人工智能与健康未来——乔尔·巴拉尔

我希望人工智能能帮助重拾行医的乐趣。最终，凭借医生和所有医疗专业人员花费大量时间录入的数据，我们将能够从中获得洞见并帮助他们。

I'm hoping AI contributes to bringing back the joy of practicing medicine. Finally, with all of the data that physicians and all their health care professionals have taken so much time to enter, we are going to be able to derive insights and help them.

欢迎回到《谷歌DeepMind播客》，我是汉娜·弗莱教授。在本期节目中，我将与谷歌DeepMind研究高级总监乔埃尔·巴拉尔探讨人工智能在健康领域的应用。多年来，我们一直开玩笑说互联网是疑病症患者最好的朋友，只需快速搜索和点击一下，就能把小头痛变成绝症，反之亦然。但在幕后，算法在医疗保健中的角色一直在悄然转变。

Welcome back to Google DeepMind, the podcast. I'm Professor Hannah Fry. In this episode, I'm talking to Joelle Baral, Senior Director of Research at Google DeepMind about AI for health. Now, for years we have joked about how the internet is the hypochondriac's best friend, capable of turning a small headache into a terminal illness and vice versa, with a quick search and a click of a button. But quietly, behind the scenes, the role of algorithms in medical care has been shifting.

我们在这个播客中已经多次讨论过人工智能对药物发现和蛋白质研究的影响。但现在，人工智能也有望改变诊断和治疗。欢迎来到播客，谢丽尔。我的意思是，人工智能已经带来了一些相当大的变化，但我们应该期待医疗保健在十到十五年后与现在看起来会有很大不同吗？

We've already talked a lot on this podcast about the impact of AI on drug discovery and research into proteins. But now, AI promises to change diagnosis and treatment too. Welcome to the podcast, Cheryl. I mean, there have been some pretty big changes already with AI, but should we expect that healthcare will look very different in ten or fifteen years' time to how it looks now?

绝对是的。我相信会这样。我认为医疗保健确实准备好被人工智能彻底改变，但可能看起来不会太不同。我把医疗保健想象成一个相当稳固的生态系统，你知道吧？所以十到十五年后，我们可能仍然会去看我们的初级保健医生，然后进行专科就诊等等。

Absolutely. I believe so. I think health care is really posed to be drastically changed with AI, but it may not look too different. I picture healthcare as a fairly sticky ecosystem, you know? So ten, fifteen years from now, we will likely still go to our primary care doctor and then follow through with specialist visits, etcetera.

但他们每个人都将被一个AI智能体增强，如果你愿意这么说的话。所以系统底层将会非常不同，但可能看起来与我们今天所知的不会有太大区别。

But each of them is going to be augmented, if you wish, by an AI agent. So underneath the system will be very different, but it may not appear as different from what we know today.

你在医疗保健领域从事人工智能工作多久了？

How long have you been working in AI in healthcare?

哦，哇。我想说我的整个职业生涯都是。我记得当时我还是个学生，应该是CS229课程对吧？那门机器学习课程正火热进行。我记得我当时——我已经在读博士了——和学生们一起研究如何分割喉部，对吧？

Oh, wow. My whole career, I would say. I was still a student when I think it was CS229, right? The class on machine learning was ticking off. And I remember working with, I was further along in my PhD, but working with students on how to segment the larynx, right?

而人工智能，你知道，是一种新工具，能真正帮助我们更好地做到这一点。从那以后我就一直没停下。但为什么是

And AI was, you know, this new tool that could really help us do that better. And I haven't stopped ever since. But why this

人类这个领域？它对你有何吸引力？

area of human? What's the appeal of it for you?

在医疗领域，你总是面临挑战，对吧？比如有很多事情我们做得不够好，或者你没有无限的数据。所以你永远无法做到完美。因此你总是试图利用最好的工具来完成目标。然后我在十年前加入谷歌，从事外科机器人工作。

In healthcare, you're always challenged, right? Like there are lots of things that we don't do that well or you don't have infinite data. And so you're never perfect. So you're always trying to leverage the best tools to do what you're trying to do. And then I joined Google a decade ago to work on surgical robotics.

我清楚地记得，那是我们真正开始意识到，对于图像，人类能做的任何解读工作，机器都将能够做到。于是我坐下来和我的外科医生同事一起，我们取了100例胆囊切除术的数据，也就是切除胆囊的手术。我们手动分割了肝脏和胆囊，然后我输入了一个相当简单的神经网络，看看算法是否能够区分这两者。我记得当它完美做到时，我实际上非常惊讶，因为我们手头的其他大多数工具从来都不是完美的，对吧？它们能完成任务，但并不总是完美。

And I remember vividly, it was the beginning of us really realizing that with images, anything that a human could do in terms of interpreting those images, machines were going to be able to do. And so I sat down with my surgeon colleague, and we took 100 cholecystectomies, which are, you know, when you remove the gallbladder. We manually segmented livers and gallbladders, and I fed, you know, a pretty simple neural net to see whether the algorithm would be able to decipher between those two things. And I remember being incredibly surprised actually when it did that perfectly, because most of the other tools we had at our disposal were never perfect, right? Like it did the job, but not always.

现在，你知道，肝脏和胆囊看起来真的不一样。这是一个非常简单的任务。任何学生都能告诉你哪个是哪个。但即便如此，对吧？对我来说，那真的意味着，好了，现在我们有了比之前我研究过的任何算法都强大得多的东西。

Now, you know, livers and gallbladders really don't look the same. It's a very easy task. Like any student will be able to tell you which one is which. But still, right? Like that really, for me, meant, okay, now we have something that will be infinitely more powerful than, you know, any of the algorithms I had been working on previously.

人们忘记了，对吧？但当我还是学生的时候，用计算机视觉解读二维码都很困难，对吧？那就是我们当时的水平。所以再次说明，人工智能确实开启了一个能力的世界，远远超出了我们以前的想象。

And people forget, right? But when I was still a student, it was hard with computer vision to decipher a QR code, right? Like that was where we were back then. And so again, AI has really opened up a world of capabilities far beyond anything we could envision before.

好的，那我们开始谈谈人工智能已经取得很大进展的一些例子。我想到的是诊断，特别是医学影像方面。跟我聊聊这方面的情况吧。

Okay, well let's start off with some of the examples of places where AI has already made a lot of headway. I'm thinking about diagnosis here and maybe medical imaging in particular. Talk to me a bit about what's been going on.

是的，在过去十年里我们看到了很多这样的例子。这就是我们通常所说的狭义人工智能。之所以称为狭义，是因为它只解决单一任务，对吧？你提到了医学影像，所以你可以处理胸部X光片或核磁共振成像，我们还做过糖尿病视网膜病变方面的工作。例如，这些工具已经获得FDA批准，并在临床中部署使用。

Yeah, we've seen a lot of that over the last ten years. That's what we typically call narrow AI. Narrow because it's solving one task, right? You mentioned medical imaging, so you can take a chest x-ray or an MRI, or we've done also like diabetic retinopathy work. Those are already tools that have been FDA cleared, for example, and deployed in the clinic.

它们正在辅助放射科医生，因为在那些特定任务上，机器在为医生标注图像方面做得非常出色。

And they are augmenting radiologists in the sense that for those specific tasks, the machine does a pretty damn good job at annotating the image for the physician.

让我确认一下我是否理解了。那么以糖尿病视网膜病变为例，这是糖尿病患者的一种病症，也是导致失明的原因，对吗？

Let me make sure I understand this then. Okay, so if we take diabetic retinopathy as an example, so this is people who have diabetes and it's a cause of blindness, is that right?

完全正确，这是全球可预防性失明的主要原因。过程很简单：用眼底相机拍摄眼底图像，获得视网膜图像后，眼科医生会按1到5级进行评分，以指示患者是无病变、中度病变还是严重病变。当我们开始研究这个特定问题时，令人惊讶的是，实际上很难获得我们所谓的'真实标注'——也就是在训练机器学习模型执行任务时，我们首先需要提供一组真实标注图像，即图像和标签的配对。

Absolutely, it's the leading cause of preventable blindness worldwide. And it's pretty simple. So you take an image of the back of the eye with a fundus camera, and that gives you an image of the retina that then ophthalmologists will grade on a scale from one to five to indicate whether patients have no disease, moderate disease, or severe disease. And what was striking when we started working on that particular problem was the fact that it was actually really hard to get to what we call ground truth, meaning, you know, as we're training our machine learning models to learn a task, we first need to provide them with a set of ground truth images. That means the pairs of the image and the label.

对于这个特定任务，我们可能会咨询一、二、三位眼科医生，但经常得到不同的答案。我认为我们做得特别好的一点是非常严谨地尝试建立真实标注。最终我们与50位——五零——眼科医生合作，发现实际上在某些图像上，他们的意见确实不一致。有些人会给出一分、两分、三分，每个分数都可能成为患者得到的诊断，因为患者通常只找一位医生就诊，对吧？他们不会让50位医生反复评估自己的图像。

And for that particular task, you know, we ask maybe one, two, three ophthalmologists, and often we got different answers. And I think one thing we did particularly well was to be very rigorous in trying to establish ground truth. And so we ended up working with 50, five-zero ophthalmologists, And we realized that actually on some images, they really didn't agree with each other. Some would have given one, two, three, each one of the scores could have potentially been the diagnosis a patient would have received because patients only see one, right? They don't go and have their images graded by 50 times,

是啊。天哪，所以有些眼科医生会说这是轻度或无病变，而其他医生对同一张图像却会诊断为严重病变。

yeah. Boy, so some ophthalmologists would say this is a mild or no disease and others would say all the way up to severe on the same image.

正是如此，这有时并不太令人惊讶，因为大多数眼科医生——实际上大多数医生都是如此——主要接触自己社区的常见病例，而罕见病例按定义来说他们不常遇到，所以如果不经常见到那种类型，就很难正确标注图像。

Precisely, which sometimes is not too surprising, meaning that, you know, most ophthalmologists and that's true for most physicians see what's in their community and the rare cases by definition they don't see them very often so it can get very hard to properly annotate an image if you don't see that type very often.

如果连专家都无法达成一致，那你们是如何做到的呢？如何决定正确答案是什么？

How do you do that then if even the experts don't agree? How do you decide what the correct answer is?

这正是为什么我们必须扩展到50人的原因，对吧？这样就能获得所有眼科医生的共识，有时还会咨询更资深的专家，以确保为图像确定正确的标签，从而正确训练算法。此外，我们还进行了大量严格测试，确保算法值得信赖。

That's exactly why we had to go to 50, right? Then you have this consensus among all ophthalmologists and sometimes you go to even deeper experts, right? To really get to what is the correct label for that image, and that way you properly train your algorithms. And then, you know, we also did a lot of rigorous testing to make sure that we had algorithms in which we could trust.

但反过来想，尽管这让标注变得相当困难，但既然知道专家之间存在分歧，实际上你们可以创建一种新的黄金标准，甚至比专家更胜一筹。

But then I suppose on the flip side of that, even though that makes the labeling quite difficult, if you know that experts are disagreeing, then actually you can create something that is a new gold standard really, like better than experts.

完全正确。我认为这正是人工智能发挥作用的领域。对于那些单靠人类难以完成、但AI却非常擅长的任务来说尤其如此。

Absolutely. And I think that's where the AI has really its role to play. Right? It's where for tasks that it is really good at that are really not very easy to be done by a single human.

我记得读过一篇论文，研究人员收集了大量眼底图像，出于好奇想看看是否能通过眼底血管判断患者性别。我认为世界上没有眼科医生能可靠地做到这一点，但AI可以。

I seem to remember reading this paper where some researchers had all of these images at the back of the eye and were like, you know what, let's just see for fun if you can tell the sex of the patient based on the blood vessels at the back of the eye. And no ophthalmologist in the world I think could do this reliably, but the AI can.

是的，完全正确。随后对这些图像进行糖尿病视网膜病变之外的研究也很有趣，发现模型的表现远优于随机猜测。虽然我不确定是否需要模型来解读患者性别，但对于潜在心脏病的指示等，这可以成为有用工具。既然已经进行糖尿病视网膜病变筛查，现在还能获得额外信息，就有可能筛查更多疾病。这不仅科学上很有趣——让我们意识到这些图像包含的信息远超最初想象，还可能衍生出实际应用。这意味着...

Yeah, absolutely. It was very interesting to then interrogate those images for other things than just diabetic retinopathy and see that the model performed better than flipping a coin, right? And for some things, I'm not sure you need a model to, you know, decipher the sex of your patient, but I do think for indications of potential cardiac disease, for example, that can be a useful tool because if you're going to do that screening for diabetic retinopathy anyway, but now you have a test that can also provide you with additional information, you could potentially screen for more diseases. So indeed, and it was, you know, kind of both interestingly scientifically to realize that those images contained a lot more information than we initially thought, and with potentially also applications that could derive from that. What does it imply maybe that even though

虽然你们创建算法是为了完成特定任务，但实际上AI有潜力帮助提升对人体整体认知的理解。

you've created an algorithm to do this narrow task, actually there is potential for the AI to help improve the overall understanding of the human body.

我可能会从两个方面来回答。首先，你知道，在放射学中有一种叫做偶然发现的现象。因为如果现在用MRI从头到脚对我进行筛查，很可能会发现一些问题。你会发现结节，一些异常的东西。这会让我感到担忧。

I would answer maybe in two ways. The first is, you know, in radiology there's this thing called incidental findings. Because it's very often that if you screen me head to toe with MRI right now, you'll find things. You know, you'll find nodules, things that are abnormal. They will get me worried.

但实际上，我们从未进行过临床试验来验证体内存在这些东西是好是坏，对吧？很多人一辈子都带着这些东西生活而没有任何问题。所以你并不真的想进行那种会让每个人都担心的筛查。另一个我想说的是，当你做血液检测时，你是为了某个特定目的而做的。负责检测特定病毒的实验室并不负责检查所有项目，对吧？

But actually we never did clinical trials to see whether having that thing somewhere is good or bad, right? And many people live with those things forever without any issues. So you don't really want to do that type of screening that will get everyone worried. The other thing I would say is when you do a blood test, you do a blood test for a very particular thing. The lab who did the blood test for finding a particular virus is not responsible for checking everything, right?

如果他们没注意到你还有其他不是抽血目的的问题，他们对此不负责任。所以在影像学方面，如何处理这些偶然发现总是有些模糊，对吧？对于狭义AI的这些方面，我认为陪审团仍在审议：当我们让AI查看特定图像时，是否应该让它总是寻找所有其他东西，即使它能提供相当有用的警报。

And if they haven't noticed that you had something else that wasn't what, you know, you got blood drawn, they're not responsible for that. And so for imaging, it's always a little ambiguous what to do with those incident findings, right? And for those aspects of narrow AI, I think the jury's still out whether or not we want the AI to always look for everything else when it's looking at one particular image, even if it could provide pretty helpful alarms.

这些是我们讨论过的图像，但是否还有其他输入可以用于这种非常专业的AI类型诊断？

Those are images that we've spoken about, but are there other inputs that you can use for this very narrow AI type diagnosis?

当然。我认为所有类型的模态都可以，对吧？例如，我们在声音方面做了很多非常有趣的工作，利用声音从咳嗽中检测肺结核。显然，在COVID期间人们也对此进行了大量探索。所以我们确实在声音中发现了良好的生物标志物，这也能为诊断这类疾病提供一些相对廉价的方法。

Absolutely. I would say all types of modalities, right? We've done a lot of very interesting work with sound, for example, leveraging sound to be able to detect tuberculosis from cough. People have also explored that quite a bit during COVID, obviously. And so we see good biomarkers actually in sound that can help also provide some relatively cheap ways to diagnose those types of diseases.

等等，这么说真的有效？是的，你录下咳嗽声，它就能告诉你是否患有肺结核？

So wait, it actually works then? Yes, You record a cough and it can tell you if you've got TB?

简而言之，是的，正是如此。

In short, yes, that's exactly what it is.

好的。但这东西也有局限性，对吧？我的意思是，它们每次都能准确无误吗？

Okay. But there are limitations to this stuff though, right? I mean, are they getting it right every single time?

不。我认为这是一个很好的观点。就像任何算法一样，存在敏感性和特异性，所以它可能出现假阳性或假阴性。也就是说，它可能误以为存在某种情况而实际上没有，或者漏掉了本应检测到的东西。但总的来说，我们所做的，以及在糖尿病视网膜病变论文中所做的，实际上是检查它与最佳医生或最佳医生小组相比表现如何。

No. I think that's, you know, a very good point. Like with any algorithms, right, there is sensitivity and specificity, so it can have either false positives, false negatives. So thinking that something exists when it doesn't or missing something that it should have picked. In general though, what we do and what we did with the diabetic retinopathy paper is really checking how it does with respect to the best physicians or the best panels of physicians.

然后让始终与算法协同工作的人类来决定他们希望在特异性与敏感性的权衡中处于什么位置。这样他们可以决定自己真的无法承受误报，或者反过来，无法承受漏掉任何情况。但在这两种情况下，我认为一个算法要获得监管机构的批准，都需要展现出非常强大的性能。

And then letting the human who is always working with the algorithm decide where on that specificity sensitivity trade off they want to operate. And so they can decide that really they cannot afford false alarms or the other way around that they cannot afford missing anything. But in both cases, I think for an algorithm to be cleared by regulatory authorities, it needs to have shown very strong performance.

好的，让我来分解一下。所以，算法可能会误将非结核病诊断为结核病，但也可能漏掉真正的结核病例而判断为正常。显然，这两种情况你都不希望发生。但是，在你认为这个模型能增加价值之前，你几乎愿意接受多高的准确率百分比？

Okay, let me break that down then. So it could be that an algorithm could say that it's TB when it isn't, but it also could be that an algorithm could miss a genuine case of TB and say that it was fine. And obviously, you don't want either of those things to happen. But what percentage accuracy almost are you willing to accept before you say, okay, this model is adding value?

这真的要看情况，对吧？如果你要部署你的，比如说，在一个以前什么都没有的地方进行筛查测试，那情况就与替换现有筛查同一人群的替代方法大不相同。如果你是在替换某种东西，你最好至少表现得更好。如果你进入的是一个空白领域，那么公共卫生当局将决定他们认为什么是可接受的。这也取决于被给出该诊断的患者后续会怎样，对吧？

So it really depends, right? If you're going to deploy your, let's say it's a screening test in places that had nothing before, it's a very different story than if you're replacing an existing alternate way of screening that same population. If you're replacing something, you'd better be at least better. If you're coming in what there was before a void, then public health authorities are going to decide what they deem is acceptable. It also depends what will happen to the patients that are provided with that diagnosis, right?

如果他们回家了并且你再也不会见到他们，那情况就与你的工具是一个预筛查工具、然后你会将患者转介到另一个筛查工具进行确认的情况不同。

If they go home and you never see them again, it's a different story than if your tool is a pre screening tool and then you're going to reroute your patients to an additional screening tool for confirmation.

当你推出这些东西时，例如结核病模型或糖尿病视网膜病变模型，这是否意味着你们最初的目标是去那些目前没有这些现有筛查测试的地方？

When you launch these things, for example, the TB model or diabetic retinopathy, does that mean that you aim initially to go to places where they don't have these existing screening tests in place?

这真的要看情况，对吧？比如糖尿病视网膜病变，我们确实已经在泰国部署了相关技术，并且经常与合作伙伴一起将其推广到能发挥实际作用的地方。在泰国我们已经筛查了70万人，并计划在未来几年内将这个数字扩大十倍。

It really depends, right? For diabetic retinopathy, for example, it is something that we've indeed deployed in Thailand, and then we often work with partners to really bring it where it makes a difference. We've already screened 700,000 people in Thailand, and we're 10xing that over the next few years.

为什么选择泰国？

Why Thailand?

泰国是那种每位眼科医生需要照顾的患者群体相当大的国家之一，对吧？在很多地方，医生数量确实不足。因此，这是一个很好的例子，说明人工智能筛查确实可以帮助改善医疗结果。

Thailand is one of those countries in which the patient population that each ophthalmologist has to care for is fairly large, right? There are many places in which there aren't really enough doctors. And so it's a good example of a place in which AI screening can really help improve outcomes.

假设有一种情况，某种医学影像技术与医生一起使用。如果两者意见不一致会怎样？

Let's say that you have a situation in which some sort of medical imaging is being used in conjunction with a doctor. What happens if the two disagree?

我认为归根结底，医生才是真正做决定的人，对吧？所以非常重要的一点是，他们必须保持控制权，他们是做决定的人，也是在报告上签名的人。因此，你知道，他们需要解释，对吧？如果他们说是机器错了，可能有很多原因。在其他情况下，你可以认为算法会让医生重新思考，对吧？

I think at the end of the day, the doctor is really the one making the call, right? So it's very important that they retain control and they are the one making the decision and they are the ones signing their names on the report. And so, you know, they would have to explain, right? If they're saying the machine is wrong, it could be for a variety of reasons. In other cases, you know, you could think of the algorithm making the physician rethink, right?

所以，如果你愿意的话，它就像拼写检查器一样检查你的工作，对吧？有时候我会不同意我的拼写检查器对法语单词重音的处理，或者它并不完美，但我会再检查一遍。

So it's checking your work like a spell checker if you wish, right? Sometimes I disagree with my spell checker with accents on words in French or things like it's not perfect, but I will double check.

好吧，我们来谈谈一些更高级的东西，因为到目前为止，我们使用的例子像是核磁共振、眼底照片或者咳嗽的声音。那么更全面的视角呢？我的意思是，真正好的医学不是把人看作一系列有趣的或其他医学问题的集合，而是把人看作一个完整的人。你能用人工智能进行更全面的思考吗？

Well, let's talk about some of the more advanced stuff here because because up until now, I mean, the examples that we've been using here is like an MRI or a photo of the back of an eye or, you know, the sound of a cough. What about the more holistic view? So I mean, really good medicine doesn't see a human as a collection of interesting or otherwise medical problems, Sort of sees a human as a human. Can you use AI to think more holistically?

人工智能确实非常擅长将我们迄今为止主要只能单独观察的事物整合起来审视，并可能带来我们之前未曾获得的见解。

AI is indeed very good at looking together at things that up until now we were mostly capable of looking at, you know, individually and maybe bring insights that up until now we didn't have.

比如呢？我的意思是，如果你真的能把我们对细胞的理解、对器官的理解，最终与对人体整体的理解联系起来，你能在这些联系中发现什么？

Like what, mean, if you do manage to connect up everything we understand about the cell to everything we understand about an organ to eventually everything we understand about human, what do you find in those connections?

是的，过去我们在虚拟染色方面做了不少工作。当你有一张H&E切片时——这是手术中从体内取出的任何组织通常会被切片并在显微镜下观察的方式。我们通过假设可能发生的情况并进行相应染色，使这些现象可见。你可以把其中一些技术视为无需染色就能使其显现，对吧？所以实际上是揭示组织中已有的信息，而无需引入消耗该组织样本的额外物质。

Yeah, back in the days we did quite a bit of work in virtual staining, right? So when you have an HNE slide, so that's what happens when any piece of tissue removed from your body in surgery typically will be sliced and looked at under the microscope. And the way we do that is we make hypothesis as to what might be happening and we stain accordingly so that it makes those things visible. And you can think of some of those techniques as making that visible but without having to stain, right? So really revealing some of that information if it's already in the tissue but without having to bring kind of additional things that are consuming that piece of tissue.

会破坏组织，没错，违背了初衷

Will destroy the tissue, Exactly, defeat the

正是。所以在我看来，我们现在用越来越多仪器（如转录组学、基因组学、单细胞技术）检测组织的方式也蕴含着深刻意义。这对AI来说是个绝佳的应用场景，因为AI可以审视每种模态，并为我们提供针对特定科学仪器的‘眼睛’。更进一步的是，它还能在不同类型的科学仪器之间建立桥梁。

exactly. So to me, there's also something quite profound in the way we're now interrogating tissue with more and more instruments, with things like transcriptomics, genomics, single And cell, so that's really a wonderful application for AI because the AI can look at each modality and if you wish provide us with eyes for that particular scientific instrument. And then it goes beyond because it's also capable of bridging between those different types of scientific instruments.

那么你们应该可以开始整合这些相当复杂的AI工具了，比如在影像学和基因组学方面。这是否真的能推动你们对疾病的理解？我想到的是癌症护理这方面。

So then you presumably can start to combine some of these quite complex AI tools, you know, on imaging, on genomics. Does that actually allow you to advance in your understanding of diseases? I mean, I'm thinking of cancer care here for instance.

是的，这是个很好的例子。我们正与巴黎研究所就此开展具体合作，他们非常擅长利用所有这些最新模态来更好地理解癌症——具体来说是女性癌症，如子宫癌或乳腺癌。尽管我们数十年来竭尽全力，但对许多女性仍缺乏解决方案。我们真心希望通过整合这些模态，在细胞层面更深入地理解病情，最终能够攻克例如三阴性乳腺癌或其他目前尚无良策的晚期癌症。

Yeah, it's a great example. And that's what we are actually precisely working on with the Institute in Paris that is very advanced in exploiting all of those most recent modalities to better understand cancer, in that case, women's cancer. So uterine cancer or breast cancer, which, you know, despite our best efforts for many, many decades, we're still short of answers for many women. And we're really hoping that by combining those modalities, by getting to a deeper understanding at the cellular level of what's going on, we'll finally be able to crack, you know, for example, triple negative breast cancers or some of the, again, more advanced cancers that today we don't have good solutions for.

除了细胞、基因组学和影像学，人工智能还能整合其他数据源到这个等式中，对吧？

As well as cells and genomics and imaging, there are other data sources that the AI can bring into the equation here as well, right?

绝对如此。我认为这或许也是人工智能能够破解的秘密之一——健康蕴含在我们所做的一切事情中，对吧？比如你今天走了多少步，早中晚餐吃了什么，是与朋友互动还是独自孤独一整天。通常很难将这些转化为健康习惯，或理解它们对身体健康的影响程度。我们正在传感器方面做大量工作。

Absolutely. And I think that's also maybe one of the secrets AI will be able to decipher, which is health is in everything we do, right? It's how many steps you walked today, what you ate for breakfast, lunch and dinner, whether you've interacted with friends or been lonely all day long. And so, it's typically really, really hard to translate that into either healthy habits or understand how much of that is factoring into something physical that is going on. And we're doing a lot of work with sensors.

我今天就戴着这款Fitbit。你知道，它只是众多数据源之一，可以与其他所有数据结合使用，既帮助更好地理解健康（有时仅针对个人），也在我们尝试改变行为时陪伴我们走过这段旅程，对吧？

I'm wearing this Fitbit today. You know, it's just one of the data sources that can really be leveraged in combination with everything else to try to both better understand health, sometimes just as an individual. And also when we're trying to change behavior, right? Accompany us on that journey.

你这里谈论的数据质量差异很大，对吧？既有活检切片和基因组数据，又有步数计数。感觉边界有点模糊。这真的会带来实际差异吗？

I mean, you're talking about quite different quality of data here, right? You've got sort slides from biopsies and genomic data, and then step count. It sort of feels like a quite fuzzy thing around the edges. Would it actually make a difference?

确实会。我的意思是，这已被反复证明，例如睡眠对心血管健康和肿瘤学（即预防癌症）的重要性，对吧？所以这是我的一大期望。

It does. I mean, it's been shown, you know, over and over again, for example, how much sleep matters, right? For both cardiovascular health and in oncology, right? In terms of preventing cancer. So it's one of my hopes, right?

人工智能将以传统医疗系统难以实现的方式，也将我们个人难以做到的方式，将这些因素整合起来，因为这种影响在群体层面才显现。作为个人，很难说服自己提前一小时睡觉，尽管这实际上是对健康最有益的事。

That AI will bring those two together in a way that is harder to do in our traditional healthcare systems and also very hard to do for us as individuals because you see that type of impact at a population level. As an individual, it's really hard to convince yourself that you should go to bed one hour earlier because that's actually the best thing you can do for your health.

目前关于数字孪生的讨论很多。这是否也增添了健康的整体视角？是否有为自己创建医疗用途的数字孪生的想法？

There's quite a lot of buzz at the moment about digital twins. Does this also add to the whole holistic view of health? Is there an idea of making a kind of digital twin of yourself for healthcare purposes?

是的，我们经常看到这种情况，它对不同的人可能意味着不同的东西，对吧？比如，数字孪生可以被视为对具有相似人格的个体的模拟，这可以作为探究不同类型干预措施潜在影响的一个良好替代。而在制药行业，数字孪生可能意味着完全不同的东西，例如，有很多工作试图探索我们在临床试验方面能走多远。如果我们能够组建虚拟队列，让我们能够获得关于药物安全性和有效性的尽可能多的知识，但不需要那么多的人参与特定的临床试验。

Yeah, we see that a lot and it can mean different things for different people, right? There is the idea that the digital twin is kind of a simulation of someone with a similar persona, for example, and then that can be a good proxy to interrogate the potential impact of different types of interventions. And then digital twin can mean something quite different for the pharma industry, for example, where there's a lot of work trying to see how far we can go with clinical trials. If we manage to assemble virtual cohorts that are allowing us to gain as much knowledge about the safety and efficacy of a drug, but without needing as many people for that particular clinical trial.

哦，这太有趣了。所以这是关于，我不知道，创建一种几乎像模拟人类的队列，也许不是完整的个体，也许只是一个器官或你特别关注的任何部分，这样你就不必在临床试验中使用那么多人？

Oh, that's so interesting. So is this about, I don't know, making a kind of cohort of like simulated humans almost, maybe not the whole thing, maybe just an organ or whatever it is you're particularly focusing on, but so that you don't necessarily have to use as many people in your clinical trial?

没错，对于临床试验来说，正是这样。

Exactly, for clinical trials that would be exactly that.

但要有效地做到这一点，你必须非常、非常了解真实人类的样子，这意味着在某个过程中拥有来自个体患者的大量数据。人们对于为这类目的、为医学研究贡献自己的数据感觉如何？

But then to be able to do that effectively, you have to really, really understand what real humans look like, which means having a wealth of data from individual patients somewhere along the way. How do people feel about contributing their own data for this kind of end, for the research of medical purposes?

我认为，你知道，当我们谈论隐私时，我们在世界不同地区看到不同的担忧。在欧洲，有很多主权方面的担忧。人们希望数据，我的意思是，虽然不是所有国家，但都留在他们的国土上。我们有解决方案，对吧？例如，所有可信的公共云解决方案都遵守该主题的最高监管级别。

I think, you know, when we talk about privacy, we see different concerns in different parts of the world. In Europe, there are a lot of sovereignty concerns. People want to know that the data is, I mean, not in all countries, but is staying on their soil. And we have solution for that, right? All of the trusted public cloud solutions, for example, are complying with the highest levels of regulation on that topic.

在世界其他地区，我们看到很多人渴望不觉得自己生病是白费的。我的意思是，实际上，知道如果你生病了，但你正在为研究贡献数据，你正在帮助让未来遭受同样命运的人病得轻一点，这是相当有吸引力的。所以我认为，真正促成这种良性循环才是我们必须做到的。

In other parts of the world, we see a lot of appetite for people to not think that they are sick for nothing. By that, I mean, it's actually quite compelling to know that if you're sick but you're contributing data to research, you're helping making the less person that is going to have the same fate a little less sick, if you wish. So I think really enabling that virtuous cycle is really where we have to be.

因为医疗数据确实感觉是一个特别敏感的案例。一方面，正如你所描述的，它有巨大的潜力来推进我们的理解并改善后代的状况。但另一方面，如果医疗数据落入错误之手或被滥用或不负责任地使用，我认为人们对潜在的后果有真正的担忧。那么，我的意思是，你如何平衡这一点？

Because it does feel like healthcare data is a particularly sensitive case. Because on the one hand, exactly as you described, there is huge potential to advance our understanding and improve conditions for future generations. But on the other hand, if you know, health care data gets into the wrong hands or is misused or used irresponsibly, I think people have real concerns about about the potential ramifications of that. So, I mean, how do you strike that balance?

一旦你拥有了真正保护数据的技术，也就有了问责制。这意味着，如果我们声称这项研究将有益于人们，那么最终它确实应该是对数据来源地的人们有益的研究，对吧？我认为这是一个非常重要的原则，就像你不应该在世界上某些地方进行临床试验，然后让药物在世界其他地方受益一样。

Once you have the technology to actually protect the data, there's also accountability, That meaning that if, you know, we're saying that this research will be helpful to the people that indeed, at the end of the day, it is research that is helpful to people where the data originated, right? And I think that's a very important principle that, you know, in the same way that you shouldn't do clinical trials in places of the world, in some places of the world, and then the drugs benefit in other places of the world.

所以你不是简单地从一方提取资源给予另一方？正是如此。好的，我还想多问一些关于患者在这方面的体验。大型语言模型是否已经在改变游戏规则？我的意思是，你是否担心人们正试图用生成式AI来自我诊断？

So you're not just extracting from one to give to another? Precisely. Okay, I also want to ask you a bit more about the patient's experience in all of this. Are large language models changing the game here already? I mean, are you concerned that people are trying to diagnose themselves with generative AI?

是的，我想说，即使在此之前，你真的也不应该那样做。就像你不应该自己当医生一样。但人们还是会这么做。

Yeah, I mean, would say even before, you really shouldn't do that, Like you shouldn't play your own doctor. People do

他们确实会，不是吗？说实话。

they, don't they? Let's be honest.

我认为人们想要了解情况，尤其是在看医生等待时间成为真正问题的地方，患者当然会尝试尽可能多地学习，解读正在发生的事情，有时他们在现场时做得相当有效，因为他们是自己最好的倡导者，对吧？或者父母试图为孩子弄清楚一些事情。有时候，你知道，我们有一些例子，对于罕见疾病，实际上我们看到患者做得非常出色。现在我想说，在谷歌，我们一直努力为用户提供最有帮助的答案。例如，对于症状，我们有你可能注意到的知识卡片，它们会告诉你关于疾病的信息，并提供权威内容，对吧？比如试图清晰地解释常见症状和潜在的治疗选择，但从不越界告诉你你自己的诊断可能是什么。

I think people want to know, and especially in places where waiting time to see a physician is becoming a real issue, then of course, you know, patients are trying to learn as much as possible and to, you know, decipher what's happening, and sometimes quite effectively when they are there, because they are their best advocate, right, or a parent trying to figure something out for their child. Sometimes, you know, we have a couple of examples where for rare diseases, actually, we've seen patients do a remarkable job. Now I would say, you know, at Google, we have always tried to provide the most helpful answers to our users. And so with symptoms, for example, we have the knowledge cards that you've probably noticed that are telling you about a disease and with authoritative content, right? Like trying to be clear in the explanation of common symptoms and potential options for treatment, but never crossing that line and telling you what your own diagnosis might be.

现在有了大型、长寿命的模型，我们的Gemini模型也不会告诉你你的诊断是什么，对吧？它会告诉你：抱歉，我不是医生。你应该去看医生。但如果你问哪些潜在情况可以解释这个症状，你可以得到更有帮助的答案，对吧？它会给你更教科书式的答案。

Now with large, long lived models, our model Gemini will also not tell you what your diagnosis will be, right? It will tell you, Sorry, I'm not a medical doctor. You should go and see a doctor. But you can get more helpful answers if you're asking, you know, what potential conditions could explain this, right? And it will give you more of the textbook answer.

我喜欢把它看作，你知道，那些家庭医疗指南，那些厚厚的书，你可能在有小孩的时候会有一本，试图知道在特定情况下该怎么做。嗯，它就是那个水平。如果我们相信大型语言模型将能够提供诊断，我们需要继续研究。这就是我们与AIMI（Articulate Medical Intelligence Explorer）合作的工作。这是一个研究项目，我们试图了解大型语言模型在建立诊断方面可以具备哪些对话能力。

I like to think of it as, you know, those family guide for healthcare, those, you know, I don't know, the big, thick books that you might have where you have small children and you're trying to know what to do in those particular cases. Well, it's that same level. And if we believe that large language models are going to be able to provide diagnosis, we need to continue the research. And that's what we're doing with our work with AIMI, the Articulate Medical Intelligence Explorer. It's a research project in which we are trying to see which conversational abilities a large language model can have to establish a diagnosis.

所以它能像医生那样提出正确的问题，对吗？就像与病人对话一样。

So can it ask the right question like the way a physician would do, right? Like dialoguing with a patient.

我们距离看到像AIMI这样的系统真正应用于医疗环境还有多远？

How close are we to seeing a system like AIMI being used actually in medical settings?

我们已经在这个项目上工作了相当长一段时间，你知道我们的第一批研究论文都是基于与患者演员和模拟场景进行的研究。我们现在正在哈佛大学贝斯以色列医疗中心进行一项获得RB批准的临床研究，看看在医生监督下、非常受控的环境中会发生什么。这是我们下一步的计划。很难预测这样的系统需要多长时间才能真正在临床上帮助医生。但我们已经看到，在世界某些地区，类似系统通常被用来回答低风险问题，并且始终受到监督。

So we've been working on this project for quite a while already and you know our first research papers were all based on research done with patient actors and simulated scenarios. We're now doing a clinical study under RB approval with Harvard at the Beth Israel Medical Center to see what happens, you know, again in a very controlled environment with physician supervision with the system. So that's our next step. Hard for me to predict, you know, how long it is before such a system would really help physicians in the clinic. But we are already seeing, you know, in parts of the world, similar systems that are typically being used to answer low risk questions, if you wish, and that are always supervised.

所以可能有一位医生，你知道，在一段时间内，比如模型回答病人后的十分钟或十五分钟内，会再次检查答案。但你可以看到这确实是一个循序渐进的方法。绝对是的。我认为这非常重要，因为如果一个模型十次中有九次说得对，但第十次却做了非常糟糕的事情，那么可能它对所有情况都不太好。

So there's a physician maybe, you know, double checking the answer within a period of time, a short amount of time, like, say, ten minutes of a model or fifteen minutes from a model answering a patient. But you see how it's really a step by step approach. Absolutely. And I think that's very important again, because if you have a model that says the right thing nine out of 10 times, but the tenth, it's doing something really bad, well, maybe it wasn't so good at all for none of those cases.

收益不值得付出代价。确实如此。不过这很有趣，因为你说得对，当你与真正的医生交谈时，他们会告诉你病人并不会带着一份非常清晰、精炼的相关信息清单来找你。我的意思是，他们会进来然后说，感觉好极了之类的。

The benefits don't outweigh the costs. Precisely. That is interesting though, because you're right that when, if you're talking to a real physician, I mean, they will tell you that patients don't come in with a sort of very clear, refined list of relevant information for you. I mean, they'll come in and say, sort of, what? It feels so great.

然后你的工作就是去探究并确切地找出问题的核心，哪些信息是相关的，哪些不是。这在人工智能中模仿起来一定非常困难。

And it's like it's your job then to interrogate that and actually find out precisely what's at the heart of it, which bits of information are relevant, which bits aren't. That must be a very difficult thing to mimic within an AI.

确实如此。而且，你知道，我经常喜欢把未来的人工智能想象成你家庭中的医生。你知道，每个家庭通常都有一位医生，会被问到所有与家庭成员健康有关的问题，所有堂兄弟姐妹和每个人。即使他们是皮肤科医生，也会被问到心脏病学问题，然后他们会帮助人们 navigating 医疗系统等等。但并非所有家庭都有医生。

Indeed. And also, you know, I like to often think of the AI of tomorrow as something that will be like your, you know, the physician in your family. You know, there's often this physician in your family that gets asked all questions about anything that has to do with health care of all family members, you know, all the cousins and everyone. Even if they are a dermatologist, they will be asked, you know, cardiology questions, then they will help people navigate the system, etcetera. But not all families have physicians.

因此，我愿意这样设想，我们明天将构建的人工智能将为每个人提供类似家庭医生般的服务。但关键在于，这位家庭成员真正了解你，并且是长期了解。所以他们知道你是否非常焦虑，比如当你提到过去三周一直头痛时，他们清楚你过去二十年都这么说，从而可以放心地忽略。或者，如果你打电话给他们，但你其实从未主动联系过，他们就知道这需要引起重视。他们能识别你的语调等等，我认为这让事情变得——我之前举了皮肤科医生的例子，最终虽然要对所有情况发表意见，但总体上还是相当安全的，对吧。

And so I'd like to think that, you know, the AI we will build tomorrow will provide everyone with that, like the equivalent of a physician in your family. But the key element is that that member of your family actually knows you, and they know you on the long term. So they know if you're very anxious and, you know, when you're saying that you have a bad headache for the last three weeks, you've said that for the last twenty years and they can safely discard it. Or if, you know, if you're calling them but you've actually never called them, they know that it's something they should pay attention to. They will recognize the tone of your voice, etcetera, which I think makes this, you know, I gave the example of the dermatologist that ends up, you know, having to say something about everything, but pretty safe, right, in the end.

它实际上不会给患有心脏疾病的人提供错误建议，因为它了解这个人，会在适当的时候建议他们寻求心脏病专家的意见。所以我们希望人工智能系统也能做到同样的事，对吧？但为此，它们需要足够了解你，而不仅仅是快速说出最新症状然后得到答案。这些系统必须被专门构建。

It's actually not going to provide bad advice to someone that has a cardiac condition because they know the person, and they will tell them to actually seek cardiologist advice at the right time. So we would want our AI systems to do the same, right? But for that, it means they need to know you enough, and it's not just, you know, very quickly saying your latest symptoms and getting an answer. Those systems have to be built.

我想家庭医生的一个不同之处在于，他们不太容易出现幻觉。

I guess one of the differences of a physician in your family is that they're not really prone to hallucinations.

有些可能会，但你 有些

Some of them might, but you Some

可能会。但当我们仍处于生成式AI确实存在这些幻觉、或忘记时间线、或信息错误——所有你在生成式AI中常见的错误——的世界时，如何创建这样的东西呢？

of them might. But how do you create something like that when we are still in a world where generative AI does have these hallucinations or is forgetful about timelines or mistakes information, all of the kind of common mistakes that you see in generative

是的，这就是为什么你不能仅仅拿一个现成的大型分析模型应用到医疗保健领域，并期望它是可接受的，对吧？我认为我们首先在MedPal上做了很多工作。那是我们第一个基于医学语料库微调的模型。我们证明了它在回答医疗执照考试方面比最初的平均学生、然后是专家小组表现得更好。然后我们开发了MedGemini。

Yeah, that's why you cannot just take a large analytic model off the shelf and apply it in healthcare and expect that to be acceptable, right? And I think we've done a lot of work first with MedPal. That was our first model fine tuned on a medical corpus. We demonstrated that it was doing better at answering medical license exams than initially an average student and then a panel of expert. And then we did MedGemini.

MedGemini是什么？

What was MedGemini?

哦，MedGemini是我们的大型语言模型Gemini，我知道你已经体验过了。

Oh, MedGemini is our large language model Gemini, which I know you've played with.

嘿，没完没了地玩。

Hey, endlessly.

我们在医学语料库上对其进行了微调。因此它真正继承了Gemini的长上下文能力、推理能力以及原生多模态特性。但在此基础上，它还以Gemini未曾有过的方式接触了医学数据。并且它正在针对特定医疗用途进行评估。

And we fine tuned it on a medical corpus. And so it's really inheriting the long context of Gemini, its reasoning capabilities, its native multimodality. But on top of that, it has seen medical data in a way that Gemini hadn't. And it's also being evaluated for specific medical purposes.

不过，人机协作还有另一个方面。我的意思是，如果其中一些工具是为医生设计的，是否存在医生可能失去某些技能或开始过度依赖这类模型的风险？

There is this other aspect of the human AI collaboration though. I mean, if some of these tools are designed for doctors, is there a risk that doctors, I guess, lose some of their skills or start to become overly reliant on these kind of models?

你知道，总是存在那种风险，对吧？在医疗保健领域，我会说，对我们而言，我们其实没有选择，因为世界许多地方都面临医疗专业人员严重短缺的问题。所以，问题是我们如何解决这个问题，对吧？我们如何确保全球人们尽可能健康，不遭受那些我们已有解决方案的问题困扰？还有很多问题我们仍没有答案，但对于许多医学确实能提供答案的情况，我认为我们有责任利用现有技术将其带给那些人。

You know, there's always that risk, right? In healthcare, I would say though that for me, we don't really have a choice in the sense that we have a big shortage of healthcare professionals in many parts of the world. So, the question is, how do we address that, right? How do we make sure that people globally can be as healthy as possible and don't suffer from things for which we have answers, right? There are a lot of things for which we still don't have answers, but for a lot of others where medicine can actually provide an answer, I think we have a responsibility to leverage the technology we have to bring that to those people.

对我来说，问题不在于过度依赖，而在于我们将如何培训下一代，对吧？这在我们的对话中对医生来说是真实的，在我们所处的Gen AI新时代背景下，对其他职业也可能如此。但对医生来说，可能有些事AI确实做得非常好，而当它做不好时，它也会告诉你它不知道等等。所以你可以依赖AI，并且不擅长AI真正擅长的那些任务也没关系，因为还有很多其他事情你需要学习并做好。

Rather than overreliance, for me, the question is how are we going to train the next generation, right? What is It's actually of true of doctors within our conversation. It can be true of other professions within the context of the new era we are in with Gen AI, but for doctors, there might be things where actually the AI is really doing a really good job, and when it's not, it's also telling you that it doesn't know, etcetera. So you can rely on the AI, and it's okay to not be as good at those tasks that the AI is really good at because there are so many other things that you need to learn and you need to do well.

那么，有些事你认为AI根本不会触及？比如同理心之类的东西。

So there are some things then that you think the AI just isn't going to touch? I mean, things like empathy, for instance.

嗯，实际上，这一点很好，因为我们总是检查我们的模型是否具有良好的医患沟通技巧，而它们确实做得不错。它们一点也不差，你知道吗？它们能够以很少有人能做到的方式适应听众。它们可以为五岁、七十岁或四十岁的人使用恰当的语言。如果你不是母语者，它们还能用你自己的语言与你交流。

Well, actually, that one is a good one because we always check that our models have good bedside manners, and they actually do. They're not bad at all, you know? They can adapt to their audience in a way that few human beings can. They can leverage the right language for a five year old or a 70 year old or a 40 year old. And if you're not a native speaker, they can speak in your own language.

所以实际上，当它们在同理心方面被评判时，它们表现得相当不错。

So they're actually, when they're judged on empathy, they actually do pretty well.

你们到底是怎么训练它更具同理心的？

How on earth do you train it to be more empathic?

嗯，就像你训练它们做其他所有事情一样，对吧？它们基于训练数据中看到的大量对话。如果这些对话具有同理心，那么它们也会学会使用相同的词语、相同的语气等等。你确实需要对它们进行微调或提供更具体的例子，因为并非网络上的所有内容都具有同理心。

Well, in the same way you train them for everything else, right? Like they are based on a lot of conversations that they've seen in the training data. And if those conversations are empathic, then they learn to also leverage the same words, the same tone, etcetera. And you do need to fine tune them or provide them with more specific examples because not everything on the web is empathic.

不过，你认为医疗保健中是否有一些方面是人工智能真正无法触及的？

Do you think though that there are some aspects of medical care that really AI can't touch?

我建造过手术机器人。所以即使是触觉反馈之类的，我也倾向于认为在某个时候我们会有办法的。今天，你知道，每当需要进行体格检查时，对吧？比如检查某人的腹部，对吧？很多事情，你绝对需要医生。

I've built surgical robots. So even, you know, the haptics and all of that, I tend to think that at some point we will have ways. Today, you know, whenever you need a physical exam, right? Like if you're checking someone's stomach, right? And a lot of things, you absolutely need physicians.

我停顿了一下，因为你知道，我认为每次我们说技术永远做不到这一点时，我们都错了。而且，你知道，在某个时候，事情会以某种方式改变，实际上它们在这些情况下也能有所帮助。它是作为增强医生能力的东西而来的，对吧？它并不是取代他们日常所做的所有任务。我更将人工智能视为为数字行业对医生所做的事情偿还债务的工具，数字行业多年来要求医生在电脑中输入大量数据，相当大地改变了他们的工作，在某些情况下让他们苦不堪言，对吧？

I paused because, you know, I think every single time we say that technology will never do this, we get wrong. And, you know, at some point, things change in a way that actually they can also be helpful in those circumstances. It's coming as something that is augmenting physicians, right? It's not replacing them in all of the tasks they do daily. I see AI much more as the one paying the debt for what the digital industry has done to physicians, which is asking them to enter lots and lots of data in computers for years, changing their job fairly drastically, making them miserable in some cases, right?

我确实认为我们在这一行看到了很多职业倦怠，我希望人工智能能够帮助重拾行医的乐趣。但对我来说，这就像是在偿还债务，对吧？就像最终，凭借医生和所有医疗专业人员花费大量时间输入的所有数据，我们将能够从中获得洞察或知识，并帮助他们。

I do think that we see a lot of burnout in the profession, and I think, I'm hoping AI contributes to, you know, bringing back the joy of practicing medicine, But to me, it's paying that debt, right? It's like finally, with all of the data that physicians and all their healthcare professionals have taken so much time to enter, we are going to be able to derive insights or knowledge and help them.

我想这就是为什么这一切都值得。不过听你谈论这个很有趣，因为我在这档播客中进行的许多对话都是关于，我不知道，比如AGI以及那些长期的、全面的系统，以及朝着与我们一贯做法完全不同的方向发展的进步。但你描述你的愿景的方式，几乎让人觉得这里的AI是一种将改变医学各个方面的工具，但医患关系的基本理念并未改变。

Here's why it was all worth it, I guess. But that is interesting listening to you talk about it though because, I mean, so many of the conversations that I get to have on this podcast are about, I don't know, like AGI and sort of long term sort of these holistic systems and and kind of advances towards completely doing things in a different way than we always have. But the way that you're describing your vision of this, it's almost like the AI here is a tool that will change every aspect of medicine, but fundamentally the idea of a patient doctor relationship is unchanged.

你说得对。我想我可能比我们在社会其他方面所设想的要更谨慎一些。这也再次取决于我们在世界何处设想技术的影响，因为我认为技术将为世界上那些可能根本无法获得医疗保健，或仅能以非常有限的方式获得医疗保健的地区带来医疗服务，其方式可能是我们尚未设想的，甚至可能超越我们今天所描述的任何情况。但我总是对医疗保健在一夜之间被彻底革命持怀疑态度。而且，你知道，明天的情况将与昨天完全不同。

You're right. I think I might be a little more cautious than, you know, what we're envisioning in other aspects of society. And it also depends, again, where in the world we are envisioning that impact of technology, because I think that technology will bring healthcare to parts of the world that probably didn't have access to healthcare at all, or only in a very limited way, in ways that we're not yet envisioning and that might be beyond anything we've described today. But I also, I'm always a bit skeptical of healthcare being entirely revolutionized overnight. And, you know, it will be completely different tomorrow from how it looked yesterday.

话虽如此，我也对像Amy这样的模型感到非常震撼，对吧？想想看，你现在可以与一个大型语言模型进行对话，这种对话非常类似于与一位知识渊博、汇集了全球知识的专家进行的对话，这意味着，你知道，20世纪50年代在某个偏远地方进行的一项小型临床试验，其发现可能正开始在其他地方得到印证。直到现在，如果你患有罕见疾病，偶然性有时在能否得到治疗方面扮演了非常重要的角色。而现在，有了我们的大型语言模型，我认为我们正在彻底改变这种偶然性的游戏规则。

That being said, I'm also absolutely blown away with models like Amy, right? Like if you think about it, you can now have a conversation with a larger lit model that is very similar to the conversation you would have with an incredibly well educated specialist pooling knowledge from all around the world, meaning, you know, this very small clinical trial that happened in the '50s in some remote place echoing some new things that have, you know, that are starting to be seen somewhere else. Up until now, serendipity sometimes played a very big role if you had a rare condition as to whether, you know, you would be treated or not. And now with our large language models, I think we're completely changing that serendipity game.

我的意思是，以非常积极和乐观的基调结束一档播客总是很好的，但我认为未来有很多值得期待的事情。

I mean, it's always nice to finish a podcast on a very positive and optimistic note, but I think there's a lot to look forward to.

绝对是的。谢谢。谢谢你。我

Absolutely. Thank you. Thank you. I

认为医学是一个特例，因为在假阳性和假阴性之间，尤其是两者都可能带来的潜在危害，这使得达到正确标准的要求非常高。我认为Joelle在这里采取科学方法、谨慎小心地评估使用AI的好处、以确保我们最终能为尽可能多的人带来最佳医疗结果的务实决心，是令人安心的。您刚才收听的是由我Hannah Fry教授主持的Google Deep Mind播客。如果您喜欢这一集，请订阅我们的YouTube频道。您也可以在您喜欢的播客平台上找到我们。

think medicine is a special case, because between the false positives and the false negatives and crucially the potential harms of either, this is something that has a very high bar to get right. And I think there is something reassuring about Joelle's no nonsense determination to take the scientific approach here, to carefully and cautiously evaluate the benefit of using AI, to make sure that we end up with the best healthcare outcomes for as many people as possible. You have been listening to Google Deep Mind the podcast with me Professor Hannah Fry. If you enjoyed that episode then do subscribe to our YouTube channel. You can also find us on your favourite podcast platform.

我们还有很多关于各种主题的剧集即将推出，也请务必查看。下次再见。

And we have plenty more episodes on a whole range of topics to come. So do check those out too. See you next time.