AI 战略与技能：十年回顾

大家晚上好，欢迎参加 WCIT 信息技术敬拜公司讲座，以纪念前会长 Barney Gibbons。首先，感谢汇丰创新银行对本次讲座的大力支持。

Good evening everyone and welcome to this WCIT the worship company of information technology lecture in memory of pastmaster Barney Gibbons. Firstly a couple of thanks. Thank you to HSBC Innovation Bank for being such great supporters of this lecture.

很高兴来到这里。谢谢。

Great to be here. Thank you.

所以，我在想如何开场，Demis 和 Wendy 的每个视频都会读出他们的完整履历。我问了 Gemini 先生 Demis 的最佳定义是什么。所以 Demis，如果这里有什么不对，那也不是我的错。

So, I was thinking about how to start this and every video of both Demis and Wendy has their whole CV read out. I asked Mr. Gemini what the best definition for Demis is. So, Demis, if anything's a miss here, then I would... Okay. I'm not to blame.

Demis Hassabis 爵士，Google DeepMind 的联合创始人兼 CEO，在我看来这是世界领先的研究实验室之一。我特别兴奋它就在这个国家。你的使命是推进 AI 并将其应用于重大科学和社会挑战。你开发了结合 AI 与经济学的电子游戏《主题公园》，售出 1500 万份。那还是你上大学之前的事。你打造了 AlphaGo 并击败了世界围棋冠军，还有 AlphaFold，它革新了蛋白质结构预测。你是多领域的：计算机科学家、神经科学家、企业家、前电子游戏设计师，为 AI 研究带来多领域的洞察。我们非常高兴你来到这里，Demis。谢谢。

Sir Demis Hassabis, co-founder and CEO of Google DeepMind, and in my opinion one of the world's leading research labs. I'm particularly excited as well that it's in this country. Your mission is to advance AI and to apply it to major scientific and social challenges. You built a video game called Theme Park which combined AI with economics and sold 15 million copies. That's before you even went to university. You built AlphaGo and defeated the world champion Go player, and AlphaFold which revolutionized protein structure prediction. You are multi-domain: a computer scientist, neuroscientist, entrepreneur, former video game designer, and bring insights from multiple fields in AI research. We are very pleased to have you here, Demis. Thank you.

Wendy Hall 女爵士，英国最杰出的计算机科学家之一，网络科学的全球先驱，跨学科研究网络如何塑造社会以及社会如何塑造网络。她是南安普顿大学的皇家计算机科学教授，最早从事多媒体和超媒体系统重大研究的人员之一，与 Tim Berners-Lee 爵士共同创立了网络科学信托基金会。她推动了万维网及其社会影响的理解。她是南安普顿第一位工程学女教授，曾任 BCS 和 ACM 主席。她在 AI、互联网治理、数据政策和数字伦理方面拥有国际尊重的发言权，帮助政府和机构思考技术如何影响社会。Wendy 女爵士，非常欢迎你。

Dame Wendy Hall, one of Britain's most distinguished computer scientists, a global pioneer in web science, cross-disciplinary study of how the web shapes society and vice versa. Regius Professor of Computer Science at the University of Southampton, one of the first researchers to do major work in multimedia and hypermedia systems, a founding leader of the Web Science Trust along with Sir Tim Berners-Lee. You advanced the understanding of the World Wide Web and its social impact. The first woman to become professor of engineering in Southampton, former president of the BCS and the ACM. You are an internationally respected voice on AI, internet governance, data policy, and digital ethics, helping shape governments and institutions to think about how technology impacts society. Dame Wendy, you are very welcome.

这引出了我主持敬拜公司这一年的主题——社区。我选择这个主题是因为我试图理解 AI 和技术将如何影响我们的社区、人民和行业，这涉及到本组织的四大支柱：慈善、工业、教育和联谊。所以今天我将向两位提问关于技能和 AI 素养的问题。距离你们的 2017 年评估已近十年，从现在起，一个严肃的国家 AI 战略应该是什么样子，从学校到大学到职业生涯中期再到我们人员的再培训？我想知道，英国是否仍然忽视或误解了这一技能挑战？

Which leads me to my theme in my year of mastership of the Worshipful Company, which is community. I chose the theme because I was trying to understand how AI and technology are going to impact our communities, our people, our industry, which touches the four pillars of this organization: charity, industry, education, and fellowship. So today I'm going to pose my first question on skills and AI literacy to both of you. Nearly a decade after your 2017 review, what does a serious national AI strategy look like now from schools to university to mid-careers to reskilling our people? And I was just wondering, is the UK still missing or misreading this skills challenge?

哦，这实际上是一问三个问题。真不敢相信已经十年了。Demis 当时也在场。我本想和他共同主持，但他太忙了。你记得吗？是的，我记得。我们尝试提出意见。确实如此。但他不肯共同主持。这很合理，他当时在打造 DeepMind，实现梦想。所以不幸的是，我的回答不太乐观。英国做得很好。实际上，我喜欢说我们是第一个制定国家战略的。其实加拿大比我们早，但我认为我们的更知名。我们在 2017 年完成了整个 Hall-Pesenti 报告，然后从中成立了 AI 办公室和 AI 委员会，制定了国家战略。该战略在 2021 年新冠期间演变而来。我曾担任过一段时间的 AI 技能冠军，但现在没有 AI 技能冠军了。没有 AI 委员会，没有独立机构监督 DC 那 400 人在 AI 部门的工作。我认为英国失策了。许多国家采纳了我们的战略，复制了它。你可以在其他国家的战略中看到我们的元素和措辞。例如，新加坡遵循了我们的路线图，并且做得更好。他们在开发 AI 的小岛国中正在超越我们，实力远超其规模。他们在排行榜上超越我们。很明显：美国、中国，然后我们曾是 AI 战略和采用率排名的第三位。我们下滑了，而新加坡做得非常好。其他国家也在做。我们在一些方面做得不错，稍后可以讨论。但技能方面非常零散。Hall-Pesenti 审查提出了几项建议，所有建议都得到推进。MSC 的博士项目，学校方面很少。学校几乎没有关于 AI 的指导。你会看到事情以相当零散的方式出现。目前这个国家没有对 AI 技能的监督。很抱歉这么说，但这是事实。

Oh, well, that's three questions in one. I can't believe it's 10 years. Demis was there when we did that. So, I wanted to co-chair it with him, but he was already too busy. Do you remember? Yes, I remember. We tried to input into it. You did. But he wouldn't co-chair it. That's fair enough. He was building DeepMind, living the dream. So unfortunately my answers aren't very positive. The UK did very well. In fact, I like to say we were the first to develop a national strategy. Actually Canada beat us to it, but I think ours is better known. We did the whole Hall-Pesenti report in 2017, and then the national strategy was developed by what we set up out of that: the Office for AI and the AI Council. The national strategy evolved from that in 2021 during the COVID years. I was AI Skills Champion for a while, but there is no AI Skills Champion now. There is no AI Council, no independent body overseeing what the 400 people in DC do in the AI unit. I think the UK lost a trick. Many countries picked up on our strategy, copied it. You can see elements, words of it in other countries' strategies. Singapore, for example, followed our playbook and did it better. They are now overtaking us in the league of developing AI for a tiny island, and they punch hugely above their weight. They are overtaking us in league tables. It's very clear: it's US, China, and then we were third in the tables of AI strategies and adoption. We've slipped, and Singapore has done it incredibly well. Others are doing it too. We do well in some things, and we can talk about later. But the skills piece is very patchy. Several things came out of the Hall-Pesenti review; all recommendations were taken forward. MSC's PhDs, very little for schools. Very little guidance for schools on AI. You see things coming out in a fragmented way. There is no oversight of AI skills in this country at the moment. Sorry to say that, but it's true.

你呢，Demis？你对此怎么看？

How about you, Demis? What's your view on that?

嗯，我可能会建议一些更激进的做法。我同意 Wendy 的看法。我们在 2017 年有一个有趣的先发优势，但 AI 领域的十年相当于大多数行业的一个世纪，甚至两个世纪或一千年。所以很多都变了。我认为我们做得不错。我记得我们资助了一些硕士学位，这帮助很大。我们看到今天伦敦和英国在初创企业方面做得很好。我认为我们试图以 DeepMind 为基础，吸引了许多人才。现在十年后，其中许多人正在创办其他公司，这很棒。我们需要做得更多。但具体到技能和学校，我完全同意 Wendy。我们需要关注中学阶段，甚至可能是小学阶段。

Yeah, well, I'd probably suggest something much more radical. So, I would agree with Wendy. We had an interesting head start in 2017, but 10 years in AI is like a century in most industries, or maybe two, maybe a millennia. So a lot has changed. I think we did a good job. I remember we sponsored some of these masters degrees, and that helped a lot. We see today London and the UK are doing really well in things like startups. I think we tried to anchor that with DeepMind and brought a lot of talent into the tree. Now, 10 years later, a lot of those people are spinning out other things, which is great to see. We need to do more of that. But specifically on skilling and schools, I absolutely agree with Wendy. We need to look to the secondary school level, maybe even the primary school level.

马已经跑了，意思是大家都在用大语言模型和聊天机器人做作业之类的事，但非常不成体系。我觉得政府政策和教育体系有点像说：“哦，我希望这能消失”，但它不会。我们必须接受它，帮助老师、学校、家长和学生弄清楚如何从中获益，而不是像现在这样让孩子们用它来作弊或走捷径。我们知道这会导致传统技能缺失，更不用说以后的 AI 和计算机技能了。但我认为这是一个机会。也许是时候让你写一份新的评估报告了，Wendy。我的建议是：我们应该颠覆课堂。如果我来负责教育，我会这样处理。我会说学校应该着重培养人际交往能力、项目能力、创业精神、创造力、判断力——这些是你认为 5 或 10 年后 AGI 到来时重要的技能。我知道我们会争论 AGI，但即使只是 extrapolation 今天 AI 的能力，也会有 AI 擅长的事情和需要人类判断与品味的事情。但这些现在学校里并不教。所以我认为课堂上应该注重人际接触。如何把人组织成项目小组，就像 Montessori 学校那样？或者像牛津剑桥和顶尖大学那种一对一辅导，但把它带入课堂。课外时间则进行死记硬背，由个性化 AI 辅助，根据学生的具体速度定制。课堂上搞一刀切的死记硬背是很疯狂的。行不通，因为你必须迁就平均水平。落后或超前的孩子怎么办？两边都不好过。有了 AI，你可以个性化，让每个人得到他们需要的定制课程。然后学校里全是项目。我觉得老师们会喜欢的。你可以有世界级的老师，像明星一样讲课。为什么我们期望每个学校老师都精于讲授同一门课？那是冗余的工作。我和我的老师朋友聊天时，他们每个人都要教 GCSE 数学，然后自己做幻灯片。为什么我们还停留在这个世界？应该统一。然后在家以自己节奏用 AI 远程学习。这就是我会做的。但我不知道像新加坡或阿联酋这样的国家会不会做。

The horse has bolted in the sense they're all using the LLMs and chatbots in their spare time and for homework and things like that, but it's very unsystematic. I feel like government policies and education systems are a bit like, oh, I wish that would go away, but it's not. We're going to have to lean into it and help teachers, schools, parents, and students work out how to get the benefits out of it, not just probably the way that people are using it now — kids using it to cheat on their homework or shortcut things. And we know that will result in a deficit of traditional skills, let alone AI and computer skills later on. But I think we have an opportunity here. Maybe it's time for a new review for you to write, Wendy. This is what I would suggest: we need to invert the classroom. I would lean into this if I was in charge of education. I would say school should be about people skills, projects, entrepreneurialism, creativity, judgment — all the things that you might expect to be important in 5 or 10 years when AGI arrives. I know we'll debate about AGI, but even extrapolating today's AI, there will be things AI can do well and things that need human judgment and taste. But these are not taught in schools today. So I think in class it should be about human contact. How do you organize groups of people into projects, like a Montessori school plus? Or maybe you get at Oxbridge and top universities these kinds of individual supervisions, but you bring that into the classroom. Out of the classroom is where you do rote learning, aided by personalized AI that tailors the rote learning to the exact speed of the student. It's crazy that we do a one-size-fits-all approach to rote learning in class. It doesn't work because you have to cater to the average. What about the kids behind or ahead? Good luck on both sides. With AI, you can personalize it so everyone gets exactly the tailored curriculum they need. Then in school, it's all about projects. I think teachers would love that. You could have world-class teachers who are like superstars doing the lecturing. Why do we expect every school teacher to be amazing at lecturing the same course? That's redundant work. When I talk to teacher friends, each has to teach GCSE maths and create their own slide decks. Why are we in that world? It should just be one. Then remote learning you do at your own speed at home with AI. That's what I would do. But I don't know if countries like Singapore or the UAE will do this.

我能接着这个说吗？我几乎同意你说的所有内容。我认为死记硬背不一定非得在家里做，而且这没有公平性——很多孩子家里没法做。我完全同意我们可以用 AI 来打破一刀切。一个老师教 30 个学生，他们各不相同。如果学生已经做好了基础准备，老师就能教得更好，无论在家里还是在学校。我一直坚信我们可以用计算，现在是用 AI，来开发个性化的辅导系统，帮助每个孩子达到同等水平，但这需要政府巨大的投入。我很乐意帮忙实现。

Can I riff off that? I agree with almost everything you've said. I think the rote learning doesn't have to be done at home, and there's no equality around this — a lot of kids can't do this at home. I definitely agree that we can use AI to stop the one-size-fits-all. One teacher, 30 pupils, they're all different. Teachers can teach much better if the children are receptive to what they're going to say and have already done the basic stuff, whether at home or school. I've always been passionate that we can use computing and now AI to develop personalized tutoring systems for kids that help everybody get to the same level, but that takes a huge commitment from government. I'd love to help do that.

那再往前一步：上大学——计算机科学学位现在是不是变得相当多余了？当然不是。

And then taking it one step forward: going to university — is it now become pretty redundant that we have a computer science degree? Because of course not at all.

当然不是。你绝对需要投身于 STEM 和计算机科学，因为不管怎样，编程只会变成更高级的语言——你可以这样看待编程的未来。我们以前用机器码编程，然后是 C，然后是 Python，现在可能要变成英语。但这没问题。你仍然需要知道如何架构系统以及软件工程的最佳实践。那些理解深层技术的人能比没有技术知识的人更高效地使用这些工具。所以我们当然需要人应用，但也需要全方位的 AI。在南安普顿，我们正在开发一个叫做 AI 驾驶执照的东西，面向十月入学的所有学生——主要是非理科生，让他们了解如何使用 AI。我们有关于如何在教育中使用 AI 的政策。我们还给自己设定了一个目标：到下一个学年，每个学科、每个学位都会有一个专为该领域定制的 AI 模块，无论是历史、心理学、地理、物理等等。这是个巨大的任务，但这就是我们设定的目标。

Of course not. You absolutely need to lean into STEM and computer science because no matter what, it's just a higher-level programming language — that's how you can think about what programming is going to become. We used to program in machine code, then C, then Python, and now it's going to be English. But that's fine. You still need to know about architecting things and software engineering best practices. Those who understand the deep technical will be able to use these tools 10 times more effectively than those who don't have that technical knowledge. So absolutely we need people to apply, but we also need AI across the board. At Southampton, we're developing something called an AI driving license for all students arriving in October — non-science students, to get them up to speed on how to use AI. We have a policy on how to use AI in education. And we've set ourselves a target that by next academic year, every discipline, every degree will have a module on AI customized to that, whether history, psychology, geography, physics, etc. That's a huge task, but that's what we've set ourselves.

那么想必你还需要调整课程设置吧？

And presumably, you're going to have to tweak the curriculum as well.

我正想说，我认为我们需要教授 STEM 和计算机科学。这是非常有价值的培训，但课程必须改变。可能应该学习非常不同的东西。

I was just going to say that I think we need to teach STEM and computer science. That's very valuable training, but the curriculum has to change. There might be very different things that one should study.

当然。嗯，我的意思是，如果你谈论——我不知道你最近有没有听过 Mike Waldridge 的演讲，关于它将如何改变计算机科学。人工智能被作为计算机科学的一部分教授。我们开设了人工智能学位；很多大学已经有了。我们长期以来一直用人工智能教授计算机科学，但现在我们有了一套体系。我们有计算机科学，也有人工智能。而且我热衷于将人工智能的信息传递给非理科学生。

Absolutely. Well, I mean, if you talk — I don't know if you heard Mike Waldridge talk recently about how it's going to change computer science. AI is taught as part of computer science. We do an AI degree; a lot of universities already have it. We've taught computer science with AI for a long time, but we now have a set. We have computer science and we have AI. And I'm passionate about getting information about AI to non-science students as well.

特别是风险和其他的社会影响，这些也必须要纳入课程中，大概吧。

And particularly the risks and the other social effects of doing this will have to be brought into the curriculum, presumably.

是的。

Yep.

哦，抱歉。

Oh, sorry.

是的。我是说，

Yes. I mean,

而且这也是全方位的。你知道，如果我没记错的话，这已经是你最初报告的一部分——伦理和挑战与计算机科学的技术方面一起教授。我认为所有计算机科学家都需要接受这方面的培训。

And that's across the board as well. You know, I think that was already part of, if I remember correctly, your original report — ethics being taught and challenges alongside the technical aspects of computer science. I think all computer scientists need to be trained in that.

嗯，顺便说一句，我也认为现在是时候重视人文学科了，比如哲学、经济学。我认为在我们即将进入的后 AGI 世界，那种思维真的需要。那在我看来是一个新时代。所以我觉得那种思维将会非常需要。我想知道你需要多长时间。Demis 实际上提到了 AGI 和后 AGI。对于这里未入门的听众，你能让我们感受一下 AGI 会如何改变事物，你看到的益处，以及时间尺度吗？我知道你说 2030 年及以后，但多做一些详细说明会非常好。

Um, and by the way, I also believe that the time is now for the humanities like philosophy, economics. I think we really need that kind of thinking in the world we're about to enter into, post-AGI. That's a new era in my opinion. So I feel like that type of thinking is going to be really needed. I was wondering how long it would take you. Demis has actually mentioned AGI and post-AGI. For the uninitiated audience here, could you give us a feel of how things will change with AGI, the benefits as you see it, and also the time scales? I know you say 2030 and onwards, but some more detail would be very good.

是的。首先，我对 AGI 的定义是一个可以测试的系统，它表现出人类拥有的所有认知能力。这很重要的原因是，人类大脑是我们所拥有的通用智能的唯一存在证明。我相信大脑是一个近似的图灵机。图灵证明了通用图灵机可以计算任何东西，所以这是可计算的。我认为 AGI 系统应该能够做到最优秀的科学家能做到的创造性和发明性的事情。今天的系统远不及此；它们不够一致，并且有缺失的能力。它们令人印象深刻，但缺少一些组件，这些可能通过 Scaling 来解决，但也可能需要一两个更大的突破，比如 Transformer，或者将强化学习应用于深度学习，就像我们用 AlphaGo 所做的那样。尚无定论。考虑到投入的资源和人脑力，我预计 AGI 将在大约四五年后出现，也就是 2030 年左右。当这种情况发生时，它将变成一种我们从未见过的新型工具或系统，我们需要思考如何使用这样变革性的技术。它将改变一切：经济，甚至人类状况。积极的一面将是推动科学进步、治愈疾病、推进医学。这就是为什么我整个职业生涯都致力于 AI 和 AGI。但也存在挑战，尤其是随着系统变得更加自主。我们如何让它们保持在设定的护栏内？还有关于意识和其他问题的疑问，但第一步是正确制作这些工具。

Yeah. I mean, firstly, my definition of AGI is a system that you can test and that exhibits all the cognitive capabilities humans have. The reason that's important is the human brain is the only existence proof we have of general intelligence. I believe the brain is an approximate Turing machine. Turing showed that universal Turing machines can compute anything, so that's computable. I think an AGI system should be able to do the creative and inventive things that the best scientists can do. Today's systems are nowhere near that; they're not consistent enough and have missing capabilities. They're very impressive, but they lack some components that may be solved with scaling, but may also require one or two more big breakthroughs, like transformers or reinforcement learning applied to deep learning, as we did with AlphaGo. The jury is out. Given the resources and brain power going into it, I expect AGI around four or five years from now, so around 2030. When that happens, it becomes a new type of tool or system we've never seen before, and we'll need to think about how to use such transformative technology. It will change everything: economies, perhaps even the human condition. The positive part would be advancing science, curing diseases, advancing medicine. That's why I've spent my whole career on AI and AGI. But there are challenges, especially as systems become more autonomous. How do we keep them on the guardrails we set? There are also questions about consciousness and other things, but the first step is to get these tools right.

而且你有一个有趣的……嗯，我不同意 Dennis 说的所有话。我尊重他所做的，但如果每个人都像 Demis，我们会没事，但他们不是。有些人只是为了钱或权力，他们会发布任何东西，说它能做一切，没有任何测试、评估、安全检查或指南。这目前在美国正在发生。讽刺的是，我喜欢你在达沃斯与 Dario 的辩论。我称之为“Demis 和 Dario 在达沃斯辩论”。我给我的学生用这个。Demis 说了他刚才说的话，而 Dario 在那里为他的公司赚钱，现在要 IPO，说他将在年底前达到 AGI。而你试图把他拉回来，对吧？

And you have an interesting... well, I don't agree with Dennis on all this. I respect what he does, but if everybody was like Demis, we'd be all right, but they aren't. There are people in this for the money or the power, and they'll release anything and say it does everything without any testing, evaluation, safety checks, or guidelines. This is happening in the States at the moment. Ironically, I love that you did a debate with Dario at Davos. I call it the Demis and Dario at Davos debate. I use it with my students. There was Demis saying what he just said, and Dario out there to get money for his company, now going IPO, saying they'll reach AGI by the end of this year. And you were trying to pull him back, right?

嗯，实际上，让我回溯一下。对我而言重要的事情是——我不接受，但如果我们到 2030 年达到了某种形式的 AGI，我个人不站在那一边，但如果是这样，那我们就剩四年时间来搞定 AI 的全球治理。否则，我们就会有在某些方面比我们更聪明的机器，而且重要的事情是，如果我们构建的软件开始自行决定做事——那就像“对不起，Dave，我做不到”的时刻。这将不可避免地发生。斯蒂芬·霍金在 2014 年说过：“如果我们建造出聪明的机器，那可能是人类的终结。”存在威胁。我们绝对必须建立一个全球治理体系。我曾是联合国高级别咨询委员会的成员。但人数不够。联合国已经不是原来的样子了，而且它不会是联合国的，但它可能从联合国中派生出来。

Well, actually, let me roll back. The big thing for me — and I don't accept it, but if we reach some form of AGI by 2030, which I personally am not on that side of the fence, but if so, that gives us four years to sort out global governance of AI. Otherwise, we've got machines that are cleverer than us in some aspects, and the big thing is if the software we're building starts to decide to do things on its own — the 'I'm sorry, Dave, I can't do that' moment. This will inevitably happen. Stephen Hawking in 2014 said, 'If we build machines that are clever, that could be the end of the human race.' There are threats. We absolutely have to have a global governance in place. I was on the UN high-level advisory board. It won't be enough people. The UN isn't what it used to be, and it won't be of the UN, but it could spin out of the UN.

今年七月将启动一场全球对话，一些政府将聚在一起，真正讨论这意味着什么。就在本周，特朗普食言了，说我们必须开始评估即将推出的软件。他推翻了拜登的规定，发布了一项行政令，声称不进行监管，并阻止了各州的监管。但他本周又反悔了，我认为转折点是 Anthropic 的 Mythos 软件。

There's the global dialogue starting in July where some governments are going to come together and start really talking about what this means. And just this week Trump went back on his word and said we have to start evaluating the software that's coming out. He overturned Biden's rule and did an executive order that said no regulation, and he stopped the states regulating. But he went back on that this week, and I think the tipping point was the Mythos software from Anthropic.

还不止这些。很多人都在幕后投入时间。Mythos 是一个非常有趣的触发点，因为它让银行和政府感到担忧，人们这才开始认真对待。当银行开始担忧时，政府就醒了。这意味着有一个行动的机会窗口，这也是为什么最终通过了行政令。但这还不够。

It's not just that. A lot of people are spending time inputting behind the scenes. Mythos has been a very interesting trigger because it worried the banks and the government, which is when people take things seriously. When the banks get worried, the government wakes up. That means there's a window of opportunity to do something, and that's why we saw the executive order eventually get through. But it's not enough.

我确实认为需要国际标准。我的观点是，到 2030 年 AGI 出现的概率是 50%，存在一些误差。如果递归自我改进有效，它可能会更快到来；如果无效，则可能需要更长时间。但我们没有合适的机构来应对如此重大的事情。地缘政治是 30 年来最复杂的，比以往任何时候都更加碎片化。联合国已今非昔比。时间不多了，哪怕是 10 年。我们必须认真对待；这很紧迫。Mythos 的网络问题只是个开始。

I do think there needs to be international standards. My view is that there's a 50% chance of AGI by 2030, with margins of error. It could happen sooner if recursive self-improvement works; if not, it may take longer. But we don't have institutions fit for purpose to deal with something as consequential as this. The geopolitics is as complex as it's been in 30 years, more fragmented than ever. The UN is not what it was. There's not a lot of time, even if it's 10 years. We have to take it seriously; it's urgent. Mythos cyber is just the beginning.

我更担心的是即将出现的生物风险。我一直在用 AlphaFold 致力于治愈疾病，但这些工具在坏人手中也可能被滥用。我担心两件事：一是流氓行为者将通用技术用于有害目的，这甚至影响到了开源；二是技术问题，即如何确保系统在变得更自主、更强大时仍然遵守我们设定的护栏。然后是经济问题：如何让全社会广泛受益，而不仅仅是少数公司和个人？除此之外，还有关于意义和目的的哲学问题。有一整堆问题，一个比一个复杂，我们早就该动手了。

I'm more worried about things like biorisk coming down the line. I work closely on curing diseases with AlphaFold, but those kinds of tools can also be misused in the wrong hands. There are two things I worry about: misuse of general purpose technologies by rogue actors for harmful ends, which even affects open source; and the technical question of ensuring systems stick to our guardrails as they get more autonomous and powerful. Then there are economic questions: how does all of society broadly benefit, not just a few companies and people? And beyond that, the philosophical question about meaning and purpose. There's a whole stack of issues, each more complex than the last, and we've got to start working on them yesterday.

所以我和其他人已经开始给它命名了。我们将创建一个国际 AI 机构，姑且这么叫吧。它可以脱胎于联合国的工作，但不属于联合国，它需要独立。我们需要关心此事的人——包括科技公司、政府和公民社会——都参与进来。还得有人出钱。我们不能坐视不管，等着别人去做。我七月会去日内瓦参加 AI 会议。

So I've started naming it with others. We're going to create an international AI agency. Let's just call it something. It can spin out of the work the UN is doing, but it won't be of the UN; it needs to be independent. We need the people that care about this, including tech companies, governments, and civil society, to sign up. Someone has to fund it too. We cannot just sit back and say someone else will do it. I'll be at AI in Geneva in July.

我不确定。

I'm not sure.

哦，是的，没错。

Oh, yeah, exactly.

Demis，印度那边什么结果也没有。25 万人参加的会议。人们在气候变化 COP 会议上努力了多久？空谈是不够的。我只是认为联合国目前没有能力做任何事，所以需要新的机构。

Nothing came out of India, Demis. 250,000 people at a conference. How long have people been trying to do climate change COP meetings? Talking shops are not going to be enough. I just don't think the UN has the power to do anything currently, so it's going to need to be new institutions.

确实，但这就是为什么英国苏纳克政府发起的 AI 峰会有帮助；它们至少让国际社会聚在一起讨论这些话题。但我同意，空谈是不够的。我们很多人都在幕后努力穿针引线。幕后建议和策略，以及正面应对，两者都需要。

Sure, but that's why the AI summits started in the UK with the Sunak government were helpful; they at least get the international community together to talk about these subjects. But I agree, talking shops are not enough. A lot of us are working behind the scenes to thread that needle. Both behind-the-scenes advice and maneuvering, and then head-on tackling, are needed.

我们还没有提到中国。

We haven't mentioned China either.

我们很快就会谈到中国。

We will come to China soon.

但说到大型科技公司真正参与其中，这难道不像是火鸡投票支持过圣诞节吗？

But in terms of the big tech companies actually taking part in this, is it not like turkeys voting for Christmas?

不，当然不是。事实上，他们确实希望如此。所有实验室负责人——我知道他们都是有个性的人，就这么说吧——但没人希望发生灾难性事件。这对任何公司、个人、声誉或业务都没好处。所以即使有些人确实在科学上感到担忧，而且一直如此——我很早就了解他们的动机。像 Dario 这样的人，我们当博士后时经常聊这些；其他人是后来才来的，可能更关心权力和金钱。但无论你关心什么，你都不希望发生灾难性事件，政府也不希望。即使是最资本主义、最市场驱动的政府也不希望，坦率地说，威权政府也一样。所以还是有希望的，因为达成某种协议符合每个人的利益。问题在于常见的公地悲剧、囚徒困境问题，这就是为什么我们需要某种治理和政府的参与。因为对公司来说，总有人可以通过背离商定的协议来赚钱。这就是问题所在。就个人而言，我认为我与之交谈过的所有研究人员、实验室负责人和公司负责人都希望有所行动，并且希望有一些护栏和标准。问题在于这些是什么，它们应该如何运作，以及由谁来管理。但我和其他人正在幕后向一些主要政府提出建议。

No, of course not. In fact, they want it actually. All the lab leaders — I know they're all characters, let's put it that way — but nobody wants catastrophic things to happen. That's no good for any company, any individual, any reputation, or for business. So even if some of them are genuinely worried scientifically and always have been — I know their motivations from a long time back. People like Dario, we used to chat about these things when we were both postdocs; others have come lately and maybe they're more interested in power and money. But it doesn't matter what you're interested in. You don't want catastrophic things happening, nor does the government. Even the most capitalist, market-driven governments don't want that, or authoritarian ones frankly. So there is hope because it's in everyone's interest to come to some agreement. The problem is the usual tragedy of the commons, prisoner's dilemma issues, which is why we need some kind of governance and government involvement. Because companies, it's always in someone's interest to make money by defecting from an agreed protocol. That's the problem. Individually, I think all the researchers, lab leaders, and company leaders I talk to in this space do want something to happen, and do want some guardrails and standards. The question is what those are, how they should operate, and who should run them. But others and I are making proposals behind the scenes to some of the main governments.

很高兴听到这些。谢谢。

That's great to hear. Thank you.

说到——你刚才说大型科技公司，指的是美国西海岸的那些吗？

Just moving to — when you said the big tech companies, you meant the ones on the west coast of America?

是的。

I did.

好。那么，你难道不关心中国的大型科技公司以及它们在做什么吗？

Yeah. So, don't you care about the big tech companies in China and what they're doing?

我们关心。

We do.

是的。我是说，中国很乐意运营我们可能建立的任何机构。他们会这样做，因为——正如 Demis 所说——他们同样不希望 AI 出现灾难，出于许多不同的原因，和西方一样的原因。但实际上，他们会很乐意运营我们将要建立的任何机构。关于中国还有一点——抱歉我展开了，因为我认为你无法回避它——是他们很大程度上自己生产模型。他们在生产开放的代码，更加开放，这有好有坏，但他们比美国政府更欢迎这些想法。

Yes. I mean, China would love to run whatever agency we might set up. They would do it because — as Demis said — they are as keen not to have a catastrophe in AI, for many different reasons, the same reasons as the West. But actually, they would love to run whatever we're going to set up. And the other thing about China — I'm sorry I'm going into it because I don't think you can avoid it — is they largely produce their models. They're producing open code, much more open, and there's good and bad about that, but they are much more welcoming of these ideas than say the US government is.

好的。既然我们谈到了中国，我在想，AI 领域的很多投资和活动都集中在美国和中国，我们（英国）作为一个国家，会不会被抛在后面？我们的公司是否只能旁观并成为技术的使用者，而不是像你 Demis 那样成为这些技术的发明者？

Okay. Since we're talking about China, I was just wondering whether a lot of the investment and a lot of what's happening in AI is in the US and China, and are we going to just be, as a country, left behind? Will our companies just be watching and be users of the technology rather than, like you, Demis, being inventors of such technologies?

我们不能——我认为这点很重要——我告诉过许多欧洲领导人，你不能仅仅成为监管方面的世界冠军。那不可能。确实如此。我在 2017 年左右告诉马克龙，他试图做点什么。你不能只在监管上做世界冠军。为了在谈判桌上有一席之地，你还必须在技术上领先，拥有领先的公司和学术机构，这样你才能提前看到发展趋势，并且拥有某种道德权威，让别人倾听你的意见。曾经有一段时间，我试图推动英国、法国和加拿大三方合作，因为我认为这三个国家在科研上是仅次于两个超级大国（美国和中国）的。实际上，从引用、论文、研究人员来看，这三个国家加起来很强。现在你可以加入其他欧洲国家和新加坡，但这三个国家可以在两个超级大国之间建立一个温和的壁垒。由于政治等原因没有完全实现，但机会仍然存在。我认为英国仍然具有独特的优势。我们尽了自己的一份力——显然我们是 Google 的一部分，但我把 DeepMind 总部设在这里，我们在这里雇用了数千名顶尖研究人员。他们在 DeepMind 做了出色的工作。其中许多人在五、六、十年后离开并创办了新公司。一旦你在一个地方扎根，你往往会留在那里——家庭原因、其他事情、你喜欢这个城市。我认为我们正在看到这一点。所有生态系统都需要这样的锚点。硅谷在惠普、英特尔等公司入驻之前并不是硅谷，而且它们也来自附近的强大大学：斯坦福、伯克利。我们这里也有：金三角——剑桥、牛津、帝国理工、伦敦大学学院。我们拥有这么多强大的大学。这些东西需要几个世纪才能建立起来。我们拥有这些资产，拥有非常聪明的人。我们只需要有更大的雄心并在这里利用他们，希望政府能鼓励或者至少不阻碍。我认为我们在这里做得不错，但可以做得更好。下一个阶段是公司成长阶段的融资，这目前缺失。我认为证券交易所需要更新。我不知道为什么公司不再在伦敦证券交易所上市，它们都去美国了。显然存在一些问题需要解决。我认为我们可以利用这个时刻。我们有所有正确的要素。另一件大事是我们必须解决能源成本和能源依赖问题，部分出于安全原因。也许好事会在这里汇聚。但我们拥有西方世界最昂贵的能源。问题在于这将直接转化为智能——字面意思是从瓦特到美元到 tokens。所以你不能发展我们的产业。这是新的工业基础，对吧？如果你相信 AI 将带来什么——数据中心、信息、tokens。

Well, we can't be — I think it's important that — and I've told this to various European leaders — you can't just be world champions at regulation only. That's not a thing. Exactly. I told Macron that back around 2017 and he tried to do something about it. You cannot be world champions at regulation only. So to have a seat at the table, you've got to also be leading in the technology and have leading companies and leading academics, so that you can see what's coming down the line ahead of time, and also have a sort of moral authority to be listened to. There was a time when I was trying to encourage a sort of trilateral thing with Britain, France, and Canada, because I would say those three countries were the next best at research after the two superpowers, US and China. Actually, as a three, if you look at citations, papers, researchers, those three countries together are pretty strong. Now you could include some other European countries and Singapore, but those three could create a kind of moderate bulwark between the two superpowers. It didn't quite happen for political and other reasons, but there's still an opportunity. I think the UK is still uniquely positioned to take advantage of this. We've tried to do our bit — obviously we're part of Google, but I kept DeepMind headquartered here, we hired thousands of top researchers here. They did great work at DeepMind. Many of them then after five, six, ten years went and started other things. And once you embed yourself in a location, you tend to stay there — family reasons, other things, you like the city. And I think we're seeing that. All ecosystems need anchors like that. Silicon Valley was not Silicon Valley until Hewlett Packard, Intel, all these companies seeded it, and they also came from strong universities nearby: Stanford, Berkeley. We have that here: the Golden Triangle — Cambridge, Oxford, Imperial, UCL. We have so many strong universities. These things take centuries to build up. We have those assets. We have very smart people. We just need to have more ambition and utilize them here, and hopefully government will encourage that or at least not get in the way. I think we're doing pretty well here, but we could do better. The next stage is financing the growth stage of companies, which is missing. I think the stock exchange needs updating. I don't know why companies don't float on the London Stock Exchange anymore; they go to the US. There's obviously some problem there that needs to be addressed. I think we could take advantage of this moment. We have all the right ingredients. The other big thing is we've got to address energy costs and our energy dependence partly for security reasons. Perhaps there's going to be a confluence of good things here. But we have the most expensive energy in the Western world. And the problem with that is it's going to directly translate into intelligence — it's literally watts to dollars to tokens. So you cannot build out our industry. This is the new industrial base, right? If you believe what's going to happen with AI — data centers, information, tokens.

每个单位成本的推理能力能做到多少？这事现在至关重要，对所有行业都很关键，对未来的行业也一样。如果我们谈论气候和环境，虽然数据中心之类的东西让人担忧，但实际上我觉得那是个伪命题——如果我们把技术做对，AI 本身就会帮我们实现解决气候问题的技术。我们在做核聚变、材料设计、更好的太阳能等等，我们将推动这场新的复兴，那会彻底改变气候等式，而不是现在试图节省 1% 或 2%。

How much can you inference per unit cost? And so this is vital now. It's vital for all our industries, but it's going to be vital for the industries of the future. So, and then that will pay itself if we're going to get to climate and environment, but obviously there's a worry about that with the data centers and things like that. But actually I just think that's a red herring because if we build these technologies right, AI will help us deliver the technologies that will solve climate change. So we're working on fusion, material design, better solar power, all sorts of things. We're going to advance this new renaissance, and that will completely change the climate equation, not trying to shave off 1% or 2% right now.

我能接一下话吗？顺着你的话，我认为未来我们都会希望城市或小镇里有数据中心，因为它们会成为发电站。这是未来的方向。另一点我想说的是，我们在这方面很擅长，接下来几周你会看到不少动静。首先，感谢里希·苏纳克，我们有了世界上最棒的 AI 安全研究所。他们在为评估科技公司的产出制定基准。我希望他们也覆盖中国，但那是另一个问题。我们是全球 AI 评估、测量与科学网络的一部分（因为特朗普不喜欢“安全研究所”这个名称，所以改了名）。我们还资助了国家物理实验室新的 AI 测量中心。这推动了 AI 科学本身的发展——不是 AI 辅助科学，而是 AI 的科学——我认为这个国家非常善于引领这类新科学。这一切都属于 AI 保证的范畴，我们也很擅长。下周伦敦科技周上，政府会宣布将其扩展为一种职业，不仅面向技术人员，也面向将运营 AI 保证行业的人。如果我们继续推进，英国可以在这方面领先，我认为本届政府做得对。

Can I come into that? Riffing off of that, I think in the future we will all want a data center in our city or town because they will become generators of electricity. That's where it's going to go. The other thing I want to say is that we are good at here, and you're going to see quite a bit about it over the next few weeks. First, thanks to what Rishi Sunak did, we have the best AI security institute in the world. They are setting the benchmarks for evaluating what's coming out of tech companies. I wish they covered China as well as America, but that's another issue. We are part of a global network now called the network of AI evaluation, measurement and science (they changed the name because Trump didn't want them called security institutes). We also funded a new center for AI measurement at the National Physical Laboratory. This drives the whole science of AI, not AI for science but the science of AI, which I think we are very good at leading in this country. And it all comes under the umbrella of AI assurance, which we are also very good at. You'll see an announcement at London Tech Week next week from the government about expanding that as a career, not just for technical people but for those who will run the AI assurance industry. We can lead on that in the UK if we keep going, and I think this government has got that right.

完全同意。我也想表扬一下 AI 安全研究所（在英国现称安全研究所），它是世界领先的，我们在这方面非常强。我认为我们的学术基础也能为此做贡献。这类工作确实需要非企业组织来做。这正在引领方向，也是为未来任何标准机构储备技术能力的一种方式。你需要高水平的技术人员，而这些人员可以来自这些安全研究所网络。

100% agree on that. I also want to praise the AI safety institute, now called the security institute in the UK. It is world-leading, and we are extremely good at that. I think that's something our academic base can feed into as well. You actually want non-company organizations doing that type of work. That is leading the way, and it's one way towards having the technical capability for whatever standards body there is. You need the highly technical folks, and those could come from these security institute networks.

太好了，谢谢。我要换个话题。我想问你，德米斯，你提到了我们这里的人才力量。我们有一个不错的人才库。我特别想知道你为什么在英国创办 DeepMind？你仍然在伦敦运营 DeepMind，这对我们公司来说也很棒。但我想问的是：我们如何找到新的英国突破，并把它们变成全球性的企业？

Super, thank you. I'm going to switch to a slightly different topic. I want to ask you, Demis, you mentioned the power of the brain we have here. We have a good talent pool. I was very keen to understand why you set up DeepMind in the UK? You still run DeepMind in London, which is phenomenal for us as a company as well. But I wanted to ask you: how do we find new British breakthroughs and create global businesses out of them?

嗯，我在伦敦创办 DeepMind 有几个原因。一是我在这里出生和长大，我热爱这个国家和伦敦，我在剑桥学习，我想回馈社会。第二，从商业角度来看，英国和欧洲有很多未被充分开发的人才。我当初对早期投资者的推销是：想象你是一个剑桥的物理学博士——世界一流——但你不想留在学术界，也不想进入金融业。唯一的选择就是金融或对冲基金。我想提供另一种选择。回到 2010 年，伦敦没有人做 AI 或深科技。所以整个领域都是我的，我用年薪 5 万英镑就雇到了了不起的人。现在你连一天的实习生都请不到。这点预算非常不可思议。我们筹集了几百万英镑，雇到了世界级的人才，不仅来自英国，还有全欧洲。这种情况持续了六七年，直到大科技公司意识到这里有才，设立了卫星办公室——这对英国是好事，带来了更多人才和投资。现在国王十字区因为科技投资成了一个了不起的地方。人们待上几年，然后自己创业。最后，我想证明这一切在英国也能做成。我们一直低于应有水平。我认为 DeepMind 证明了这一点，并激励了下一代创始人投身于量子、核聚变甚至金融等硬科技。但我认为我们还可以走得更远。

Well, I started DeepMind in London for a few reasons. One is that I was born and grew up here, I love the country and London, I studied at Cambridge, and I wanted to give back. Second, on the commercial side, there was a lot of talent in the UK and Europe that was untapped. My pitch to early investors was: imagine you're a physics PhD from Cambridge—world class—but you don't want to stay in academia or go into finance. The only option was finance or hedge funds. I wanted to offer a different option. Back in 2010, no one was doing AI or deep tech in London. So I had the whole field to myself, hiring amazing people for £50k a year. Now you can't even get an intern for a day for that. That was incredible on a shoestring budget. We raised a couple of million and hired world-class people not just from the UK but all over Europe. That lasted about six or seven years until big tech realized there was talent here and set up satellite offices, which is good for the UK—it brings more talent and investment. Now King's Cross is an amazing place because of that tech investment. People stay for a few years then start their own companies. Finally, I wanted to prove it could be done here. We were punching under our weight in this area. I think we proved a point with DeepMind and inspired the next generation of founders in hard tech like quantum, fusion, even finance. But I think we could go further.

我们需要在这里诞生第一家保持独立并持续增长的万亿级公司。这可能是英国生态系统的下一个挑战。但我相信我们能做到，就像 Daniel Ek 在斯德哥尔摩用 Spotify 尝试的那样，激励了新一代创业者。只需要一些条件成熟，加上政府的帮助，生态系统就能以此为基础发展壮大。

We need to produce our first trillion-dollar company that remains independent and grows here. That's probably the next challenge for the UK ecosystem. But I believe we'll do that, and there are pockets of that, like Daniel Ek trying to do this in Stockholm with Spotify, inspiring a next generation of entrepreneurs. You just need a few things to happen, and then the ecosystem, with the government's help, can build on that.

没错，但关键是政府的帮助。这正是我们得不到的。在座有人就在推动这件事。但问题出在财政部，你根本无法让他们改变做事方式，他们正在错失良机。

Yeah. But the key is the government's help. That's where we don't get it. There are people in this room working on that very thing. But it's the Treasury, and you just can't get them to change the way they do things. They're missing opportunities.

是的，我们有人才，有大学。正如你说的，Demis，所有人才都在这里，非常出色。其他国家——看看美国正在发生的事，政府停止资助研究，这很可怕。不过这对我们是个好机会。当然还有中国。我们现在有绝佳的机会乘势而上，但确实需要政府解决如何将投资注入公司的问题。我们可以培育独角兽，但如何再上一个台阶是挑战。

Yeah, we've got the brains. We've got the universities. As you said, Demis, all the talent is here. Amazing talent. Other countries—well, if you look at what's happening in America, how the government is stopping funding research, it's very scary. It's a good opportunity for us though. And of course, you have China. We have a fantastic opportunity to ride high now, but we do need the government to sort out how we get investment into the companies. We can get the unicorns, but getting to the next level is the challenge.

我们现在就需要再上一个台阶。

We need to get to the next level now.

就是现在，再上一个台阶。

The next level now.

对。

Yeah.

再问 Demis 一个问题。我很想知道，你为什么选择蛋白质折叠作为第一个突破点？

And just one more question for you, Demis. I was very keen to understand: why did you choose protein folding as the first?

是的。我一生都在研究人工智能，因为我认为 AI 是科学的终极工具。所以我一直对关于现实本质、事物如何运作的重大问题感兴趣。我的表达方式就是打造最好的工具来帮助科学家。蛋白质折叠和蛋白质结构预测，是我在剑桥读本科时接触到的。当时我有很多生物学家朋友，其中一位对此特别着迷，总是谈论它。所以我认真倾听。我倾向于认真倾听那些对某件事充满热情的人。我认为这是一个惊人的科学问题——终极谜题。一维的基因序列如何折叠成精妙的三维纳米机器？太神奇了。这也是我所说的根节点问题：一旦攻克，将开启全新的发现分支——药物发现、基础生物学。它就像生物学界的费马大定理。人们已经尝试了 50 年。即使在 90 年代，虽然还没有今天的技术，但我觉得这正是 AI 可以助力的问题。所以我把这个问题在脑海中存档了将近 20 年。直到 AlphaGo 比赛后的第二天，我从韩国回来。我们不仅赢了比赛，AlphaGo 还下出了原创性的第 37 手，在一个有限的游戏领域做出了发现。这正是我一直等待的：一个能在那种水平上表现并做出此类事情的系统。回来的那天，我们就启动了 AlphaFold 项目。

Yes. Well, I spent my whole life working on AI because I thought it would be the ultimate tool for science. So I've always been interested in the big questions of the nature of reality, how things work. For me, my expression of that was to build the best tool to help scientists. Protein folding and protein structure prediction came to my attention as an undergrad in Cambridge. A lot of my biologist friends, one specifically, was obsessed with it and would talk about it all the time. So I listened carefully. I tend to listen carefully to people who are very passionate about things. I thought it was an amazing scientific question—the ultimate puzzle. How does a one-dimensional genetic sequence fold into an exquisite three-dimensional nano-machine? Amazing. It was also what I call a root-node problem: if you cracked it, it would open up whole new branches of discovery—drug discovery, fundamental biology. I was attracted to it, like Fermat's Last Theorem for biology. People had been trying to solve it for 50 years. And even back in the '90s, though we didn't have the technology we have today, I could feel it was the right sort of problem for AI to help with. So I filed it away in the back of my mind for nearly 20 years. Then the day after the AlphaGo match, I came back from South Korea. Not only had we won, but AlphaGo had played an original move—Move 37—making a discovery in the limited domain of a game. That was what I was waiting to see: a system that could perform at that level and do those kinds of things. The day I got back, we started the AlphaFold project.

太棒了。我之前没听过这个故事。真好。

Fantastic. I hadn't heard that before. That's good.

现在我们继续推进 Isomorphic Labs，这是英国新成立的一家衍生公司。因为我认为生物学和信息科学之间存在同构性。生物学可以被看作一个信息处理系统。这是在 AlphaFold 成就的基础上——蛋白质结构只是药物发现过程中一个重要组成部分。这个领域需要彻底重塑，而这正是我们试图通过 Isomorphic 来实现的。

And now we continue with Isomorphic Labs, this new spin-out in the UK. Because I feel there's an isomorphism between biology and information sciences. I think biology can be thought of as an information processing system. That's building on what AlphaFold did—protein structure is one important component of the drug discovery process. That field needs to be completely reimagined, and that's what we're trying to do with Isomorphic.

是的。未来三到五年制药业会发生什么，会很有趣。

Yeah. It'll be interesting to see what happens in the next three to five years in the pharmaceutical industry.

是的，非常有趣的时代。我们会关注你们的工作，因为你们在材料科学方面也有动作。

Yes. Very interesting times. We'll be watching what you guys do because you're doing something with material sciences as well.

对，那是另一项工作。我们在 DeepMind 做这些。我们将 AI 应用于许多科学领域，但都处于更早期阶段——类似于 AlphaFold 之前的阶段。材料科学方面，我们正在帮助核聚变，控制聚变反应堆中的等离子体。还有数学、天气预报。我们与英国气象局合作。我们拥有世界上最好的飓风预测算法，而且比传统的流体动力学系统快得多。去年美国气象局用它预测了飓风梅利莎的路径，最终飓风转向了牙买加。如果能提前 2 到 4 小时发出预警，对当地的影响结果会有天壤之别。

Yeah. That's a separate thing. We're doing that at DeepMind. We're applying AI to many areas of science, but those are earlier stage—they're pre-AlphaFold stage. Material science, we're helping with fusion, containing plasma in fusion reactors. Mathematics, weather prediction. We work with the Met Office. We've got the best hurricane prediction algorithms in the world. They're also way faster than traditional fluid dynamic systems. That was used by the US Met Office for Hurricane Melissa last year to predict which path it would take. It ended up veering into Jamaica. Being able to give two to four hours more warning makes a huge difference for the outcome on the ground.

可惜我们今天没能预测到 RMD 封锁。

It's a pity we couldn't predict the RMD strike today.

是的，那要难得多。

Yes. That's a lot harder.

要难得多，因为涉及人。

It's a lot harder because it's about people.

人类。没错，他们是最难预测的。

Humans. Exactly. They're the most complex things to predict.

呃，我的助手 Clark 一直在给我使眼色，觉得该开始提问了，不过她现在笑了。我想谈一下你之前提到的平等问题。我很好奇，请简短回答：我们是否会为所有人打开大门，不论性别、背景或特权，还是技术有可能让我们走上一条不同的道路？

Uh, my Clark is giving me really bad looks thinking she wants to start the questions, but she's smiling now. I just want to touch on something you said earlier on, which is equality. I'm very curious, a very short answer on this please: are we going to open doors for people regardless of gender, background, privilege, or does technology risk taking us down a different path?

嗯，我得说，人们——我其实很不愿意用“服从”这个词，因为我结婚时拒绝在婚礼中使用“服从”这个词。但我在这里服从了（笑了）。所以，我一直充满热情——我先从性别问题说起。

Well, I have to say, people—I really was reluctant to use the word obedience because when I got married, I refused to have the wedding ceremony that had the word 'obey' in it. But I obeyed here. So, I've been passionate about—I'm going to start with the gender thing.

从 20 世纪 80 年代我进入计算机领域开始，我就一直对女性在其中的缺失充满热情。可怕的是，情况不但没有好转，反而恶化了。在 AI 领域，情况更糟。看看硅谷，领导层全是同一种文化。我看到坐在观众席中的我和 Sue 的年轻一代。我们努力改变这个状况已经很久了，但过去 40 年并未改变指针。这在 AI 领域更为重要。我的导师 Karen Spärck Jones 曾经穿一件 T 恤，上面写着“计算太重要了，不能只交给男人”。这不是贬低男性，而是为了我们所有人。我们需要每个人都参与进来。这不仅仅是性别多样性。我现在说“AI 太重要了，不能只交给男人”。如果我们想要达到 AGI 的圣杯，我们必须思考我们要构建什么样的社会。有些人轻描淡写地谈论 AGI，好像一切都不会改变——我们只是拥有非常聪明的机器，生活照旧。但事实并非如此。我们必须认真思考我们想要构建什么样的社会。你只能告诉人们你的工作会消失，你的孩子不会有职业生涯，或者这可能会毁灭人类。但抵抗 AI 的运动已经开始，我认为这非常危险。我记得当人们说“我的镇上不要核弹”时，他们感到无力。在美国，现在他们感到完全无力。我助理的 18 岁女儿告诉她妈妈，她们班决定教室是无 AI 的。南安普顿大学和其他大学的学生都说不想用 AI，因为公司告诉我们的事。我们希望人们抓住 AI 带给我们的巨大机遇。科技公司必须参与进来。我不知道政府应该扮演什么角色，但我们必须讨论我们要构建什么样的社会，这些非常聪明、可能具有智能的机器。我认为机器智能和人类智能有区别。我们可能会得到 AGI，但它不是人类的。

I've been passionate about the lack of women in computing since the 1980s when I started. The scary thing is it's got worse rather than better. In AI, it's far worse. You look at Silicon Valley and the leadership is all one monoculture. I see the younger generation of me and Sue there sitting in the audience. We've been trying to change this for so long, but we haven't shifted the dial in the last 40 years. It's even more important in AI. My mentor, Karen Spärck Jones, used to wear a t-shirt that said "Computing is too important to be left to men." This is not to denigrate men. It's for all of us. We need everybody involved. It's not just gender diversity. I now say AI is too important to be left to men. If we're going to get to the holy grail of AGI, we have to think about what type of society we want to build. Some people glibly talk about AGI as if it's all going to be the same—we just have very clever machines and life continues as before. It won't be like that. We have to think through what type of society we're trying to build. You can only tell people so long that their jobs are going, their kids won't have careers, or this could wipe out the human race. There is the beginning of a resistance movement to AI, and I think that's very dangerous. I remember when people said "No nuclear bombs in my town" because they felt powerless. In America, they feel totally powerless now. My PA's 18-year-old daughter told her mom that her class decided their classroom is AI-free. Students at Southampton and other universities are saying they don't want to use AI because of what companies tell us. We want people to grasp the great opportunities AI affords us. Tech companies have to get involved with this. I don't know where governments fit in, but we have to have this discussion about what type of society we're building, with these very clever, potentially intelligent machines. I think there's a difference between machine intelligence and human intelligence. We might get AGI, but it's not human.

谢谢。我收到一个现场提问。提问者是伦敦金融城政策委员会主席、我亲爱的朋友 Chris Haywood。非常感谢，大师。这个问题是给你的，Demis，如果允许的话。它分为两部分。第一部分：如果英国要在未来 12 个月内采取一项政策决策，以便不仅在构建 AI 方面领先，而且还要负责任地大规模部署 AI，尤其是在我代表的金融服务领域，那会是什么？第二部分：你认为像 DeepMind 这样的组织、生态系统中的其他领导者以及伦敦金融城这样的机构之间，最大的合作机会在哪里，以帮助加速安全采用？

Thank you. I've got a question from the floor for you. It's from the policy chairman of the City of London Corporation, my very dear friend Chris Haywood. Thank you very much, master. It's a question to you, Demis, if I may. And it's in two parts. The first part is: if there is one policy decision that the UK needs to take in the next 12 months to lead not just in building AI but in deploying it responsibly at scale, particularly in financial services which is the area I represent, what would it be? And the second part: where do you see the greatest opportunity for collaboration between organizations like DeepMind, other leaders in the ecosystem, and institutions such as the City of London Corporation to help accelerate safe adoption.

好的。两个好问题。第一个问题，我想谈两件事。一是能源状况。我认为出于安全考虑，无论如何都需要解决这个问题，但考虑到即将到来的 AI 经济，它变得加倍重要。我们必须参与其中。我们需要建设数据中心，需要在这里拥有第一方设施并加以利用。第二件事是我们之前也谈到的：如何解决从非常擅长打造独角兽到真正的大型全球影响力公司的跃迁。我们需要一些这样的公司在这里。这是下一个挑战，我认为政府必须参与其中——证券交易所、整个生态系统，你们的世界需要解决这个问题，以便促成这一切。所以我认为在合适的生态下这是可能的。第二个关于合作的问题：我认为这类论坛很好。这就是为什么我很高兴参加这次活动。和 Wendy 聊天也总是很有趣。这些论坛不仅聚集了技术专家，我们还需要更多。伦敦特别令人惊叹的是，我们在许多领域都拥有世界级的东西。美国分散在不同城市——政府 in DC，金融 in 纽约，生命科学 in 波士顿，硅谷，娱乐 in 洛杉矶。而在伦敦，这些几乎全部集中在一个城市，而且都达到世界级水平。我一直坚信多学科工作的重要性。现在更是如此。我认为 AI 可以促进这一点，因为它能让你快速成为某个其他领域的专家。我还认为，我们需要有意识地思考，在 AI 和 AGI 之后的世界里，我们想要拥有和构建什么样的社会。忘记术语，忘记时间线。这很快就会到来。就在我们有生之年。它将会发生。

Right. Okay. So, two good questions. I think the first one probably I would talk about two things actually. One is the energy situation. I think that needs to be addressed anyway for security, but I think it's doubly important given the AI economy coming down the line. We need to be part of that. We need to build data centers. We need to have that first-party stuff here on the ground and take advantage of that. The second thing is what we also talked about earlier: sorting out how we get from doing very well getting to unicorns to the really big global influential companies. We need some of those here. That's the next challenge, and I think government has to be part of that—the stock exchange, the whole ecosystem, your world needs to sort that out to help allow for that to happen. So I think it could happen with the right ecosystem. And then the second thing on collaboration: I think these kinds of forums are good. This is why I was delighted to participate in this. It's always fun chatting with Wendy too. These forums bring together not just technologists, but we need more. The amazing thing about London specifically is we actually have world-class things in many domains. The US splits across different cities—government in DC, finance in New York, life sciences in Boston, Silicon Valley, entertainment in LA. We have all of that pretty much to a world-class level in one city. I have always been a huge believer in multidisciplinary work. I'm even more of a believer in that now. I think AI can enable that because it allows you to get expert quite quickly at some other subject area. I also think we need to work out intentionally what sort of society we want to have and build in a post-AI, AGI world. Forget the term, forget the timeline. It's soon. It's in our lifetimes. It's going to happen.

所以，接受这一点吧。这是既定事实。马已经回不了马厩了。它正在发生。我们需要弄清楚我们想对此做些什么。未来不是注定的。它不是。它是由我们社会来书写的。但前提是我们不带恐惧地去面对。这是一件有挑战性的事情。它令人害怕，充满不确定性。甚至没有人有答案。没有一个大型科技公司的人，没有科学家有答案。但这并不意味着答案不存在。这要由我们社会来决定。而且我认为这需要社会所有部分的参与。我指的是人文学科、科学、技术和工业、以及政府。而且我认为英国和伦敦实际上可以成为这方面的中心。但前提是我们在经济和技术上足够成功，能有一席之地。这是让他人倾听我们声音的先决条件。

So just deal with that. That's a given. The horse is not going back into the stable. It's happening. We need to then work out what do we want to do about it. The future is not written. It is not. It's for us society to write that. But only if we approach it without fear. It's a challenging thing. It's scary. There's uncertainty. Even no one has the answers. None of the big tech people, no scientists have answers yet. But it doesn't mean there aren't answers out there. It's for us as society to decide that. And I think that requires all parts of society to be involved. I mean the humanities, the sciences, the tech and industry, and government. And I think the UK and London could be a center for that actually. But only if we're successful enough economically and technically to have a seat at the table. That's the prerequisite for having a voice that others will listen to.

好的，谢谢。下一个问题……副主管加里·摩尔。

Okay, thank you. The next question... deputy master Gary Moore.

主管，谢谢。很有趣你用了“未来不是注定的”这个词。在弗兰克·赫伯特的《沙丘》宇宙中，禁止思维机器成为了该文明的基础背景，它源于一场灾难性的星际战争，人工智能和有知觉的机器人奴役了人类。问题开头很积极。你已经提到了这一点，但当我们迈向 AGI 并考虑监管时，在有些东西已经开源的世界里，我们如何管理那些不服从监管的 rogue 行为者？

Master, thank you. Interesting you use the words 'the future has not been written'. In Frank Herbert's Dune universe, the prohibition of thinking machines became a foundational background of that civilization. It stemmed from a catastrophic interplanetary war where artificial intelligence and sentient robots enslaved humanity. Nice positive start to the question. You've touched on this already, but as we move to AGI and perhaps thinking about regulation, do you have any reflections in a world where some of this stuff's already open source, how we manage rogue actors not just those who submit to regulation?

嗯，我们必须通过有牙齿的法律。比如深度伪造，我们没有合适的法律来处理。如果有人拿了别人的图像做了可怕的事情，那应该是非法的。而且当他们被抓到时，判决应该有牙齿。这很简单。还会有更难处理的事情。我让你来接手。

Well, we have to pass laws that have teeth. We've got to actually take, for example, deep fakes. We don't have laws to deal with them properly. If someone takes someone else's image and does something awful with it, that should be illegal. And when they get caught, there should be teeth in the sentence. That's simple. There's going to be harder stuff to deal with. I'm going to let you take over.

是的，没错。深度伪造就是一个很好的例子。我们有像隐形水印这样的技术，我们开发了它，其他公司也有，现在这将成为行业标准。如果政府想要，而且我认为这会发生，因为这很明显应该做，那么不仅要禁止这种行为，还要确保任何生成媒体的模型都必须有水印作为其一部分，还有一个检测系统，政府、记者可以访问，或者最终网络应该自动标记出来，如果这是一张生成的图像。这在我看来是一件没有遗憾、无需动脑的事情。但是关于《June》（可能指《沙丘》？），你知道，如果事情出错，那可能会发生。我看了电影，读了很多科幻小说，但我没读《June》可能是因为它禁止了AI。但我看了那些伟大的电影。如果AI构建错误，我们可能没有机会恢复。我不知道在《June》的宇宙中是如何解决的，但如果事情变得非常糟糕，可能就没有挽回的余地了。所以我们必须在第一次就做对。我想说的是，在科幻小说中，有很多很好的宇宙例子，人类确实做对了，人类繁荣至星辰。所以我推荐伊恩·M·班克斯，他是我最喜欢的科幻小说家之一，以及《文化》系列。它描绘了一个可能一千年后的未来世界，人类遍布星辰，也有AI，但他们以大体和平的方式共存，互补。这只是其中之一。有很多有趣的愿景。也许我们需要更多的科幻小说家来创作关于它应该如何发展的积极故事，因为很多科幻小说确实会影响科学家们的想法。这在历史上已经被证明。所以当我与科幻小说家交谈时，我告诉他们写一些硬科幻，展示一个关于我们在未来50到100年可能达到或应该努力的目标的现实视角。我认为我们需要一个巨大的创意爆发，关于我们应该建立什么样的社会，假设我们把一些事情做对了，比如技术方面正确，那么我们就可能处于一个后稀缺世界。但那意味着什么？如果我们解决了聚变、材料科学和一些根本问题，那么在那种世界里经济学应该如何运作？

Yeah, sure. So deep fakes is a good example. There are technologies like invisible watermarking that we've developed and others that will become an industry standard now. That would allow, if a government wanted, and I think this will happen because it seems obvious that should be done, outlawing that, but also making sure that any model that generates media has to have watermarking as part of it, and also a detection system that government can have access to, or journalists, or eventually maybe the web should just automatically flag that up if this is a generated image. That seems to me a no-regrets, no-brainer thing to do. But look, as to June, which you know, that's what could happen if things go wrong. I've watched the film, I've read so much sci-fi, but I didn't read June maybe for this reason because it banned AI. But I watched the great movies. We might not get the chance to recover if AI was built wrong. I don't know how it resolved in the June universe, but if things go badly wrong, there may not be potential to just walk that back. So we got to get that right first time. What I would say is in science fiction, there are many great examples of universes where humanity did get this right, maximum human flourishing to the stars. So I would point to Ian M. Banks as one of my favorite sci-fi novelists, and the Culture series. That depicts a future world maybe a thousand years from now where humanity's spread to the stars, and they also have AIs, but they coexist in this largely peaceful way, complementary way. So that's just one. There are many interesting visions of what that could be. Maybe we need more science fiction authors to produce positive stories about the way it should go, because a lot of science fiction does affect what scientists then think to do. I mean this has been proven through history. So when I speak to a science fiction author, I tell them write some hard sci-fi that shows a realistic view of where we could get to or where we should be trying to aim for in the next 50 to 100 years. I think we need a big creative explosion of ideas of what sorts of societies should we build, assuming we get some of these things right, like the technical aspects correct, and then we might be in a post-scarcity world. But what does that mean? How should economics work in that sort of world if we solve fusion and material science and some root node problems.

我喜欢谈论团队合作。我喜欢把AI看作团队的一部分。我认为它会成为。这一切看起来发生得很快，但实际上并没有。技术可能在前进，但采用速度远没有那么快。而且我认为并不是所有的工作都会消失。我不是伊隆·马斯克那种“没有人需要工作”的垃圾观点。工作会改变。我认为我们会适应工作场所中的AI。它会取代一些人们不得不做的苦差事。AI来做这些，然后一切都会变得更好。人类可以更有创造力。但当你诊断某人的疾病时，你仍然需要人类医生。但AI会过来说：“我可以从数据中告诉你，所有个性化医疗，”这就是团队合作。正如我之前所说，我们必须认识到机器智能和人类智能是不同的，我们应该为此高兴，并利用它来改善社会。

I like to talk about teamwork. I like to think about AI becoming part of teams. I think it will become. It all seems to be happening so fast. It's not actually happening. I mean the technology might be moving on. Adoption is not happening anywhere near as fast. And I think it's not that all the jobs are going to disappear. I'm not the Elon Musk 'no one will need to work' rubbish. Jobs are going to change. I think we will get used to AI in the workplace. It will replace some of the grunt work which people have to do. AI will do that, and all will be the better for it. Humans can be more creative. But when you're diagnosing someone's illness, you still need the human doctors. But the AI will come along and say, 'I can tell you from the data, all the personalized medicine,' and it will be teamwork. As I said before, we have to recognize that machine intelligence and human intelligence are different, and that we should rejoice in that and use it for the betterment of society.

我猜在过去，在六七十年代，人们也是用同样的方式谈论计算机。

And I guess in the olden times in the 60s and 70s people talked about computers in the same light.

嗯，我经常举我父亲的例子，他是一名会计师。他完成工作后，经营着一家公司的会计部门。他们一切都用手工做。没有计算器，没有计算机。他能在脑子里做英镑、先令和便士的除法。全是账簿。他们抄写账簿。那就是他们所做的。

Well, I always use the example of my father was an accountant. He finished his work and he ran an accounts department in a company. They did everything by hand. They had no calculators, no computers. He could divide pounds, shillings, and pence in his head. It was all ledgers. They copied the ledgers. That's what they did.

这些工作岗位都已经消失了。但现在金融行业的工作岗位更多了，因为计算器和计算机能完成许多人类无法做到的事情。这正是 Demis 之前说的：AI 将能完成我们以前甚至无法想象的事情。我们应当珍惜这个机会，但同时也必须把安全和护栏问题处理好。我们不能袖手旁观，指望别人来做。我们现在就得行动起来。我们行业协会的一个支柱是慈善，我们有一个关于慈善的问题，我觉得会非常有意思。这个问题将由 AI 小组主席 Paul Excel 提出。小组就像委员会，帮助成员讨论和改进。有请 Paul。

And those jobs have all gone. But there are now more jobs in the finance industry because calculators and computers can do so much that humans can't. This is what Demis said earlier: AI will do things we couldn't even dream of before. We need to relish that opportunity, but we must also get the safety and guard rails sorted. We can't sit back and assume someone else will do it. We have to do it now. So, part of our livery company's pillars is charity, and we have a question on charity that I think will be quite interesting. It will be from the chair of the AI panel, Paul Excel. A panel is like a committee that helps members discuss and improve. Paul.

谢谢主席。我的问题是：在我们的 AI 慈善社区中，现在有 90 多家慈善机构正在用 AI 做一些非常出色的基础工作，为所服务的社区带来实实在在的好处，但他们普遍非常担忧。他们所支持的人群非常脆弱。既然我们所有人都有机会塑造我们、我们的孩子和孙子孙女的未来，那么从负责任 AI、伦理、信任、同理心、安全保障等方面，您会提倡哪一到三个负责任行动来防止伤害并产生积极影响？这是给在座各位、政府和公民社会的作业，以确保全球解决方案既能满足社会需求，又能带来安全、可持续的增长和所有人的机会。这几乎像一个目标，也许这个愿景太乐观了，但在伦敦和英国，我们肯定有可能抓住它。

Thank you, Master. So my question is: across our AI for Charities community, we now have over 90 charities doing fantastic fundamental work using AI, unleashing real benefits for the communities they serve, but they are generally very concerned and worried. They support very vulnerable people. Given that we all have a chance to shape the future we, our children, and our grandchildren live in, what one to three responsible actions would you advocate—from responsible AI, ethics, trust, empathy, safeguarding—to prevent harm and deliver positive impact? This is homework for everyone in this room, governments, and civil society to ensure global solutions serve society's needs while delivering security, sustained growth, and opportunity for all. It's almost like a goal, maybe too positive a vision, but surely we can grasp it here in London and the UK.

这个问题很难回答。我首先想说，我的第一要务是多样性。抱歉，但有人问我什么让我夜不能寐时，我的答案不是 Demis 所说的那些，而是这个世界缺乏多样性。我认为如果没有多样性，用 AI 建立一个平等主义的世界将更加困难。我主要是从性别角度考虑，但我指的是最广泛意义上的多样性。我正在变老——我们都在变老，这真是个蠢笨的同义反复——我开始觉得有些事情我再也做不到了。我们需要思考像我这样的老年人如何从 AI 中受益，以及无法接触科技的学生。在许多方面，我们需要创造一个公平的竞争环境。

It's quite a hard question to answer. I'll start by saying my number one would be diversity. I'm sorry, but when I was asked what keeps me awake at night, it's not any of the things Demis talks about; it's that we don't have enough diversity in this world. I think it will be harder to build an egalitarian world using AI without diversity. I'm looking at it through a gender lens, but I mean diversity in the widest sense. I'm getting older—we all are, that's a silly tautology—and I'm beginning to feel there are things I can't do like I used to. We need to think about how the elderly, like me, will benefit from AI, and school kids with no access to technology. There are so many things where we need to create a level playing field.

这正是我所说的社区主题。正是您刚才阐述的。

And that's exactly where my theme of community came out. Exactly what you've just been articulating.

好的。

Okay.

我们只想到一个。你的首要建议是什么？

Well, we only came up with one. What's your top one's question?

这是一个棘手的问题。我主要考虑的是全球治理层面需要发生的事情，这显然会影响一切。但自下而上的草根层面也至关重要，正如 Wendy 所说。我们谈到了许多主题：教育、重新培训技能。这些都是社区层面的事情。课程需要更新。我告诉过你我会怎么做。也许你同意，也许你能施加影响。我没有时间，但应该有人做。所以这取决于你的成员的兴趣。但帮助慈善机构快速掌握并使用这些工具，将有助于处理他们花钱的后台事务，从而释放资源用于一线工作。慈善机构在筹款和后台方面花费很多，但如果这些资金能用于一线会更好。因此我们既需要全球治理层面——领先实验室和超级大国，也需要社区层面。也许我们可以在英国引领这一方向。

Well, it's a tricky one. I'm thinking mostly about what needs to happen at the global governance scale, which obviously affects everything. But the bottom-up grassroots aspect is critical too, as Wendy said. We've touched on many themes: education, reskilling. These are community-level things. The curriculum needs updating. I've told you what I would do. Maybe you agree, maybe you can influence it. I don't have time, but someone should. So it depends on your members' interests. But helping charities get up to speed and use these tools would be great for back-office tasks they spend money on, freeing resources for frontline work. Charities spend a lot on fundraising and back office, but it would be better if that went to the front line. So we need both global governance with frontier labs and superpowers, and community scale. Maybe we can lead the way here in the UK.

我们在南安普顿有一个公民项目，目标是将南安普顿打造成 AI 卓越城市，但从基层做起。在慈善领域，我们举办快闪活动——实际上是我举办——我们邀请专家来主持工作坊。我们为慈善机构举办了一个名为“关于 AI 你想知道但又不敢问”的活动。说真的，我们鼓励那些感到脆弱或害怕的人来参加，我们作为一个慈善工作者社区进行交流，并分享使用 AI 的慈善机构的最佳实践。我们现在已经在三个领域开展了这样的活动，并且正在扩大范围。这关乎意识和教育。

We have a civic project in Southampton where we talk about making Southampton an AI city of excellence, but from the grassroots up. In the charity sector, we do pop-ups—I do pop-ups—where we bring in experts and run workshops. We have one for charities called 'Everything You Wanted to Know About AI but Were Afraid to Ask'. Seriously, we encourage vulnerable or scared people to come, talk as a community of charity workers, and share best practices from charities using AI. We've done that in three sectors and are expanding. It's about awareness and education.

太好了。下一个问题来自 Liverman Nicholas Bill。Nicholas，请。

Super. I have another question from Liverman Nicholas Bill. Nicholas, this is...

非常感谢两位。几个月后，所有西方大型 AI 开发者都将成为对投资者负责的上市公司。全球投资者是唯一能与这些企业抗衡、运营规模达数万亿美元的参与者。他们共同拥有避免任何投资失败的巨大利益。所以我的问题想请教两位，特别是 Demis：投资者可以做些什么来鼓励负责任的行为？他们或许能够非常有效地帮助解决博弈论问题。

Thank you very much indeed both of you. In a few months, all Western big AI developers will be publicly listed companies responsible to investors. Global investors are the only players operating at a trillion-dollar scale comparable to these companies. They have a common interest in avoiding blow-ups. So my question to both of you, especially Demis, is: what can investors do to encourage responsible behavior? They could help with the game-theoretic issue quite effectively.

是的，我认为这是个很好的观点。我认为投资者可以施加比现在大得多的影响力。

Yeah, I think that's a great point. I think investors could exercise a lot more influence than they probably do.

我认为问题在于，参与那些融资轮次的回报太高了，条款基本上是既定的，要么接受要么放弃，而且你没法错过，因为这种火箭式增长的公司有多少？并不多，可能只有几家。所以我觉得这有点打破了平衡。此外，其中许多公司拥有不寻常的控制结构，即便上市了也不对股东负责。

I think the problem is it's so lucrative to get into those funding rounds that the terms are basically take it or leave it, and you can't really miss out because how many of these rocket ship companies are there? Not that many, maybe just a handful. So that skews the balance a bit. Also, a lot of them have unusual control structures, where they're not beholden to the shareholders even when they float.

也许吧，但公平地说，普通投资者最关心的是他们的回报。所以我不确定我们是否还需要更多纯粹的经济激励。

Maybe, but the average investor, to be fair, mostly cares about their returns. So I don't know if we want any further purely economic incentive.

我认为这已经是问题的一部分了。或许有办法解决，而我期待的是市场在智能体时代能自我修正。想象一下，如果一个领先的前沿模型实验室想为金融公司提供智能体，金融公司肯定会要求对这些智能体的护栏有所保障——它们会如何处理客户数据。你不能让其中一个智能体因目标函数设定不当而执行一笔交易，导致你第二天损失十亿美元。所以我猜测，那些没有适当护栏的公司会发生类似的错误，然后市场会做正确的事，激励负责任的行为，尤其是在受监管的行业，这些行业同时也是最有价值的。所以我认为这可能是好事。这就是一线希望：这些技术变得更具商业实用性和可行性，尤其是在企业领域，我们现在第一次看到了这一点。我们之前看到了消费者方面，但现在企业领域也出现了，这可能会以真正有利于负责任部署的方式调整激励。我认为这很可能会发生。投资者和企业——大银行的CTO和CIO——应该利用他们的购买力来支持那些负责任行动的供应商。显然，你不能在质量上妥协，但可以支持那些你认为负责任的公司。我也对普通民众说，作为消费者，你们可以用金钱投票，利用市场力量推动好的方向，施压那些以你们认同的价值观行事的公司和实验室。随着这些系统变得更加商业化，我希望投资者和企业能做到这一点。这可能与我们共同建立的全球AI治理组织挂钩。投资者可以说，如果你的公司不签署这个协议，我们就不会投资你。

I think that's already part of the problem. Perhaps there's some way around that, and what I'm hoping is the market will self-correct a bit in the agentic era. If you imagine that one of the leading frontier model labs wants to provide agents for a financial service company, I would imagine the financial service companies will want some guarantees around the guard rails of those agents—what they'll do with their client data. You can't have one of them putting on a trade that loses you a billion dollars because of a misspecified goal function. So I suspect that some sort of error like that will happen with companies that don't have the right guardrails, and then the market will do the correct thing and incentivize responsible behavior, especially in regulated industries, which are also the most valuable. So I think it could be good. That's the silver lining: these technologies becoming more commercially useful and viable, especially in enterprise, which we're seeing now for the first time. We saw the consumer part, but now there's enterprise, and that may align incentives in a way that's really good for responsible deployment. I think that's pretty likely to happen. Investors and enterprises—CTOs and CIOs of big banks—should use their purchasing power to support providers that act responsibly. Obviously, you shouldn't compromise on quality, but support those you believe are acting responsibly. I also say to the average person that as a consumer, you can vote with your dollars and use market forces for good, putting pressure on companies and labs that reflect the values you want to see. As these systems become more commercial, I hope that's what investors and enterprises will do. It could be tied to whatever organization we collectively set up to govern AI globally. Investors could say, if your company doesn't sign up to this, we won't invest in you.

我来问最后一个问题，这个问题来自英国历史最悠久的分析师，理查德·豪威尔。嗯，确实。嗯，昨天我写了一篇文章，庆祝 Computer Centre 进入富时 100 指数，这是第二家进入该指数的科技公司。实际上，按市值排名的前 100 家科技公司中，英国只有一家，那就是 ARM，而且它还在纳斯达克上市。如果 DeepMind 当初进行了 IPO，我敢保证你现在肯定在富时 100 指数里。事实上，你可能还会领跑。你可能会成为催化剂，带动许多其他英国公司走同样的路，并围绕它建立基础设施和生态系统，就像硅谷那样。我想知道你对这件事的看法。我想大家都知道我的看法。

I'm going to go to one last question, and this question is from the city's oldest analyst in the UK, Richard Hallway. Um yes indeed. Um yesterday I wrote an article celebrating Computer Centre entering the Footsie 100, only the second tech company to be there. Indeed, in the list of the top 100 tech companies by market capitalization, the UK only has one, and that is ARM, and it's actually listed on NASDAQ. Now, if DeepMind had gone for an IPO, I guarantee that you would currently be in the Footsie 100. In fact, you might even be leading it. And you might have acted as a catalyst to many other UK companies to follow the same route and you would have built the infrastructure and ecosystem around it, just as they have in Silicon Valley. I wonder what your views were about that. I think everybody knows what mine are.

嗯。好吧，另一种历史。嗯，当然。听着，我不知道 2014 年你在哪里。我不知道当时你是否愿意给我们几十亿美元的支持。也许你愿意，但我找不到。回到 2014 年，在 AI 领域那是另一个世界。我们是唯一在做这件事的人。那时 OpenAI 还没有成立，我们还没有做出 AlphaGo。我们只有 Atari 游戏，人们会说，‘这是什么？你在玩《太空侵略者》。这有什么重要的？’但谷歌理解其重要性。特别是拉里·佩奇，对吧？所以我非常感激他，因为他一直对 AI 感兴趣。他把谷歌视为一家人工智能公司。我卖掉公司的原因是，我知道需要 70 亿美元的投资才能尝试成为那些公司之一。而我当时几乎凑不齐 1000 万美元的融资。问题是，Atari 项目已经惊动了足够多的硅谷巨头。他们开始挖我的人。我们没有业务，也没有产品。我自己零薪水，员工年薪大约 10 万美元，然后他们收到了年薪 1000 万美元的 offer。所以我想，然后他们还会回来说，这里是年薪 5000 万美元。所以迟早会出问题。当时，软银和一些大型风投还没有开始做数十亿美元的融资。我在时机上有点不走运。如果我们坚持到 2015 或 2016 年，我本可以进行那样的融资。不幸的是，不会是英国投资者，但我可以从一些更大胆的、有实力的投资者那里获得融资。但在 2014 年，那就是没有。另一个我不后悔的原因是，我想回到研究和科学本身。很难同时做两件事：为初创公司融资、担心产品，然后又做出像 AlphaGo 和 AlphaFold 这样的东西。我永远不会。所以我一点也不后悔，因为对我来说，始终是研究和技术，以及它能做什么，而不是金钱或权力。

Yes. Well, the alternate history. Yes, sure. Look, I don't know where you were in 2014. I don't know if you would have backed us with several billion dollars back then. Maybe you would have, but I couldn't find it. Back in 2014, which is another world in AI terms, we were still the only people doing it. This was before OpenAI was set up, before we did AlphaGo. We just had the Atari games, and people were like, 'What is this? You're playing Space Invaders. Why is that important?' But Google understood the importance. Larry Page specifically, right? So I give him a lot of credit because he always was interested in AI. He saw Google as an AI company. The reason I sold was I knew it would require $7 billion of investment to try and become one of those companies. And I was barely able to scrape together $10 million rounds at that point. The problem was that the Atari thing had alerted enough of the Silicon Valley titans. They were coming after my people. We had no business, no product. I was paying myself zero and people were being paid like $100k a year, and then they were getting offers of $10 million a year. So I could think, and then they would come back and say here's $50 million a year. So at some point that was going to go wrong. At the time, this was before SoftBank and some of the big VCs were doing billion-dollar rounds. I was slightly unlucky with the timing. If we'd held on until 2015 or 2016, I could have done a fundraising round like that. It wouldn't have been British investors unfortunately, but I could have got it from some of the more out-there investors that had the firepower. But that just wasn't available in 2014. The other reason I don't regret it is I wanted to get back to the research and the science. It's very hard to do both: fundraise for a startup, worry about product, and then do something like AlphaGo and AlphaFold. I would never. So I don't regret it for a moment because for me it was always about the research and the technology and what it could do, rather than the money or the power.

我知道这对英国更好，但也许对我的银行账户更好。但我不会用我的诺贝尔奖来交换任何数量的钱，真的。你知道吗，我已经看到全球 300 万研究人员在他们的重要生物医学研究中使用了 AlphaFold。我的一位领先农民朋友说，未来不会有任何一种药物在研发过程中没有使用 AlphaFold。我为此感到非常自豪。这是英国建立的东西。我认为我们应该为这一科学事实感到自豪。谁知道呢？也许我会用 Isomorphic 完成这个工作。这是计划的一部分，对吧？总部设在这里。但我们需要解决现在证券交易所的任何问题。为什么公司不在那里上市？我不知道。所以我们需要解决这个问题。几年后我会来找你，我们可以谈谈。

I appreciate it would be better for Britain but maybe for my bank balance but I wouldn't swap my Nobel for example for any amount of money you could give me literally any amount so you know and the what I've seen 3 million researchers around the world have used AlphaFold in their important biomedical research. A farmer, a leading farmer friend of mine said there won't be a drug invented in the future that hasn't used AlphaFold somewhere in their process. So I'm very proud of that. I think it's something built here in Britain. I think we should be proud of that scientific fact. And who knows? Maybe I'll complete the job with Isomorphic. That's kind of part of the plan, right? Headquartered here. But we need to fix whatever the problem is with the stock exchange right now. Why are companies not floating on there? I don't know. So we need to fix that. I'll come to you in a few years time and we can talk about it.

我想赞扬 Demis，我认为英国从 Demis 的成就中获益巨大。很明显，我认为我们不应该探讨平行宇宙或另一个你，因为英国从你与 DeepMind 的合作中受益。下周我要去加州。我们在加州因这个人而自豪。你现在正在东南亚开始发展。是的。但真正出现的是，虽然感觉所有公司都在做类似的技术，但它们实际上正在成为不同类型的玩家。我一直这么说。你看，我不知道 OpenAI 发生了什么，但微软，我们通过使用 Office 就得到了所有东西。Anthropic 有它的专长。我们会看到这种类型。我不喜欢……我的钱一直押在谷歌上，不管有没有 Demis，因为我喜欢他们的价值观。他们有搜索引擎，他们有数据。实际上，我喜欢 Gemini。它是我最喜欢的。我的钱一直在那里。而且我这么说不是因为我坐在 Demis 旁边。但这不是一场只有一个赢家的竞赛。会出现许多不同类型的 AI 公司。我觉得这一切都很令人兴奋，我们将用它做些什么。

I'd just like to say in praise of Demis that I think the UK's done hugely well out of what Demis has done. It's obvious and I don't think we should explore the parallel or the alternate you because the UK benefits from what you've done with DeepMind. I go to California next week. We ride high in California because of this man. And you're starting out in Southeast Asia now. Yeah. But what's actually emerging is that while it feels like all the companies are doing the same sort of technology, they're actually emerging as different types of players. I've always said this. You see, I don't know what's happening to OpenAI but Microsoft, we're getting all that through using Office. Anthropic has its particular specialties. We're going to see this type. I don't like to make it... My money has always been on Google with or without Demis because I like their values. They've had the search engine, they've got the data. I love Gemini, actually. It's my favorite. It's always been where my money is. And I'm not just saying it because I'm sitting next to Demis. But it isn't a race that there will be one winner. There will be many different types of AI companies that emerge. I find that whole thing really exciting about what we're going to do with it.

好的，在我致谢之前，我想邀请汇丰创新银行 CEO Emily Turner 说几句。

Well, before I say my thanks, I'd just like to invite Emily Turner, CEO of HSBC Innovation Banking, for a last few words.

我们要感谢你们的充电宝，是吗？谢谢。谢谢。我是个学者，我总是会拿免费的东西。嗯，如果你想要更多周边，我想后面还有一些。所以我们可以安排。请和我一起感谢我们今天的尊贵小组成员，这是一场精彩的讨论。所以 Demis 爵士和 Wendy，非常感谢。显然，我还要感谢信息技术公司公会今天接待我们。我相信每个人都像我一样，在这样一栋美丽的建筑里，在组织的历史背景下讨论今天的主题，形成了巨大的对比。对于那些不了解我们的人，汇丰创新银行是一家英国专门为创新经济服务的银行。我们与这个国家的 4000 名创新者、企业家和投资者合作。正如今天大家谈到的，现在是在创新经济领域工作的绝佳时机，尤其是在英国，我们拥有人才、思想多样性、领域和专业知识的交汇，正如 Demis 所说，还有雄心。这是一个奇妙的时刻。仅第一季度，这个国家 74% 的风险投资资金流向了 AI 公司，这比几年前大幅增加。我知道有一个理论问题：如果 DeepMind 留在这里会怎样？我要指出，仅第一季度就有超过 10% 的资金来自由离开 DeepMind 的人创办的英国公司。所以我认为 DeepMind 的故事及其对这个经济的重要性将决定下一个不仅独角兽而且十角兽是否会来到这里。它的指纹将与 DeepMind 相连。所以我想指出这一点。

We've got to thank you for the power banks, have we? Thank you. Thank you. I'm an academic. I'll always take a freebie. Well, if you want more merch, I think we have some in the back. So, we can work that out. Please join me in thanking our esteemed panelists today for what was an amazing discussion. So Sir Demis and Wendy, thank you so much. I obviously also want to thank the Worshipful Company of Information Technologists for hosting us today. I'm sure everyone like me felt the great juxtaposition of coming into a beautiful building like this and the history of the organization along with the topic today. For those of you who don't know us, HSBC Innovation Banking is a UK purpose-built bank to serve the innovation economy. We work with 4,000 innovators, entrepreneurs and investors in this country. As everyone talked about today, this is an amazing time to work in the innovation economy and importantly in the UK where we have the confluence of talent, diversity of thinking, diversity of domains and expertise as Demis mentioned, and ambition. It's an amazing moment. In the first quarter alone, 74% of VC dollars in this country went into AI companies, which is a massive increase from just a few years ago. I know there was the theoretical question on what would have happened if DeepMind had stayed here. I will call out that over 10% of the funding in the first quarter alone came from companies in the UK started by people who left DeepMind. So I think the story of DeepMind and its importance to this economy will shape whether the next not just unicorn but decacorn comes here. Its fingerprints will be linked to DeepMind. So just wanted to call that out.

当然，对我个人而言，但我确信在座的每一位都觉得今天能参与这场讨论非常棒，听到那些把我们带到今天这个领域的人发言，更重要的是，你们三位都在发挥巨大作用，关系着我们的未来。所以请加入我们，喝一杯。让我们继续讨论，如果你有问题没有得到回答，希望人们会在那里可以继续。所以感谢大家，感谢我们的嘉宾。最后，我要说，会众 Demis Hassabis 爵士和会众 Wendy Hall 女爵士，我希望这不是我们最后一次在公司和会众中见到你们，但请随时参加我们的任何活动。你们将会非常受欢迎。女士们先生们，我能否请你们再次用热烈的掌声感谢 Demis 爵士和 Wendy 女爵士？

Um certainly for me but I'm sure for everyone else here it's been amazing to be part of the discussion today to hear from people who got us to where we are today in this space but importantly are playing such a huge role all three of you and where we go tomorrow. So please join us for drinks. Let's continue the discussion and if you didn't have questions answered, hopefully people will be around to continue it then. So thank you all and thank you to our panelists. So all that remains for me to say is Liveryman Sir Demis Hassabis and Liveryman Dame Wendy Hall, I hope this is not the last time we see you in our company and our livery, but please feel free to join us at any of our events. You'll be more than welcome. Ladies and gentlemen, can I ask you one more time to give Sir Demis and Dame Wendy a big hand?