Tortus' Dom Pimenta: "AI is the answer to NHS doctor burnout"
发布时间 2024-05-24 05:41:55 来源
In this week's episode of Your History from The Times and Sunday Times, we look back on the extraordinary life of Dr Ruth Westheimer, who escaped Nazi Germany as a child and found fame as a pioneering sex therapist in the 1980s. We also celebrate Eddie Spence, renowned for his skills as a cake decorator for the British royal family. Join me and Atemkin for Your History. Find us on TheTimes.com and wherever you download your podcasts.
在本周的《你的历史》节目中，由《泰晤士报》和《星期日泰晤士报》呈现，我们回顾了露丝·韦斯特海默博士的不凡人生。她小时候逃离了纳粹德国，后来在1980年代作为一名开创性的性治疗师成名。我们还庆祝了埃迪·斯彭斯的成就，他以为英国皇室装饰蛋糕的精湛技艺而闻名。请和我以及阿特姆金一起收听《你的历史》。你可以在TheTimes.com以及任何可以下载播客的平台找到我们。
Yo. Tikmology. What is it all about? I think it's not a pipe dream. That's number one. AI doctor in your pocket. That's today, right? You literally have GPT-O in your pocket right now. Is a better diagnostician than 99% of physicians based on that one study looking at New England General Medicine? And that's not to say that I'm saying we should replace it, but I do think it's a really interesting time in medicine when the problem for all of time has always been about distribution of expertise. And now potentially we have an infinite distribution of expertise. That's quite a hard concept for us to get our heads around.
嗨，Tikmology，这是怎么回事？我觉得这不是异想天开。首先，口袋里的AI医生，这已经是现实，对吧？你现在真的有一个GPT-O在口袋里。根据《新英格兰医学》上那篇研究，GPT-O的诊断能力比99%的医生都要好。当然，我不是说我们应该用它来取代医生，但我确实认为医学界现在处于一个非常有趣的时代，因为一直以来，我们面临的问题都是专业知识的分布不均。而现在，我们可能拥有无限分布的专业知识，这对我们来说是一个非常难以理解的概念。
Hello and welcome to Danny and the Valley of Weekly Dispatch from Behind the Scenes and Inside the Minds of the Top People in Tech. I'm your host, Danny Forts, and the West Coast correspondent for The Sunday Times. And this week, we're talking about healthcare. But specifically, the problems in healthcare like Dr. Burno, high cost, poor results, the general brokenness of the system. And how AI might just help solve a lot of those problems, but, and here's the twist, not in the way you think.
您好，欢迎来到《丹尼与每周幕后科技领袖之谷》。我是你的主持人，丹尼·福茨，《星期日泰晤士报》的西海岸记者。本周我们讨论的是医疗保健，特别是医疗保健中的问题，比如医生疲劳、高昂的费用、糟糕的结果以及系统本身的种种弊端。我们还会探讨人工智能可能如何帮助解决许多这些问题，但有个转折——它的解决方式可能和你想象的不一样。
So depending on who you ask, doctors spend anywhere from three to five hours in front of a computer every day. Writing up notes from patient visits, writing letters, doing general admin. And the first I had heard of these numbers, which sound kind of realistic, was earlier this month when I went to Stanford, they were having an event there. It was an AI in medicine symposium, and it was kind of a first of its kind, gathering of the top minds in healthcare and AI. There's obviously lots of excitement about AI. And the thing that kind of emerged that was most interesting to me was that while there was some talk about this fantastical idea of, you know, we've been talking, hearing about for a while now, of the AI doctor in your pocket, the ability of AI to, you know, train on every single medical case in history and thus become better than any human doctor. That idea, obviously, people are very excited about, that's the promise, that is the hype. But there are lots of reasons why that future, if it does arrive, will arrive very slowly.
所以，视问的人不同，医生每天要在电脑前花费三到五小时不等的时间。他们需要记录病人就诊的笔记、写信以及处理各种行政事务。我第一次听到这些看起来很真实的数据是在本月早些时候，当时我去了斯坦福参加一个活动。那是一个AI与医学的研讨会，也是首次举办的类似活动，汇聚了医疗和人工智能领域的顶尖人才。显然，大家对人工智能非常兴奋。而对我来说，最有趣的一点是，即使有些人谈到了这个听起来像幻想的想法——也就是说，我们已经听说了一段时间的"口袋中的AI医生"的概念，人工智能可以通过训练所有历史上的病例，从而变得比任何人类医生都更优秀。这个想法无疑令人兴奋，是一种承诺和炒作。但也有很多原因表明，这样的未来即使到来，也会非常缓慢。
On the other hand, AI could break the paperwork log jam like now. Because the combination of real-time transcription plus large language models that can understand, summarize, write notes, means that doctors could be freed from hours of administrative work, which, in effect, would be like instantly hiring millions more doctors by liberating them from that three to five hours a day of busy work. You know, so more free time to, well, have more free time, and also to see more patients, not typing. And that is the promise of this week's guest. Dom Pimenta is the co-founder of a company called Tortoise. They are a UK startup that recently raised a bunch of money from coastal adventures to do exactly what I just described. They've created a note-taking app that, you can probably think of it like a medical co-pilot. You can autonomously write notes, generate letters, it listens into the consultations to summarize them, etc. And Pimenta is a trained cardiologist, and he recognizes can wipe away hours of work, which, you know, if you look at the NHS, this is a very big deal. There are record levels of Dr. Burnout. There are strikes going on for doctors arguing for higher pay, better conditions. And if you could effectively, with this magical tool, dramatically increase the quality of your job or, like, your ability to actually do the thing you trained to do, that feels like a pretty big deal.
另一方面，人工智能可以像现在这样打破文书工作的瓶颈。因为实时转录加上能理解、总结和写笔记的大型语言模型的结合，意味着医生可以从数小时的行政工作中解放出来，这实际上就像立即雇用了数百万名医生一样，将他们从每天三到五小时的繁忙工作中解放出来。你知道，这意味着有更多的自由时间，不仅可以享受更多的休闲时光，还可以看更多的病人，而不是在打字。这正是本周嘉宾所承诺的内容。Dom Pimenta 是一家名为 Tortoise 的公司的联合创始人。这是一家英国初创公司，最近从沿海风险投资公司筹集了一大笔资金，正是为了实现我刚才描述的目标。他们创建了一款记笔记的应用程序，你可以把它想象成一个医疗副驾驶。它可以自主撰写笔记、生成信件、监听咨询过程并进行总结等等。而 Pimenta 是一名受过训练的心脏病医生，他认识到这可以消除数小时的工作，如果你看看英国国家医疗服务体系（NHS），这是一件非常了不起的事情。医生的倦怠水平创下历史新高，医生们正在为更高的薪酬和更好的工作条件而罢工。如果你能够有效地使用这个神奇的工具，大幅提高你的工作质量，或者说真正做你所受训练去做的事情，这看起来就像一件大事。
In short, AI could revolutionize healthcare, but do it by abstracting away admin, not necessarily by this more sexy idea of, you know, these AI doctors in your pocket. So, I think it's just a really instructive example that kind of separates out the AI hype from what is still a very, potentially very compelling reality and the difference it might make. So, anyhow, I think you're going to really enjoy this conversation, so I will now hand you over to my chat with Dom Pimenta, co-founder and CEO of Tortoise. Enjoy.
简而言之，人工智能可能彻底改变医疗保健，但这种改变更多地体现在简化行政工作上，而不是我们常想的那些装在口袋里的人工智能医生。因此，我认为这是一个很有启发性的例子，它帮助我们区分出人工智能的夸大宣传和它很有可能带来的真正变革之间的区别。不管怎样，我相信你会非常喜欢这次对话，所以现在把时间交给我与Tortoise公司联合创始人兼首席执行官Dom Pimenta的对话。请享受。
We actually talked initially a week ago when I was writing about a story on AI and the NHS and how AI may or should be able to kind of make some serious dense in how healthcare works and more importantly doesn't work today. So, I'd love to just understand basically what is tortoise, because I imagine most of our listeners will not have heard of you. Tortoise is a AI company. We are a healthcare company based in the UK, co-founder by myself. I'm a practicing physician. I co-founder Chris Tan, he's a machine learning engineer. And our thesis is building an AI interface between the doctor and the computer.
我们大约一周前讨论过，当时我正在写一篇关于人工智能和英国国家医疗服务体系（NHS）的文章，探讨人工智能如何或应如何在医疗保健工作中的一些关键环节中发挥作用，特别是解决目前医疗体系中的一些不完善之处。因此，我很想了解一下Tortoise究竟是什么，因为我猜大多数听众可能没有听说过你们。
Tortoise是一家人工智能公司。我们是一家位于英国的医疗保健公司，由我共同创立。我是一名执业医生，我的联合创始人Chris Tan是一名机器学习工程师。我们的理念是构建一个在医生和计算机之间的人工智能界面。
So, we really have seen a massive explosion of AI capability in the last two years, but very specifically and very interesting to me, the ability of large language models to understand clinical contexts, like the actual language of medicine. Medicine fundamentally is the language, and now we have technology that can interact with that. So then the question becomes what does AI clinician co-working actually look like? And that's the question we want to answer in the long term as a company.
因此，我们真的看到在过去两年里人工智能能力有了巨大的飞跃，但令我非常感兴趣的是，大型语言模型理解临床环境的能力，以及理解医学实际语言的能力。医学本质上是一种语言，而现在我们有了能够与这种语言互动的技术。那么问题就变成了，人工智能与临床医生的合作到底是什么样子？这一点是我们公司希望长期回答的问题。
In the short term, doctors spend 60% of their time on computers. We don't want to, we never, we never wanted to. And those three tasks are summarization, documentation and doing things like tasks and actions. So what we're building is essentially as an AI agent, we call our agent, Ozzler, that takes over that interaction piece, allows the physician to spend twice as much time with their patients, and still actually gets the same, in fact, probably better quality data into the system, and closes a big productivity problem in actually workforce in healthcare systems everywhere, which is just as just not people.
从短期来看，医生花60%的时间在电脑上。我们不想这样，我们从来不想这样。这些任务主要包括总结、文档处理和执行任务等。所以我们正在开发的是一个人工智能代理，我们称之为Ozzler，它将接管这些互动任务，使医生能够多花一倍的时间与患者交流，同时还能录入同样甚至更高质量的数据，从而解决全球医疗系统工作效率低下的问题，因为目前医疗系统中人力短缺。
So you have to be much cleverer about what those people are actually spending their time doing. So that's all, yeah. And it's interesting because I kind of came upon the story initially because I was at Stanford last week at this AI in medicine symposium. It was the first one they were doing because as we all know, there's been all this hype around AI over the past, you know, 18 months in particular about how it's going to kind of revolutionize everything, right?
所以你必须更加聪明地去了解那些人实际在做什么。所以，这就是全部了，对吧。而且这很有趣，因为我最初接触到这个故事是因为我上周在斯坦福参加了一个医学人工智能研讨会。这是他们第一次举办这个活动，因为我们都知道，过去18个月中，围绕人工智能的炒作非常多，大家都在说它将彻底改变一切，对吧？
And one of the big examples people keep rolling out is this idea of like an AI doctor in your pocket. And that is something that people have been getting very excited about and kind of marketing is like, oh my goodness, you can have this thing that is trained on every documented medical case in the textbooks and, you know, potentially patient history is all this stuff. And it will be able to administer health in a very cheap, accessible way to anybody who has a smartphone or whatever.
其中一个大家常提到的重要例子，就是“口袋里的AI医生”这个概念。这让人们非常兴奋，甚至有人将其大力宣传为“哦天哪，你可以拥有一个训练过所有医学书籍中记录的医疗案例，甚至包含潜在患者历史信息的东西”。它可以以一种非常廉价、便捷的方式为任何拥有智能手机的人提供医疗服务。
And so that's a really exciting vision. And then we go to where we are today with the healthcare system, which is kind of pretty broken in some pretty fundamental ways and we can talk about the NHS because, you know, there's industrial action. There's doctors striking for better pay. There is burnout. There is stress. And I think depending on who you talk to, it's like anywhere from three to five hours a day is spent just putting in electronic health records, making notes, doing letters, all that kind of stuff. And so it's fair to say that what you guys are doing is initially at least is targeting that piece.
这确实是一个非常令人振奋的愿景。现在我们来看看当今的医疗系统，它在一些基本方面已经有些崩溃。我们可以谈谈英国国家医疗服务体系（NHS，因为目前存在一些工业行动，医生们正在罢工要求更高的薪酬。这种情况下有疲劳过度的问题，有压力。而且根据与你交谈的对象不同，每天大约有三到五个小时都花在输入电子健康记录、做笔记、写信件等事务上。所以可以说，你们正在做的事情至少在初期是针对这一部分问题的。
Because if you think about, if you produce that five hours to one, then that's four more hours to actually see humans and deliver actual healthcare. Yeah, exactly. And I think we'll probably look back at this period in history in the 200, 300 years of modern medicine. And look at all the human to human interaction that basically constituted the majority of those interactions for the first two centuries. And then look at these weird 10 years where we just typed on computers instead of looking at our patients as very weird as like, why did you do that? Why did you have to learn to touch type to be a doctor?
因为，如果你想一想，如果你能把五个小时减少到一个小时，那么就多了四个小时来真正接触病人并提供实际的医疗服务。是的，没错。我想我们可能会在现代医学的200到300年的历史中回顾这段时期，看看前两个世纪里基本上都是人与人之间的互动构成了大多数医疗行为。然后再看看这奇怪的十年，我们只是在电脑上打字，而不是看着病人，感觉非常奇怪，为什么要那样做？为什么当医生还得学会触摸打字？
That doesn't make sense. But that is actually an essential and a requisite skill, at least in today's world. The intention with digitizing our workflows was never, was never this. It was always about better record keeping, understanding patients, sharing data. None of those benefits ever, ironically, ever actually materialize. Better record keeping maybe as an exception. But certainly the sharing of information amongst healthcare systems in the NHS, your primary care tech still doesn't know what your secretary care doctor is doing.
这不合逻辑。但是，这实际上是当今世界的一项基本且必需的技能。我们数字化工作流程的初衷从来不是这样。初衷是为了更好的记录保存、了解病人情况、数据共享等。讽刺的是，这些好处从未真正实现。也许在更好的记录保存方面有所例外。但毫无疑问，在英国国家医疗服务体系（NHS）中，医疗系统之间信息共享仍然存在问题。你的初级医疗技术人员仍然不知道你的专科医生在做什么。
Your secretary care doctor still doesn't know what the other secretary care doctor did in the other hospital, even if that was two days ago and 100 yards down the road. But what we've also done is a sort of boiling frog approach constantly added more and more digital work to the clinicians because who else is actually going to do that work? Who is going to file the codes? Who's going to do the problem? Doctors like me, we're actually rebellious. We weren't doing it anyway.
您的秘书负责的医生到现在还不知道另外一个医院的秘书负责的医生做了什么，尽管那是两天前发生的事情，而且仅仅相隔100码。但我们做的也像煮青蛙一样，不断给临床医生增加更多的数字工作，因为没有其他人会做这些工作。谁来归档代码？谁来处理这些问题？像我这样的医生，其实是有点反叛的，本来就不打算做这些工作。
I deleted every button I could find on Epic to be able to actually use it the way I wanted to use it to do my actual job until one day I deleted the button bags, which was the labs button, and then I couldn't look up blood tests. And that was a bit of a disaster. So like there is an extreme version of that. But also, you know, the idea that we can go back to being on paper, it's nice in principle and I experienced that. So when I was cardiologist, what about six years ago now, I was working during the cyber attack on the NHS.
我删除了Epic系统中我能找到的所有按钮，好让它能按照我想要的方式来使用，以便完成我的实际工作。直到有一天我误删了"按钮包"，也就是查看化验结果的按钮，然后就没法查看血液检查结果了。这真是个灾难。所以这就是一个极端的例子。但你知道的，认为我们可以回到使用纸质的方式，这个想法在原则上是好的，我也有过这种经历。大约六年前，当我还是心脏病专家时，我在NHS（英国国家医疗服务体系）遭遇网络攻击期间工作过。
So we had about two years of computing. And then for a weekend, every computer in the hospital was down. Closed day and he, my boss called me on the Friday night and he was like, do you remember all the patients? And I was like, yeah, been with them all week. And he's like, did you write them down? I was like, well, no, he's like, we'll write them down now because tomorrow that's how we're going to do the water out. So we did it basically at the top of my head. But I have to say, the care was better, the patients loved it.
所以我们当时有大约两年的计算机使用时间。然后在一个周末，医院里的所有计算机全都瘫痪了。医院关闭了一天，周五晚上我的老板打电话给我，他问我还记得所有的病人吗？我说记得，我一周都和他们在一起。他问我有没有把他们的情况写下来，我说没有。他说，那现在就写下来，因为明天我们需要用这些记录来进行工作。所以我们把这些病人的信息都凭我的记忆写了下来。说实话，护理工作反而更好了，病人们也很满意。
We finished the water and had half the time. And I think the realization was the computers are actually getting in the way for most of what we're actually trying to do. Has really escaped us because we just kept adding adding more and more cognitive load. And that is a massive contributor to burnouts. About 50% of physicians globally now are self-reporting symptoms of burnout. I myself have been burnt out at least twice in my career. But we talk about that very facetiously.
我们喝完了水，但时间只过去了一半。我认为我们意识到，电脑实际上在妨碍我们完成大部分工作。这个问题一直没有被我们注意到，因为我们不断地给自己增加认知负担。而这种负担是导致职业倦怠的一个重要原因。现在全球约有50%的医生报告自己出现了倦怠症状。我自己在职业生涯中至少有两次经历过倦怠。但是，我们对这个问题谈得非常轻描淡写。
But what does that actually mean? What does it mean to be burnt out? It means that your doctor doesn't care. Now that is terrifying. And every time I say that out loud, it still gives me chills to think about that as a potential. And we're seeing the collaries of that right now in the healthcare system. So what is the solution? More workforce fine. It is the solution. But like, they're 10 years away. And actually, when you get to that point in 10 years, all of the pre-existing problems, increasing chronic disease, reduced shortage of resources, inefficient systems have just got worse.
但这究竟意味着什么呢？疲惫不堪到底是什么意思？这意味着你的医生不再关心你。这真的很可怕。每次我大声说出这句话时，想到这有可能发生，我依然感到不寒而栗。而我们现在在医疗系统中看到了这种情况。那么解决方案是什么呢？更多的劳动力，好吧，这确实是个解决办法。但是他们需要十年的时间。当你十年后真正达到那个点时，所有的现有问题，比如慢性病增多、资源短缺、系统低效等，只会变得更糟。
So training more people now is probably just staying still in 10 years time. So technology like ours is not the answer. I genuinely am asking this question now to any of your listeners. I don't know what is. If it's not going to be technology that helps us solve at least the short-term problem. And the power of these models is insanely impressive.
所以，培训更多人现在可能只是为了在10年后保持现状。所以像我们这样的技术并不是答案。我真诚地向你们的听众提出这个问题。我不知道答案是什么。如果不是技术帮助我们解决至少短期问题的话。这些模型的强大之处真是令人印象深刻。
And I remember using GPT three about two years ago and just being shocked that I could ask it, what is heart failure? And it would give me a really coherent answer, better than most medical students. Now, GPT four and GPT omnis, whatever they want to call it, has just come out. The knowledge that's encoded in these models is incredible. But they are not designed and never have been designed actually for knowledge retrieval or for accurate knowledge.
我记得大约两年前，我使用GPT-3时感到非常震惊。我可以问它“什么是心力衰竭？”，而它会给我一个非常连贯的回答，比大多数医学生的回答还要好。现在，GPT-4和GPT-omnis（不管他们怎么称呼它）刚刚推出。这些模型中包含的知识令人难以置信。但它们并不是为了知识检索或准确知识而设计的，实际上从来都不是。
They're transformer models. They're designed for translation tasks, one to the other. The biggest problem we have right now is how do we clinically evaluate these models and bring them in safely? And that's kind of why we're called tortoise because we realize that, and we'll live in this very strange world for the next 20, 30, 40 years, probably. Technology moves exponentially fast. And all our societal systems are just linearly progressing very slowly because there's so much legacy and infrastructure that just doesn't, is not ready for AI yet.
它们是转换器模型，专门用于翻译任务，从一种语言翻译到另一种语言。目前我们面临的最大问题是如何对这些模型进行临床评估并安全地引入应用。我们之所以被称为“乌龟”，正是因为我们意识到这一点，我们可能会在未来的20、30、40年里生活在一个非常奇怪的世界里。技术进步得非常快，而我们所有的社会系统却缓慢地线性发展，因为存在太多传统和基础设施，它们尚未准备好迎接人工智能。
There are hospitals in the UK still on paper, right? That's not going to work. Well, so that was going to ask, like you were talking about this idea of like when we look back and be like, what were we doing then? What is it about that this transition that has happened as you say over the past 10, 15 years or whatever it was that has gone so wrong or that is so clunky because it feels like on just, you know, any kind of normal person who uses computers versus uses, you know, is old enough to use a typewriter, you're like way better experience, obviously, computer.
英国还有些医院在使用纸质记录，对吧？那是行不通的。所以我想问一下，当我们回顾过去，会觉得当时在干什么？你提到的这种转型，就是过去十年、十五年间发生的事，为什么会出现这么大的问题，或者说为什么会这么笨重？因为对任何一个用过电脑的普通人来说，电脑的体验显然比使用打字机要好得多。
But it feels like something about the kind of implementation of, you mentioned Epic, which I think is the biggest electronic health company in the world or the software. What is it about that system that has made this so much worse than just writing notes, paper notes down? Because it feels like it should be actually far more efficient. Yeah, it's a super good question. I suppose, I mean, and to be fair to Epic, that's actually one of the better ones. There's lots of EHRs that are way worse. But I think it's the fundamental imposition of a very complex task when you're actually trying to do a very complex task, which is to see a patient.
但是，似乎有种感觉是，你提到的Epic，我认为它是全球最大的电子健康公司，或者说软件公司。这种系统到底有什么问题，让情况比写纸质笔记还糟糕？因为感觉它应该实际上更高效才对。是啊，这是个非常好的问题。我想，公平地说，Epic实际上是比较好的系统之一，还有很多电子健康记录（EHR）系统更差。但我认为问题的根本在于，当你实际上在尝试完成一个非常复杂的任务，也就是看病时，它却强加了另一个非常复杂的任务。
And there's this myth that human beings can multitask. It's not actually true. Human beings cannot actually multitask. No, I mean, the statistic, like they've asked people, all sorts of, they've tested this in many ways. Trying to even talk to someone on the phone and hold a conversation is about the same reaction speed as if you're drunk when driving a car. Our brains are not these fantastical multitasking beats. We're very good at concentrating on one thing. So there's a cognitive workflow. And then there's also the imposition of the distraction of having to document.
有一种人类能同时做多件事的说法，这其实是个神话。人类实际上不能真正地同时处理多项任务。科学统计和各种测试都证明了这一点。比如，当你在开车时打电话和人交谈，你的反应速度跟醉驾差不多。我们的脑子并不是那种神奇的多任务处理器。我们更擅长专注于一件事。因此，在认知工作流程中，记录的干扰也是一个问题。
So I'm typing, right? And I'm not looking at you because I'm trying to capture what you're saying. The irony is I'm missing most of what you're saying. And I'm now interrupting you to type down what you're saying. So you lose your flow. The majority of consultations, if at any healthcare system in the world, fundamentally comes down to the consultation between a patient and a clinician at any given moment and any given point. That breaks down into three steps. Talking, so taking a history, doing an examination, which is like physical examination, which is, you know, AI is going to struggle to do that, at least for a little bit, and ordering some tests or some x-rays.
所以我在打字，对吧？因为我试图记录你所说的话，所以我没有看你。讽刺的是，我反而错过了你大部分的话。而且现在我为了记录你说的东西，不得不中断你说话，这样你就失去了说话的连贯性。在任何医疗系统中，大多数的咨询基本上都是在某个时刻和某个点上，患者和临床医生之间的交流。这个交流可以分为三个步骤：谈话，也就是病史采集；做检查，也就是体检，这一点，至少目前AI还很难做到；以及安排一些测试或拍X光片。
Now you ask anyone on the street what they think the most important part of those three buckets is, they will always say it's the test, right? It's got to be the test, the test of the objectivity. Yeah. It's wrong. It's completely wrong. The tests are almost completely useless in almost every circumstance. It's about 70, 75% the actual talking, the interaction, really listening the right symptoms, asking, listening for the things that you don't say, checking that you actually asked about family history and social history and allergies. We don't spend enough time doing that bit already. The examination adds about 15% and the tests themselves are about 10%.
现在，如果你在街上问任何人他们认为这三个方面中最重要的是什么，他们总是会说是测试，对吧？一定是测试，测试的客观性。对，但这是错的，完全错了。在几乎所有情况下，测试几乎是无用的。其实，有70%到75%的重要性在于真正与人交谈、互动，仔细倾听正确的症状，问问题，聆听那些你没有说出口的事情，检查你是否询问了家庭病史、社会史和过敏史。我们在这方面花的时间还不够。体检大约占15%，而测试本身只占约10%。
So now take that knowledge that that is the single most important diagnostic part of that interaction. And the diagnosis is basically what defines how efficiently you flow through any given healthcare system. Right diagnosis, right time, you get the right test, the right treatment, you're out the door. Wrong diagnosis, you're going round and round, your heart, your cum, right diet. And now I'll say, okay, now do all of that, but also type, look at screens with 50 to 60 buttons. Don't miss anything, prescribe something, but don't make a mistake, fill in the forms, otherwise it doesn't get done. If you don't fill in this form with 55 fields, that patient doesn't get their x-ray and that patient might die.
所以现在理解一下，这点知识是该互动中最重要的诊断部分。诊断基本上决定了你在任何医疗系统中的流畅程度。正确的诊断，及时的诊断，你会得到正确的测试和治疗，然后就结束了。错误的诊断，你会反复检查，影响你的健康，甚至饮食。现在我再说，好，现在做所有这些，但还要打字，看着有50到60个按钮的屏幕。不要错过任何东西，开处方时不要出错，填写表格，否则事情无法完成。如果你不填写这张有55个字段的表格，那个病人就得不到X光检查，而那个病人可能会因此丧命。
Right. And that's the burnout, right? It's the stress of the meaningless work that actually is important, that loses importance and then causes this really big friction point. And it's funny, I've been doing this for about a few years now, like talking to physicians all over the world. It's a universal problem. And I have not yet defined a single physician that really thinks this was a good idea for clinical care. Systems love it, right? Better data, better billing, more audit capabilities. But they're also under the illusion that they're actually getting the data that they're asking for. I can always continue there not, right? We are not, we're giving up on both sides and just trying to get through the day. And that's actually what's happening to most physicians right now.
对的，这就是倦怠，对吧？这是因毫无意义的工作所产生的压力，这种工作其实很重要，但却失去了重要性，导致了巨大的摩擦点。有趣的是，我已经做这个好几年了，和世界各地的医生交流。这是一个普遍问题。我还没找到一个真的认为这对临床护理是好主意的医生。系统很喜欢，对吧？更好的数据、更好的收费和更多的审计能力。但他们也在错觉中，以为自己真的得到了想要的数据。但实际上，我们没有获得，我们双方都在放弃，只是想熬过一天。这实际上是现在大多数医生的状况。
But going back to, you know, the best job I ever had was when I had no login for the computer at all. So I had a little composition, it rolled a contract every two weeks, got paid by the hour. So to go and do IT training was half a day that I didn't get paid. So that was the point, I'm only here for two weeks. But every two weeks, they would renew my contract. I ended up working there for like maybe four months. But I couldn't log into the computer, I didn't have any passwords. So what I did was I took a junior doctor with me on the ward, saw 20 patients, took another junior doctor, saw another 25 patients around the hospital every single day. And I didn't realize this, but like as a cardiologist, I was like actually unblocking a lot of people's discharge plans. Like they could go home after being after seeing cardiology. So I was saying 45 patients a day and that seems, that is a lot for a secondary care doctor far more than I've ever saw before or since.
回想起来，你知道的，我有过最好的工作是当时完全不需要登录电脑。我只需要每两个星期写一个小报告，按小时计算报酬。所以去参加IT培训对我来说意味着半天没有收入。于是我认为，反正我只在这里工作两个星期。然而，每两个星期他们都会续签我的合同。最终我在那儿工作了大概四个月。但由于我无法登录电脑，没有任何密码，我采取的做法是每天在病房里带着一名初级医生，查看了20位病人，然后再带着另一名初级医生，在医院里查看另外25位病人。我没有意识到的是，作为一名心脏病学家，我其实在帮助很多病人完成出院计划，他们在看完心脏病科后就可以回家了。所以我每天要看45名病人，这对于一个二级护理医生来说很多，这是我以前或以后都从未达到的数量。
But the irony is I loved it. It's just pure medicine. See a patient make a diagnosis, make a plan, check the blood, move on. Actually doctors don't want to work less hard. We just want to do the bit that we actually enjoy and we train for and thought would give impact to the world. Not the bit that seems to only really care if you fill in the forms or the appointment time. So how does it work? So what have you built? It's called Oslur, which I think you mentioned is there's a reason you called it Oslur. But so what is it and how does it work? Yeah, so well it's interesting. So it's called Oslur. It's named after Sir William Oslur who was the father of modern medicine. And he's very famously ascribed to quote, listen to the patient. They are telling you what is wrong, which is exactly the point I was going to make about listening to the history. But also the irony is here stands for an acronym because I like I cranums operating system leverage in electronic records. And it's all about the leverage. How do you do more with the people that you have by using AI to be able to increase their capabilities? So what that looks like, it's a desktop app. It's installed, we're live in a bunch of primary care and secondary care settings in the UK right now. It listens to the consultations. So it takes the audio of your ambient consultation. So you walk in, I'm a patient, right? Yeah. And you, Dr. Pimenta, walk in. And I'm like, do you have to say, okay, I'm turning this, or do you consent to be recorded? Or like, is that a requirement?
但是讽刺的是，我很喜欢它。这就是纯粹的医学。看看病人，做出诊断，制定计划，查验血液，然后继续。实际上，医生并不想少工作。我们只是想做我们真正喜欢的部分，并且我们为之接受了培训，并认为这会对世界产生影响的部分。而不是那种似乎只关心填写表格或预约时间的部分。那么它是怎么运作的？你们开发了什么？它叫做Oslur，我想你提到过这个名字的背后有原因。那么它到底是什么，又是如何运作的呢？是的，这很有趣。这个系统叫做Oslur，以现代医学之父威廉·奥斯勒爵士命名。他非常有名的一句话是，“倾听病人，他们在告诉你问题所在。”这正是我要强调的，倾听病史的重要性。但这里有一个讽刺的是，它还是一个首字母缩写，我喜欢首字母缩写，Oslur代表“操作系统在电子记录中的杠杆作用。” 这就是它的核心，通过利用人工智能，以提升现有人员的能力来做更多的事情。那么它看起来是什么样子呢？它是一个桌面应用程序，已经在英国的许多初级和次级医疗环境中上线了。它会监听咨询记录，记录你咨询时的音频。比如说，我走进诊室，我是病人，对吧？然后你，皮门塔医生，走进来。我是否需要说，“好，我正在录音”，或者询问是否同意被录音？这是不是一个硬性要求？
Yeah, I mean, it's, again, interesting open question and lots of our physicians are posting a bit differently. I mean, we do provide patient information consent forms and things like that. Interestingly, patients don't seem to mind. And in fact, they love it. I have one user who's used it, I think 700 times in the last two months. And not single patients ever asked for the actual information consent form about the technology. The doctor's there, you know, we have all our security, cybersecurity badges, data protection, we don't store any data. But other than that, they're getting a doctor who's looking them in the eye and talking to them the whole time with no typing. And they're really like, what is this magical experience? Oh, that's medicine, actually. That's what it's supposed to be. So I'm the patient. You come in with your computer, you set it on whatever the table next to us. And it's recording or you hit record or whatever. And then it is in real time, presumably, because, you know, we have a lot of these tools as journalists, which we use, like kind of automatic real time transcription, which, you know, you're not going to be able to do it. You know, as I've mentioned before, in this poem, many times, has been the biggest productivity unlock in my professional career. Because transcription alone is, you know, for old people like me who used to have to do it where you stop the tape, you reverse it, you press play, you reverse it, you press play again. You're like, what did they say, all that stuff? To create a real-time transcription that it becomes a searchable document, it kind of saves, I don't know how many hours and hours and hours and actually produces a better product.
是的，我的意思是，这依然是一个有趣的开放性问题，我们的许多医生对此有不同的看法。我是说，我们确实提供患者信息同意书之类的东西。有趣的是，患者们似乎并不在意，事实上，他们很喜欢。我有一个用户，在过去两个月里用了大约700次。然而，没有一个患者曾经要求查看关于这项技术的信息同意书。医生就在那儿，我们有所有的安全、网络安全徽章和数据保护措施，我们不存储任何数据。除此之外，病人在整个过程中得到的是一个时刻注视他们并与他们交谈的医生，而不是在打字。他们真的会觉得这是一种神奇的体验。哦，这其实就是医学，本该如此。所以我是患者，你带着电脑进来，把它放在我们旁边的桌子上。然后你开始录音，或者怎么操作，然后它是实时工作的，因为我们作为记者，拥有许多类似的工具，可以进行自动实时转录。对我这样的老人来说，转录一直是我职业生涯中最大的生产力解放。因为以前我们要不停地倒带、播放、再倒带、再播放，搞清楚他们到底说了什么。现在通过创建一个实时转录，成为一个可搜索的文档，不知道能节省多少小时，并且还能生成更好的产品。
So you have that, and is that it, basically? No, so, and actually, we don't even use real-time, and I'll tell you why in a second, but like, so, listen to the consultation, and then also any dictation. So, you know, it's a lot more like having a colleague in the room with you that you're sort of talking, they listen to the console, they have the audio of that, then they take any orders for tests and things like that. And then it concatenates that together, and then we transfer that from a transcript, first of all, so we can infer that very fast now. So an hour of audio in about five to ten seconds, which is completely nuts, and it's a really crazy world that we now live in. And then we transfer that to a large language model, we have a stack of them, and we do lots of clinical evaluations for accuracy and performance, and constantly changing the models now as well as things are moving so fast. That makes your medical note, in your style and your templates, your structure, that's super important to physicians. We realize that the output of these models actually sounds and feels like theirs, not because they're trying to pretend, but because of a pattern recognition, that's actually how they understand that this is accurate or complete, like it's a pattern recognition instinct.
所以，情况就是这样，仅此而已吗？不，还不止这些。实际上，我们甚至不使用实时处理，稍后我会解释原因。你需要听一下咨询内容，还有任何口述记录。这样做非常类似于有一个同事在房间里与你交谈，他们听着咨询内容，记录下音频，然后接受任何检验或测试的指令。接着，我们把这些内容串联起来，首先进行转录，现在我们可以非常快速地完成转录，一个小时的音频大约只需五到十秒，这简直是太疯狂了，这个世界真是变化迅速。
然后，我们将这些内容传输到一个大型语言模型中，我们有一系列这样的模型，同时也进行大量的临床评估以确保准确性和性能。由于变化速度太快，我们不断调整这些模型。这能生成符合你的风格、模板和结构的医疗记录，这对医生来说非常重要。我们意识到，这些模型生成的结果听起来和看起来确实像是医生自己写的，不是因为它们在模仿，而是因为模式识别的关系。这才是真正让医生感觉到准确和完整的方法，就像是一种模式识别的本能。
Sorry, how do they learn the doctor's style? Do you have to, like, as a doctor, do you have to submit, I don't know, a bunch of notes or things you have written, so it kind of gets your cadence and kind of things like that? Yeah, exactly. And again, back in the day, this would have been to submit 100, 1000 notes. Now we can pretty much do it with one note, and we're building that system at the moment, it's still very much in beta, but it seems to work pretty well. So you give an example of what you're trying to achieve. It generates a template for you, then it will apply that template to future models. And I think over time, we'll probably build a bit more intelligence where it will learn the little corrections that you make and the little voice, the different contexts. Lots of doctors have different templates for different contexts. They have different voices for different situations. But again, it's one of those interesting problems that seems very complicated, but actually there's an acceptance gap where you'll go, okay, it's not 100% me, but it's 85% me, and that's good enough, because I didn't have to write this myself, right? And I recognize it. And I've had that moment, actually, when I use the system, I put my cardiology concise template into it. And I was like, oh, that's me. That's how I write. And it didn't really sound like me, but it looked like me, and it felt like me, and it had all the information that I cared about. And I think that's the important thing for physicians. It's a very personal game.
对不起，他们是怎么学习医生的风格的呢？作为医生，你要提交很多笔记或者写过的东西，以便掌握你的节奏和风格之类的东西吗？对，没错。以前你可能得提交上百甚至上千份笔记。但现在我们差不多只需要一份笔记就能做到这一点。我们正在构建这个系统，目前还是测试阶段，但效果看起来相当不错。你可以提供一个你想要达到的示例，系统会生成一个模板，然后将这个模板应用到未来的模型中。我认为随着时间的推移，这个系统会变得更智能，能够学习你做的小修改和不同语境下的表达。许多医生在不同语境下会有不同的模板和表达方式。
但说到底，这其实是一个看似非常复杂的问题，但实际上存在一个接受度的问题——你会发现，虽然它可能并不是100%的你，但85%的相似度已经够好了，因为你不用自己去写这些。我自己在使用这个系统的时候也有过类似的体验，我把我的心脏病简单模板放进系统里，当生成结果出来的时候，我觉得，哦，这就是我。这就是我的写作风格，虽然听起来不完全像我，但看起来像，感觉也像，而且包含了我关心的所有信息。我认为这对医生来说非常重要，因为这是一个非常个人化的事情。
So yeah, it creates notes, and then also letters. So patience is in the UK. The letter is the medical legal document of record. But now we have the capability to paralyze these functions. So what I mean by that is it produces a note and it produces a letter, but it can also produce a patient letter in the patient language, or a translated note, or a note referral, and all simultaneously. So these aren't any more sequential tasks for a physician. All the documentation can be parallel. And that list could be in the future, maybe 100 different outputs, audits and registers and clinical trial searches, and who knows what else. And coding now as well. So coding is an interesting AI task, because it's a classification task, but it requires clinical reasoning. And we have a bunch of coding tools in the system as well. And the plan is to extend that to taking over the downstream tasks as well. So ordering the blood tests, not just documenting them, ordering the prescriptions and having you review and approve them, and also collecting information.
所以是的，它不仅能生成笔记，还能生成信件。在英国，信件是医疗法律文件记录的一部分。但现在我们有能力使这些功能并行处理。我的意思是，它可以生成笔记和信件，还能以患者的语言生成患者信件、翻译笔记或转诊笔记，并且都能同时进行。也就是说，这些任务对于医生来说不再是依次完成的了，所有文件记录可以并行进行。未来，这可能会包括多达100种不同的输出，像审计、注册以及临床试验搜索等，谁知道还能做些什么，包括编码。编码是一个有趣的人工智能任务，因为它是一个分类任务，但需要临床推理。我们系统中也有一系列编码工具，并计划进一步扩展，接管下游任务。例如，不只是记录血液检查，还能下单进行血液检查，开具处方让你进行审核和批准，还能收集信息。
So before I even see, I'd like to know a lot more about you. I spend a lot of time trying to find information on healthcare record systems. It's usually pretty badly kept. It's kept for medical legal purposes, but not for user capabilities. So trying to pull information out at speed in the way that you want it is like the other big time sink. So basically, by the end of the year, we'll have built a system that essentially means that you don't touch the computer regularly as a physician at all. And only to do specific tasks or things that aren't, we haven't added yet or thought about yet or in emergency or something. And it does feel much lighter. I mean, we built a prototype 18 months ago, and I ran a clinic at the accelerator where we in Chris met, like a simulated clinic with simulated patients. It just felt like magic.
在我开始之前，我想更多地了解你。我花了很多时间研究医疗记录系统的相关信息。这些信息通常保存得很糟糕，主要是出于法律目的，而不是为了用户的便利。因此，快速提取你想要的信息可能会耗费大量时间。基本上，到今年年底，我们将构建一个系统，让医生几乎不再需要频繁使用电脑，只需在执行特定任务或紧急情况时使用。这个系统感觉轻松了许多。18个月前我们构建了一个原型，并在加速器中进行了模拟诊所和模拟病人的测试。感觉就像变魔术一样。
I really did feel like I'm just doing my job and that relief of the burden of documentation that is no longer my responsibility, but it's done for me. It's a kind of a magical feeling. And now definitely feeling that with our customers who were going home early, you know, doctors in the UK never come across that before going home 15 minutes early from your shift as opposed to two hours late. That is, you know, a phenomenal change and really something worth, you know, digging deep much deeper in. Yeah. Your history is a new podcast brought to you from the times and it brings together the real life stories from our obitories desk, which have been published for over a century. In this brand new show, we build on this legacy and explore the endlessly fascinating lives who have enriched and informed our own. Join me and our sponsor, Ancestry, as we journey through your history.
我确实感到自己只是在完成工作，而且文档负担不再是我的责任，而是由系统替我完成。这感觉非常奇妙。现在我们的客户也有这样的体验，他们能够提早回家。你知道，英国的医生以前从未有过提早15分钟下班的经历，相反，他们通常会晚两个小时下班。这是一个巨大的变化，真的值得进一步探讨。
《你的历史》是一档由《时报》推出的新播客节目，带来了已有数十年历史的讣告部门的真实故事。在这个全新节目中，我们将延续这一遗产，探索那些丰富了我们生活的无尽迷人故事。和我以及我们的赞助商Ancestry一起，踏上探索你历史的旅程吧。
You say you're a cardiologist, so could we go back like where are you from and how did you end up as a doctor and then what made you decide to do this? Well, we've got to get right back to the beginning. Okay. So hi, I'm done. Actually, I do do this, right? I do. We do this thing in the company. We do it. We call it Lifeline and it really helps people understand where they're going to understand how to predict it. I don't give you the whole story, but so you guys have your own podcast. So yeah, yeah, log story short, Catholic parents, both laps or chains, but grew up essentially with the knowledge that life is not about being happy, but there was no like carry over to that. It was like not happiness, but there was a bit of an open question like, okay, what is life important?
你说你是心脏科医生，那我们能不能回到最初，聊聊你是从哪里来的，以及你是怎么成为医生的，还有是什么让你决定从事这个职业的？好的，我们得从头说起。嗯，大家好，我已经做完了。其实我真的做这行，对吧？我们公司有个活动，我们称之为“生命线”，这真的帮助人们理解他们的方向并预测未来。我不会讲整个故事，但你们有你们自己的播客，对吧？总之，简短来说，有天主教信仰的父母，他们都是虔诚的信徒，但基本上我是在一个知道生活不是为了追求快乐的环境中长大的，只是没有具体的延续。当然，问题还有就是，究竟什么才是生活的重要意义。
And then if you're if you're raised Catholic, you spend a lot of time thinking about death in the afterlife. Right. So I decided that, you know, money doesn't carry over. So that's not that's not that's pointless. Therefore, what is not pointless? Helping people. And then you might see them in heaven and therefore the logic was, you know, that's a good career because that's the riches you take with you. So that was my four year old decision to become a doctor. So it doesn't sound like you're very lapsed. This sounds this all sounds very Catholic and not lapsed. Well, I mean, I mean, I mean, I'm still pretty religious in that sense, not Catholic at all, but I do believe in God. So I think I just didn't change that plan, which is an interesting sort of operating system problem that I have that I make a plan. I just tend to execute it until something else changes my mind. And nothing ever did.
然后，如果你是天主教徒，你会花很多时间思考死亡和来世，对吧？所以我决定，金钱是带不走的，这是没有意义的。因此，什么才是有意义的呢？帮助别人。然后你可能会在天堂见到他们，所以逻辑上这是一个不错的职业，因为这是你能带走的财富。所以这就是我四岁时决定成为医生的原因。听起来你并没有远离宗教，这一切听起来都很天主教，而且并没有脱离信仰。嗯，我是说，我还是蛮宗教的，不是完全的天主教徒，但我相信上帝。所以我觉得我没有改变这个计划，这是个有趣的“操作系统”问题：我制定了一个计划，就会执行它，直到有其他事情改变我的想法，而这样的事情从来没有发生过。
So a four year old self decided to become a doctor. Just did that basically flash forward to 28 years later or something. And I've been a cardiologist in the NHS for about 10 years, planned to go out to do an ML research PhD in cardiovascular machine learning for diagnosis of under diagnosed conditions. So specifically looking at like populations like women, for example, tend to be under diagnosed for heart attack when they go to hospital because the human beings aren't very good pattern recognisers. And sorry, the machine the thing with machine learning right is that it's really good at recognising patterns. Yes. That's the whole thing of that like and consistent as well. And actually, I think kind of sift through mass amounts of data.
所以一个四岁的孩子决定成为一名医生。基本上就是这样，然后快进到大约28年后。我在NHS担任心脏病医生已经有大约10年了，计划出去攻读一个关于心血管机器学习的ML研究博士，以诊断那些未被充分诊断的疾病。具体来说，比如像女性这样的群体，当他们去医院时，往往心脏病发作未被充分诊断，因为人类在识别模式方面并不是很擅长。抱歉，说到机器学习，它在识别模式方面非常出色，是的，这就是它的强项，而且非常一致。实际上，我认为它可以筛选大量的数据。
Yeah, exactly. And I think it's the consistency that I would argue is actually even more important. So if you have a machine learning algorithm, there's 75% accurate, but it's 75% accurate all the time, you can meaningfully make that better. But if you have a human being that's 90% accurate, but has bad days, sometimes we're going to 40%, you can't actually iterate on human beings. And this is something that's very fundamental to healthcare where it's about repeatable motions done boringly and done well is actually the fundamental to care.
对，完全正确。我认为，一致性实际上更加重要。比方说，如果你有一个机器学习算法，它的准确率是75%，但始终保持75%的准确率，你就能有效地改进它。但如果一个人的准确率是90%，但有时遇到糟糕的日子掉到40%，你就无法对人进行优化。在医疗保健领域，这一点尤其重要，因为稳定重复地执行某些操作，并做好这些操作，才是护理的基础。
So that's why I really like AI, because I think even if it's bad today, you can make it better, but you can't make humans better in the same way. So that was the idea. I didn't get anywhere because all the funding dried up because COVID came along. So this was like early 2020 in the first wave. And I got redeployed to COVID IT. I worked there for six months and at the same time founded a charity. So that was like my first founding experience. So we raised some money.
所以这就是我特别喜欢人工智能的原因，因为我觉得即使它现在还不够好，你也可以让它变得更好，但你无法以同样的方式改进人类。这就是我的想法。但是我没有成功，因为所有的资金都枯竭了，因为遇到了新冠疫情。这大概是在2020年初，第一波疫情期间。我被调派去处理与新冠相关的IT工作，在那里工作了六个月，同时创办了一家慈善机构。这算是我第一次创办机构的经历。我们筹集了一些资金。
What was the charity? So it's now called the Healthcare Workers Foundation. It's still ongoing. At the time we called it heroes, which again was an acronym and I can't remember familiar life for me what it was. I think it was something like healthcare, extraordinary response organization or something. But the point really was like a lot of people wanted to help. There wasn't really an obvious channel. I was working in IT, you're seeing that we didn't have PPE and then lots of colleagues reaching out to me saying, how do we get more PPE because we don't have any protection.
是什么慈善机构呢？现在它叫做医护人员基金会，还在继续运作。当时我们叫它英雄，这也是一个缩写，但我实在记不清具体是什么了。我觉得好像是类似于医疗特别响应组织之类的名字。但重点是，很多人想要帮忙，却没有一个明确的渠道。我当时在IT部门工作，看到我们没有足够的个人防护装备（PPE），然后很多同事联系我说，我们该怎么获得更多的PPE，因为我们没有任何防护措施。
So we built reusable PPE as she was our thesis and we got just a day that we could to put in. And that was really exciting. And the team I had at that moment, I made all the founder mistakes to go to your point about founders. So we had five co-founders. It was me, my wife, my sister, my wife's cousin and one of my good mates. And we fought like cats. Oh my God. Really discoordinated. I didn't know what I was doing. I was a CEO. We had an amazing team, like genuinely the team that worked for that company. I mean, the branding guy now represents massive brands. I think we've just actually been acquired that company. Comms, I think she works for Procter & Gamble Global now. The social media person works for Co-Coder. Like literally the talent that was in that team was nuts. And I didn't recognize it at all. I was like, oh, these are just people who do some stuff. I don't know anything about the world.
所以我们当时在做毕业论文时，开发了一种可重复使用的个人防护装备（PPE）。那之后我们只有一天的时间来完成它，但真的非常令人兴奋。而且当时那个团队，我犯了所有创业者可能犯的错误，比如你提到的那些错误。我们有五个创始人，分别是我、我的妻子、我的妹妹、我妻子的表妹和我的一位好朋友。我们争吵得厉害，像猫狗一样，真的非常不协调。我当时是首席执行官，但我根本不知道自己在做什么。我们的团队非常棒，说真的，在那家公司工作的人才非常出色。比如我们的品牌负责人现在代言了许多大型品牌；我们好像刚被收购了。负责通讯的人现在为宝洁全球工作；负责社交媒体的人在为Co-Coder工作。团队的这些人才简直太惊人了，而我当时根本没有意识到这一点，我只是觉得他们不过是做做事的人罢了，我对这个行业一无所知。
So I've been a doctor at that point. And are all of those relationships intact because your wife, your brother, your sister, I mean, that seems like a high risk endeavor. Yeah, I will do it. I'm not going to lie. It was hairy. I've got some scars from it. But no, 100% all intact. And actually, Rosh, my good friend, is the current chairwoman of the charity that's been running it for the last few years herself and doing a really, really good job. I think one of the really amazing moments about it was, yeah, it was super tough. And to be fair, COVID-19, the first two months was a very grim, very grim job. We were very fortunate where I worked. But in other places, it was dark. It was a lot of death. A lot of death.
所以当时我已经是一名医生了。而且你妻子、兄弟、姐妹的关系都还好吗？因为这看起来风险很大。是的，我会去做的。我不会撒谎，这确实很艰难。我从中也留下了一些伤疤。但不，百分之百关系都还在。而且，Rosh，我的好朋友，现在是那个慈善机构的现任主席，这些年一直在运行，并且做得非常非常好。我认为其中一个非常令人惊叹的时刻是，对，这的确非常艰难。老实说，COVID-19的头两个月，工作非常悲惨，非常悲惨。我们工作地点算是非常幸运的。但是在其他地方，情况很糟糕。那是一段充满了死亡的时期。大量的死亡。
Yeah. Yeah. Well, yes. And I think that was, I mean, part of talking about burnout, that was probably a big contributor to my burnout, on that stage. But I think what actually was a release for me was the charity because there was a positive channel for energy. There was a lot of good things that we were doing, delivering meals, getting celebrities to like, you know, Jamie Lange from Candy Kittens donated 80 grand and gave loads of candy kittens to hospitals. It was just those mad things. But I think the really interesting thing was the learning curve as a founder is infinite. Like you're constantly learning how to do things that you have no idea.
对，是的。嗯，是的。我认为，谈到倦怠，这可能是我在那个阶段感到倦怠的一个主要原因。但是，我认为真正让我解脱的是做慈善，因为这是一个正能量的输出渠道。我们做了很多好事，比如送餐、与名人合作。比如Jamie Lange从Candy Kittens捐了8万英镑，还给医院送去了一大堆Candy Kittens糖果。这些事情虽然疯狂，但真的很有意义。我觉得作为一个创始人，真正有趣的是学习曲线是无限的—你不断地在学习如何做那些你完全不知道的事情。
Like, what's Trello? What's Scrum? How do I do charity governance? You know, why am I fighting with the BSI about the thickness of plastics in their, you know, certification for eyewear? Like, there's some bad thing to do. Right. But, you know, I genuinely loved it. And I think one of the things that happens to medics in general is that you train for years and then you plateau because there really isn't much more to learn. I mean, you can become an academic, but actually the fundamentals are, okay, now you've got to see a bunch of patients for 50 years. And that actually does burn a lot of people out because unless they find some other thing to keep them occupied, what's here was this thing where you could just keep growing and learning and growing and learning.
就像，Trello是什么？Scrum是什么？我该如何做慈善机构的治理？你知道的，我为什么要和BSI争论他们认证眼镜用的塑料厚度？好像这些都是些很头疼的事，对吧？但其实，我真的很喜欢。而且，我认为对医生来说，经常会发生的一件事就是，你们经过多年的训练，然后达到顶峰，因为真的没有更多东西可以学了。我是说，你可以成为学者，但实际上，基本上就是，你现在得看五十年的病人。而这实际上会让很多人感到疲惫，因为除非他们找到其他可以让他们保持活力的东西，否则总是重复这些工作。而在这里，你可以不断地成长和学习，不断地成长和学习。
And I thought, wow, this is, this is probably what I actually want to do. I didn't really have an avenue for that. Anyway, a bit burnt out, quit after the first wave. When COVID was over, that was a good, that was a good moment. Do you remember that? Do you remember COVID was over? That was a joke. So I left, I left because COVID was over. Sorry. So you quit, you quit the NHS. You quit your job as a cardiologist.
我心里想，哇，这大概就是我真正想做的事情。我之前一直找不到途径。无论如何，我有点筋疲力尽，在第一波疫情后我辞职了。COVID结束后，那是个不错的时机。你还记得吗？你记得COVID结束的那个时候吗？开个玩笑，其实COVID并没有结束。对不起，我辞职了，因为COVID结束了，对不起，开个玩笑啦。所以你辞去了 NHS 的工作，你辞去了心脏科医生的职位。
Yeah. Yeah. Big thing. It was a big thing at the moment. And my mum, especially, was like, what are you doing? But I just think I was super burnt out, like six months of really hard work. And like, yeah, for sure, lots of reward. I mean, not to be, you know, but also just a lot of really thinking what is happening and just needing a break to like put my head back together after this absolutely crazy period. And then my wife was like, okay, it's great. You quit your job, but you do have three kids.
是的，是的，当时发生了一件大事。这事儿当时很重要。我妈妈尤其问我，你在干什么？但我觉得我真的精疲力竭了，连续六个月的努力工作。就是说，收获确实很多，不是说没有，但是也有很多困惑，想搞清楚到底发生了什么，所以需要休息一下，让自己冷静下来，整理一下思绪。这段时间实在是太疯狂了。然后，我妻子就说，好吧，你辞职了，但你得记住我们还有三个孩子。
So I didn't know two kids at that time. So what's the plan? I was like, okay, let me get another job. So I ended up becoming a pharmaceutical physician and I worked on clinical trials. And again, that was really interesting work, worked in first in human trials. I worked on CRISPR as one of the few doctors that worked on the first CRISPR trial in human beings, um, sign the singer RNA. Really loved that work. But you know, black's one event, my wife got sick. I mean, she's fine now. She's back away with our third child, but it meant I had to stay at home.
当时我不知道有两个孩子。所以，下一步计划是什么呢？我心想，好吧，那就找另一份工作吧。于是我最终成了一名制药医生，参与了临床试验工作。这份工作非常有趣，我首先参与了人体试验，甚至是工作在CRISPR技术上，成为了少数参与首个人体CRISPR试验的医生之一。我真的很喜欢这份工作。但是，有一件意外，我的妻子生病了。不过她现在已经没事了，并且怀上了我们的第三个孩子，但这意味着我必须待在家里。
So became an AI healthcare academic. I was really grateful to run the AI, sorry, the academic part of the company where I was working. And at this moment, we had to learn to code to run the studies because we didn't really have a lot of engineering support. So I learned to code. And then I was like, Oh my gosh, you can talk to computers, which anyone who's listening is vaguely techy is going to think that's the stupid thing that they realized when they were like three.
于是我成为了一名人工智能医疗学者。我非常感激能够负责我所在公司的人工智能，哦不，学术部分。那时候由于工程支持不足，我们必须学会编程来开展研究。所以我学会了编程。当时我心想，天啊，你竟然可以和电脑"对话"，而任何一个有点技术知识的人一定会觉得这是他们三岁时就意识到的"愚蠢事情"。
And here I am at like 35 or whatever going, Oh, you don't need to do software. You can do stuff and code. And it's really powerful to really like that. And again, just got really into the autonomy of creation of running studies and finding things out and discovering things. And I think at some point, and I can't really remember what happened. I realized two things. One, there's not a lot of impact in pharma for doctors at that stage, right? You're giving drugs that someone else invented 10 years ago.
于是我到了大约35岁的时候，才意识到，哦，你不一定非要做软件。你可以做其他事情，也可以编程。而且这样做真的很有力量，让人非常喜欢这一点。我还迷上了自主地进行研究、发现新事物和探索事物。有一次，我突然意识到两个问题。第一，对于医生来说，在制药行业影响力并不大，因为你只是在给病人使用别人十年前发明的药物。
If I didn't want, if I didn't go to work, someone else just did my job. And in fact, I can send to the second ever human being for CRISPR. But the only reason I did that is because the person who's supposed to do it was sick. And I just turned up and did his job for him. So your leverage is like one, if minus not maybe minus point five. But health tech allowed me to leverage my knowledge as a clinician, 10, 100, 200 times over by designing studies that make clinical sense, but then using technologies to deliver them. And that was phenomenally powerful. I was like, wow, this is really heavy impact and doing really good work when it's done properly, which has always been the hard thing to do. And I think the other thing I realized was I just got too old to have a boss. I just remember being in a meeting being like, what, why am I in this meeting? I'm 35. I've got three kids. What are you talking about? Yeah. So I sort of did a quiet quit and I got into a accelerator called entrepreneur first, which is pretty well known in London now. I've actually got an office out here in SF, which I might go visit later.
如果我不想工作或不去上班，有人会替我做我的工作。事实上，我甚至可以安排第二个接受CRISPR实验的人类，但我这么做的唯一原因是原本应该做这件事的人生病了。我只是去代替他完成了他的工作。所以你的杠杆作用就像是1，如果减去的话可能是负0.5。但健康科技让我利用作为临床医生的知识，可以成倍地提升效率，10倍、100倍、甚至200倍，通过设计有临床意义的研究，然后使用技术来实现这些研究。这真的非常强大。我感到很震撼，这种力量非常大，能够在正确执行时做出非常好的工作，而这一直是难点。还有一点让我意识到自己变老了，不再适合有个老板。我记得有一次在会议上，我心想，我为什么要在这个会议上？我已经35岁了，有三个孩子。你们在说些什么？于是我选择安静地退出，然后加入了一个叫“Entrepreneur First”的加速器，在伦敦很有名。我在旧金山这边也有一个办公室，可能稍后会去看看。
Was there a plan? Or were you just like, I'm just kind of in. Am I creating and et cetera? But I'm just going to go in there and see what happened. Yeah, you saw my wife. What is there a plan? You know, I think the thesis was AI is cool. It's super powerful. What can we do with it in healthcare where we can unlock a lot of value? And actually I had an early thesis about patient records and decentralizing them and having AI as a companion. I hope your pitch to your wife was better than AI is cool. I mean, it wasn't far off to be honest, but like it's an accelerator. Right. So actually it's a really interesting one. So it's not like why I see where it's an established company.
有没有一个计划？还是你只是觉得，我只是在参与。我是在创造等等，但我只是想进去看看会发生什么。是的，你看到了我的妻子。究竟有没有计划呢？我觉得当时的想法是，人工智能很酷，它非常强大。我们能在医疗保健领域做些什么，从而释放出巨大的价值？实际上，我一开始的设想是关于病人记录以及如何去中心化它们，并把人工智能作为一个辅助。我希望你对你妻子的企划要比“AI 很酷”更好。老实说，其实也差不多，但就像是一个加速器项目。所以，这其实是个很有意思的事情。它并不像是 Y Combinator 那样的成熟公司。
They look for individuals who show really good, found a potential, but actually aren't wedded to any idea. So maybe that was where I lucked out by not really having strong thesis. I hope this is cool. And I was building something else that I met Chris and Chris, Tans, machine learning engineer, he's an academic and he'd always worked on sort of AI human co-working very specifically about teaching AI to use the human environment of computers. So the virtual environment, like your computer desktop is designed for humans like this. You have to into it a lot to actually utilize the system and the mouse and keyboard are actuators.
他们寻找那些表现出好潜力，但实际上并没有拘泥于任何想法的人。所以，也许我运气好在于我并没有特别坚定的主题。我希望这可以算作优秀表现。在我搭建另一项目时，我遇到了克里斯和克里斯-谭，他是一位机器学习工程师，身为学术界人士，他一直在研究AI与人类协作的问题，特别是教AI如何使用人类设计的计算机环境。也就是说，这种虚拟环境，比如你的电脑桌面，是为人类设计的。你必须直觉性地去使用系统，而鼠标和键盘就是操作工具。
So if you give that over to an AI model, can you train it to do? And now we'd call them large action models, but this was a couple of years ago when really this didn't even have words for what this was trying to describe it. And then he came up to me one day, he's like, don't doctors spend a lot of time on computers? And I was like, yes, we do. And it's awful. Let's build that. So it's Chris's idea. I'm just along for the ride if I'm honest with you. And that's what we've been doing for the last 18 months is trying to build out these tool sets.
所以，如果你把这些交给一个AI模型，你能训练它做什么呢？现在我们称它们为“大动作模型”，但这还是几年前的事，当时甚至没有词汇来描述这是什么。他有一天走到我面前，说：“医生不是花很多时间在电脑上吗？”我回答：“是的，我们确实花很多时间在电脑上，而且这很糟糕。我们来解决这个问题吧。”所以这是Chris的想法，如果我要跟你说实话，我只是随波逐流而已。这就是我们过去18个月一直在做的事情——试图构建这些工具集。
In February, I think it was announced that coastal eventures, Vinod Kosla, who's extremely well known out here, billionaire investor, very, very smart, the first investor in open AI, various other things. I know it was announced in February, which probably means you raised it a year ago. But how did you end up getting Kosla to invest? That's a really good question. No one's asked me that question before. So yes, very astutely we did raise it very close on. So we got some money from the accelerator and then very quickly realized it's a very competitive market. So we really did need to raise quite quickly.
二月份的时候，我记得宣布了海岸风投和Vinod Kosla的投资。他在这里非常有名，是一位亿万富翁投资者，非常聪明，是OpenAI的第一位投资者，还有很多其他投资。我记得是二月宣布的，这可能意味着你们在一年前就已经融资了。但你们是如何让Kosla投资的呢？这是一个非常好的问题，以前没有人问过我这个问题。是的，我们确实是很快就进行了融资。我们从加速器那里拿到了一些资金，然后很快意识到市场竞争非常激烈，所以我们确实需要尽快融资。
And I think one of the things I've realized about EF is really good at this is it's much easier to be incredibly ambitious than it is to be highly ambitious. So like shooting your shot for the moon shots, because you only have to land one of those moon shots, right? So I remember I met Ross Harper, who's the CEO of Limbik as part of entrepreneur first. And he'd been invested in Kosla a few years prior to me. And we talked about something else. We didn't really make much ends into what we were talking about there. But I did say, you know, can I get an intro to Kosla ventures? Which is a completely random thing. Because you know, London based, I don't know anything about the VC networks, like looking at some West Coast VC. Anyway, made an introduction. Matt, Adina Techley, who's our partner here at KV, and a really, really incredible person. And ended up like having five meetings.
我发现，EF（Entrepreneur First）有一点非常突出，那就是人们要做到极度有雄心实际上比有高度雄心要容易得多。就像去追求那些“登月计划”（指非常宏大的目标），因为你只需要成功实现其中的一个“登月计划”就够了。我记得我遇见了Limbik的CEO Ross Harper，他是作为Entrepreneur First的一部分被投资的。而且他在我之前几年就已经得到了Kosla的投资。我们聊了一些事情，尽管没有深入探讨，但我还是问他能否介绍我去Kosla Ventures。这完全是个随机的请求，因为我在伦敦，根本不熟悉风投网络，更别提了解西海岸的风投了。无论如何，他还是帮我引荐了。我认识了Matt Adina Techley，他是KV（Kosla Ventures）的合伙人，非常了不起的人。最终，我们进行了五次会议。
We're in the middle of a round. So it was competitive. But we sort of went from first meeting to close very, very quickly. So you've built Oslur. But inertia is a powerful thing, especially in the healthcare industry. How has it been trying to actually, because as you said, you have it being used in doctors, offices and hospitals, I think. Was it hard to get in? And like, what is that like when you're trying to like inject something new into this system that is very stayed or stagnant? Yeah, it's a really good question. So I think there's a few elements to that. Was it hard? Yes. It was probably it remains the hardest part. The technology that we're deploying today, it's more or less exactly what we built 18 months ago. But we spent a year building product, getting compliant, figuring out a lot of the compliance workflows. I think AI is a fascinating technology, like, because it's new and old at the same time. And what I mean by that is it's new, for sure, super powerful. But it's starting to replicate situations that are old, for example, having a human assistant. And now an AI assistant. But actually lots of the older doctors recognize that there's a cognitive workflow that they already fit in, where they have a scribe, a human doctor sitting there doing their work. So by emulating that was a very sort of good way of getting into the system. So by using a chat interface, producing notes in your style, so really closely matching what you'd expect a human being to be doing and even having the interaction, which we haven't built out yet, but planning to do.
我们正处在一个融资阶段，所以竞争非常激烈。但我们从第一次会面到达成协议的整个过程非常迅速。你们已经创建了Oslur，但惯性是一股强大的力量，尤其是在医疗行业。当你试图将一个新事物引入这个非常固定或停滞的系统时，你觉得有多难？你的产品现在已经在医生办公室和医院里使用了，这是难以进入的吗？这过程是什么样的？
这是一个非常好的问题。我认为这个问题有几个方面。难吗？是的，这可能仍然是最难的部分。我们今天部署的技术大致上与18个月前我们构建的内容相同。但我们花了一年的时间来开发产品、确保合规性、理清很多合规工作流程。我认为人工智能是一项引人入胜的技术，因为它既新颖又古老。我的意思是，人工智能确实是新的，非常强大，但它开始复制一些旧的情景，例如有人类助手，现在是人工智能助手。实际上很多年长的医生认识到，这是一种他们已经适应的认知工作流程，他们习惯有人类记录员，另一个医生坐在那里做记录。因此，模仿这种情况是进入系统的很好方式。通过使用聊天界面，生成符合医生风格的笔记，非常接近于你预期人类助手会做的事情，甚至包括互动，这方面我们还没有完全开发出来，但计划要做。
And I think that's what a lot of people have realized. For example, I've seen some comments reasoning about people interacting with GPT Omni through voice. And for the first time talking to it, I think it's a really good example of how the interaction in the workflow can fundamentally change how you appreciate the technology. And it's been exactly the same mentality. Right. But again, it's really interesting. Some of the older doctors never used to type. So we're bringing them back to a world that they used to live in. This is almost like a nostalgic thing for some of them. And I think it's really interesting to be like, have you ever heard that expression? How do you get a donkey down from a minaret? Have you heard this? No. You just find the part of the donkey that really wants to get down. And I think that's a really good ethos for AI and healthcare. You just need to find the part of the system that's suffering the most. And that is workforce productivity that is burnout. So you come with solutions to fix the biggest pain point, but at the moment, the only pain point that's worth trying to fix, because everything else is meaning. And people are willing to at least give it a go.
我认为这也是很多人已经意识到的问题。例如，我看到一些评论在讨论人们通过语音与GPT Omni互动的情况。第一次与它对话时，我认为这是一个很好的例子，说明互动流程如何从根本上改变你对这项技术的看法。这种心态是完全一样的。但是，再次说明，这真的很有趣。一些老医生以前从不打字，所以我们其实是在把他们带回到他们以前熟悉的世界。这对他们来说几乎是一种怀旧的感觉。我觉得这很有意思。你听说过这个说法吗？“怎样才能把一头驴从尖塔上弄下来？”你听说过吗？没有。只需要找到驴真正想下来的那个部分。我认为这对AI和医疗行业来说是一个很好的理念。你只需要找到系统中最痛苦的部分，而那就是生产力和倦怠问题。因此，你要带来解决这些最大痛点的问题方案，因为目前只有这一问题值得尝试去解决，而其他一切都是次要的。人们愿意至少试一试。
Are you aiming to get into America? Because obviously it's a, I think healthcare, whatever the number is people throughout. It's a six-year of the economy out here. I mean, it's a very different system than the centralized NHS system, but presumably there's a lot more opportunity as well. Yeah. The answer is yes, again. It's obviously where there's a massive opportunity in me. And we're looking for US design partners right now and have in early discussions. I think there is a huge opportunity at home. It's a 160 billion pound market to really build a product end to end. That's the strategy at the moment with doctors solving the same problems. Clinically, the systems are very similar at a clinician patient level.
你打算进入美国市场吗？因为显然，美国的医疗保健系统， 无论具体数字是多少，都占了整个经济的很大一部分。我认为这和集中的 NHS 系统非常不同，但显然也有更多的机会。是的，答案是肯定的。在美国市场有巨大的机会。我们目前正在寻找美国的设计合作伙伴，并且已经开始了早期讨论。我认为在本地市场也有很大的机会，这可是一个 1600 亿英镑的市场，可以真正从头到尾地开发产品。这是目前的战略，医生们会解决相同的问题。从临床上看，医生和患者之间的系统非常相似。
The functions are the same. In fact, the child is the same epic persona, take up 100% of secondary care. Most of US over here, the ratios are slightly different, but this is exactly the same set up. But I also think that the opportunity here is interesting because you do have models that are similar to the NHS, like Kaiser, for example, value-based care systems actually do care about getting patients out in empty beds, whereas lots of other systems are incentivized the other way here. But I think it's really interesting that I don't know if I have you and your sisters, but like, I do think, and I do think, as my company or me, another company, there is real opportunity for a startup to fundamentally disrupt the US healthcare system entirely with the combination of a new healthcare model, with services, with human clinicians, and AI, because the leverage you can then achieve actually does make that cost-structural work. And that actually frees people from insurance and from pre-awth and all the awful stuff that's happened to show that.
这些功能是一样的。事实上，孩子就是同一个史诗般的人物，承担了100%的二级护理。在我们大多数人的地方，比例稍有不同，但设置完全一样。不过，我也认为这里的机会很有趣，因为你们确实有类似NHS的模式，比如凯撒医疗系统，基于价值的护理系统实际上确实注重让病人出院腾空床位，而许多其他系统在这里却有相反的激励。但我认为，尽管我不确定你和你的姐妹们，但无论是作为我的公司还是另一家公司，我确实认为，一个创业公司通过结合新的医疗模式、服务、人工医生和人工智能，完全颠覆美国医疗系统的机会是真实存在的，因为这样获得的杠杆效应确实能够使成本结构奏效。这实际上也将人们从保险、预授权和所有糟糕的事情中解放出来。
But I think incremental change is not going to work. It's just going to actually probably make the problems worse for some of the incentive system. If I speed up your clinicians and I save you time, I mean, we've had US clinicians tell us this, if I save that clinician time, they're like, no, no, I want to. I want to build for that time. Don't speed that up for me. I like that time. And that as a UK clinician, that's been a super-alien to navigate. But healthcare systems are having exactly the same problems in terms of efficiency, capability, audit reports, errors, things like that. So lastly, just kind of going back where we started this idea, if we kind of future-scape a little bit, you're designing these tools with these very rapidly developing AI models, these large language models. And the visions of people like Vinod Kosla is an AI doctor in your pocket for everyone. That could be the future. Do you think that is realistic or do you think this is, again, more of kind of like this Silicon Valley, Hocus Pocus, Hype Machine stuff that we're like, actually, that may be deliverable 20 years from now, but it's just a lot going to be a lot harder than we think.
但是我认为渐进式的改变是行不通的。这实际上可能会使某些激励机制的问题变得更糟。如果我加快你的临床医生的工作速度，为你节省时间，我们曾经有美国的临床医生告诉我们，如果我为他们节省时间，他们会说，不，不，我希望利用那个时间来做事，不要为我加速，我喜欢那段时间。作为一名英国临床医生，这种反应让我感到非常困惑。但是在效率、能力、审核报告、错误等方面，医疗系统面临着同样的问题。
最后，再回到我们之前提到的这个想法，如果我们稍微展望一下未来——你们正在使用这些迅速发展的AI模型，尤其是大型语言模型，来设计这些工具。像Vinod Kosla这样的人设想未来每个人口袋里都有一个AI医生。这可能是未来的趋势。你认为这是现实的吗，还是认为这只是硅谷的炒作，认为这可能要二十年后才能实现，而实际上会比我们想象的困难得多？
Yeah, I think I might be allergic to that word, realistic. So let's discount that for a second. But I think it's an interesting thesis to dissect. So I think it's not a pipe dream. That's number one. AI doctor in your pocket, that's today, right? You literally have GPT-0 in your pocket right now. And there's a lot of statistical evidence. I mean, you quoted one of the papers in that newspaper article showing that GPT vision, or whatever you want to call it, VEO, whatever. So I just call it four because I can't remember what it's actually called. Omni, 40. 40, OK, why do they make it say that? Is a better diagnostician than 99% of physicians based on that one study looking at New England General Medicine? And that's not to say that. I'm saying we should replace it. But I do think it's a really interesting time in medicine when the problem for all of time has always been about distribution of expertise. And now potentially we have infinite distribution of expertise. That's quite a hard concept for us to get our heads around. In that perspective, if you wanted a diagnosis and you were worried about your symptoms, I wouldn't recommend this because it's not clinically validated and I'll come on to that point in a second. But you do have a much better capability to get some pretty sound advice than you ever used to ever in the history of humankind.
嗯，我认为我可能对“现实”这个词过敏。所以我们先把它撇开一会儿。但我认为这个理论很值得深入探讨。所以我认为这不是一个白日梦。这是第一点。口袋里的AI医生，这已经是现实。你现在实际上就有GPT-0在你的口袋里。有很多统计数据为此提供了一定的证据。我指的是你在报纸文章中引用的一个研究，显示GPT视觉，或者你想叫它什么都行，VEO，随便。我就是叫它四代，因为我记不住它的具体名字。哎呀，不对，叫它40吧，40，好吧，为什么它们要这样命名呢？根据那个研究，它在诊断方面比99%的医生都要好，研究发表在《新英格兰医学杂志》上。这并不是说，我认为我们应该用它替代医生。但我确实认为在医学领域这正是一个非常有趣的时代，因为历史以来的问题一直都是如何分配专业知识。而现在我们可能拥有了无限分配专业知识的潜力。这个概念对我们来说确实很难理解。从这个角度来看，如果你想要一个诊断并对自己的症状感到担忧，我不会推荐这个方法，因为它还没有经过临床验证，我稍后会详细解释这一点。但确实相比于历史上的任何时候，你现在有了更好的能力来获得一些相当可靠的建议。
So in many sense, it's not even a dream, it's a reality. But the medical legal system, the clinical evidence, the clinical validation, all of these systems that we've built to do things safely in medicine are actually called medicine medicine and not like nutrition or wellness or something else doesn't exist yet for AI. And that's the lag. That's the bottom. So if I could prove to you that you asking one question to 100 human primary care physicians who like talk to you and have a diagnosis and examine you and I can give the same data to an AI model and the AI model is better in a clinical trial head to head, I don't know what physicians would I would accept that. But then the medical legal system doesn't have an answer for that. Who's liable if that AI is wrong, even in that one out of 10 million cases, let's say it's egregiously incorrect. So that's why we've always gone for the at least for the next five to 10 years, AI augmentation of existing human clinicians is going to be the paradigm. I really can't see because of the limitations, not the technology at all.
所以，在很多意义上，这不仅仅是一个梦想，而是现实。但是，医疗法律系统、临床证据、临床验证，这些我们为了在医学中安全行事而建立的系统，实际上是被称为医学医学的，而不是像营养或健康之类的东西，对AI来说这些系统还不存在。而这就是滞后的地方。这是底线。所以如果我可以向你证明，你向100位人类初级护理医生提一个问题，他们会与你交谈、进行诊断和检查，然后我把同样的数据交给一个AI模型，并且在临床试验中，AI模型的表现更好，我不知道医生们会不会接受这一点。但是，医疗法律系统对此没有答案。如果那AI犯了错误，即使在一千万例中的一个，这种情况下谁该负责？如果这种错误非常严重。这就是为什么我们始终认为在未来五到十年内，AI帮助现有的人类临床医生是主流。我真的看不出，因为不是技术的限制，而是其他方面的限制。
But having said that, you know, people might just get so used to the user technology that you use to use Google that these systems actually evolve and organically become part of the system and then we retroactively have to find a way to make it work as the medical professional it backwards. I mean, that's also a reality. Right. I do know, having taught myself this, this would definitely happen and it will probably happen sooner, we think, as of all these things. But it will take a lot more change than just the technology itself. And I think that's the bit that we're missing and people are kind of waking up to you like with CHI and a few of these other organizations and companies like ours, we're basically building clinical frameworks, assessment systems, evaluation. That's the bit that we're pioneering to actually evaluate the systems at scale to automate that evaluation now as well to create guardrails.
但是话说回来，你知道，人们可能会如此习惯于你用来使用谷歌的用户技术，以至于这些系统实际上会演变并自然地成为系统的一部分，然后我们再回过头来找到一种方法让它以医疗专业的身份工作。这也是现实，对吧。我确实知道，自己亲自学习过这些东西后，这种情况肯定会发生，而且它可能会比我们想象的更快发生，就像所有这些事情一样。但要做出这种变化，不仅仅是技术本身的问题。这也是我认为我们忽略的部分，但人们正在逐渐意识到，比如与CHI及其他一些机构和公司合作。我们基本上在构建临床框架、评估系统、评价机制。这部分是我们在开创性的领域，实际上是为了在大规模评估系统中实现自动化，创建防护栏。
And then actually to run trials. So we've been running trials in the UK and HS now for eight months with one of our partner site hospitals, phase one, phase two, phase three, increasing number of patients, increasing risk, but actually de-risking the accuracy. And that's just for ambience. So this isn't diagnostic support. This isn't decision making. But I think that's the right way of doing it. Setting up actual systems of evidence where you can reliably look at these models in a clinical way. Eventually, it will converge into something like a pharma model and we won't allow AI in healthcare that hasn't been through a rigorous process of testing, has to have provenance. So I think the pipe dream for technologists is that technology is the only is going to be the solution. Well, there is a hostile subware that I talked to recently that they're on paper. So what do they do? Right? Where's their AI going to live when they don't have computers? And that's the reality of healthcare. So that's the delta of societal change moving much slower than exponential technology.
然后实际上进行试验。所以我们已经在英国和HS与我们的一家合作医院进行了八个月的试验，覆盖了第一阶段、第二阶段和第三阶段，患者数量逐渐增加，风险也在增加，但实际上是在降低准确性的风险。这只是为了环境。因此，这不是诊断支持，也不是决策支持。但我认为这是正确的做法，建立实际的证据系统，可以可靠地以临床方式观察这些模型。最终，这将收敛为类似药品的模式，我们不会允许未经严格测试过程的AI进入医疗保健领域，必须具有来源可靠性。所以我认为技术人员的梦想是技术是唯一的解决方案。好吧，我最近与一家使用纸质记录的医院交谈过。那么，他们怎么办？在没有电脑的情况下，他们的AI将何处存活？这就是医疗保健的现实。所以这是社会变化的速度远远慢于指数型技术发展的差距。
At one point, I think we'd just be like, forget what the AI is saying. It's too complicated. We might just go back to just talking to each other as boring humans. Do you know what I mean? Well, that might be the answer. It just spins off and it's like, AI is over there, but it's got too much now. Yeah, exactly. Just makes some runs or something. Well, look, I really appreciate taking the time. It's fascinating. And yeah, we'll have you back on as things develop and it kind of percolates more into the system, but I think especially with given all the challenges in the NHS and how the NHS in particular is such a kind of a national, it's a national source of angst and debate constantly. It feels like things could move quite quickly there because they kind of they have to, right? Yeah, I have to. For lots of different reasons. So it'll be interesting to watch. But thank you for taking the time. I appreciate it. And good luck.
有那么一刻，我觉得我们会说，别管AI在说什么了，太复杂了。我们可能会回到仅仅作为无聊的人类互相交流。你懂我的意思吗？也许这就是答案。AI在那里运行，但太复杂了。是啊，说得对。我们就让它运行就好。好吧，非常感谢你抽出时间来讨论。这很有趣。随着事情的发展，我们会再请你回来，因为这种技术逐步渗透到系统中，特别是在面对NHS面临的各种挑战时，NHS作为一个国家级的、不断引发焦虑和辩论的机构，变化速度可能会很快，因为他们必须做到。是的，需要做到，出于多种原因。所以，这将是很有意思的一次观察。再次感谢你的时间，祝你好运。
No, it's great. Well, thanks for having me. And that is all the time we have. I want to thank Dom for making the time on his West Coast swing to sit down and chat. I want to thank you all for listening, for the ratings, for the reviews, for telling your friends and neighbors about this fantastic program. I will be writing this week a little bit off piece. So do check that out. I won't spoil it, but it's a fun one. I think especially for our UK readers, you're going to find it very, very interesting. So do check out the times.co.uk. You can find me on Twitter at Danny Fortson. That is it for me this week. Thank you so much, as always, for listening and rating and reviewing. And we'll talk to you very soon. Bye bye.
不，不需要改动，非常好。好了，非常感谢你们邀请我。而我们今天的时间也到了。我想感谢Dom在他西部行程中抽出时间来和我们聊天。我还想感谢大家的收听、评分、评论，以及向你们的朋友和邻居推荐这个精彩的节目。本周，我会写一些不太寻常的内容，所以请一定要关注。我不会提前剧透，但保证很有趣。我觉得特别是我们的英国读者会发现这篇文章非常有意思。所以请访问 times.co.uk 查阅。你们也可以在推特上找到我，用户名是 Danny Fortson。这就是我这周的内容。非常感谢你们一如既往地收听、评分和评论。我们很快再聊。再见。
Our history is a new podcast brought to you from the times, and it brings together the real life stories from our obituary's desk, which have been published for over a century. In this brand new show, we build on this legacy and explore the endlessly fascinating lives who have enriched and informed our own. Join me and our sponsor, Ancestry, as we journey through your history.
我们的历史》是一档全新播客，由《泰晤士报》推出，汇集了我们讣告部门百余年来发表的真实人生故事。在这档全新节目中，我们继承这一传统，深入探讨那些精彩的生命，他们丰富和启迪了我们的人生。请和我以及我们的赞助商 Ancestry 一起，踏上这段探索你自身历史的旅程。