本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
大家好，欢迎收听由汉娜·弗莱教授主持的谷歌DeepMind播客。
Hello, and welcome to the Google DeepMind Podcast with me, professor Hannah Fry.
正在观看的朋友会注意到今天我们不在演播室，因为我们正在英国皇家学会与谷歌DeepMind联合举办的'AI for Science论坛'后台。
Now those of you who are watching will notice that we are not in the studio today, and that is because we are backstage at the AI for Science Forum, a very special event that's been cohosted between the Royal Society and Google DeepMind.
作为活动环节之一，我有幸在台上采访了德米斯·阿比斯，我们觉得您可能会想听这段对话，所以特别将其收录为本期节目。
And as part of the day, I had the opportunity to speak to Demises Arbis on stage to interview him, and we thought that you might like to hear that conversation, so we have, included it here as a special episode.
更特别的是，一位诺贝尔奖得主还不够，我们还邀请了另外三位诺奖得主与我进行加长版小组讨论。
And then even more special than that, as though one Nobel laureate wasn't enough, we have three further Nobel laureates joining me on stage for an extended panel for you to enjoy.
好的。
Okay.
接下来的环节中，能采访今天的联合主持人德米斯·哈萨比斯爵士，对我而言总是莫大的荣幸，也常常是令人震撼的体验。
Now for the next session, it is, always an absolute treat for me and often a mind boggling experience to get to interview your your cohost for today, sir Demis Hassabis.
德米斯是计算机科学家。
Now, Demis is a computer scientist.
他是人工智能研究员。
He's an artificial intelligence researcher.
作为企业家，他于2010年联合创立了DeepMind并至今担任CEO。
He's an entrepreneur who co founded DeepMind in 2010, where he is still CEO.
我不清楚各位对德米斯背景的了解程度——除了诺贝尔奖、爵士头衔、皇家学会院士身份，还有AlphaFold、AlphaGo的成就和15万次论文引用量。
Now, I don't know how much you know about Demis' background because as well as the Nobel Prize and the knighthood and being a fellow of the Royal Society and AlphaFold and AlphaGo and having a 150,000 citations.
他早年还曾是杰出的神经科学家、国际象棋冠军和电子游戏设计师。
He also had a previous life as a very successful neuroscientist, chess champion, and video games designer.
说实话我有个理论：德米斯早就破解了通用人工智能的奥秘，一直把它藏在地下室里。
I have to say, I have a theory that actually, Demis worked out the secrets to AGI a little while ago, and has been hiding it in his basement.
然后像这样一点一点地向我们逐步释放这些重大突破。
And then just, like, slowly eking out all of these massive breakthroughs one by one to the rest of us.
但有一点可以确定，与他的对话总是令人着迷。
But one thing is for sure, our conversations are always genuinely fascinating.
那么，请大家欢迎德米斯上台。
So please, welcome Demis to the stage.
谢谢。
Thank you.
谢谢你，德米特。
Thank you, Demet.
非常感谢。
Thank you so much.
好的。
Okay.
我听说你事先不知道诺贝尔奖的事。
I heard that you didn't know that the the Nobel Prize was coming.
是真的吗？
Is that true?
不是。
No.
是的。
Yes.
不是。
No.
真的吗？
It's Really?
真的？
Really?
是啊。
Yeah.
你是怎么发现的？
How did you find out?
嗯，这其实是个挺有趣的故事，因为我们之前听到些传言说我们团队被提名了，但你根本不会预料到这种事。
Well, it was quite a funny story, actually, because we'd heard some rumor that our fold had been put put sort of nominated, but you never expect something like that.
那天早上，我正照常进行日常工作。
And in the morning, I was just getting on with my normal work.
其实我妻子当时也在家工作。
Actually, my wife was working at home as well.
到了大约10点半，我们想着，哦，看来今年没戏了，因为我们没收到任何消息。
And it got to about 10:30, and we thought, oh, well, obviously, it's it's not happened this year because we hadn't heard.
突然之间，我妻子的电脑通过Skype响了起来。
And then suddenly, my wife's computer started ringing on her Skype, I think it was.
我当时还在想，这烦人的声音是什么？
And there's I was like, what's that annoying noise?
结果发现是来自瑞典的电话，他们正拼命想联系上我。
And then it turned out to be a call from Sweden, and they were desperately trying to get hold of my number.
他们也没有约翰的联系方式。
And they didn't have John's number either.
所以所有人都有些慌乱，但这反而增添了当时的戏剧性。
So everyone's in a kind of panic, but it all added to the drama of the moment.
就在宣布前几分钟才联系上。
Just You had to minutes before yes.
没错。
Exactly.
就在即将公布前的最后一刻。
Just before it was all announced.
简直太不可思议了。
Absolutely amazing.
是啊。
Yeah.
我还看到你们通过举办扑克之夜来庆祝，参与者包括一些国际象棋冠军，比如马格努斯·卡尔森和Hikara。
And and I also saw that you celebrated by having a poker night with some chess champions, including Magnus Carlsen and Hikara as well.
我想知道谁赢了？
I I wanna know who won?
你的虚张声势策略是什么？
What's your bluffing strategy?
嗯，确实如此。
Well, it turned out yeah.
这件事发生在周三，当时伦敦正在举办一场大型国际象棋锦标赛。
So that happened on I think it was on the Wednesday, and it turned out that there there was a big huge chess tournament going on in in in London.
我的一位国际象棋老友——从我年轻时起就认识的——在第二天晚上组织了一场扑克与国际象棋之夜。
And one of my chess friends, old chess friends from when I was young, was hosting a poker and chess evening the net the the day after.
所以我觉得这是庆祝的完美方式。
So it was the I thought it was the perfect way to celebrate.
但如果你想赢钱，我不建议在家里和几位世界扑克冠军及前国际象棋冠军玩扑克。
And but it's I don't recommend a poker home poker game with couple of world poker champions and a couple of ex world chess champions if you want to win some money.
不过这就是我心目中的乐趣。
But it's it's that's my idea of fun.
那真是个美妙的夜晚。
It was actually an amazing night.
我知道这听起来很书呆子气，但这确实是我庆祝时理想中的天堂。
I know it sounds pretty pretty pretty nerdy, but it's it was actually my my idea of of heaven to celebrate So
听起来像是
Sounds like
是啊。
a Yeah.
我不能告诉你虚张声势的策略，否则我就再也赢不了马格努斯了。
I can't tell you the bluffing strategy, otherwise, I'll never beat Magnus again.
是啊。
Yeah.
但这是事实。
But this is true.
这是事实。
This is true.
好的。
Okay.
所以，我是说，除了诺贝尔奖之外，这句话没人说过。
So, I mean, aside from aside from the Nobel Prize, a sentence that no one has ever said.
但在此之前，AlphaBold的研究成果已被引用超过28,000次，这令人难以置信，因此你也被评为高被引学者。
But the before that, in advance of that, you also were named as a citation laureate because the AlphaBold work has now been cited over 28,000 times, which is incredible.
我是说，这项成果问世其实并不久。
I mean, this is really not very long ago that it that it came through.
有哪些突出的应用案例真正引起了你的共鸣？
Are standout applications for you that really resonate?
嗯，说实话这确实令人震惊。
Well, I mean, it is mind boggling really.
这正是我们希望通过公开AlphaFold、开源代码、与欧洲生物信息研究所（EBI）的优秀合作伙伴（比如在场的珍妮特和尤恩）共建社区所实现的目标——让所有人都能使用这项技术。
It's everything we hoped would happen by putting AlphaFold out there and open sourcing it and putting out community with our fantastic partners at Embol EBI, like Janet and Ewen in the audience as well, to get it out there to everyone.
我是说，有太多应用案例难以一一列举。
And I mean, there's so many applications we you know, to to mention.
但或许可以重点说说我最喜欢的两三个案例，比如确定核孔复合体结构——这是人体内最重要且体积最大的蛋白质之一。
But I think maybe I can mention two or three of my favorites were, you know, the the the this the kind of determining the structure of the nuclear pore complex is one of the most important and largest proteins in in the human body.
它之所以关键，是因为掌控着进出细胞核的分子和营养物质。
And it's really important because it governs, you know, the molecules and nutrients that go in and out of your cell nucleus.
实际上，多个研究团队结合AlphaFold的预测数据和自己的实验数据，最终拼凑出了这个极其复杂的蛋白质结构。
And, actually, several teams used alpha fold predictions as well as well as their own experimental data to finally piece that very, very complex protein structure together.
那真是太棒了。
So that was amazing.
我特别喜欢的另一项工作是张锋在博德研究所开发的分子注射器，它能将药物有效载荷输送到体内难以触及的部位。
Another piece of work that I really like is Feng Zhang's work from the Broad developing a molecular syringe to deliver sort of drug payloads to hard parts of the body to reach.
他还利用AlphaFold帮助改进了这种分子注射器的设计。
And he also used AlphaFold to help modify how that syringe molecular syringe was designed.
最后可能要说的是，我最喜欢的另一个项目来自朴茨茅斯大学John McGeehan团队设计的塑料降解酶。嗯。
And then maybe finally, my other favorite project is from the University of Portsmouth, John McGee McGeehan's group designing pro plastic heating enzymes Mhmm.
他们使用了AlphaFold。
Using using AlphaFold.
但我觉得这只是研究人员用它做出的所有惊人成果中的一小部分样本。
But I think it's just, you know, a small sample of all the incredible work researchers are doing with it.
不过这类项目应该算是你的最爱吧？
But these kinds of projects, I mean, are your favorites, right?
就是那些不仅仅停留在技术构想本身的案例。
The ones where it's not just the sort of the technical idea itself.
而是能产生深远下游影响的潜力。
It's the potential impact that it can have further downstream.
没错，正是如此。
Yeah, that's right.
我一直对蛋白质折叠问题着迷的原因就在于，蛋白质结构预测问题就像我们常说的'深度思维根节点问题'。
I mean, that's why I was always intrigued by the protein folding problem and that, know, protein structure prediction problem was because I felt, you know, we sometimes call it a deep mind, a root node problem.
这么说的意思是，如果把知识体系比作一棵树，有些问题就像根节点——一旦攻克它们，就能开辟全新的研究分支。
What we mean by that is if you think of the whole tree of knowledge, you know, there are certain problems where, you know, if they're root node problems, if you unlock them, if you discover a solution to them, it would unlock a whole new branch or avenue of discovery.
我始终认为，通过这种方式解析蛋白质结构将带来疾病认知、药物设计等领域的突破。
And I always felt that, you know, determining protein structure in this way would do that, you know, lead to disease understanding, drug design, and much more.
现在事实似乎正是如此。
And that seems to be what's happened.
好的。
Okay.
所以，不要贪心。
So, not to be greedy.
是的。
Yeah.
但接下来呢？
But what's next?
难道还有另一个Alpha雾吗？
What is there another alpha Fog?
我是说，GNOME是我个人非常喜欢的一个项目。
I mean, GNOME is a personal favorite of mine.
也许可以给我们简单介绍一下那个项目。
Maybe tell us little bit about that project.
是的。
Yes.
关于GNOME，你看，它涉及太多领域了——实际上James早上提到过，Pushmi也谈到过——我们几乎触及了科学的每个领域。
So, GNOME, look, there's so many areas and actually James mentioned it in the morning and Pushmi talked about as well that we're touching on nearly every area of science.
而Gnome是我最喜欢的项目之一，它专注于材料设计。
And Gnome is one of my favorite projects, which is on material design.
我认为材料设计具有与AI适用问题相同的某些特征。
And I think material design has the same char-, some of the same characteristics that we look for in a problem that's suitable for AI.
这是一个巨大的组合空间。
It's massive combinatorial space.
你需要尝试建立一个能理解自然现象物理和化学特性的模型。
You need to try and build a model that understands the physics and the chemistry of the natural phenomena.
如果你有了这样的模型，或许就能用它在这个组合空间中进行高效搜索，从而找到最优解。
And then if you have a model like that, you can then maybe use it to do a very efficient search through that search space, combinatorial space, and then find an optimal solution.
在材料科学领域，我认为这同样具有突破性意义。
And in materials, I think that would be also just as groundbreaking.
你可以想象设计新型电池，或者有朝一日发现室温超导体——这一直是我的梦想之一。
You could imagine designing new batteries or maybe one day discovering a room temperature superconductor has always been one of my dreams.
因此我认为我们显然还处于这个领域的早期阶段。
And so I think we're obviously at the early stages of that.
我可能会将其描述为类似AlphaFold第一代的水平。
I would characterize it at sort of alpha fold one level maybe.
我们需要将预测能力提升到AlphaFold第二代的水准。
And we've got to get to alpha fold two level of predictions.
但我们能清晰地看到实现路径。
But we can see a clear path from there.
而Gnome项目就是这项工作的开端。
And Gnome is the beginning of that work.
我们在《科学》杂志上发表了相关成果，应该是去年的事。
We published it in, I think it was in Science last year.
我们发现了20万种前人从未见过的新型晶体结构。
And we discovered 200,000 new crystals that no one had ever seen before.
所以你看，我认为这展现了AI在材料设计等领域的部分潜力。
So, you know, I think that shows some of the potential of AI and things like material design.
另外让我兴奋的是AI在数学领域的应用，或许能解决重大猜想——比如千禧年难题之一，让AI成为解题过程中不可或缺的重要工具。
And then maybe the other thing I'm excited about is AI applied to mathematics and perhaps solving one of the great conjectures, maybe one of the millennium prize problems and using AI as a material, as a big part of that solution.
关于Gnome和AlphaFold的理念在于，你实际上是在让合成世界为你所用，从而走捷径。
So that idea though about Gnome, also about AlphaFolders is, you know, you're sort of shortcutting or you're making the sort of synthetic world really serve your advantage.
这样就不必完全依赖实验手段。
So you're not having to do everything through experimentation.
没错。
Yeah.
我知道你和保罗·纳斯一直在讨论虚拟细胞这个概念已经很久了。
I know that you and Paul Nurse have been talking about a virtual cell for a long while.
给我们简单讲讲这个吧。
Tell us a little bit about that.
保罗一直是我的导师之一。
Well, Paul has been one of my mentors.
他非常慷慨地指导我学习生命科学，至今已有二十五年左右了。
He's very generously mentored me in the life sciences for you know, more than twenty five years now or something.
能定期和保罗探讨这些话题，总是让人感到不可思议的鼓舞。
And it's been incredible sort of inspiring talking to Paul very regularly about these topics.
特别的是，我认为保罗作为生命科学家和生物学家，
And unusually, I think Paul is for a life scientist is biologist.
他长期将生物学视为信息系统来思考。
He's thought a lot about biology as an information system.
他在这方面发表过许多有趣的研究论文。
He's written a lot of interesting research papers on that.
所以我们始终在背景中保持着这样的讨论。
So we've always had we've always had this discussion in the background.
我大概每五年就会思考一次：我们是否已掌握足够的技术和知识，来真正尝试构建虚拟细胞这个如同攀登珠峰般的难题？
And I've always wondered every sort of five years or so, I've thought of thought, do we have enough technologies, enough know how to actually really attempt this, you know, Mount Everest of a problem of trying to build a virtual cell?
本质上就是建立一个能真实预测细胞行为的模拟系统。
A simulation of a cell basically that would be predictive of something that's really going to happen.
但每隔五年，我都觉得我们还没准备好，技术储备仍不足。
And and every five years, I thought we're not, you know, we don't really have enough technologies yet.
不过现在终于可以说答案是肯定的。
But finally now, I think the answer is yes.
我们很可能已经掌握了足够的知识和技术，可以正式启动这个项目了。
We probably do have enough know how and techniques that we could attempt this seriously now.
也许在未来五到十年内，我们最终能够构建出一个虚拟细胞的图景，可能先从酵母细胞开始，正如保罗一直致力于此，将其作为模型生物。
And maybe in the next five to ten years, we could eventually build up a picture of of a virtual cell, perhaps a yeast cell to start off with, as Paul's always been working on this, as the model organism for this.
我的思考方式是，你可以把AlphaFold二代视为本质上解决了蛋白质静态结构的问题。
And my way of thinking about is, you can think of AlphaFold two as essentially solving the static picture of of what a protein looks like.
但当然，我们知道生物学是一个动态系统。
But of course, we know biology is a dynamic system.
这正是生物学中所有有趣现象发生的地方。
That's where all the interesting things happen in biology.
而AlphaFold三代是我们尝试模拟这些相互作用的第一步。
And AlphaFold three is our first step towards trying to model those interactions.
所以AlphaFold三代可以模拟蛋白质与蛋白质、蛋白质与RNA、蛋白质与DNA之间的成对相互作用。
So alpha vol three can model, you know, pairwise interactions between proteins and proteins, proteins and RNA, proteins and DNA.
接下来的升级可能是模拟整个通路。
And then maybe the next step up from that will be to model a whole pathway.
最终或许我们能构建出完整的细胞模型。
And then eventually, maybe we can make it to an entire cell.
太惊人了。
Amazing.
我想量子计算的出现将会改变这一领域，使得分子层面的模拟成为可能。
I mean, I imagine that the advent of quantum computing will make a difference to this, that's possible to do simulations down at the molecular level.
是的。
Yeah.
量子计算确实非常令人兴奋。
So look, quantum computing is very exciting.
它本身正在加速发展。
It itself is accelerating.
詹姆斯今早也提到了人工智能与量子计算之间有趣的交叉融合现象。
And and James also mentioned earlier this morning about the interesting cross pollination that's happening between AI and quantum computing.
事实上，我们与谷歌的量子计算团队合作密切，他们在纠错码等领域是全球顶尖的量子计算团队之一。
Actually, we collaborate a lot with our quantum computing group at Google, which is one of the world's best quantum computing teams on things like error correction codes and things like that.
当然，量子计算机的用途之一就是模拟分子、原子等量子系统及化合物，从而可能产生大量合成数据。
But and and of course, one of the uses of a quantum computer would be to simulate quantum systems like molecules and and atoms and things like that and compounds, and then produce a lot of synthetic data potentially.
但有趣的是，我有个略带争议的观点——我曾与多位世界顶级量子计算机科学家探讨过——实际上我认为经典图灵机（传统计算机）的潜力远超我们过去的认知。
But interestingly, I also have a slightly controversial take, which is and I've I've talked to some of the world's top quantum computer computer scientists on this is that, actually, I believe that classical Turing machines, classical computers are capable of lot more than we previously thought.
我认为AlphaFold的工作成果证明了这点，此前的AlphaGo（我们的围棋世界冠军程序）也是例证。要知道，围棋的复杂度远超国际象棋。
And I think that's what the work that we've been doing has showed both with AlphaFold, but also previously with AlphaGo, our program to beat the world champion at the game of Go, which, you know, just as an indication, the complexity of Go, much more complicated than chess.
其可能的棋盘布局数量甚至超过宇宙原子总数——10的170次方。
And has more possible board positions than there are atoms in the universe, 10 to the power 170.
这意味着你无法通过暴力计算找到特定棋局的最佳落子。
So what that means is you can't possibly brute force a solution to find the best move in a particular position.
必须采用更精妙的方法。
You have to do something much cleverer.
蛋白质折叠同样存在海量可能性，如果采用最原始的方法尝试每种组合（这是最笨拙的解决方式）。
And so both protein folding also has enormous amounts of possibilities if you were to do it naively and just try every combination, which is the naive way of doing it.
但如果你进行大量预计算，在提问前先构建系统模型——比如'这个围棋局面该如何落子？'
But, yeah, if you do things like you do a huge amount of precompute, to build a model of the system before you ask it the question that you're interested in, you know, what move should I play in this go position?
'这个新型蛋白质会如何折叠？'
How does this novel protein fold?
事实证明，你可以在几秒内得出接近最优的围棋落子，或在几分钟内完成蛋白质折叠——这些在蛋白质空间领域原本被认为需要量子计算机或量子算法才能实现。
It turns out you can actually come back with a near optimal move and go in a few seconds or fold a protein in a matter of minute minutes, which you might have expected in in protein space at least that you might need a quantum computer for or quantum algorithm for.
但事实证明并非如此。
And it turns out you don't.
因此我认为应该认真看待这个发现：如果使用方法得当，经典系统或许能模拟更复杂的系统——甚至反直觉地模拟量子系统。
So I actually think we should take this very seriously that, classical systems, if you use them in the right way, may be capable of modeling a lot more complex systems, perhaps even counterintuitively quantum systems.
因为通常人们认为只有量子计算机才能模拟任何经典系统。
Because normally you talk about needing quantum computers to model model any kind of classical system.
但经典系统或许能够模拟量子系统。
But it may be that classical systems can model quantum systems.
我已经和一些人测试过这个想法，比如最近因开创性量子计算工作获得诺贝尔物理学奖的蔡林格教授。
And I've tested this out with some of the, you know, people like professor Zeilinger who won the Nobel Prize in Physics recently for his pioneering quantum computing work.
他觉得这个想法非常有趣。
And, you know, he thought it was very interesting.
还有大卫·多伊奇——我心目中的科学英雄之一，量子计算的奠基人——他说这想法很疯狂，但是正确的疯狂。能得到他这样的评价，我视作一种赞美，也暗示或许值得继续深入研究。
And David Deutsch, also one of my scientific heroes who's basically invented quantum computing, said it was crazy, but the right sort of crazy, which coming from him, I take as a compliment and as also a sign to maybe pursue it further.
那么让我确认下是否理解正确。
So, let me make sure I understand this then.
传统计算机通常被认为是确定性机器。
So I guess the classical computers sort of thought of as deterministic machines.
而你在讨论概率性...但可以用确定性机器来...是的。
And you're talking here about probabilistic But you can use the deterministic ones to Yes.
核心观点是：量子系统或任何自然现象，如果用经典系统去穷举所有可能性来建模，很快就会耗尽算力。
The idea is, you know, the quantum systems, anything you try to model, any natural phenomena, if you start trying to model it by every possibility that it could take that system, then you quickly run out of computation in a classical system.
所需的比特数会超出极限。
You need too many bits to model it.
但这本就不是经典系统的正确使用方式。
But that's just not the way that you would attempt to do it with a classical system.
你会先建立一个学习模型。我的猜想是：任何自然现象都存在某种结构规律。
You you would build a model first that learns You know, my my conjecture on it would be that any natural phenomena tends to have structure.
如果存在结构规律，理论上就能用经典机器学习系统来学习它。
And if it has structure, you could potentially learn it with a classical machine learning system.
学习出它的高效模型。
Learn an efficient model of that.
然后用这个模型来高效地搜索可能性。
And then use that to search the possibilities in an efficient way.
因此我认为这可能规避一些原始方法中的低效问题。
So I think that might get around some of the inefficiencies from a naive way of doing this.
所以，我的意思是，这是个相当大胆的主张。
So, I mean, it's a pretty big claim.
因此我选择用温和的方式提出。
So I'm I'm making it in a soft way.
但你知道，研究这个领域算是我的个人爱好，我认为这个方向前景广阔。
But I it's, you know, it's just kind of hobby of mine to to look at this this area and it's something I think that could be quite promising.
你的另一个爱好现在已经独立发展成为同构公司了。
Another hobby of yours that has now spun off into its own into its own business of isomorphic.
因为这项工作是将AlphaFold应用于药物研发。
Because, I mean, this is work using AlphaFold to apply to drug discovery.
是的。
Yes.
你现在还拥有一些相当高端的合作伙伴。
You've got some pretty prestigious partnerships now as well.
能告诉我们同构公司目前的主要研究方向吗？
Can you tell us what isomorphic is focusing on at the moment?
是的。
Yes.
同构是我们成立的子公司，旨在用AI从第一性原理彻底革新药物研发，从根本上重新构想整个药物发现流程。
So, isomorphic is our spinout to try and revolutionize drug discovery from first principles using AI as a sort of from the ground up, trying to reimagine the drug discovery process.
我们在开发AlphaFold时就有这个构想。
And I had this in mind when we were doing AlphaFold.
我认为治愈疾病显然是AI最佳的应用场景之一。
So, I think it's one of the most obviously good use cases of AI is to cure diseases.
还有什么比用AI来治病更好的用途呢？
I mean, what better use of AI could there be?
因此，这始终是我希望AI成熟后能实现的首要目标之一。
And so that has always been one of the number one things I wanted to do with AI once it got mature enough.
当然，AlphaFold是基础研究和基础生物学研究的绝佳工具。
And so AlphaFold, of course, is a great tool for fundamental research and fundamental biology research.
你知道，全球已有超过200万研究人员使用了AlphaFold和我们发布的结构数据。
You know, over 2,000,000 researchers around the world have used AlphaFold and the structures we put out there.
但它也能实际应用于辅助药物研发。
But it's also a practical use to help with drug discovery.
当然，了解蛋白质结构只是整个药物研发过程中的一小部分。
Of course, knowing the structure of a protein is only one small part of the whole drug discovery process.
所以在完成AlphaFold2后，我们分拆成立了Isomorphic公司，以延续这项工作并构建新的机器学习系统，拓展到相关领域。
And so we spun out isomorphic after we did AlphaFold two to build on that work and and extend and build new machine learning systems into adjacent areas.
比如设计化学化合物和药物化合物、毒性测试，以及预测ADME特性（药物在体内的吸收、分布、代谢和排泄特性）等关键属性，这些都是确保药物有效并减少副作用的重要因素。
Things like designing chemical compounds and drug compounds, testing for toxicity, and predicting things like ADME properties, important properties that you need for drugs to work in the body and minimize side effects and things like that.
我们正在构建——你可以理解为在这些相关领域开发更多类似AlphaFold的模型。
And we're building, you can think of as building up further alpha fold like models in these adjacent areas.
最终我们会把所有系统整合起来。
And then eventually we'll stick them all together.
我们希望在不远的将来，能将药物设计时间从数年甚至十年缩短到数月，或许几周就能完成。
And we hope that, you know, one day in the near future, actually, we'll reduce the time down from years, maybe even decade to design a drug down to months or perhaps even weeks.
我认为这将彻底改变药物研发的进程。
And that would revolutionize, I think the drug discovery process.
这确实是那种无需在造福人类和实现盈利可持续发展之间做选择的事业。
This really does seem like one of those endeavors where you're not having to make a choice between doing something that benefits humanity and doing something that that can be profitable and and and sustain itself.
你们最初会如何寻找这类项目？
How much do you look for those kind of projects at the outset?
或者说，你们是先进行自由探索性研究，然后期待最终能达成这样的结果吗？
Or how much of it is that you do some blue sky research and then hope that it'll turn out that way in the end?
我认为我们两者兼顾。
I think we do both.
所以我们遵循的是，你知道，我们团队一直以研究为导向。
So we follow, you know, we're research led as a group, always have been.
因此我们试图确定研究的正确下一步，比如AGI（人工通用智能）或类似AlphaFold的项目，这些真正致力于解决科学挑战的领域。
So we try to do what's the right next step for researching, you know, AGI, artificial general intelligence, or something like AlphaFold, where you really want to solve the the scientific challenge.
但在内心深处，我也非常务实。
But then in the back of my mind, I'm also pretty practical.
所以我希望解决的问题也能对世界产生直接的实际积极影响。
So I I want to solve things that will also have a practical positive impact on the world pretty directly.
如果能找到两者兼具的项目，那简直就是圣杯般的存在。
And so if you can find projects that do both, then that's, you know, the holy grail really.
值得投入大量时间和精力的事情。
Something that really worthwhile investing a lot of time and effort.
AlphaFold正是这样的项目。
And AlphaFold was exactly like that.
我对这个问题本身非常着迷。
So I was fascinated by the problem itself.
正如珍妮特之前提到的，越是深入研究蛋白质，越能发现它们是精妙的生物纳米机器。
As Janet mentioned earlier, these, the more you get into proteins, they're exquisite bio nano machines.
我的意思是，它们美得难以置信。
I mean, they're unbelievably beautiful.
在这一点上我完全同意珍妮特的看法。
I totally agree with Janet on that.
当你开始研究它们时，就会不由自主地爱上它们。
And you sort of fall in love with them when you start working on them.
这简直不可思议。
And it's just unbelievable.
自然是经过设计的。
Nature's engineered.
而且，但你知道，他们一直认为那会是一种社会公益，但同构性可以兼顾两者。
And, but then, you know, you, they always had in mind that that would be a social good, but then isomorphic can do both.
对吧？
Right?
我认为我们可以借助AI治愈许多疾病。
I think we can, you know, cure many diseases with the help of AI.
但它同时也应该是一家极具价值的公司。
But also it should be an incredibly valuable company.
当然，如果事实证明如此，那将为我们提供更多资金投入基础研究。
And that, course, if that's turns out to be the case, that will give us more money to invest in fundamental research.
所以这是一个良性循环。
So, it's all a virtuous cycle.
在基础研究领域有这么多不同的项目同时进行。
With all of these different projects going on in fundamental research.
你觉得我们现在是否正处于一个即将迎来垂直腾飞的时刻，就像
I mean, do you think that we are at this moment where we're going to have a sort of a vertical takeoff as it
在进步方面？
were on progress?
我认为我们正处在这个临界点。
I think we are on the cusp of that.
我真的感觉我们正站在新发现黄金时代的门槛上，就像今天整个研讨会的主题所说的那样。
I really do feel like we're on the brink of a new golden era of discovery, like the whole of today's, you know, symposium is called.
我认为我们需要的是更多跨学科的科学合作。
And I think what we need is a lot more interdisciplinary science.
因此，要善用AI，以正确的方式引入，与领域专家共同提出正确的问题。
So, using AI, bringing in the right way, asking the right questions with domain experts.
我认为它的应用潜力几乎是无限的。
And I think it's almost limitless what its applications could be.
当然，人工智能本身作为一门科学学科，也在不断进步。
And of course, AI itself is a scientific discipline, is improving all the time.
所以现在可以直接将现有技术应用到其他领域。
So there's applying today's technologies directly to the other fields.
同时我们也在持续改进人工智能本身。
And then there's also continuing to improve AI itself.
这种改进本身也是一种指数级的进步。
And that as well is a sort of exponential improvement.
所以，我认为未来几年内将会取得大量进展。
So, you know, there's a lot of, I think, progress to be made in just the next few years.
我认为跨学科性绝对是今天的重要主题之一。
I think interdisciplinarity has definitely been one of the big themes of today.
但其他主题则关乎未来科学的面貌。
But I think that the other themes are about what science looks like in the future.
在人工智能时代，从事科学研究意味着什么？
You know, what does it mean to do science in an era of AI?
我知道你非常推崇科学方法。
And I know that you're a sort of big advocate for the scientific method.
能否请你简单谈谈这个问题？
But just talk to me a little bit about that.
随着我们不断前进，科学方法会呈现怎样的形态？
What does that look like as we kind of move forwards?
嗯，你看，
Well, look,
我认为科学方法可以说是人类有史以来最伟大的构想。
I think the scientific method is arguably maybe the greatest idea humans have ever had.
我认为这显然支撑着所有科学领域，同时也正是由于这些科学技术，才有了现代文明。
And I think underpins, obviously, all of science, but then also because of that science and technology, you know, modern civilization.
我认为在当今世界，我们比以往任何时候都更需要围绕这种方法来定位。
And I think more than ever, we need to anchor around that method in today's world.
特别是对于像人工智能这样强大且具有变革潜力的技术，我认为重要的是我们应更多采用科学方法，而非依赖常规的技术AB测试。
And I think especially with something as powerful and potentially transformative as AI, I think it's important we use more the scientific method than perhaps lean on normal sort of technology AB testing out in the wild.
新技术往往都是如此。
That is often the case with new technologies.
如果可能的话，我觉得我们应该将其更多地视为一项科学探索。
And I I feel we should treat this more as a scientific endeavor, if possible.
当然，这项突破性技术也具备所有新技术通常具有的特点——在采用速度和变革速度方面。
Although, obviously has all the implications that that that breakthrough technologies normally have in terms of the speed of adoption and the speed of change.
因此我们正处于一个有趣的局面。
So it's an interesting situation that we're in.
我认为我们需要运用科学方法来更好地理解这些系统的运作，建立基准和严格评估以了解其能力边界及可解释性。
And I think we need to use a scientific method to to to do things like better understand what these systems do, build benchmarks and rigorous evaluations for understanding the limits of the capabilities, interpretability.
实际上，我认为应该使用神经科学技术——这些本就是为理解真实大脑而设计的方法——来建模虚拟大脑。
Actually, I think we should be using neuroscience techniques to which obviously built for understanding real brains to model virtual brains.
我有时称之为虚拟大脑分析。
You know, I sometimes call it virtual brain analysis.
这些神经网络的功能磁共振成像等效技术是什么？
You know, what's the equivalent of an fMRI of these neural networks?
对吧？
Right?
我认为有很多知识可以从自然科学借鉴转化到这门本质上属于工程科学的领域。
And I think there's a lot we can learn and bring across and translate from, from the natural sciences, to what is basically an engineering science.
我称之为工程科学是因为与自然科学不同，你必须先构建出目标人工制品。
I call it an engineering science because, unlike the natural sciences, you have to build the artifact of interest first.
一旦掌握了它，你就可以运用科学方法将其分解并理解其组成部分。
And then once you have it, you can then use the scientific method to to reduce it down and understand its components.
所以，这是一个艰巨的挑战。
So, so it's a hard challenge.
这一点是肯定的。
That's for sure.
在我看来，这些系统与我们通常想要研究的自然现象一样复杂。
And these systems are as, in my opinion, are as complex as the natural phenomenon we normally want to study.
因此，没有人应该认为理解这些人工系统会比理解自然系统更容易。
So no so no one should think that it will be easier to understand one of these artificial systems than it would be to understand a natural system.
我认为它们在许多方面同样复杂。
I think they're just as complex in many ways.
所以我们面前还有很多工作要做，我认为这是产业界、学术界和民间社会需要共同努力来更好地理解的课题，包括如何部署这些技术。
So we got a lot of work in front of us, and I think this is something that industry, academia, and civil society needs to come together to understand better and including how to deploy these technologies.
科学家的直觉在这其中处于什么位置？
Where does a scientist's intuition fit in all of this?
我认为科学家的直觉和创造力至关重要。
Well, I think scientists' intuition and creativity is critical.
我认为目前人工智能系统只是工具。
I think right now the AI systems are just tools.
它们非常擅长发现数据中的关联性、模式和结构。
I think they're great for finding correlations and patterns and structures in data.
但就目前而言，它们还无法提出自己的假设或问题。
But for the moment, they're not able to come up with their own hypotheses or their own questions.
我想在座的科学家都知道，科学最难的部分就是提出正确的问题。
And I think as all the scientists in the room know, I think that's the hardest thing about science is asking the right question.
而且，在可预见的未来，我认为这仍将来自人类科学家。
And, you know, that has to still come from human scientists and for the foreseeable future, I think that will be the case.
所以我认为这可能是，你知道，这就是为什么我对使用这些AI系统感到非常兴奋，它们或许能成为帮助我们加速科学发现的终极工具。
So I think it could be, you know, that's why I'm very excited about the use of these AI systems as maybe the ultimate tools to help us accelerate scientific discovery.
我想关于人类这个话题，最近有研究表明近年来科学发现的进展速度有所放缓。
I guess on that topic of humans, I mean, there has been some recent research saying that the progress of scientific discovery has slowed down over recent years.
你对此有何看法？
What's your take on that?
可以采取哪些措施来改变这种状况？
And what could be done to change that?
嗯，我也看过类似的研究报告。
Well, look, I think things I've seen studies like that too.
我认为关于原因的各种推测很有意思。
And I think there's interesting conjectures as to why that is.
要知道，科学已经变成一项更庞大的事业。
You know, science has become bigger endeavor.
需要组建更大的团队，购置更昂贵的设备等等。
You need to have bigger teams, more expensive equipment, and so on.
这自然会导致进展放缓。
So that leads to some slowdown.
而且我们现在要解决的问题可能比以往更加复杂。
And perhaps the questions we're now tackling are ever more complex.
所以这也很棘手。
So that's also tricky.
我的建议是，我认为未来十年的重大突破将主要来自跨学科合作。
I think what I'd recommend is, again, I think a lot of the advance in the next ten years are gonna be interdisciplinary work.
就是把来自两个或多个领域的专家聚集在一起。
But sort of, you know, bringing together experts from from two or more fields.
然后在这些领域的交叉处实现重大突破。
And and then that making the big advances in between those areas.
事实上这就是DeepMind的故事，最初我认为它是神经科学理念与机器学习理念的结合。
And actually that's the story of DeepMind, which originally I think was a combination of neuroscience ideas with machine learning ideas.
这也是AlphaFold的故事，它是我们团队中生物学家、化学家与机器学习专家和工程师的协作成果。
It's also the story of AlphaFold, which is combination of biologists and chemists on our team, as long along with machine learning experts and engineers.
因此我一直认为这是能最快取得重大进展的途径。
So I've always found that's the place to get the most advances most quickly.
实际上我们今天宣布，谷歌.org将提供2000万美元基金，用于资助学术界的这类跨学科研究工作。
And actually we're announcing today, you know, dollars 20,000,000 fund from google.org to kind of fund this kind of interdisciplinary work in academia.
我希望其他资助者也能加入这项事业。
And I hope that other funders, you know, join that effort.
我认为我们需要培养新一代的博士生和博士后，让他们掌握这些不同领域的交叉知识。
And I think that's what we need is to train a new generation of PhDs and postdocs in these kind of combination of these different areas.
作为结尾，这真是个非凡的消息。
An extraordinary piece of news to finish on.
2000万美元用于支持研究和跨学科团队。
$20,000,000 for research and interdisciplinary teams.
德米斯，真的非常感谢你。
Demis, thank you so much indeed.
非常感谢你，德米斯。
Thank you very much, Demis.
好的。
Okay.
德米斯，在我们布置舞台时，请你先下去与其他获奖者会合。
I think, Demis, if you go down to join the other laureates while we sort out the stage.
现在我要告诉你们，实际上'获奖者'这个词源自月桂花环，那是古希腊授予胜利者的荣誉象征。
Now I I should tell you, actually, the the laureates the word laureates, by the way, comes from the laurel wreath, which was given to victors in ancient Greece as a sign of honor.
每年都会颁发五个独立奖项，授予那些被认为为人类带来最大福祉的人，这要归功于阿尔弗雷德·诺贝尔。
And every year, there are five separate prizes that are awarded to those who are considered to have conferred the greatest benefit to humankind, courtesy of of Alfred Nobel.
有传言说，数学界没有诺贝尔奖是因为阿尔弗雷德·诺贝尔的挚爱跟一位数学家私奔了。
There is a rumor that there is no bell no Nobel Prize for mathematics because Alfred Nobel's great love ran away with a mathematician.
你听过这个说法吗？
You heard this?
我是说，我们确实很有魅力。
I mean, we are charming.
我能说什么呢？
What can I say?
现在，与我一同登台的还有三位这一崇高奖项的获得者。
Now, so joining me on the stage, we have three further recipients of this esteemed prize.
约翰·詹珀，我们此前还未有幸邀请他登上这个舞台。
John Jumper, we've not yet had the pleasure of having John Jumper on the stage thus far.
他是谷歌DeepMind的董事，曾领导团队开发AlphaFold，并持续致力于将机器学习应用于蛋白质生物学的新方法研究。
He is a director at Google DeepMind, where he led the team that built AlphaFold and continues to work on new methods to apply machine learning to protein biology.
约翰因其工作获得了无数奖项，包括拉斯克奖、生命科学突破奖、加拿大国际奖，当然还有今年的诺贝尔化学奖。
John has won numerous awards for his work, including the Lasker Prize award, the Breakthrough Prize in Life Sciences, the Canada International Award, and of course, this year's Nobel Prize in Chemistry.
我们还有幸邀请到保罗·纳斯爵士，他是弗朗西斯·克里克研究所的CEO，因在蛋白质——特别是控制细胞分裂周期的蛋白质分子方面的研究，获得了2001年诺贝尔生理学或医学奖。
We are also joined by Sir Paul Nurse, who is CEO of the Francis Crick Institute and winner of the 2,001 Prize in Physiology of Medicine for his work also on proteins, specifically the protein molecules that control the division of cells in the cell cycle.
最后是詹妮弗·杜德纳，她早些时候与詹姆斯分享了CRISPR技术的研究心得。
And finally, Jennifer Doudna, who earlier shared her lessons from CRISPR with James.
她于2020年获得了诺贝尔化学奖。
She won the Nobel Prize in chemistry in 2020.
请大家和我一起欢迎四位诺贝尔奖得主登台。
Please join me in welcoming our four Nobel Laureates to the stage.
谢谢。
Thank you.
好的。
Okay.
我知道人们经常问你，当你听到消息时你在哪里，当时在做什么。
So I I know that often you get asked about where were you when you heard, what were you doing when you heard.
我想问你一个稍微不同的问题。
I wanna ask you a slightly different question.
我想问你是否曾有过这样的时刻，意识到你正在做的工作是真正具有开创性的。
I want to ask you about whether there was a moment when you realized that the work that you were doing was genuinely groundbreaking.
是否有某个瞬间让你意识到它的重要性？
Was there sort of a moment when you realized the significance of it?
约翰？
John?
我认为
I think
对我来说有两个关键瞬间。
there were kind of two moments for me.
一个是发布成果后刷推特时——每当向世界发布研究成果，你都会焦虑地刷新推特，搜索AlphaFold这个词，看看有什么推文出现。
One was actually like watching Twitter after, you know, whenever you release some work into the world you anxiously refresh Twitter and stick the word AlphaFold and just see what tweets popped up.
我记得数据库开放时看到很多推文，许多震惊的研究生说，其中一条我记得是'他们怎么得到我的结构的？'
And I remember seeing so many when the database became available, so many astounded grad students saying, I think one was, How did they get my structure?
这还没发表呢。
It hasn't been published.
对吧？
Right?
这怎么可能？
How in the world?
展开剩余字幕（还有 387 条）
还有大量令人震惊的反馈，比如很多人说'我们预测了走廊那头那个人的结构'。
And just had absolute astounding, like the number of people that said either, you know, we predicted their structure, this person down the hall we predicted their structure.
我能感觉到他们完全被震撼到了。
So I thought they were absolutely blown away.
我认为第二个重要时刻是《科学》杂志出版了一期关于核孔结构的特刊，这是人类细胞蛋白质链中最大规模的蛋白质集合。
I think the second moment was there was a special issue of science related to the structure of the nuclear pore, the largest collection of proteins in the human cell protein chains.
当时共有四篇论文发表在那期《科学》特刊上，其中有三篇都大量使用了AlphaFold技术。
They had all, there had been like four papers in a special issue of science and three out of the four had made huge uses of AlphaFold.
类似的情况还有，《科学》杂志上提到AlphaFold这个词超过100次，而我们却完全不知情。
Something like more than a 100 mentions of the word AlphaFold in Science and we had nothing to do with it.
我们根本不知道发生了什么。
We didn't know what was happening.
它就那样突然出现了。那一刻我才真正意识到，人们正在用我们的工具做出值得发表在《科学》——这本世界顶级期刊上的研究成果，而且完全不需要我们的参与。
It just appeared one day and that was the moment when I really knew that people were doing you know, science worthy of appearing in science, right, one of the most prestigious journals in the world, on top of our tools and without us.
当人们开始在我们构建的基础上做出这些发现时，那才真正标志着我们作为工具开发者的成功。
And the moment at which people start, you know, making these discoveries on top of what we've built, that's when you really make it as someone who makes tools.
詹妮弗，你也有这样的时刻吗？
Jennifer, did you have a moment?
嗯，我想我们确实也有两个这样的关键时刻。
Well, I think for us, yeah, maybe also two two moments.
第一个是在2011年秋天，当时我们开始与埃马纽埃尔·沙彭蒂耶合作研究CRISPR——一种细菌免疫系统，我们想知道它的工作原理。
One was in the fall of, probably 2011 when we had started a collaboration with Emmanuelle Charpentier to work on CRISPR, which is a bacterial immune system, and we wondered how does it work.
我们共同发现这是一个RNA引导的系统，能够靶向切割DNA，更重要的是，由于我们理解了其化学原理，我们实际上可以对这个系统进行编程。
And we had, together figured out that it's an RNA guided system that targets DNA for cutting, for cleavage, and furthermore, that we could actually program the system because we understood the the chemistry of how it worked.
这真是令人惊叹的时刻。我个人完全被震撼到了——细菌竟然进化出了这样的机制，而我们现在竟能驾驭这个系统来以新方式操控DNA。
And it was just it was really one of those moments of, you know, I I personally, I was just astounded that bacteria had figured out how to do this and that furthermore, we now understood how to harness that system to manipulate DNA in new ways.
这是第一个时刻。
So that was one.
第二个时刻与你说的类似，约翰，那大概是在一年后。
And then the second one was sort of akin to what you said, John, where I think it was then fast forwarding about a year.
我们在2012年发表了那项研究成果。
So we published that work in the 2012.
到了秋天，我开始收到世界各地读过这篇论文的人发来的邮件，他们说，天哪。
And by the fall, I was starting to get emails from people all over the world who had read the paper and said, oh my gosh.
这太令人兴奋了。
This is so exciting.
我现在正用它来测试果蝇的基因。
I'm now using it to test genes in Drosophila.
我开始测试斑马鱼的基因。
I'm starting to test genes in zebrafish.
我对我在人类细胞中的研究感到非常兴奋。
I'm I'm excited about what what I'm doing in human cells.
我收到了一些几乎不认识的人发来的信息，他们都表现得非常非常激动。
I was getting messages from people that I often barely knew who were just very, very excited.
你能感觉到这个领域正在积聚的势头。
And you could start to feel the momentum that was building in the field.
保罗？
Paul?
好吧，我是这里的老前辈，恐怕要追溯到1985年了。
Well, I'm the oldie here, so it's 1985, I'm afraid.
如你所知，我研究的是酵母菌。
And I work on yeast, as you've already been told.
我的实验室已经找出了控制细胞周期的基因。
And my lab had worked out the genes that controlled the cell cycle.
这是导致一个细胞分裂成两个的过程，对所有生物的生长和发育都至关重要。
That is the process that leads to the reproduction of one cell into two, fundamental to growth and development in all living organisms.
我们已经识别出了那个基因。
And we'd identified that gene.
但直白地说，谁会在意酵母菌呢？
But the blunt thing is that who cares about yeast?
说实话，我确实关心酵母，但世界上大多数人并不关心酵母。
I mean, being perfectly honest, I care about yeast, but most of the world does not care about yeast.
因此我发现我们的工作并未受到多少关注，于是我想，人类是否拥有相同的基因？
And so I found our work wasn't really being taken much notice of, and so thought, do humans have the same gene?
那时距离人类基因组测序等技术的出现还很遥远。
Now, this is long before the human genome sequence and so on.
所以我们做了一个疯狂的实验。
So we did a crazy experiment.
我想告诉你们这个实验有多疯狂。
I want to just tell you how crazy it was.
我们有一种酵母突变体，由于这个基因缺陷而无法生长。
We had a mutant in yeast that couldn't grow because it was defective in this gene.
我们使用了首个人类cDNA文库，这是当时已有的Plat基因。
We took the first human cDNA library, it's a Plat gene that had ever been made.
文库不是我们制作的。
We didn't make it.
几个月后我们获得了文库，并将人类基因撒在缺陷酵母上——我们推测：如果人类基因能替代酵母缺陷基因的功能，如果酵母细胞能吸收它，如果基因能表达并起作用，那么这些细胞就能生长分裂，我们就能回收这个基因并证明人类拥有它。
We got it a few months afterwards And fundamentally sprinkled the genes onto the defective yeast, arguing that if there was a gene in humans that could do the same job as the defective gene in yeast, if the yeast cell took it up, if it could be expressed, if it worked, then those cells would grow and divide and we could get the gene back and show that humans had it.
这个实验本不该成功。
That experiment had no right to work.
酵母与人类大约在15亿年前就分道扬镳了，而我们却要求经过15亿年后基因功能仍能保持。
Yeast of humans probably diverged 1,500,000,000 ago, and we were demanding that it still worked after fifteen hundred million years.
但你们知道吗？它确实成功了。
You know, it did work.
当我们得到结果时——顺便说，当时测序基因花了几个月时间——我常常想：我要回家去相信这个结果有效。
When we got that result, and by the way, there were a couple of months when we had to sequence the genes, it took that long then, I used to think, I'm just going to go home and believe this works.
明天上班时我们就会发现实验其实失败了。
Tomorrow, I will go into work and we will have shown it didn't work.
但它确实成功了，那时我想或许它会被认可，因为这不是酵母，而是人类。
But it did work, that's when I thought perhaps it might be recognised because it wasn't yeast, it was humans.
多么精彩的故事集锦啊。
What an amazing collection of stories.
现在，我知道在座的各位一定对我们的专家小组有很多问题要问。
Now, I know that all of you must be bursting with questions for our panel.
所以我可能会从观众中选取一些问题。
So I might take some questions from the audience.
我们有麦克风在传递。
We have mics running around.
哦，那边有一个。
Oh, there's one there.
我可能会把它们分成小批次处理。
I might take them in a little cluster.
还有其他人想提问吗？
Is there anyone else who would like to pop them?
好的。
Okay.
我们这边有一个问题，那边也有一个。
So we have one here, another one here as well.
收到。
Roger.
今天真是令人印象深刻的一天，感谢你们以这样的方式结束。
It's it's such been a such an impressive day, and thank you for closing it out like this.
我有一个很有趣的问题：如果你能回到过去，遇见18岁或21岁的自己，你会说什么来让那个年轻人确信你正在做出正确的人生选择？
I've got a very playful question, which is if you could take yourself back in time and met your 18 year old or your 21 year old self and you could say something that would reassure reassure that person that you were making good life choices, what would you say?
而且不准说'你会赢得诺贝尔奖'。
And you're not allowed to say, you're gonna win a Nobel Prize.
还好吗？
Alright?
那个或与之相关的任何事。
That or anything to do with that.
那么你会对18岁、20岁时的自己说些什么呢？
So what would you say to the your your 18 year old, 20 year old former self?
好的。
Okay.
我们会收集另一个问题。
We'll collect another question.
我想罗杰·海菲尔德刚才这里也有个问题。
I think Roger Heifield just here had a question too.
罗杰·海菲尔德，科学博物馆的。
Roger Heifield, the Science Museum.
只是提个意见。
Just a comment.
我们已经看过梧桐（Sycamore）的图片，那是谷歌非常酷的量子计算机。
We've seen images of Sycamore, the very cool looking Google quantum computer.
能否请给我们一些看起来很炫的人工智能硬件，让我们科学博物馆可以收藏？拜托了。
Can we please get some sexy looking AI hardware that we can collect at the science museum, please?
实际上更严肃的一点是，德米斯你也提到过，就是人工智能擅长给你答案，却不擅长提供机制性见解。
Actually, more serious point is, and you've kind of alluded to it, Demis, is this thing about how AI is great at giving you answers, but it's not very good at giving you mechanistic insights.
你有数十亿个参数，但它们没有物理意义。
You've got billions of parameters, they've got no physical meaning.
这对公众信任人工智能有多大障碍？
How much of a barrier is this to the public trusting AI?
我也和约翰讨论过这个问题。
And I've talked to John about this as well.
我们究竟进展到哪一步了，才能开发出能揭示生物学定律的人工智能，就像我们已经掌握的物理定律那样，从而获得真正的机制性洞见？
Where are we in actually getting an AI that can give us the laws of biology, just like we've got the laws of physics, so we can get true mechanistic insights?
太棒了。
Fantastic.
这里有两个非常好的问题。
Two very good questions there.
我们从第一个问题开始。
We'll start with the first one.
德米西，我们先请你回答。
Demissey, we go to you first.
你会对18岁的自己说什么呢？
What would you say to your 18 year Well,
这有点复杂。
it's a little bit complicated.
我的意思是，我18岁时确实有过这个计划。
I mean, did actually have this plan when I was 18.
所以，神奇的是它居然实现了。
So, what is amazing is it worked out.
但我想告诉自己的是，我骨子里是个棋手，总是像下棋那样提前很多年做规划。
But but what I would have told myself would I mean, was a chess player in me as I always plan like many, many years ahead.
不过这就是从四岁开始下棋的后果，最终会变成这样。
But but that's what happens if you play chess from the age of four, you end up like this.
但你可能...我可能会对自己说，多享受这个过程，因为事情总会解决的。
But you you you you probably what I would have said to myself is just enjoy the journey a bit more because it will, like, work out.
因为在当时看来，这一切怎么可能实现呢？
Like, because at the time, it's like, how is this ever gonna work out?
我要处理这些梦想，这可能就是我想给18岁的自己的建议。
I have to deal with these dreams and then that that's probably what I'd recommend to my 18 year old self.
多姆？
Dom?
我
I
其实在想两件事。
think two things actually.
有趣的是，我在想，别担心，卡罗琳，我们会娶你的。
Amusingly, was thinking, Don't worry, Carolyn, we'll marry you.
但这就是人生的梯度下降。
But it's the gradient descent of life.
当下做正确的事效果非常好，要对即将为你开启的有趣事物保持开放心态。
Do the right thing right now has worked out really well and like be open to the interesting things that will open up for you.
我认为我们正处在生物学与AI的黄金时代之一，能亲身经历这些真的很有趣。
Then I think we're in, you know, one of the golden ages of biology and AI and biology and it's really fun to live through that.
所以我觉得，不要害怕局部最优。
And so I think like, don't be afraid to be locally optimal.
基本上就是与德米斯的建议相反。
Just basically the opposite of Demis' advice.
是
Are
你是梯度下降型还是提前战略规划型？
you are you a gradient descent or are you planning strategically in advance?
可能更接近约翰。
Probably closer to John.
是的。
Yeah.
我想我会告诉18岁的自己：追随热情，永不放弃。
I I think I would tell my 18 year old self to follow my passion and never ever give up.
别听那些唱反调的人的话。
And don't listen to people that are naysayers.
这非常重要。
That's really important.
我并非学术背景出身。
I came from a non academic background.
我当初简直不敢相信这还能赚钱。
I couldn't believe that you could get paid.
所以只管追随你的好奇心。
So just following your curiosity.
四十年后——不，五十年后的今天，我依然觉得难以置信。
I still don't believe it now forty years later, fifty years later.
其实还不止。
Actually, more than that.
五十五
Fifty five
年了。
years later.
请务必坚持相信我们能做到——作为
Just just hold on to believing that that we can, okay, as president of
皇家学会主席
the Royal Society.
我们确实可以。
We could.
罗杰提出的第二个问题是关于AI系统的直觉。
The second question from Roger there about about intuition from AI systems.
对。
Yeah.
我觉得Ketchy相当有趣。
Look, I think Ketchy is pretty interesting.
我不像其他人那么担心，因为我认为我们正处在一个特殊的历史时刻。
I'm not so worried about it as other people are because I think we're in a we're in a moment in time.
正如我之前在演讲中提到的，我相信AI是一门工程科学。
So what I said earlier, actually in my talk was what I believe, which is that AI is an engineering science.
这意味着你必须先构建出值得研究的实体，然后才能用科学方法进行分析。
So what it means is you have to build the artifact first, worthy of study, and then you can break it down with the scientific method.
过去五到十年间，我们看到的是构建出真正值得投入研究的实体——比如现在的Transformer模型和Alpha系列。
And so what you've seen in the last five, ten years is is the building of the artifacts that are even worth putting any effort into study, you know, until you have things like today's, you know, transformer models and alpha falls, alpha goes.
早期的系统可能根本不值得投入研究，因为它们还不够成熟。
The earlier systems are probably not worth really putting the effort into study because they were not sophisticated enough.
但现在不同了。
Now they are.
所以现在人们开始认真研究现有系统。
So now people are seriously studying the current systems.
更重要的是，这些系统还能自我改进。
On top of that, you've got the additional benefit of, these systems improving themselves.
至少我认为，我们将面临这样一种局面：系统能够用语言、数学或代码来解释自己。
And you're at the at the minimum, I think we'll be in a situation where firstly, the systems might be able to explain themselves in language or mathematics or code.
我们正在接近这个目标。
So we're getting close to that.
你可以对系统说：'好的，你已经理解了这个概念。'
So you could sort of say to a system, okay, you've understood this.
现在，用数学公式尽可能解释清楚这个概念。
Now now, you know, explain that in a mathematical equation to the extent that it can be.
顺便说一句，我不确定生物学能像物理定律那样被解释清楚。
And I'm not sure biology can be explained like like the laws of physics, by the way.
我认为这要混乱得多。
I think it is a lot messier.
这更多与相互作用有关。
It's more to do with interaction.
所以我认为模拟会比牛顿的运动定律之类的方法更适合用于探索。
So I think a simulation would be more appropriate that you probe than than like Newton's, you know, laws of motion or something like that.
我不认为生物学能被简化到那种程度。
I don't think biology could be reduced to that.
它太复杂了。
It's too complex.
另外一点是，我之前提到的将神经科学技术和分析技术应用于这些人工神经网络。
And then the other thing is, what I also said earlier about applying neuroscience techniques, analysis techniques to these artificial neural networks.
我们应该至少能获得与自然大脑相同程度的洞察。
And we should be able to get at least the same level of insights into them as we do with natural brains.
如果把这两者结合起来，我们就能取得很大进展，更不用说我们还将通过进一步的工程努力来分解这些系统。
So if you combine that together, we should get pretty far already, let alone with further engineering efforts that we're gonna put on top to to decompose these systems.
所以我认为在未来五年内，我们将摆脱目前这种黑箱状态。
So I think in the next five years, we'll be out of this era that we're currently in of kind of black boxes.
真有意思。
Fascinating.
好的。
Okay.
还有问题。
Further questions.
明白了。
Alright.
我们这里有几个问题。
We've got a couple here.
那我们开始，一、二，然后请保持手举高让我能看到。
So let's go one, two, and then if you can keep your hands up actually so I can see.
我们先做一、二，然后我们往那边移动。
We'll go we'll go one, two, and then we'll go over there.
如果可以的话，请分段朗读出来。
Read them out in chunks if that's okay.
开始吧。
Go for it.
谢谢。
Thank you.
我是谢菲尔德大学的丹尼·纽恩格拉菲斯特。
Denny Newengrafist from University of Sheffield.
接着早前关于社会科学与AI的话题，有人提出AI可能变革社会科学的观点。
Just following up on the topic earlier came up of social sciences and AI, and the idea was raised about AI transform potential for AI to transform social sciences.
我想反过来请教：在您看来，社会科学研究将如何帮助塑造AI的未来发展？
I wanna ask you about flipping that around and thinking about, in your understanding, how can you see the work of social sciences helping to transform these next futures of AI?
嗯。
Mhmm.
同时也在探索如何将工程科学转化为跨学科实践——正如我们讨论的，将AI工具与对应用场景的深刻理解相结合。
And, you know, building the ways to kinda transform that engineering science into the interdisciplinarity that 've been talking about where AI tools are being combined with really deep understanding of application areas and contexts.
这是个引人深思的问题。
Fascinating question.
你旁边的这位。
Your neighbor next to you.
哦，一个很简单的问题。
Oh, a very simple question.
注意力机制是否仍然是我们所需的一切？
Is attention still all we need?
克利福德。
Clifford.
我是来自美国国家联盟研究所的迈克尔·张。
Michael Chang from the National Alliance Institute in The US.
非常感谢。
Thank you very much.
这是一次非常棒的会议。
This was an awesome conference.
本次会议的主题是'用AI推动更优科学'，我认为我们常做的一件事就是可以多教科学家AI知识，让他们运用起来。
The theme here is AI to do better science, and I think one of the common things that we do is, well, we can just teach scientists more about AI, and they can use it.
我的问题是要反过来思考。
My question is flipping that around.
你认为是否有AI做不到的事情，科学家应该通过加强培训来最大化他们的附加价值？具体会是哪些方面？
Do you think there are things that AIs do not do that scientists should be getting more training on to to maximize their added value and what what those would be.
很想听听您对这个问题的看法。
Just love your perspectives on that.
真有意思。
Fascinating.
好的。
Okay.
明白了。
Alright.
那么我们先从你开始吧，保罗，因为你之前谈到了社会科学。
So we'll start off with I might start with you, Paul, actually, because you were the one who was talking about social sciences earlier.
是的，AI与社会科学。
So, yeah, AI and social sciences.
我们需要在多大程度上考虑人类与机器之间的这种交互关系？
How much do we have to think about that sort of interface between humans and the machines?
首先，我认为我们需要更多地关注社会科学来帮助我们科学家。
Well, first of all, I think we need to focus more on the social sciences to help us scientists.
不过我们必须认识到，人类互动是相当复杂的。
We have to appreciate, though, that this is pretty complicated, the human interactions.
所以我不确定一开始它能帮到我们多少。
And so I'm not sure how much it's going to help us at the beginning.
对此不太确定。
Not sure about that.
但我们确实应该接纳它，应该思考它可能带来的影响。
But we really ought to be accommodating it, ought to be thinking about how it might.
我能想象到某些社会科学问题，我猜你是研究交通这类问题的社会科学家吧。
I could sort of imagine certain social scientist problems, and I suspect you're a social scientist, to do with transport and things of this sort.
所以我认为有些机械性工作需要理解人类如何互动和工作。
So I think there are some mechanical things trying to understand how human beings interact and work.
嗯，你得问卡罗尔，对吧？
Well, you have to ask Carol, was it?
是的，当然。
Yes, sure.
约翰。
John.
对，没错。
Yes, exactly.
哦，卡罗琳。
Oh, Caroline.
是的，就是卡罗尔。
Yes, that's Carol.
换句话说，抱歉，我以为这很简单明了。
In other words, sorry, I thought it was straightforward.
情绪会变得难以处理。
Emotions to be difficult.
好吗？
Okay?
他当时正看着我。
And he was looking at me.
这让人加倍困惑。
It was doubly confusing.
是啊。
Yeah.
是啊。
Yeah.
给你，约翰。
To you, John.
我是说，
I mean,
我该回答这个还是该去
should I answer this or should I go to the
对。
Yeah.
去关注那个。
Go to the attention.
去关注那个。
Go to the attention.
我认为'注意力就是你所需要的全部'这种说法过于简单化了。
I think I think attention is all you need is a is a is an oversimplification.
我觉得真正有趣的一点是，AlphaFold并不是简单地从模型库拿个现成的Transformer就来预测蛋白质结构，明白吗？
I think one of the things that's really interesting is AlphaFold isn't just like grab a transformer off the shelf from the transformer store and apply it to protein structure prediction, you know?
这花了几年时间，因为工作量巨大，有许多关于注意力机制的新想法需要整合。
It took a couple years because there's a lot of work, there's a lot of new ideas where attentions are, is a component.
但我们拥有所谓的'邪恶前任'（指旧方法），在此基础上又有新思路，还有各种...德米斯运营着一个出色的AI研究机构，他们每天投入工作，从不满足于现状——比如仅依赖注意力机制。
But we have what we could call evil former, we have new ideas on top, we have all sorts of, you know, Demis runs an incredible AI research org which goes into work every day and doesn't just say, well, we've got all we need, there's attention, right?
我们所有人都在从事这项工作，我认为人们低估了当前AI领域真正新颖而激动人心的进展——这些工作正在使系统发生质的飞跃，既能解锁新数据源，又能从现有数据中学习更多。
We all are doing this work and it's really, I think people underestimate how much really novel and exciting work is going on in AI right now that are making these systems transformatively better, both unlocking new data sources and learning a lot more from the existing data sources.
AlphaFold的故事就是典型案例：使用与其他人相同的数据，却对蛋白质结构有了突破性认知。
AlphaFold is the story of having the same data as everyone else and learning tremendously more about protein structure from it.
因此我认为我们将持续看到AI研究带来的这些红利。
And so I think we'll continue to see these dividends from AI research.
我们会不断见证激动人心的新突破，也许永远会沿用'注意力'这个称呼——这让我想起计算机科学家的老笑话：'未来的科学计算语言会是什么样？它肯定会被叫做Fortran。'
We're going to continue to see exciting new things, and we may always call something attention, but it reminds me of this old computer scientist joke, I don't know what the scientific computing language of future will look like, but it will be called Fortran, right?
所以我们倾向于保留概念标签，而在其内涵上进行更新迭代。
And so, we tend to keep the labels of ideas and then update within it.
我认为我们不应低估当下正在发生的那些精彩、前沿且充满智慧的研究工作。
And I think we shouldn't underestimate how much really wonderful and exciting and clever research is going on these days.
或许我可以快速补充一点。
Maybe I could just add quickly to that.
我同意这个观点。
I agree with that.
实际上Transformer架构——就是那篇论文发明的技术——确实令人惊叹。
Actually the transformer architecture, which is, you know, what was invented with that paper, has been amazing.
我认为它将成为未来AGI系统的主要基础组件之一。
And I think will underpin be one of the main components of a future AGI system.
但我的预测是：仅靠它自身是不够的。
But my my prediction is it's not gonna be enough on its own.
我认为我们还需要几个类似量级的突破性进展，而这些突破仍在路上。
I think we're gonna need a couple of other big breakthroughs like that in addition, and they're still to come.
珍妮弗，我可能要请教你关于我们第三个问题中AI能力的不足之处，我们需要真正聚焦于人类所需的技能组合。
Jennifer, I might come to you just to pick up on on our third question there about the gaps in in in what AI can do, and we need to really focus on the skill set from humans.
是的。
Yeah.
谢谢你的问题，迈克尔。
And thanks for that question, Michael.
关于这一点，我一直在思考的是今天多次提到的数据问题，以及训练这类模型所需的数据类型。
So I guess what I've been thinking about with regard to that is there's been a lot of mention today about data and the kind of data that are necessary for training models like this.
生物学领域的挑战之一当然是数据质量，但同样重要的是数据量——目前至少需要大量高质量数据来训练模型。
And one of the challenges in biology, of course, is the quality of data, but it's also the quantity of data and the fact that typically one needs a lot of data that's of high quality to train models, at least currently.
因此我希望AI能教会我们科学家如何更智能地采集稀疏数据，使其覆盖面足够广泛，从而为训练提供合适的平台。
And so what I'd like to see AI do for us scientists is to educate us about how to collect data, perhaps sparsely, but smartly so that your sparse data is, you know, broad enough that it actually does provide the right platform for training.
我认为作为实验人员，我们目前设计实验时不会考虑这点，但我们其实可以做到。
And I think that's something we don't at least as an experimentalist right now we don't think about that when we design experiments, but we could.
我想这又回到了我们之前讨论的'提出正确问题'这个话题。
I guess that kinda comes back to the asking the right questions thing that we were talking about.
确实如此。
It does.
最精彩的部分是什么？
What is the the best part this?
关键问题中的。
Of the key questions.
对。
Yeah.
好的。
Okay.
我们大概还有时间再讨论一个吧。
We probably have time for another right.
我们后面有件红色套头衫。
We've got a red jumper at the back.
你可以先看一号，然后我需要从战略角度考虑麦克风的问题。
You can go view number one, and then I need to think strategically about microphones.
好的。
Okay.
如果我们过去，这边有个小集群。
If we go there's a little cluster over here.
那我们接下来走二、三、四号。
So we'll go two, three, four.
再往后些。
Further back.
哦，没错。
Oh, yes.
就在那儿，好了。
Just there we go.
我们找到了。
We found it.
好的。
Okay.
开始吧。
Go for it.
一号。
Number one.
你好。
Hi.
我是南安普顿大学的温迪·霍尔。
It's Wendy Hall, University of Southampton.
一位长期与社会科学家共事的人。
Someone who's been working with social scientists for a long time.
最大的问题在于数据收集以及相关的隐私问题。
The big issue there, collecting the data and the issues around privacy.
这使得它比你们从事的科学研究要困难得多。
It makes it so much harder than the sort of science you're doing.
在这方面我们做得越多越好。
The more we can do about that, the better.
不过我的问题是，作为一个刚患过感冒的人，就像许多人一样，人工智能能否解决普通感冒的问题？
My question though, as someone who's just been suffering from a cold, as many people please can AI sort out the common cold?
各位，你们最近在忙什么？
Guys, what have you been doing?
好的，我们这里有三个非常紧密相连的问题。
Okay, we've got three very closely together here.
好的，请继续。
Yeah, go ahead.
你好，我是托马斯·克兰普顿。
Hi, Thomas Crampton.
显然，我们讨论了很多伟大的科学，很多精彩的科学研究。
There's been a lot of talk, obviously, about great science, a lot of fantastic science discussed.
但对于那些不了解科学甚至可能对其持怀疑态度的外界人士，你们怎么看？
What about the people outside of this room, though, who don't understand science and might be suspicious of it?
你们有多担心社会可能会拒绝接受这些即将问世的重要突破？对此你们认为我们应该采取什么措施？
How much concern do you have that society might reject a lot of these great breakthroughs that are coming to the world, and what do you think we should do to address that?
很好的问题。
Great question.
下一位。
Next.
是的。
Yeah.
我是威尔弗里德·尼凡，来自非洲数学科学研究所。
I am I'm Wilfrid Nifan from the African Institute for Mathematical Sciences.
非常感谢这次精彩的讨论。
So thanks very much for a wonderful discussion.
如果您允许的话，我有两个问题。
I have two questions if you permit.
第一个问题是，当人们谈论通用人工智能时，普遍假设其目标是达到人类智能水平。
The first is that when people talk about AGI, think the common assumption is that the target is the human intelligence.
但人类智能可能并非最优，因为这是历史上偶然进化过程的结果。
But human intelligence may be suboptimal because this is the result of an evil historically contingent evolutionary process.
那么您认为通用人工智能在多大程度上会超越人类智能，成为某种超级优化存在？
So to what extent do you think AGI is going to be something super optimal with respect to human intelligence?
第二个问题关于包容性。
The second is about inclusion.
到2050年，非洲将拥有全球最庞大的青年成年人口。
Africa by 2050 would have the largest population of young adults in the world.
这些人口将为世界服务。
This population would be serving the world.
人类要继续进步，就必须依靠他们。
It will have to in order for humanity to continue its ascent.
当前业界在确保非洲人参与方面做了哪些努力？
To what extent is the community ensuring that Africans are included?
值得一提的是，谷歌表现非常出色，谷歌DeepMind在培养非洲青年人工智能能力方面给予了惊人支持。
I should mention that Google has been amazing, and Google DeepMind has been an amazing supporter of capacity building for young Africans in AI.
整体而言，业界还在采取哪些其他措施？
As a whole what else is the community doing?
谢谢。
Thank you.
很好，然后我认为你身后也有一小段内容。
Great and then I think there's one bit immediately behind you as well.
谢谢。
Thank you.
你。
You.
看到Kundee，来自数字科学。
Sees Kundee, Digital Science.
我是一名正在转型的学者，专业背景是纳米化学。
I'm a recovering academic, a nanochemist by training.
在化学领域，显微镜等技术发展总是为许多其他研究领域打开大门。
And in chemistry, technological developments and things like microscopy have always opened doors to many, many other areas of research.
人工智能也是如此，但从今天许多演讲者和讨论中我们看到，很多发展来自工业界而非学术界。
AI is very much the same, but I think we've seen from a lot of the speakers and the discussions today that a lot of the developments come about through industry rather than academia.
保罗，你谈到研究文化，其他演讲者也提到当前我们奖励成功研究的方式可能不利于为创新腾出空间。
So Paul, know you talked about research cultures and how maybe some of our other speakers talked about how maybe the way that we reward successful research at the moment isn't conducive to making space for innovations in the same way.
当没有人愿意第一个改变时，我们如何改变全球文化？
How do we change a global culture when nobody wants to be the first to do that?
因为他们可能会吃亏。
Because they may be losing out.
好的。
Okay.
明白。
Right.
四个极其精彩的问题。
Four extremely good questions.
各位，我们只有四分钟时间来回答这些问题。
We have precisely four minutes to answer them, everybody.
明白吗？
Okay?
好的。
Alright.
那么，保罗，我们从你开始。
So, Paul, we'll start with you.
为什么普通感冒至今还没有治愈方法？
Why isn't there a cure for the common cold yet?
嗯，你知道，这很难，不是吗？
Well, you know, it's difficult, isn't it?
我会去那个...请，可以
I'll go to the Please, can
我？
I?
当然。
Of course.
你当然可以。
You absolutely can.
让公众保持支持。
Keeping the public on board.
这真的非常关键。
This is really critical.
至关重要。
It's crucial.
我们真的需要集中精力处理这件事。
We really need to concentrate on it.
当然，这不是第一次了。
It is, of course, not the first time.
几乎每次新技术和变革出现时，都会引发担忧。
Nearly always, when there's been a new technology and changes, there has been concern.
正如我之前提到的，我们必须与合适的人对话，必须与公众沟通。
And I think I mentioned it earlier, we have to talk to the right people, we have to talk to the public.
但这类讨论经常被利益集团劫持——他们声称代表公众发声，实则往往是为自身特殊利益或个人狂热主张代言。
And too often these discussions get hijacked by interest groups and people who say they talk to the public, when they talk often for their own particular interests or their own particular passion that they have.
因此我们必须找到既能与公众对话，又能理性探讨的方式。
So we have to work out ways in which we can discuss with the public and discuss in a sensible way.
我提到过协商民主这个概念。
And I mentioned deliberative democracy.
实施成本很高，但我认为这非常重要。
It's expensive to do, but I think it's a really important thing to do.
坦率地说，若不能赢得公众支持，我们将无法充分实现其潜在价值。
Because bluntly, if you don't take the public with you, we won't be able to see all the merits that can come out of this.
这些价值既体现在增进对世界的认知，也体现在将发现成果真正造福公众。
Merits in terms of understanding the world better around us, merits in terms of actually using those discoveries for the public good.
我们必须积极沟通，让公众相信这些事业的正确性。
We have to engage and we have to convince the public these things are right.
我想就苏茜的问题稍作补充，其实这也涉及你问题的部分内容。
I think I want to actually just add on a little bit to Suzie's question there, which I think also incorporates part of your question.
今天我们在座各位讨论了很多关于科学家的成功标准和激励措施。
For all of you, we were talking a lot today about measures of success and incentives for scientists.
当然，你们都已经获得了科学界的最高荣誉——诺贝尔奖。
And, of course, all of you have the ultimate prize of of, you know, being Nobel laureate.
你们对成功的看法发生了怎样的变化？
How how has your view of success changed?
我们该如何影响全球科学界的未来一代，以确保获得正确的研究成果？
And how do you how should we kind of influence future generations across the world of scientists to make sure that we're getting the right outcomes?
约翰，你想先说说吗？
John, do you want to go first?
我认为，在我的科研生涯中，对我影响最深的一点就是团队合作开展科研工作的力量与乐趣。
I think that's, I think one of the things that's really been informing me over my scientific career is just the power and the fun of working on a team doing science.
而我
And I
在Google DeepMind就曾实践过这种模式。
was able to do it at Google DeepMind.
我在攻读博士学位前也这样做过一段时间。
I did it for a time before my PhD.
不过读博时期会相对孤独一些，对吧？
And then the PhDs are a little bit lonelier, right?
那时你只能独自钻研某个特定课题。
You're working on a particular individual thing.
但我认为团队协作的力量确实至关重要。
And I think, really, you know, the power of working within a team is really, really important.
这种模式本身就能产生强大的驱动力。
I think that provides also its own motivation.
科学研究本就是由无数失败和偶尔的辉煌成功构成的——在座的各位都是那些成功案例的见证者。
Science is about, you know, loads of failure and occasional dramatic success, and we're all up here, right, from those dramatic successes.
虽然迪马斯可能已经将科研流程标准化了，但我始终认为鼓励团队协作不仅能产出更优质的科研成果，也能让科研过程充满乐趣。
But I think that, and maybe Dimas has it down to a repeatable science, but I think that really, you know, encouraging people to work in teams, work together, and I think that helps provide the kind of motivation that leads to better science, that leads to more fun science.
如果做科研不快乐，人们就不会坚持下去。
If you don't have fun doing it, you won't do it.
所有伟大的科学家看起来都乐在其中。
All the great scientists seem to be having fun at it.
珍妮弗。
Jennifer.
完全同意。
Couldn't agree more.
回顾至今的职业生涯，我必须说，看到我培养的学生和他们现在从事的工作，我感到无比喜悦与自豪。
And I have to say that when I think back on my career so far, I really feel great joy and pride in the students that I've trained and the work that they're now doing.
这其中蕴含着难以言喻的满足感。
And it's just there's something incredibly satisfying about that.
其实，我能谈谈另一个问题吗？
And actually, can I speak to one of the other questions?
你之前提到关于非洲的问题，涉及非洲科学家和那些想做科研的年轻人。
So, you you asked about the question about Africa and involving African scientists and, you know, young people who want to do science.
我很自豪创新基因组研究所目前在肯尼亚持续开展项目，至今已连续进行了三年。
So I'm really proud that the Innovative Genomics Institute has an ongoing effort right now in Kenya where we've now, I think we've done three years running.
我们派遣团队前往肯尼亚各地，与当地科学家合作，帮助他们真正理解CRISPR技术。
We've sent a team to different parts of Kenya where they've been working with scientists there to work with them and really help them understand CRISPR.
看到反馈视频中，这些本土科学家回到社区后与学生互动，他们充满热情，开始在实验室里进行富有创意的科研，这非常鼓舞人心。
And they they it's really been really motivating to see some of the videos that come back where these local scientists then go back to their communities and they work with students and they get excited and they start doing interesting creative science, you know, in their own laboratories.
我希望看到更多这样的场景。
So I'd like to see more of that.
我认为那里蕴藏着非凡的机遇。
I think there's an extraordinary opportunity there.
我也对谷歌正在推进的项目感到非常兴奋。
I'm also very excited about the things that I hear about Google doing.
这是我们所有人携手合作的大好机会。
So I think a big opportunity for all of us to work together on that.
越来越多人正因追寻好奇心而获得报酬。
More and more people are being paid to follow their curiosity for.
你几乎说出了我想说的话。
You almost said what I was going to say.
哦，抱歉。
Oh, sorry.
我会稍微补充一下。
I'll amplify it slightly.
我们生活在这个大数据的世界里。
We live in this world of big data.
有时我们的标准很低，觉得只要报告大量大数据就够了。
And sometimes we have low standards that it's enough just to report lots of big data.
某些非常高调的期刊，似乎除了报道大量数据外什么都不做。
There are certain journals, very high profile journals, that seem to do nothing but actually report lots and lots of data.
我认为在大数据世界中关注创造力是值得的，因为在这种情境下创造力可能会被埋没，而事实上如果我们采取创造性方法，大数据蕴含着巨大机遇。
And I think it's worth giving some attention to creativity in the world of big data, because creativity can get lost in this scenario when in fact there is enormous opportunities with big data if we actually take a creative approach.
我们需要思考一下，首先，究竟什么是创造力？
And we need to think a bit, one, what is exactly creativity?
关于这点我可以继续展开，但时间所限就不多说了。
And I could go on about that, but I won't because it's near the end.
但我们需要鼓励同事和学生进行创造性思考。
But we need to encourage in our colleagues, in our students, creative thinking.
这与收集大量数据是有所不同的。
And that is a bit different from collecting lots and lots of data.
但如果采取创造性方法，海量数据确实能带来成果。
But lots and lots of data really will deliver if they take a creative approach to it.
戴维斯？
Davies?
有太多要点需要讨论了，但时间所剩无几。
So many things to pick up on, and there's not much time.
但我觉得，创造力这个话题值得专门组织一次小组讨论，或许可以下次安排。
But I mean, creativity would be a very interesting thing to have a whole panel discussion on it, maybe next time we do this.
不过现在先来回答几个问题吧。
But maybe just to pick up on a couple of questions.
在激励下一代方面，对我个人而言，费曼是我的偶像之一。
I think in terms of encouraging the next generation, for me, one of my heroes was Feynman.
但不仅仅是因为他那本著名的物理学著作。
But it wasn't just his physics books, which are very famous.
对吧？
Right?
实际上，是他的科普读物激发了我投身科学的兴趣。
But actually, was his lay person's books that inspired me to get into science.
我真的认为所有学生都应该读读这些书。
And I really think all school kids should read them.
比如《别闹了，费曼先生》。
So surely you're joking, mister Feynman.
还有《发现的乐趣》。
And the the great joy in finding things out.
因为在我看来，这些书比其他任何读物都更能展现身处知识前沿的激动人心。
Because I think those books more than any other I've read, and maybe there are others like that, show how exhilarating it is to be at the frontier of knowledge.
以及这种体验的意义。
And what that means.
光是谈论这个话题就让我激动得起鸡皮疙瘩。
I'm feeling getting boost goosebumps even just talking about that.
这些思想在我大约10岁或11岁时就深深影响了我。
And I and it instilled that in me when I was, you know, I can't remember what age, like 10 or 11 or something.
我认为让学生接触这些内容会非常有益。
And I think it would be great for school kids to be exposed to that.
从事科学工作该是多么不可思议又充满乐趣啊。
How incredible and fun doing science should be.
我认为费曼就是那种在从事惊人科学研究中获得极大乐趣的人，他是这方面的绝佳榜样。
And I think Feynman was one of those people who had a lot of fun doing his incredible science and as a great role model for that.
非常感谢在座各位，真的非常感激。
Thank you very much to all of you very much indeed.
谢谢。
Thank you.
和往常一样，Demis的演讲涵盖了一系列令人眼花缭乱的话题。
Well, as always with Demis, there was a a dizzying array of topics that were covered there.
我们谈到了室温超导体。
We had room temperature superconductors.
我们讨论了核孔复合体。
We had nuclear pore complex.
我们提到了分子注射器。
We had molecular syringes.
内容涉及塑料降解酶、药物设计，甚至扑克虚张声势的策略。
There was plastic eating enzymes, drug design, even poker bluffing strategies.
但对我来说，今天对话中最突出的时刻是关于人工智能时代成为科学家意味着什么。
But I think for me, the most standout moment from from the conversation today is about what it means to be a scientist in the era of artificial intelligence.
因为这次活动是由皇家学会联合主办的。
Because, you know, this this event was co hosted by the Royal Society.
皇家学会创立于'站在巨人肩膀上'的伟大孤独天才时代。
The Royal Society, which was founded during the era of great lone geniuses standing on the shoulders of giants, as the phrase goes.
但我认为或许我们需要开始接受孤独天才的时代真的已经过去了。
But I think that maybe we need to start accepting that those days of lone geniuses really are behind us.
因为如果要应对社会面临的最大挑战——气候变化、能源、疾病、我们对宇宙的理解——我们真正需要的是跨越传统学科界限的优秀科学家团队。
Because I think if we are going to address the biggest challenges that society faces, climate change, energy, disease, our understanding of the universe, then what we really need is big talented teams of scientists who are working across traditional discipline boundaries.
要知道，政治从未像现在这样需要科学，但我想整个世界也是如此。
You know, politics has never needed science more, but I think neither has the world.
我认为我们需要公共机构之间、政府之间、医疗保健提供者以及私营部门之间的这种合作，因为创新既是为人追求的事业，也是由人推动的事业。
And I think we need this collaboration between our public institutions, between governments, our health care providers, and also the private sector because innovation is a pursuit for people but also by people.
您正在收听的是由我——汉娜·弗莱教授主持的《谷歌DeepMind播客》。
You've been listening to Google DeepMind the podcast with me, professor Hannah Fry.
如果您喜欢这期节目，请订阅我们的YouTube频道。
If you enjoyed that episode, do subscribe to our YouTube channel.
您也可以在您喜爱的播客平台上找到我们，我们还将推出涵盖各类主题的更多节目。
You can also find us on your favorite podcast platform, and we have got plenty more episodes on a whole range of topics to come.
所以也请关注那些内容。
So do check those out too.
下次见。
See you next time.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。