we built our foundation model to be really good multimodally, so it could understand vision, image, the world around us. So, it has a very good understanding of the physical world. And that's exactly what you'd want for robotics,

<a id="t-68b49c951ff5"></a>
## 00:24:45

**中文**

——一种能理解自身所处物理环境的通用系统。未来两三年，我们可能会看到机器人技术出现一些很有意思的突破时刻。不过仍有相当多研究要做，并没有达到某些机器人公司所宣称的程度。

**English / Original**

is a kind of general system that understands the physical context the robot's in. So, I think we're probably in the next 2-3 years, we're going to see some very interesting new breakout moments for robotics. I think there's still quite a bit more research to be done. I don't think we're there yet like some robotics companies are claiming.

<a id="t-9abb3ddfb663"></a>
## 00:25:03

**中文**

我认为人形机器人和非人形机器人都会出现，两者都有用途。未来几年，机器人技术会迎来真正的突破。因此，这是一个非常令人兴奋、值得现在就进入的领域。主持人：接下来是一个有点棘手的问题。很多人都在散播对 AI 的恐惧，而我们知道其中很大一部分并无必要。

**English / Original**

And I think we'll have humanoid robots and also non-humanoid. I think both will be useful. But I think in the next couple of years there'll be some really breakthrough moments. So I think it's a very exciting space to watch and a good area to get into right now. So a little tricky question. So a lot of people are I mean that is a lot of fear mongering around AI. I mean we know that most of it is unnecessary.

<a id="t-addff34af132"></a>
## 00:25:30

**中文**

如果开始出现由基础模型驱动的人形机器人，你认为人们的恐惧会加剧吗？有这种可能。它取决于我们怎样设计这些机器人；风险也会增加，具体又取决于部署用途。尤其当人形机器人能力很强、身体也很重时，确实会带来危险和风险。因此，在大量机器人四处活动之前，我们需要落实刚才讨论的那些护栏。

**English / Original**

But if you start getting humanoids that are running on foundation models, do you think the the fear factor would go up? Potentially. I mean it depends how we design those humanoids, but I think some risks go up too, right? Again, depends on what you deploy them for and I think increasingly especially if the humanoids are pretty capable and they're heavy. You know, there are dangers and risks with that. So I think we need to have those guardrails that we were discussing earlier in place by the time there's a lot of robots roaming around.

<a id="t-42fdec233fbb"></a>
## 00:26:08

**中文**

谢谢。风险已经谈了不少，现在多谈谈积极面。拥有更多资源的国家似乎仍能获得前沿 AI 的大部分好处，例如拥有 GPU，可以在本地运行模型。那么，要让 AI 惠及全球南方和更多人口，需要怎样做？

**English / Original**

Thank you. So we talked enough about risks. So let's talk a little bit more about the positive positivity of things. So I know a lot of the benefits of of all this cutting edge AI right now is still seems to accrue to you know, the countries that have more resources, right? that can that have GPUs that can run their models there and so on and so forth. So what do you think would take for AI to reach the global south, benefit much more larger fraction of the population.

<a id="t-fa85690082ff"></a>
## 00:26:38

**中文**

我们应该关注哪些举措？刚才其实已经谈到：全球最领先的基础模型大约有三四个，如果把中国模型算上，也许有五六个。几个月后，它们通常就能以很高的性价比被广泛使用；此外还有开源。我们也在开发自己的开放模型 Gemma，很快会发布新版，它在边缘设备上非常强大。这是个很有意思的方向：高效模型可以运行在——

**English / Original**

So what kind of initiatives that we should be looking at? I think we kind of touched on it earlier, but um these the leading foundation models, maybe there's three or four of them and and perhaps five or six uh if you include the Chinese models, too. Then, um they're pretty much available uh very uh cost-effectively for um uh you know, a few months behind. Also, open-source. I mean, we work on um our own open-source models, Gemma, which uh uh we'll be releasing a new version of soon, which are very powerful for edge devices. So, I think that's very interesting area. It's really efficient models uh for you know, computing on the

<a id="t-354b9b87f143"></a>
## 00:27:17

**中文**

——边缘端，例如手机、单台笔记本电脑，最终也包括机器人。在优化这类模型的能力，以及在其之上构建产品或应用方面，都存在巨大机会。我认为，这类工作拥有广阔空间。

**English / Original**

edge, whether that's um you know, your phone uh or a lap single laptop or you know, eventually robotics. I think uh there's huge opportunities there for optimizing uh what those kinds of uh models do and the types of products or applications you can build on top of that. So, I think there's um you know, a lot of uh potential there for for for those types of um work to happen.

<a id="t-3a26876f9d20"></a>
## 00:27:42

**中文**

我看到整个礼堂第一次突然暗了下来。[笑声] 不，不，这绝不是什么不祥信号，大家不用担心。[笑声] 这真的很了不起。你们早年在 NeurIPS 的一场会外活动中首次演示游戏系统时，我也在场；当时只是一间小屋，但同样挤满了人。

**English / Original**

I know that the entire the auditorium went dark for the first time. >> [laughter] >> No, no, no. That's that is not signaling any kind of ominous uh things, so don't worry about that. [laughter] So, I mean, it's amazing. So, I was actually there when you did the first uh game-playing demo at one of the NeurIPS side events. And that was in this tiny room. Even then, it was packed.

<a id="t-1e64f587fcf5"></a>
## 00:28:07

**中文**

如今这座大礼堂也座无虚席。你觉得自己能坐满的最大场馆有多大——麦迪逊广场花园吗？我不知道。这里已经相当大了，而且听说还有许多人在线观看。我清楚记得那次 NeurIPS 活动：场地或许只有这里的三分之一，全是站席，门外也挤满了人。那确实是深度强化学习系统第一次取得成功：我们开创了能够玩 Atari 游戏的系统。以今天标准看游戏很简单，但系统只接收屏幕像素，没有任何额外信息，目标只是——

**English / Original**

And and this large auditorium is packed. So, what's the largest size hall you think you can pack if you are talking about Madison Square Garden? I don't know. This is um pretty big one and I I hear it's streaming to many, many people online. So, um but yes, I remember that that uh at that NeurIPS event very well. It was a it was a hall maybe maybe a third of this size, but it was I think it was standing room only. It was packed outside the door, and that was really the first success that we had with um these deep reinforcement learning systems that we sort of pioneered that could play Atari games. Very simple games now, but um at but just directly from the pixels, not giving any other information, just

<a id="t-7e7d81be2c1e"></a>
## 00:28:49

**中文**

——把得分最大化。它接收屏幕像素并采取行动。我认为，这也许是现代 AI 时代第一次展示一个智能体系统完成具有挑战性且有趣的任务。这里的任务本来就是为人类设计的，既有趣、令人愉快，又具有一定挑战。对我们而言，这是一个分水岭；对整个行业或许也是如此。我记得那是在 2013 年。这项关于学习系统和学习算法的工作证明了通用性的设想：无需针对信息作特殊处理，也无需给予系统——

**English / Original**

maximize the score, here the pixels on the screen, and it was the first demonstration I would say maybe of the modern AI era of an energetic system doing something kind of challenging and interesting. That in the case in this case a task that was designed for humans to find interesting and and enjoyable and somewhat challenging. And I think it was kind of a watershed moment of course for us but also for maybe for the industry it was back in 2013 I think. And that actually this thesis of learning systems, learning algorithms, this idea of generality that you don't special case the information or give give it

<a id="t-3a1eddeae9f6"></a>
## 00:29:30

**中文**

——带特权的先验信息；而在此之前，传统或老派 AI（例如专家系统）通常正是这样做的。它表明这种方法确实可以扩展到有意思的任务。Atari 屏幕大约包含 2 万个像素，按今天标准很小，但对当时的系统而言，已经构成非常大的动作空间和数据空间。

**English / Original**

privileged prior information to the system that maybe was the way traditional AI had been done or old good old fashioned AI had been done until then expert systems that actually they could scale to something interesting. In this case an Atari screen with 20,000 pixels on the screen, you know, trivially small by today's standards but very you know, a very large action space and data space for the types of systems we had then.

<a id="t-21b740272619"></a>
## 00:29:59

**中文**

如今，Atari 已经成了强化学习的“Hello World”。是的。它当然鼓励我们继续开发 AlphaGo。2016 年 AlphaGo 让整个领域乃至行业都刮目相看，是又一个重大分水岭，也引发了对这些技术的大量商业兴趣。我们证明深度强化学习和学习系统可以扩展到在韩国的比赛中击败世界冠军、传奇棋手李世石。

**English / Original**

And now they've become like the hello world of reinforcement learning now. Yes. Yes. And then of course that encouraged us to then do go on with AlphaGo which was I think really the the big watershed moment that made the field and the industry sit up and notice in 2016 and and I think started a lot of the commercial interest in in these technologies that we could scale this kind of deep reinforcement learning and learning systems to actually you know, beat the world champion and the legendary you know, Lisa Doll in the in our South Korean match.

<a id="t-b133f68696c3"></a>
## 00:30:29

**中文**

有一点我必须指出：你们完整公开了 Atari 游戏系统、后来又公开 AlphaGo 的工作，使强化学习社区的其他人能够跟进，所以谢谢你。当然，我个人最早的“成功”只是能待在房间里面，而不是在门外争着挤进去看演示。[笑声] 暂时离开更宽泛的 AI，我想问强化学习。Rich Sutton 一直在讨论，David Silver 和 Rich 写过一篇很精彩的文章，谈强化学习如何推动 AI 前进。你怎么看？

**English / Original**

So and one thing I really have to say that I mean you wrote up everything about the Atari game player and also later AlphaGo that allowed rest of the reinforcement learning community to catch up so thank you for that. And of course my first success was I was actually in the room and not outside clamoring to get into the room to watch the demo, but so that was that's that's that's great. So again, so let's forget AI for a minute. Let me ask about reinforcement learning. So you know that Richard has been talking a lot about how how and and and David and Rich wrote this amazing amazing article on how reinforcement learning is going to drive AI forward. So what's your take on that?

<a id="t-0ea53a6d167e"></a>
## 00:31:10

**中文**

我们多年来当然有过许多讨论。不妨更一般地回答：我怎样看今天的基础模型与强化学习？强化学习已是这些模型后训练不可或缺的一环。我认为，推理时计算——也就是模型的“思考”部分——还能更多借鉴我们在 AlphaGo、蒙特卡洛树搜索等工作中开创的思想。许多方面都需要把 AlphaGo 的思想与今天的基础模型结合起来。当然，这更困难，因为我们并没有一个完美的——

**English / Original**

Well, yeah, obviously we've we've had many debates over the years and I think it's for me if I was to say and maybe we can take this question more generally like um, what do I think about today's foundation models and reinforcement learning? Of course reinforcement learning is an integral part of the post training of these these models and I think the influence time compute the thinking part of the the models could actually benefit a lot more from the ideas we pioneered in AlphaGo, the Monte Carlo tree search and other things. So in actually many respects we need to combine the ideas we had with AlphaGo with today's foundation models. Of course it's harder because you don't have a perfect model of the

<a id="t-e4f65a9bd723"></a>
## 00:31:49

**中文**

——世界模型；我们需要更好的世界模型。在游戏中，状态转移矩阵是明确的，这个问题很简单。若要我判断今天的路线，我认为 Gemini 这样的基础模型将是最终 AGI 方案的关键组成部分，随后还会叠加许多有意思的强化学习方法。

**English / Original**

world, right? You need a better world model. In games it's trivial, right? The the transition matrix. So I think that's an issue, but if you're talking about if I was to suggest today, I think the foundation models are going to like Gemini are going to be a critical part of the ultimate AGI solution. And then I think we'll have a lot of interesting reinforcement learning on top.

<a id="t-3d4efe082701"></a>
## 00:32:13

**中文**

也许 20 年后，我们最终会拥有更接近 AlphaZero 的系统，由强化学习从零学会一切。但我认为，这实际上并不是实现 AGI 的最快路径。更合理的方式是利用基础模型学习既有信息，形成对世界运行方式的模型，再在其之上叠加强化学习与规划。至少在第一阶段，这样会更高效。

**English / Original**

I I think eventually one day maybe 20 years from now we'll have a kind of more like an AlphaZero type system where you know, reinforcement learning can learn everything from scratch, right? But I think actually that's not going to be the fastest way to AGI. I think it makes sense to use the foundation models and you know, all all information that's already out there and learn that as almost like a model of how the world works and then do your reinforcement learning and planning uh on top of that. I think that will be uh more efficient in the first case.

<a id="t-2c16c6474aee"></a>
## 00:32:46

**中文**

主持人：强化学习仍只是“蛋糕上的樱桃”吗？不是。至于“樱桃”的说法，你得去问 Yann，我相信他能详细解释。我从来没有真正同意这种说法；强化学习是基础组成部分。如果用比特衡量，有人会问：“你从强化学习中究竟获得了多少比特的信息？”但在我看来——你或许也同意——不同信息比特的价值并不相等。例如，“是否赢得比赛”这一比特，就比屏幕上的某个随机像素重要得多。

**English / Original**

>> Is it still going to be the cherry on top? No, I think it's going to be I think it's I mean, you know, you have to ask Jan about his cherry comments. I'm sure he's he can talk at length on that. But I I've never really agreed with that. I think it's just a fundamental. I don't really, you know, I think um obviously if you measured it in terms of bits, uh then one can say, "Well, how much bits of information you're getting from the reinforcement learning?" But in my view and presumably your view is that not all bits are equal in terms of information. Of course, if you get a bit about, you know, did you win the game or not win the game, that's much more important than some random pixel on the

<a id="t-5289a1de3f6b"></a>
## 00:33:21

**中文**

——屏幕上的像素，对吧？因此，以一种简单方式把这些比特的信息价值等同起来，显然并不正确。我确信基础模型将是 AGI 的重要部分；问题只是，它们是否已包含所需的一切，还是仅仅构成关键部分。我认为毫无疑问，它们至少会是首批 AGI 系统的核心组成。

**English / Original**

screen, right? So, to equate the the informational value of those bits um in just, you know, a trivial way is clearly incorrect in my opinion. But I do think that the foundation models um are going to be The question is is are they um going to be all that's needed or just a critical part of what's needed? I think there's no question they're going to be at least a critical component of um the first AGI systems.

<a id="t-2cd8881bb0f5"></a>
## 00:33:51

**中文**

[笑声] 我们快没时间了。你想向峰会参与者传达什么信息？我的信息是“谨慎乐观”。我们正处于一场惊人变革的前夜，它将为科学和医学带来巨大收益。人类健康尤其是我热衷的方向，我能看到我们应对人类健康问题的方式发生革命性变化。围绕这些系统，还可以打造许多杰出的公司、工具和产品——

**English / Original**

Uh so >> [laughter] >> So, I I I So, to be So, we're almost out of time. So, I just want to ask you, what's your message for the attendees of the summit? Well, look, I think my message is uh one of I would say cautious optimism. So, I think we're on the cusp of an absolutely incredible transformation um that's going to uh bring incredible benefits in science and medicine specifically is what I'm passionate about and I can see uh revolutionizing the way uh we we we deal with human health. Um I think there's many amazing companies and and and tools and products to build [clears throat] on top of uh these systems and I think

<a id="t-e42f806a63ae"></a>
## 00:34:33

**中文**

——全世界的人都能在这些 AI 系统之上进行创造。但我也要补充一句警示：只要有足够时间和足够多优秀头脑投入，我相信人类的智慧能够解决这些技术问题和技术风险。

**English / Original**

uh everyone in the world can can build on these AI systems to do that. But then also I would just add a note of caution which is I think we will solve these technical issues given enough time and enough brain power on it. I believe in human ingenuity and if the best minds work towards that I think we'll solve the technical risks.

<a id="t-7554896b7e57"></a>
## 00:34:54

**中文**

但这也必须在国际层面推进。因此，社会层面的挑战最终可能比技术挑战更难。谢谢。主持人：感谢 Demis。谢谢。[掌声] 演讲者：感谢部长和各位教授同事邀请我发表主题演讲。能来到这里，我深感荣幸。

**English / Original**

But we also need to do this internationally. So the societal challenges of that may actually end up being the harder problem than the technical ones. Thank you. So so there is a service everyone. Thank you. >> [applause] >> Thank you for the minister our fellow professors for inviting me here to give this keynote. It's a huge honor to be here.

<a id="t-604044dca1ac"></a>
## 00:35:25

**中文**

看到印度各地如此踊跃地参与、对 AI 表现出浓厚兴趣，我非常激动；同样令人振奋的是，来自世界各地产业界和学术界的许多同事也汇聚于此。我期待未来几天围绕 AI 前进方向展开精彩讨论。我们在 2010 年创办 DeepMind，距今似乎只有 16 年，但以 AI 的发展尺度来看，几乎已是漫长的一生。

**English / Original**

Very excited to see the incredible turnout from the country and the interest in AI of course but also from many of my colleagues from around the world in both industry and academia who have convened here. So I'm looking forward to really amazing couple of days of discussions about the way forward for AI. When we started DeepMind in 2010 seems only 16 years ago now but it's almost a lifetime in terms of AI.

<a id="t-1e3186e2aa4e"></a>
## 00:35:56

**中文**

当时几乎无人研究 AI，学术界只有少数人，产业界更几乎没有人。因此，我们在短短十多年里取得的进展令人惊叹。但如今，我认为我们已到达一个临界点：AGI，也就是通用人工智能，可能在未来五至八年内出现。因此，本次峰会召开在一个关键节点：我们开始看到自主性更强、能力也强得多的智能体式 AI 系统，并初步见识这类系统能够做什么。当然，机遇——

**English / Original**

Almost nobody was working on AI back then few people in academia but almost nobody in industry. So it's been amazing to see the progress we've made in just over a decade. But at this point I think we're at a threshold moment where AGI artificial general intelligence is on the horizon maybe in the next five five to eight years. So this summit comes at a critical moment as we start seeing more autonomous agentic AI systems much more capable. I think we're seeing the beginnings of of what those kinds of systems can do. And of course, the opportunities are

<a id="t-4b5f3d56b200"></a>
## 00:36:35

**中文**

——极其巨大。我个人最热衷的是以 AI 推动科学，并借助 AlphaFold 等系统促进科学与医学进步。我认为 AI 能够彻底改变药物发现和人类健康，也能助力材料科学、气候变化等领域。不过，它当然也伴随着许多风险。AI 是一种双重用途技术，将成为人类历史上变革性最强的技术之一，甚至可能没有之一。

**English / Original**

incredible in for my my personal passion is using AI for science and to advance science and medicine with systems like AlphaFold. I think we can revolutionize drug discovery and human health and also areas like material science and climate change with help of AI. But of course, it also comes with many risks, too. AI is a dual-purpose technology. It's going to be the most one of the most, if not the most transformative technology in human history.

<a id="t-3ad038862888"></a>
## 00:37:08

**中文**

而如今，它已近在眼前。这项技术会影响全世界每一个人，也会跨越国界。因此，举办这样的国际峰会极其重要：让国际社会共同讨论，如何确保这些机遇惠及全世界——

**English / Original**

And that's just on the horizon now. And the thing about this technology is it's going to affect the whole world, everybody in the world. It's going to cross borders. So it's very important that we have summits like this, international summits to bring the international community together to discuss how to make sure the opportunities benefit the whole world.

<a id="t-625bfac3b9f4"></a>
## 00:37:31

**中文**

我们又该如何降低风险？我认为，要降低其中一些风险，必须开展国际合作，而第一步就是国际对话。从英国、巴黎、韩国到如今的印度，这些峰会正是为此而设；它们有效召集了许多重要国家领导人和该技术领域的重要领袖，共同讨论即将到来的变化。感谢邀请我出席，我期待未来几天在这里的讨论。谢谢。

**English / Original**

And also, how do we mitigate the risks? And I think in order to mitigate some of the risks, it's going to need international cooperation. And the first step for that is to have international dialogue. And I think that's what these summits starting in the UK and then in Paris and then in Korea and now here in India are for and have been really helpful in convening many of the important heads of state and many of the important leaders in this technology together to discuss what is on the horizon. So with that, thank you for having me and I look forward to the discussions here in the next couple of days. Thank you.

<a id="t-b75d690a489f"></a>
## 00:38:11

**中文**

[掌声] 大家早上好。[叹气] Demis，你刚才谈到 AI 的进步令你多么惊讶。你在 2010 年创办 DeepMind，愿景是解决智能。那么，我们已经做到这一步了吗？还没有，我认为尚未实现。

**English / Original**

>> [applause] >> Yeah, so good morning everyone. >> [sighs] >> And so Demis, you you are at the progress that AI has made and you started DeepMind in 2010 with a vision to solve intelligence. So, are we there yet? Um no, I don't think we're there yet.

<a id="t-9be073a758f4"></a>
## 00:38:54

**中文**

如果“解决智能”是指构建通用人工智能，我的门槛相当高：系统必须展现人类拥有的全部认知能力，包括创造力、长期规划等。今天的系统令人印象深刻，但也仍有许多缺陷和做不到的事情，包括缺乏通用系统应具备的跨任务一致性。

**English / Original**

Uh I I have quite a high bar for if we talk about solving intelligence meaning building artificial general intelligence, um then, you know, my measure for that is having a system that can exhibit all the cognitive capabilities humans can, including in creativity, long-term planning, things like that. And today's systems are very impressive, but um as we know, they still have many flaws uh and things they can't do, uh including consistency across different tasks that you would um want from a general system.

<a id="t-88942c9bbca8"></a>
## 00:39:28

**中文**

你刚才提到 AGI。那么，从我们目前所处的位置出发，怎样才能迈向你所定义的通用人工智能？在我看来，当前系统距离通用智能尚缺少持续学习等能力，也就是在完成训练、部署到现实世界后仍能继续学习。今天的系统经过各种训练后，基本就被冻结下来，再投放到现实世界；而我们希望它们能持续在线地从经验及——

**English / Original**

So, you did mention AGI, right? So, what do you think really that we can do to take where we are right now and move it towards artificial general intelligence as you define it? When I look at the um current systems and uh what's missing from uh them being a kind of general intelligence, I would say things like continual learning. So, learning after they've been trained and put out into the world. The today's systems, we train them, um we do various different types of training on them, and then uh they're kind of frozen and then put out into the world. Um but what you'd like is uh for those systems to continually learn online from experience, uh to learn from

<a id="t-535bc9ea486d"></a>
## 00:40:10

**中文**

——所处情境中学习，并针对具体场景和交给它的任务实现个性化；今天的系统还做不到。它们也难以制定长期且连贯的计划：能做短期规划，却不像人类那样能够规划数年。不过，最大问题之一或许是我所说的“一致性”。今天的系统像是能力参差不齐的智能：某些方面极强，另一些方面却很弱，有时甚至在同一项能力上表现不一。例如——

**English / Original**

the context they're in, maybe get maybe personalize to the situation that uh and the and the task that you have for them and today's systems don't do that. Um also they have difficulty in things like doing long-term coherent plans. They can plan over the short term, but over the longer term like the way that we plan can plan over years, um they don't really have that capability at the moment. Um but I think probably the one of the biggest issues is uh what I would call consistency. So, um today's systems they're kind of like jagged intelligences. They're very good at certain things, um but they're very uh uh you know, very poor at other things including sometimes the same thing. So,

<a id="t-53b8baf3302d"></a>
## 00:40:51

**中文**

——它们能在国际数学奥林匹克竞赛中取得金牌、解决极难问题，但如果换一种提问方式，却仍可能在初等数学上出错。真正的通用智能系统不应呈现这种参差不齐。主持人：但人类的能力不也参差不齐吗？

**English / Original**

for example, today's systems can get gold medals in the International Math Olympiad, really hard problems, but sometimes can still make mistakes on elementary maths if you pose it uh the question in a certain way. Um a true general intelligence system shouldn't shouldn't have that kind of jaggedness. But but aren't humans jagged as well?

<a id="t-9aa42d11d1be"></a>
## 00:41:11

**中文**

不，我认为并不是这种情况。如果你是数学专家——在座许多人都是——或物理学等领域的专家，就不会在严格来说更简单、更基础的问题上出错；你总有办法推导出答案。同样，以我十分关心的游戏为例：DeepMind 早期就曾用游戏训练 AI 系统。今天的通用基础模型其实很不擅长下国际象棋，甚至低于较弱的业余选手水平，这并不是——

**English / Original**

Uh no, not in that way. I don't think so. So, um I think if you're an expert in maths and many of you in the room are and or in physics or something like that, um you're not going to uh sort of make a mistake on uh a strictly easier, more trivial problem, right? You would have some way to work through that problem. Um similarly, you know, uh let's take the example of games, which is very close to my heart and we've used in the early days of DeepMind to train our AI systems, uh you know, today's general uh foundation models are very poor at for example, playing chess. Um you know, below kind of weak amateur level. Uh and that is not what you would expect from

<a id="t-387a4d463ee2"></a>
## 00:41:52

**中文**

——AGI 系统应有的表现。我还会用另一项标准测试它：真正的创造力。AI 作为科学工具和助手非常有用，也能像 AlphaFold 解决蛋白质结构那样处理专门领域的问题。但真正区分伟大科学家与优秀科学家的，是创造力，以及可以称为“判断什么才是正确问题”的品位。

**English / Original**

an AGI system. Um there are other tests too that I would have is uh true creativity. So, AI is very useful as a scientific tool, an assis- an assistant, um or for solving specialized areas like AlphaFold does with protein structures, but really what separates great scientists from good scientists is their creativity and their sort of maybe you could call it taste for what's the right question.

<a id="t-e8f7c2b33abd"></a>
## 00:42:16

**中文**

什么是正确的假设？提出正确的问题和假设，比证明一个猜想困难得多。我认为这代表最高层次的创造力，而今天的系统还不具备这种能力。一种可能的测试方法是：设想训练一个知识截止于 1911 年的基础模型，再观察它能否像爱因斯坦在 1950 年那样提出广义相对论。

**English / Original**

What's the right hypothesis? And it's much harder to come up with the right question and the right hypothesis than it is to solve a conjecture. And you know, I would call that the highest level of sort of creativity. And so far today's systems, you know, they don't have that capability. You know, one way we maybe would test that is you could imagine training a foundation model with a knowledge cutoff of something like 1911 and then see if it could come up with general relativity like Einstein did in 1950.

<a id="t-e1977994db1d"></a>
## 00:42:47

**中文**

我认为，这会是检验 AGI 的一项好测试；今天的系统显然做不到，不过我相信这个问题终将得到解决。好的，这让人放心。你刚才说 AI 是科学发现的工具。我们常看到 AI 在科学上取得重大突破的新闻，但这怎样才能成为常态？你认为 AI 很快会成为用于发现的科学工具吗？

**English / Original**

But that would be a good test I think for AGI. And I think today's systems clearly would not be capable of doing that. But I think this will be solved in time. Okay, that's that's good. So you talked about scientific tool for scientific discovery, and so we see a lot of you know, headlines about AI making great breakthroughs in science sometimes, but you know, how would this become the norm? Right? I mean, so do you see AI becoming the scientific tool for discovery soon?

<a id="t-4706d3eae484"></a>
## 00:43:22

**中文**

我之所以将一生和整个职业生涯投入 AI，是因为我很早就意识到：如果能构建擅长模式识别的通用模型，它们会成为极其有用的科学工具，甚至可能是科学的终极工具。毕竟，科学在很大程度上就是从海量数据中发现洞见和结构，这恰好非常适合 AI。

**English / Original**

Yeah, the reason I spent my whole life and career working on AI is I saw it quite early on as if you could build these kind of general models that were good at pattern recognition, they would be incredibly useful scientific tools. Maybe the tool for science. That's what I was thinking. You know, you know, a lot of science is about finding insights and structure in vast amounts of data. That's perfect for AI.

<a id="t-41939ba51cd1"></a>
## 00:43:50

**中文**

我认为，未来大约十年，我们将进入科学发现的新黄金时代，借助这些非凡工具迎来近似新文艺复兴的时期。AlphaFold 是其中之一，我希望它只是众多此类工具中的第一个，能够大幅加快研究，并推动几乎所有学科领域的科学发现。下一阶段会以把这些系统作为工具为主；再往后，随着系统自主性提高，我们将看到它们能否成为与你合作的“共同科学家”，像博士生那样。我认为目前距离这一步还很远，但十多年后可能——

**English / Original**

And so I think we're going to enter in the next probably 10 years this new a new golden era for scientific discovery. Almost a new renaissance using these tools. Incredible tools like AlphaFold, but I hope that will be the first of many that can massively speed up our research and accelerate scientific discovery across almost any subject area. So, I think that's going to be the next period is is using these systems as tools, and then after that we'll see as they become more autonomous, can they be kind of like co-scientists with you, like a PhD student something like that. I think we're still quite a way from that, but maybe in 10 plus years that will be

<a id="t-4455d2b575e9"></a>
## 00:44:34

**中文**

——成为可能。

**English / Original**

possible.

<a id="t-e7e48f99d6b2"></a>
## 00:44:35

**中文**

为了让在座相当一部分人放心：在那样的未来，人类仍有角色，对吧？是的。下一阶段对人类专家和科学家来说会非常精彩，他们能够完成的工作量会大幅增加。我尤其期待跨学科科学。它很难，因为你必须掌握不止一个领域，也许是两、三甚至四个领域，再找出它们之间有意思的联系。我认为未来几年许多真正有价值的进展将出现在跨领域结合之中，而 AI 这类工具会——

**English / Original**

So, just to reassure a good fraction of the audience, you still see a role for humans in that future, right? Yes, I think you know, the next phase is going to be incredible I think for human experts and scientists. The amount of work they'll be able to do and I'm actually really excited about cross-disciplinary science, which is quite hard because you have to understand, you know, more than one subject area, maybe two, three, four subject areas, and then find some interesting connections between that. I actually think that's where a lot of the really valuable advances are going to happen in the next few years. These combinations of subject areas, and I think having a tool like AI will really

<a id="t-681c8ffd126f"></a>
## 00:45:15

**中文**

——帮助科学家学习、理解并处理来自多个领域的信息。从某种意义上说，科学是一个定义明确的领域，成功与否可以辨认。那么，在政策及其他公共决策等更抽象的领域，AI 会发挥什么作用？

**English / Original**

help scientists learn about and be able to understand and process all of that information from, you know, multiple different domains. So, so in some sense science is is is a well-defined field, if you will. I mean, you can recognize success and so on and so forth. So, what do you think would be the role of AI in more abstract domains like policy and other kinds of public decision-making?

<a id="t-8f9d746acdd2"></a>
## 00:45:43

**中文**

科学，尤其是编程和数学，更适合今天的系统，主要因为这些领域——以及国际象棋等游戏——的结果可验证。AI 输出的答案能够检查对错；训练时可以准备问题数据库，并百分之百确定答案是否正确。

**English / Original**

I think where I think things like science, but especially if you take coding and maths, are more amenable to the current systems we have today. Mostly because especially coding and mathematics, but also things like games like chess, they are verifiable. So, the answer that the AI system outputs can be checked for correctness. And so, it's very useful when you're training these systems. You can have databases of questions and check to see if 100% if it's right or not right.

<a id="t-dd642df0bb2b"></a>
## 00:46:19

**中文**

当然，一旦进入艺术、人文学科、决策和政策等领域——我不确定你具体指哪一种场景——问题就主观得多，同一个实验也很难重复两次。因此，很难获得数据来判断什么才是好的决策，AI 对这些领域的建模也会困难得多。

**English / Original**

I think of course when you get into the arts and the humanities, things like decision-making, policy, you know, I'm not sure what you necessarily had in mind, but they're much more subjective. They're hard to run the same experiment twice. So, it's it's difficult to get data about what is a good decision in those cases. So, I think they'll be those areas will be a lot harder for AI to to sort of model.

<a id="t-fcc9f61768f4"></a>
## 00:46:45

**中文**

作为神经科学家，你认为 AI 的这些进步让我们对人类智能本身有了哪些认识？现代 AI 阶段的早期——大致可以说是过去 20 年，以 Geoffrey Hinton 等人的深度学习工作为开端——

**English / Original**

So, as a neuroscientist, so what do you think we have learned about human intelligence itself with all these advances in AI that we see? Well, I think when you're in the early days of of of this I guess modern AI phase, which maybe you could say is the last 20 years with the advent of of you know, things like deep learning from from Jeff Hinton.

<a id="t-d95d0c978654"></a>
## 00:47:08

**中文**

DeepMind 创办之初，我们也从大脑获得了许多启发。这是高层次的系统启发，而不是直接照搬大脑的具体机制，例如情景记忆、海马体的功能——这正是我研究的课题——以及神经网络和强化学习；我们知道，大脑中的多巴胺系统实现了强化学习机制。我们以大脑为起点，因为它是我们所知唯一的通用智能实例，甚至可能是宇宙中唯一的实例。因此，我们至少知道通用智能原则上是可以实现的。我认为，这也是——

**English / Original**

A lot in the when we started at DeepMind too, we took a lot of inspiration from the brain. High-level systems inspiration, not not the direct mechanics of the brain, but things like episodic memory, what does the hippocampus do? Which is what I studied. How obviously neural networks, reinforcement learning, your area, which we know the brain dopamine system in the brain implements. So, we kind of took inspiration from the brain, which is the only general example of a genuine intelligence that we have, know of, maybe in the universe, as as a starting point. And so, we sort of knew that it was possible in the limit, right? And and I think that's why

<a id="t-bb3e6ef5cfed"></a>
## 00:47:48

**中文**

——神经网络和强化学习如此成功的原因：关键在于“学习”。系统不是像 20 世纪 90 年代的专家系统或 Deep Blue 那样被直接编入答案，而是能够直接从数据中学习。如今再回看神经科学，我最关注的是大脑何其高效。

**English / Original**

neural networks and reinforcement learning have been so successful because I think learning is the key to the these modern AI systems working, not being programmed with the answer like expert systems were in the '90s, like Deep Blue, but actually allowing the systems to learn directly from data. And I think now what I've what I've I view looking back at neuroscience is, you know, how efficient the brain is.

<a id="t-ccde4b7ea152"></a>
## 00:48:15

**中文**

大脑的样本效率非常高，不需要吞下整个互联网才能理解事物。今天构建的系统虽采用了一些相同原理，但最终呈现为一种与大脑很可能截然不同的系统。谢谢。人们经常谈论 AI 的安全与安保风险。Google DeepMind 对此持什么立场？你们怎样确保 AI 安全？

**English / Original**

Sample efficient, uh it doesn't need, you know, to ingest the whole of the internet to understand things. So, it's a it's a different what we've built now today. It uses some of the same principles, but it's been manifested in a in a very different type of system than probably the way the brain works. Okay. Thank you. So, uh so one of the things that people always talk about is the safety risk, security risk from AI and so on and so forth, right? And so, what does GDM's stand on this? And how do you approach making sure that AI is safe?

<a id="t-11c022c638f1"></a>
## 00:48:50

**中文**

从创办 DeepMind 之初，我们就在为成功作准备。我们的使命表述是：第一步解决智能，第二步用智能解决其他一切。当时这听起来像科幻小说，如今其可行性已经越来越清晰。

**English / Original**

Well, we've been um when we again, when we started DeepMind, we we actually, you know, we're planning for success. Our mission statement was solving intelligence, step one, and then step two, use it to solve everything else. Which at the time, you know, was sounded like science fiction, but I think now it's becoming clearer how that might be possible.

<a id="t-2bfd57b45e2a"></a>
## 00:49:09

**中文**

还有把 AI 应用于几乎每个学科领域的可能。但我们从一开始就在为成功作准备。尽管当时才刚刚起步、整个领域也才开始积蓄势头，但我们已经理解：如果使命真的成功，会产生什么影响。我们把它视为一项 20 年使命，而且我认为进度基本符合计划——大约到 2030 年。届时，AI 在为科学、医学和其他全球挑战带来巨大益处的同时，也会伴随相应风险。随着系统能力增强——

**English / Original**

Um and and uh applying AI to almost every subject area. But we planned for success. So, um even though we were just starting out and the field was mostly uh just building um building up a head of steam, um we understood also the implications of that if that did turn out to be the case. And we thought about it as a 20-year mission, and I think we're basically on track for that, you know, around 2030. Um then it would come with these attendant risks as well as the enormous benefits to science and medicine and all these things that I think we need as a society um to help with our other challenges, many other challenges that we have uh around the world. Um but also, as these systems become more powerful, I think

<a id="t-9cacacc1d90a"></a>
## 00:49:52

**中文**

我们一直担心至少两类风险。一是恶意行为者，可能是个人，也可能是国家行为体；他们会利用这些系统，并因其双重用途属性而将其改造用于有害目的。二是随着我们接近 AGI，我们或许正进入系统自主性更强的“智能体时代”，今年和明年会看到大量进展。我们必须确保系统护栏到位，使其按预期行事，不会偏离到我们没有——

**English / Original**

there's, you know, at least two risks that we've always worried about. One is bad actors, human actors, but individuals, but it could be also nation states using these systems and repurposing them for harmful ends because they're dual purpose. But then also as we get closer to AGI, and I think we're entering maybe a kind of agentic era, should we call it, where systems are more autonomous. I think we'll see a lot of that happening this year and next year. Um then we have to make sure as well that the guardrails in place that these systems do what we expect them to do and don't veer off into areas that we didn't we hadn't

<a id="t-4016620ba850"></a>
## 00:50:33

**中文**

——规划、可能引发问题的领域。我认为，这就是两大挑战。其中一类更偏社会层面，需要国际对话，理想情况下还应形成一套国际共同认可的最低标准；另一类更偏技术风险，即如何确保系统稳健可靠。

**English / Original**

planned for and that could also be problematic. So I think those are the two challenges. There's a very sort of societal one which I think also is going to require international dialogue and maybe ideally a minimum set of standards that is agreed internationally. Um and then the second one more of a technical risk of how can we make sure these systems are robust and reliable.

<a id="t-0daf67ad8350"></a>
## 00:50:57

**中文**

主持人：既然由我来问所有生存风险问题，那么你认为当今 AI 系统最应优先应对的两项风险是什么？我刚才已经提到两类问题。近期我们必须关注生物与网络风险。当前系统的网络能力已相当强，我们必须确保网络防御能力强于攻击手段。Google 和 DeepMind 在这方面投入了大量工作，用 AI 加强网络安全；AI 当然也是——

**English / Original**

So I I mean sitting as I all existential risk questions. So what do you think is the top two risks that we should address with with AI systems today? Well, I mean I I think I mentioned the two classes of problems I think we need to worry about things like bio and cyber risk very soon. I think that the current systems are getting pretty good at cyber and I think we need to make sure cyber defenses are more powerful than the attack vectors. And it's something that, you know, we work on quite a lot at at Google and at DeepMind is sort of using AI for cybersecurity. Um to make sure that of course it's a very useful

<a id="t-1cdc8c4434fa"></a>
## 00:51:40

**中文**

——网络防御的有力工具，但必须保证防守强于进攻。这些属于近期风险；此外还有许多风险需要思考并投入更多研究。其中一些涉及共同商定一套标准；我知道，包括你在内的许多人都在为此努力。

**English / Original**

tool for cyber defense, too, but you need to make sure that the defenses are stronger than the offenses. So, um you know, I think those are kind of near-term risks, uh, but there there are many that we need to think about and and do a lot more research on, and some of them are to do with, uh, um, agreeing a sort of set of standards, which, you know, I know many people, including yourself, are trying to work on.

<a id="t-db43c6519dd6"></a>
## 00:52:02

**中文**

很好，谢谢。再谈国际合作：你说我们必须认真应对这个问题。像今天这样的听众和聚会，正在努力让全球南方充分参与对话。你认为这类会议会怎样影响 AI 的整体方向？未来真能带来重大改变吗？我认为能。这也是为什么在世界各地召开峰会很重要：这项技术会影响每一个人。它是——

**English / Original**

Great. Thank you. So, again, talking about international collaborations, right? So, you you said that we really have to worry on that work on that. So, an audience like this or or or or or gathering like this, right? Where we are trying to involve uh the global south uh very much in the dialogue, right? So, what do you think would be the impact of such gatherings on the overall direction of AI? Do you think and are these going to make a big change going forward? I think so, and I and I think that's why, um, it's important we convene this summit around the world because it's going to affect this technology is going to affect everyone. It's, um,

<a id="t-ca3b159ea068"></a>
## 00:52:43

**中文**

——数字技术，不可能真正被国界限制。以开源为例，它通常是好事；但如果在开源软件中发现漏洞或其他问题，该如何召回、如何修补？它并不存在统一的召回机制。我们必须思考这些新问题，因为对于 AI，很难预先完全了解是否存在漏洞。对全球南方以及印度这样的国家而言，今天的年轻人拥有巨大机遇：你们几乎都能使用世界上最前沿的工具。我——

**English / Original**

it's a digital technology, so it can't really be contained by borders. Um, there's things like open source, so which is generally very good, but also one has to think about if you found a vulnerability or or some issue with an open source piece of software, how do you recall it? How do you patch it? There's no recall. So, we need to think about that. These are new issues with something like AI, where it's hard ahead of time to fully understand if there are any vulnerabilities. And, um, I think for the global south and and countries like India, I think there's huge opportunity for the youth of today. You you all have access to pretty much the most cutting-edge tools in the world. I

<a id="t-cd8d895e537b"></a>
## 00:53:24

**中文**

——认为这在过去从未发生过；即便有延迟，也可能只是在前沿实验室发明后的三至六个月。作为身处这一领域最前线的人，我可以说，我们甚至来不及充分理解这些模型能在产品和应用研究中支撑哪些惊人能力。

**English / Original**

don't think that's ever sort of happened before, maybe any three to six months after they've been invented in the frontier labs. And I think and I can say as someone working on the coalface of this, um, we barely have time to understand what are the amazing uh capabilities that could be supported by these models, you know, in products and applied research.

<a id="t-d6e55a25965a"></a>
## 00:53:45

**中文**

这里还有极其广阔的潜力有待探索，我认为我们会看到大量尝试。也希望在座的企业家能借助这些工具完成惊人的事情，产出甚至可能达到从前的十倍，因为它们能力强大，而且几乎能即时供世界各地使用。再问一个更具体的问题：你现在身处印度。许多人都注意到，本次峰会的与会者异常年轻。

**English / Original**

So, I think there's so much potential there to be explored. And I think we'll see a lot of that and hopefully many of you in the audience to entrepreneurs here, you can do incredible things, maybe 10x of what you could do before because these tools are so capable and they're available almost instantly around the world. So, I I have a question that's more specific because you are right now in India. So, so So, the Indian population, like actually one of the things that a lot of people have been remarking is that the crowd at this summit is extraordinarily young.

<a id="t-10bc96e8ead8"></a>
## 00:54:25

**中文**

这股年轻活力将发挥什么作用？考虑到印度的资源约束，同时又拥有庞大人才库，印度未来能有效扮演怎样的角色？这里展现出的活力已经令我印象深刻。部长也谈到了当代年轻人；从民调看，印度年轻人对 AI 尤其积极，我认为这很好。我建议今天的学生全力掌握这些新型 AI 工具。未来十年，它们几乎会赋予学生“超能力”，让他们能够——

**English / Original**

Right? So, this youthful energy, right? So, you think what what what would be the role that India can play effectively in the future given the resource constraints of India and but also the availability of this talent pool. Well, look, I've been incredibly impressed already about the energy that's here and we heard from the minister that, you know, the youth of today and in India especially, I think when you see the polls on this are very positive about AI, which I think is great. And what I'd recommend to the students of today is to really lean into becoming incredibly proficient with these new AI tools. And I think over the next 10 years that will almost kind of make them superpowered in terms of what they're

<a id="t-c96699e51101"></a>
## 00:55:08

**中文**

——做到更多，无论是在商业还是科学领域。这有点像计算机时代、移动时代或互联网时代的黎明：伴随技术成长、天生熟悉它的一代人，最终会做出今天只能想象的非凡事情。我认为 AI 也会如此，而印度及其年轻人可以站在这股浪潮的最前沿。

**English / Original**

able to do. You know, whether that's business or science and it's a little bit like the dawn of the computer age or maybe mobile or internet that we went through. If those the the the the the generation that grows up native with that technology will end up doing sort of incredible things that we can only dream of right now. And I think that's going to happen with AI and I think India and and the youth here can be at the vanguard of that.

<a id="t-eeb635d4f7a7"></a>
## 00:55:36

**中文**

既然这是研究研讨会，下一题能否请你稍微讲得技术一些？当然。很好。我们看到 AlphaFold 的演进：它最初利用了 Baker 实验室的既有工作，随后不断独立发展。那么，从技术角度看，这类“共同科学家”模型的下一阶段是什么？以 AlphaFold 为例，我们其实构建的是一个全新系统，但它需要 PDB，也就是蛋白质数据库。因此，我们需要 15 万个——

**English / Original**

Since this is a resource symposium, can I can I ask you to get a little technical on the next answer? Sure. Great. So, So, we we we we saw the evolution of how AlphaFold, right, started off by, you know, building on top of existing work from the Baker lab and and then moving on to evolving all on its own, right? So, what do you think is the next stage technically in the evolution of these kinds of co-scientist models? Yeah. So, well, with AlphaFold, what we what we we actually built a new completely new system, but it would it required the PDB, the protein database. So, we needed the 150,000 structures that

<a id="t-822e84c836a0"></a>
## 00:56:17

**中文**

——人类过去 50 年通过实验艰苦测出的结构；事实证明，这些数据只是勉强足以解决问题并构建 AlphaFold 这样的系统。DeepMind 和其他机构正在讨论一个有趣问题：通用系统——你可以把它看作大脑——与其使用的工具究竟有什么区别？对人类而言，心智与工具在物理上分离，所以没有争议；但若两者都是数字系统，甚至工具和负责协调的系统本身都是 AI，那么哪些能力应放进主系统——

**English / Original**

humanity had painstakingly found over the last 50 years through experimental work, right? And that turned out to be only just enough data to actually, you know, build a solve the problem and build a system like AlphaFold. The interesting thing is in the debate we have at DeepMind and and other places is, what's the difference between the general system, which you can think of like the brain, and the tools that it uses? So, for us as as humans, there's no debate about what's our minds and what are our tools because, obviously, it's physically separated. But, if both things are digital and in some cases both things are AI, the tool and the and the orchestrating system, then what do you put in the main system

<a id="t-28683d5195b3"></a>
## 00:57:00

**中文**

——哪些应保留为专用工具？我认为，在我们的方案中，Gemini 这样的基础模型最终会把 AlphaFold 等 Alpha 系统作为工具调用。假如 Gemini 需要折叠蛋白质、理解蛋白质结构，与其把所有蛋白质信息都塞进主系统，不如调用 AlphaFold。技术上的选择归结为：把数据纳入主系统是否能帮助其他任务、能否迁移，还是反而会损害其他任务的表现？

**English / Original**

and what do you leave as a specialized tool? Now, in my opinion, I think you'll end up with in our case with foundation models like Gemini, we'll use Alpha things like AlphaFold as a tool. So, if you if you if Gemini wanted to or needed to fold a protein, understand the structure of protein, I think it would be better for it to call AlphaFold as a tool than put all of that protein information into the main system. And I think the if you want to talk about it technically, I think the choice comes down to if you put that data into the main system, does it help with other tasks? Does it transfer to other tasks? Or does it actually degrade the performance on those other tasks? So

<a id="t-b684c51e8f24"></a>
## 00:57:41

**中文**

这其实是一个经验问题。例如，我们把编程和数学纳入通用基础模型，是因为事实证明，擅长编程或数学会提升一般性的规划与推理能力；这些任务既是有用的技能，也能泛化。但蛋白质折叠或许是高度专门化的能力，未必能迁移到其他领域。因此，我倾向于把它保留为专用工具。主持人：这很有意思，因为机器人路径规划中的许多经验曾被用于尝试解决蛋白质折叠。

**English / Original**

it's actually an empirical question. So that's why for example coding and maths, we put that in the general foundation models because it turns out if you get good at coding or math, you're actually better at planning and reasoning in general. So it's a it's a useful task a skill but it also generalizes. And but something like, you know, the folding proteins probably is a is a very specialized skill that that wouldn't necessarily transfer to other domains. So I would be of the opinion that we should leave that as a specialized tool. Oh, that's interesting. Because a lot of learnings from robot path planning was used for trying to solve robot path planning was tried used to try to solve protein folding.

<a id="t-4a4c7c2cde12"></a>
## 00:58:23

**中文**

那么，掌握蛋白质折叠不会反过来迁移到其他领域吗？可能会，但这需要实验验证。事实上，我们一直在较小模型上做这类实验：消融不同数据集，混入特定数据集，再观察它们是带来帮助，还是导致某些基准退步。例如，如果加入全部蛋白质数据后语言能力反而变差，那就毫无意义——而这很可能就是目前会发生的事。也许未来的 AGI 系统会把一切都纳入同一系统；但在可预见的未来，保留——

**English / Original**

So you think protein knowing how to protein fold will not transfer back to other domains? It might do but but we would I need to in fact we do these experiments all the time on smaller scale models where we ablate different data sets and we try and mix in certain data sets and see if they help or if they regress some benchmarks like it would be no use if you put all the protein data in and then it got worse at language for example. Um which is probably currently what would happen. So maybe over time with an AGI system you just have everything in in the one system but I think for the foreseeable maybe you know, future I think it'll be more efficient to still have

<a id="t-7f8fefa8423b"></a>
## 00:59:03

**中文**

——独立工具更有效率。这些工具也可能是混合系统，而不只是学习系统：它们可以像 AlphaFold 那样内置有关物理、化学和化学键的结构。这些结构原则上可以学得，但直接或间接编入系统可能更高效。

**English / Original**

separate tools. Also those tools by the way might be hybrid systems in that they might just not they might not be just learning systems. They might also have built-in structures like AlphaFold did actually about physics and chemistry and chemical bonds that you could learn but would be more efficient to just tell the system or or program that indirectly.

<a id="t-2216e3c921ce"></a>
## 00:59:25

**中文**

既然提到机器人，物理 AI 的下一步是什么？我对机器人技术越来越兴奋。十年前我没那么感兴趣，因为我认为瓶颈在算法，而不是机体的物理制造；落后的是算法。如今情况不同。我们对 Gemini Robotics 很兴奋，因为 Gemini 基础模型从一开始就擅长多模态，能够理解视觉、图像及周遭世界，对物理世界有很好的认识。而机器人恰恰需要——

**English / Original**

So now that I mentioned about robots, so let me ask you, what's next in physical AI? Uh well, look, I think I'm getting increasingly excited by robotics. Um you know, I probably wasn't so interested in that 10 years ago because I I felt that the issue was the algorithms, not the physical uh construction of the bodies. I thought the algorithms were the thing that was um behind. Now, I think algorithmically, and actually we're very excited about Gemini robotics because we built our foundation model to be really good multimodally, so it could understand vision, image, the world around us. So, it has a very good understanding of the physical world, and that's exactly what you'd want for robotics. Um is a kind of

<a id="t-2cdf18e376b6"></a>
## 01:00:08

**中文**

——一种能理解自身所处物理环境的通用系统。未来两三年，我们可能会看到机器人技术出现一些很有意思的突破时刻。不过仍有相当多研究要做，并没有达到某些机器人公司所宣称的程度。我认为人形机器人和非人形机器人都会出现，两者都有用途。未来几年，机器人技术会迎来真正的突破。因此，这是一个非常令人兴奋、值得现在就进入的领域。

**English / Original**

general system that understands the physical context the robot's in. So, I think [snorts] uh we're probably in the next 2-3 years we're going to see some very interesting new breakout moments for robotics. I think there's still quite a bit more research to be done. I don't think we're there yet like some robotics companies are claiming. Um and I think we'll have humanoid robots and also non-humanoid. I think both will be useful. Um but uh I think in the next couple of years there'll be some really breakthrough uh moments. So, I I think it's a very exciting space to watch and a good area to get into right now.

<a id="t-8ea4724da00d"></a>
## 01:00:40

**中文**

好，一个有点棘手的问题。很多人都在散播对 AI 的恐惧，其中很大一部分并无必要。但如果开始出现由基础模型驱动的人形机器人，你认为人们的恐惧会加剧吗？有这种可能。它取决于我们怎样设计这些机器人；风险也会增加，具体又取决于部署用途。尤其当人形机器人能力很强、身体也很重时，确实会带来危险和风险。因此，我们需要落实那些护栏——

**English / Original**

Okay. So, a little tricky question. So, a lot of people are you know, I mean, there's a lot of fear-mongering around AI. I mean, we know that most of it is unnecessary. Uh but um if you start getting humanoids that are running off foundation models, do you think the the fear factor would go up? Um potentially. I mean, it depends how we design those humanoids, but I think um some risks go up, too, right? Again, depends on what you deploy them for, and I think increasingly, especially if the humanoids are pretty capable, um and they're heavy, you know, there there are dangers and risks with that. So, I think we need to um have those guardrails that

<a id="t-b9e2ba9f428d"></a>
## 01:01:23

**中文**

——确保在大量机器人四处活动前就已部署到位。谢谢。风险已经谈了不少，现在多谈谈积极面。拥有更多资源的国家似乎仍能获得前沿 AI 的大部分好处，例如拥有 GPU，可以在本地运行模型。那么，要让 AI 惠及全球南方和更多人口，需要怎样做？

**English / Original**

we were discussing earlier in place by the time there's a lot of robots roaming around. Okay, thank you. So, now that we we've talked enough about risk, so let's talk a little bit more about the positive positivity of things. So, I know a lot of the benefits of of all this cutting-edge AI right now is still seems to accrue to, you know, the countries that have more resources, right? That kind that have GPUs, that can run their models there and so on and so forth. So, what do you think would take for AI to reach the global south, benefit a much more larger fraction of the population?

<a id="t-ac1407690e6a"></a>
## 01:02:00

**中文**

我们应该关注哪些举措？刚才其实已经谈到：全球最领先的基础模型大约有三四个，如果把中国模型算上，也许有五六个。几个月后，它们通常就能以很高的性价比被广泛使用；此外还有开源。我们也在开发自己的开放模型 Gemma，很快会发布新版，它在边缘设备上非常强大。这是个很有意思的方向：高效模型可以运行在——

**English / Original**

So, what kind of initiatives that we should be looking at? >> Well, no, I I think we kind of touched on it earlier, but um these the leading foundation models, maybe there's three or four of them, and and perhaps five or six, I think including the Chinese models, too. Then, they're pretty much available very cost-effectively for, you know, a few months behind. Also, open source. I mean, we work on our own open source models, Gemma, which we'll be releasing a new version of soon, which are very powerful for edge devices. So, I think that's a very interesting area. It's really efficient models for you know, computing on the

<a id="t-f8d6aa4b5cf9"></a>
## 01:02:39

**中文**

——边缘端，例如手机、单台笔记本电脑，最终也包括机器人。在优化这类模型的能力，以及在其之上构建产品或应用方面，都存在巨大机会。我认为，这类工作拥有广阔空间。

**English / Original**

edge, whether that's, you know, your phone or a single laptop or, you know, eventually robotics. I think there's huge opportunity there for optimizing what those kinds of models do and the types of products or applications you can build on top of that. So, I think there's, you know, a lot of potential there for those types of work to happen.

<a id="t-073fb405cdaa"></a>
## 01:03:04

**中文**

不，是整个礼堂突然暗了下来。不，不，这绝不是什么不祥信号，[清嗓子]大家不用担心。[笑声] 这真的很了不起。你们早年在 NeurIPS 的一场会外活动中首次演示游戏系统时，我也在场；当时只是一间小屋，但同样挤满了人。如今这座大礼堂也座无虚席。你觉得自己能坐满的最大场馆有多大——麦迪逊广场花园吗？

**English / Original**

No, the the entire audience the auditorium went dark for the No, no, no, that's that is not signaling any kind of ominous things, [clears throat] so don't worry about that. >> [laughter] >> It's So, I mean, it's amazing. So, I was actually there when you did the first game playing demo at one of the NeurIPS side events. And that was in this tiny room. Even then, it was packed. And And this large auditorium is packed. So, what's the largest size hall you think you can pack if you are talking about it? It's Madison Square Garden?

<a id="t-60704de8e67c"></a>
## 01:03:37

**中文**

我不知道。这里已经相当大了，而且听说还有许多人在线观看。我清楚记得那次 NeurIPS 活动：场地或许只有这里的三分之一，全是站席，门外也挤满了人。那确实是深度强化学习系统第一次取得成功：我们开创了能够玩 Atari 游戏的系统。以今天标准看游戏很简单，但系统只接收屏幕像素，没有任何额外信息，目标只是把得分最大化。

**English / Original**

>> I don't know. This is um pretty big one, and I I hear it's streaming to many, many people online. So, um But yes, I remember that that uh that NeurIPS event very well. It was a It was a hall maybe maybe a third of this size, but it was I think it was standing room only. It was packed outside the door, and that was really the first success that we had with um these deep reinforcement learning systems that we sort of pioneered that could play Atari games. Very simple games now, but um at that just directly from the pixels, not giving any other information, just maximize the score.

<a id="t-74d404694390"></a>
## 01:04:12

**中文**

系统接收屏幕像素并采取行动。我认为，这也许是现代 AI 时代第一次展示一个智能体系统完成具有挑战性且有趣的任务。这里的任务本来就是为人类设计的，既有趣、令人愉快，又具有一定挑战。我认为，这是一个分水岭时刻；对我们而言当然如此，对整个行业或许也是如此。

**English / Original**

Here are the pixels on the screen. And it was the first demonstration, I would say, maybe of the modern AI era of an agentic system doing something kind of challenging and interesting. That um you know, in in this case, a task that was designed for humans to find interesting and and enjoyable and somewhat challenging. And uh I think it was kind of a watershed moment uh of course for us, but also for maybe for the industry.

<a id="t-b303bb522974"></a>
## 01:04:38

**中文**

我记得那是在 2013 年。这项关于学习系统和学习算法的工作证明了通用性的设想：无需针对信息作特殊处理，也无需给予系统带特权的先验信息；而在此之前，传统或老派 AI（例如专家系统）通常正是这样做的。它表明这种方法确实可以扩展到有意思的任务。Atari 屏幕大约包含 2 万个像素，按今天标准很小，但对当时的系统而言，已经构成非常大的动作空间和数据空间。

**English / Original**

It was back in 2013, I think. And um that actually this thesis of uh learning systems, learning algorithms, this idea of generality that you don't special case uh uh the information or give give it um privileged prior information to the system. That maybe was the way traditional AI had been done or old good old-fashioned AI had been done until then, expert systems. That actually that could scale to something interesting. In this case, an Atari screen with 20,000 pixels on the screen. You you trivially small by today's standards, but very, uh, you know, very large, uh, action space and, uh, data space for the types of systems we had then.

<a id="t-917c1505b498"></a>
## 01:05:20

**中文**

如今，Atari 已经成了强化学习环境的“Hello World”。是的。它当然鼓励我们继续开发 AlphaGo。2016 年 AlphaGo 让整个领域乃至行业都刮目相看，是又一个重大分水岭，也引发了对这些技术的大量商业兴趣。我们证明深度强化学习和学习系统可以扩展到在韩国的比赛中击败世界冠军、传奇棋手李世石。

**English / Original**

Yeah. And now they've become like the hello world of reinforcement learning environments. Yes, yes. And then of course that encouraged us to then do, uh, go on with AlphaGo, which was, I think, really the the big watershed moment that made the field and the industry sit up and notice in 2016 and and I think started a lot of the commercial interest in in these technologies that we could scale this kind of deep reinforcement learning and learning systems to actually, you know, beat the world champion and the legendary, you know, Lee Sedol in in our South Korea match.

<a id="t-db2b4e3c3d92"></a>
## 01:05:52

**中文**

有一点我必须指出：你们完整公开了 Atari 游戏系统、后来又公开 AlphaGo 的工作，使强化学习社区的其他人能够跟进，所以谢谢你。[笑声] 当然，我个人最早的“成功”只是能待在房间里面，而不是在门外争着挤进去看演示。暂时离开更宽泛的 AI，我想问强化学习。Rich Sutton 一直在讨论，David Silver 和 Rich 写过一篇很精彩的文章，谈强化学习如何推动 AI 前进。你怎么看？

**English / Original**

So, and the one thing I really have to say that, I mean, you brought up everything about the Atari game player and also later AlphaGo that allowed the rest of the reinforcement learning community to catch up. So, thank you for that. And [laughter] of course my first success was I was actually in the room and not outside clamoring to get into the room to watch the demo, but, so that was that's that's that's great. So, again, so let's forget AI for a minute. Let me ask about reinforcement learning. So, you know, that Rich Sutton has been talking a lot about how how and and and David and Rich wrote this amazing amazing article on how reinforcement learning is going to drive AI forward. So, what's your take on that?

<a id="t-a5cd7dc59c59"></a>
## 01:06:32

**中文**

我们多年来当然有过许多讨论。不妨更一般地回答：我怎样看今天的基础模型与强化学习？强化学习已是这些模型后训练不可或缺的一环。我认为，推理时计算——也就是模型的“思考”部分——还能更多借鉴我们在 AlphaGo、蒙特卡洛树搜索等工作中开创的思想。许多方面都需要把 AlphaGo 的思想与今天的基础模型结合起来。当然，这更困难，因为我们并没有——

**English / Original**

Well, yeah, I obviously we've we've had many, um, debates over the years and I think it's, uh, for me if I was to say, and maybe we can take this question more generally, like, um, what do I think about today's foundation models and reinforcement learning? Of course, reinforcement learning is an integral part of the post-training of these, um, these models. And I think the inference time compute, the thinking part of the the models, uh, could actually benefit a lot more from the ideas we pioneered in AlphaGo, the Monte Carlo tree search and other things. So, in actually many respects, we need to to the ideas we have with AlphaGo with today's foundation models. Of course, it's harder because you don't have a

<a id="t-aca697d2f938"></a>
## 01:07:10

**中文**

——完美的世界模型；我们需要更好的世界模型。在游戏中，状态转移矩阵是明确的，这个问题很简单。若要我判断今天的路线，我认为 Gemini 这样的基础模型将是最终 AGI 方案的关键组成部分，随后还会叠加许多有意思的强化学习方法。也许 20 年后，我们最终会拥有更接近 AlphaZero 的系统，由强化学习从零学会一切；但我认为，这实际上并不是——

**English / Original**

perfect model of the world, right? You need a better world model. Uh in games it's trivial, right? The the transition matrix. So, I think that's an issue, but if you're talking about um uh if I was to guess today, I think the foundation models are going to like Gemini going to be a critical part of the ultimate AGI solution. Um and then I think we'll have lots of interesting reinforcement learning on top. Um I I think eventually, one day maybe 20 years from now, we'll have a kind of more like an AlphaZero type system where uh you know, reinforcement learning can learn everything from scratch, right? Um but I think uh actually that's not going to be the

<a id="t-da8d75907ffd"></a>
## 01:07:51

**中文**

——实现 AGI 的最快路径。更合理的方式是利用基础模型学习既有信息，形成对世界运行方式的模型，再在其之上叠加强化学习与规划。至少在第一阶段，这样会更高效。

**English / Original**

fastest way to AGI. I think it makes sense to use the foundation models uh and you know, all the information that's already out there and learn that as almost like a model of how the world works and then do your reinforcement learning and planning uh on top of that. I think that will be uh more efficient in the first case.

<a id="t-46a3b30f2f76"></a>
## 01:08:08

**中文**

主持人：强化学习仍只是“蛋糕上的樱桃”吗？不是。至于“樱桃”的说法，你得去问 Yann，我相信他能详细解释。我从来没有真正同意这种说法；强化学习是基础组成部分。如果用比特衡量，有人会问：“你从强化学习中究竟获得了多少比特的信息？”但在我看来——你或许也同意——不同信息比特的价值并不相等。例如，“是否赢得比赛”这一比特，就比屏幕上的某个随机像素重要得多——

**English / Original**

>> Is it is still going to be the cherry on top? No, I think it's going to be I think it's I mean, you know, you have to ask Jan about his cherry comments. I'm sure he's he can talk at length on that. But I've never really agreed with that. I think it's just a fundamental. I don't really, you know, I think um obviously if you measure it in terms of bits, uh then one can say, "Well, how much bits of information you're getting from the reinforcement learning?" But in my view and presumably your view is that not all bits are equal in terms of information. Of course, if you get a bit about, you know, "Did you win the game or not win the game?" That's much more important than some random pixel on the

<a id="t-67216a261f06"></a>
## 01:08:43

**中文**

——屏幕上的像素，对吧？因此，以一种简单方式把这些比特的信息价值等同起来，显然并不正确。我确信基础模型将是 AGI 的重要部分；问题只是，它们是否已包含所需的一切，还是仅仅构成关键部分。我认为毫无疑问，它们至少会是首批 AGI 系统的核心组成。

**English / Original**

screen, right? So, to equate the the informational value of those bits um in just, you know, trivial way is clearly incorrect in my opinion. But I do think that the foundation models um are going to be The question is is are they um going to be all that's needed or just a critical part of what's needed. I think there's no question they're going to be at least a critical component of, um, the first AGI systems.

<a id="t-46158263cc89"></a>
## 01:09:13

**中文**

[笑声] 我们快没时间了。你想向峰会参与者传达什么信息？我的信息是“谨慎乐观”。我们正处于一场惊人变革的前夜，它将为科学和医学带来巨大收益。人类健康尤其是我热衷的方向，我能看到我们应对人类健康问题的方式发生革命性变化。围绕这些系统，还可以打造许多杰出的公司、工具和产品——

**English / Original**

Uh, [laughter] so So, I I I So, to be we're almost out of time. So, I just want to ask you, what's your message for the attendees of this summit? Well, look, I think my message is uh one of, I would say, cautious optimism. So, I think we're on the cusp of an absolutely incredible transformation um that's going to uh uh bring incredible benefits in science and medicine specifically is what I'm passionate about and I can see uh revolutionizing the way we we we deal with human health. Um, I think there's many amazing companies and and and tools and products to build on

<a id="t-d09913d99ed9"></a>
## 01:09:53

**中文**

——全世界的人都能在这些 AI 系统之上进行创造。但我也要补充一句警示：只要有足够时间和足够多优秀头脑投入，我相信人类的智慧能够解决这些技术问题和技术风险。

**English / Original**

top of these systems and I think uh everyone in the world can can build on uh these AI systems to do that. Um, but then also I would just add a note of caution, which is um I think we will solve these technical issues given enough time and enough um brainpower on it. I believe in human ingenuity and if the best minds work towards that, I think we'll solve the technical risks.

<a id="t-ad0f8fd191f7"></a>
## 01:10:15

**中文**

但这也必须在国际层面推进。因此，社会层面的挑战最终可能比技术挑战更难。谢谢。主持人：感谢 Demis。谢谢。演讲者：感谢部长和各位教授同事邀请我发表主题演讲。能来到这里，我深感荣幸。看到印度各地如此踊跃地参与、对 AI 表现出浓厚兴趣，我非常激动；同样令人振奋的是，来自世界各地产业界和学术界的许多同事也汇聚于此。我期待——

**English / Original**

Um, but uh we also need to do this internationally. So, the societal challenges of that may actually end up being the harder uh problem than the technical ones. Thank you. So, so there is a summit everyone. Thank you. UH OUR FELLOW PROFESSORS FOR INVITING ME here to give this keynote. It's a huge honor uh to be here. Uh, very excited to see the incredible turnout um uh from the country and the interest in AI, of course, but also from many of my colleagues from around the world um in both industry and academia who've convened here. So, I'm looking forward

<a id="t-f201d5edef0e"></a>
## 01:10:57

**中文**

——未来几天围绕 AI 前进方向展开精彩讨论。我们在 2010 年创办 DeepMind，距今似乎只有 16 年，但以 AI 的发展尺度来看，几乎已是漫长的一生。当时几乎无人研究 AI，学术界只有少数人，产业界更几乎没有人。因此，我们在短短十多年里取得的进展令人惊叹。

**English / Original**

to a really amazing couple of days of discussions um about the way forward for AI. When we started uh DeepMind in 2010, um seems uh it's only 16 years ago now, but it's almost a lifetime in terms of AI. Uh almost nobody was working on AI back then, few people in academia, but almost nobody in industry. So, it's been amazing to see the progress we've made in uh just over a decade.

<a id="t-e83e67b31e3b"></a>
## 01:11:24

**中文**

但如今，我认为我们已到达一个临界点：AGI，也就是通用人工智能，可能在未来五至八年内出现。因此，本次峰会召开在一个关键节点：我们开始看到自主性更强、能力也强得多的智能体式 AI 系统，并初步见识这类系统能够做什么。当然，机遇极其巨大。我个人最热衷的是以 AI 推动科学，并借助 AlphaFold 等系统促进科学与医学进步。我认为 AI 能够彻底改变药物发现和人类——

**English / Original**

Um but at this point, I think we're at a threshold moment um where AGI, artificial general intelligence, uh is on the horizon, maybe in the next five uh five to eight years. So, this summit uh comes at a critical moment um as we start seeing more autonomous agentic AI systems, much more capable. I think we're seeing the beginnings uh of of um what those kinds of systems can do. And of course, the opportunities are incredible in um for my my personal passion is using AI for science and to advance science and medicine uh with systems like AlphaFold, I think we can revolutionize drug discovery and human

<a id="t-5784d70ca617"></a>
## 01:12:05

**中文**

——健康，也能助力材料科学、气候变化等领域。不过，它当然也伴随着许多风险。AI 是一种双重用途技术，将成为人类历史上变革性最强的技术之一，甚至可能没有之一。而如今，它已近在眼前。这项技术会影响全世界每一个人。

**English / Original**

health, and also uh areas like material science and climate change uh with the help of AI. But of course, it also comes uh with many risks, too. AI is a dual-purpose technology. It's going to be the most one of the most, if not the most, transformative technology in human history. Um and that's just uh on the horizon now. And the thing about this technology is it's going to affect the whole world, everybody in the world.

<a id="t-572073ea0e45"></a>
## 01:12:31

**中文**

它会跨越国界。因此，举办这样的国际峰会极其重要：让国际社会共同讨论，如何确保这些机遇惠及全世界，以及如何降低风险。我认为，要降低其中一些风险，必须开展国际合作，而第一步就是国际对话。从英国、巴黎、韩国到如今的印度，这些峰会正是为此而设，并且有效召集了许多——

**English / Original**

It's going to cross borders. Um so, it's very important that we have uh summits like this, international summits to bring the international community together uh to discuss um how to uh uh make sure the opportunities benefit the whole world. Um and also, how do we mitigate the risks? Uh and I think in order to mitigate uh, some of the risks, it's going to need international cooperation. And uh, the first step for that is to have it um, international dialogue. And I think uh, that's what these summits um, starting in the UK and then in Paris and then in Korea and now here in India uh, are for and have been uh, really helpful in convening um, uh, many of uh, the

<a id="t-46032b6f421b"></a>
## 01:13:13

**中文**

——重要国家领导人和该技术领域的重要领袖，共同讨论即将到来的变化。感谢邀请我出席，我期待未来几天在这里的讨论。谢谢。[掌声] 大家早上好。

**English / Original**

important heads of state and many of the important uh, leaders in this technology together to discuss um, what is on the horizon. So, uh, with that um, thank you for having me and I look forward to the discussions here in the next couple of days. Thank you. >> [applause] >> Yeah, so good morning everyone.

<a id="t-8f03c47e0463"></a>
## 01:13:53

**中文**

那么，Demis，你刚才谈到……

**English / Original**

And so, let me say you you started talking about how