夏洛特·布利斯：人工智能能否解决医学界的难题？

这非常具有挑衅性，夏洛特。首先，我们有很多共同点，比如我的书《深度医学》和你的新书。而且我们正处于一个人工智能备受质疑、遭遇强烈抵制的时期。现在你出版了一本书，听起来像是在说AI是解药。那么，请先分享一些精彩的病人经历案例来展开讨论。

This is very provocative, Charlotte. And to start off, there's a lot of common ground that we've covered in, like my book, Deep Medicine, and your new book. And we're also at a time when AI is getting lots of grief, lots of pushback, blowback, everything. And now you've come out with a book that, you know, this sounds like AI is the cure. So let's start off, if you would, you come up with really some great vignettes of patient experience.

比如詹·劳森，一位患有埃勒斯-当洛斯综合征的年轻女性。能否通过她的故事来展现当今医学存在的问题？

And one, is Jen Lawson, a young woman with Ehlers Danlos syndrome. Why don't you take us through that to kind of exemplify where the problems are in medicine today?

是的。埃里克，我不想陷入这种非此即彼的虚假二分法。这本书的标题在某种意义上确实很挑衅，但我的核心观点是：找出当前传统医学中哪些问题可能（或不可能）通过现有AI技术解决。詹·劳森作为埃勒斯-当洛斯患者，因一系列健康问题失去医保，经历严重摔伤后不得不动用养老金支付医疗费用。

Yeah. And I would say, Eric, I don't want to fall into this kind of false dichotomy of eitheror camps here. I mean, the book is a very provocative title in some sense. But basically, what I'm trying to do is say, what are the problems with current traditional medicine to which AI could potentially be a solution or might not, thinking about the status of current AI. But yeah, Jen Lawson is a patient with Ehlers Danlos syndrome who, through a series of problems with her health, loses her medical she has a very serious fall, she loses her health coverage and basically eats into her pension fund in order to cover her health costs.

这种故事并不罕见。我以她为例不仅讨论医疗财务壁垒——即使在全民医保或免费医疗体系下这些障碍仍存在——更想说明：即便获得诊疗机会（比如依赖年迈母亲跨州驾车送她就诊），患者仍面临多重阻碍。这种不便的就诊过程需要自费，因为经济型方案无法满足实际需求。

And as we know, that's not a very uncommon story. So what I do is I open it up with her to discuss not only the financial barriers to care, which sometimes even get hidden when it comes to thinking about universal healthcare, healthcare free at the point of access, so to speak, is going to be a solution to some of medicine's woes. It will be for some patients, but what I'm trying to say is there are multiple barriers, even given the fact that, say, she's able to get access to care because she has to rely on her elderly mother to drive her across state to see a specialist at one point. It's very uncomfortable. She has to pay out of pocket in order to do that because realistically, she can't rely on budget options that would be covered.

她遭遇的整个障碍链使舒适就医成为奢望。虽然远程医疗消除了部分困难（比如能靠枕头支撑舒适坐姿问诊），但除了实体障碍，她还遭遇隐性偏见——医护人员对特殊需求患者的理解不足。我借此案例展开更深层探讨。

And she's basically got a whole series of obstacles that prevent her from having comfortable access to care. Then I say, you know, with telemedicine, some of the challenges were eradicated and she was able comfortably to access some doctors, you know, with pillows behind her and all the rest of her where she can sit comfortably. But suffers from sort of not just sort of the literal barriers to care, but figurative barriers too. She discusses some of the biases that do kick in that make it more difficult to see patients with extra needs and to make them comfortable. And I use that as a vehicle and as a way to discuss.

这揭示了医护工作者在多任务处理时的挑战。简而言之，她完美诠释了'逆向护理法则'：最需要医疗资源的人，往往面临更多实体医疗服务的获取壁垒。

The challenges for doctors and nurses and other people to multitask and to do things to make but ultimately, she exemplifies, to cut a long story short, the inverse care law. That is, for the people most in need, there are more barriers to getting access to bricks and mortar healthcare.

确实。80岁母亲驾车两个半小时送她就诊的案例，凸显了医疗时效性问题。美国每年25万人死于医疗差错，80万人因此致残或死亡。当今医学的核心矛盾在于：医生不是超人，公众虽对医者怀有敬意，但患者常自我审查倾诉内容——平均18秒就会被打断陈述，这些现象都值得深思。

Yeah, no, I think it exemplifies the problem where her 80 year old mother had to drive her two and a half hours to get to the doctor, and there was very little time that she could have. And you get to the point about where medicine two hundred and fifty thousand people die each year in The United States from errors and eight hundred thousand from disability, total or deaths. And so this issue that we have today in medicine, summarized by that doctors aren't superheroes. Generally, we have deference, the public to doctors, and patients are often self centered, self censored about things that they would share. And there's only eighteen seconds before they're interrupted to tell their story, and all those sorts of things.

这就是AI所能发挥作用的大背景。现在让我们深入探讨，因为你喜欢将当今医学与AI进行并列比较，比如医生会疲倦，而AI不知疲倦。通过这种对比，如果你愿意的话

So that's the backdrop to what an AI do. And, let's let's get into that now because you do like a side by side comparison of medicine today versus AI, whereby, you know, for example, doctors get tired, AI is indefatigable. Us through that by comparison, if you

可以这么说。这本书基本上遵循传统医疗预约的流程轨迹。从你向医生描述症状开始，我思考了其中的挑战，因为有时这会被忽视。然后如你所说，我探讨了技术能提供什么替代方案。我也考虑了诊断环节。

will. So basically the book follows the trajectory of the traditional medical appointment. So from when you disclose your symptoms to a doctor, and I consider the challenges with that because sometimes that's overshadowed. And then I look at, as you say, what could technology do as a workaround for that? I also consider diagnostics.

医生们面临保持知识更新的挑战，医学领域存在各种干扰因素，比如医生判断时可能存在的差异——可能是性别、年龄等因素，甚至一天中的不同时段都可能微妙影响医生的决策方向。我思考了AI在这些方面能做什么，不仅是保持更新，还能消除部分干扰。当然，我并非完全站在技术乐观派这边——这也不是写给技术的情书，因为我们知道AI同样存在挑战。某种程度上，我可能同时惹恼了双方，但试图理解医生面临的处境。

So the challenges on doctors to keep up to date, the noise in medicine in terms of the range of, if you will, disparities in terms of how a doctor might judge you. It could be gender, it could be age, all of these things could be the time of day that can nudge doctors' decisions in different directions. And I consider what AI can do there as well, not just to keep up to date, but also in terms of eliminating some of that noise. But of course, I'm not giving the technologist it's not a love letter to technology either because we know that there's challenges there too. So in a sense, probably I'm irritating both sides in this to some extent, but trying to be sympathetic to what doctors have to do.

我还研究了同理心问题。这是个重要议题，我发现人们常有个假设——我为此做过大量调查——认为医患沟通技巧是种软技能。但心理学研究表明这并不简单。然而在调查中，包括最近用生成式AI做的调查，当我问医生哪些任务最可能/最不可能被取代时，他们始终认为最不可能被取代的就是同理心。

And then I also consider empathy. Empathy is a big one where I think sometimes there's an assumption that, look, and I've done a lot of surveys on this, when it comes to bedside manner, that that's a soft skill. And the psychological research shows that it's not particularly easy. But doctors routinely, when I've asked them in surveys, even laterally with generative AI surveys, which task do you think is most likely to be replaced or least likely to be replaced? The one they always say is the least likely to be replaced is empathy.

这个结论的认同度高达90%。自2018年以来多项调查都显示这是个相当稳定的数据。关键在于探讨AI能为这些多样化挑战提供什么解决方案。我正更专注地研究医学实践中的心理挑战和认知陷阱。

They're at nine and ten. That's a pretty robust figure, not a lot of surveys they've done since 2018. So the point is to say, what can AI do for each of these variety of challenges? And I'm trying to focus into on the more carefully on what the psychological challenges and pitfalls in medicine are.

是的。我认为你的研究视角非常有趣，因为你首先审视了当前医疗实践中的严重问题，然后探讨AI能在哪些方面提供帮助，同时也分析了AI的局限性。如你所说，这不是对AI的赞美诗。但现在很多人对AI抱有抗体。比如《华尔街日报》8月30日有篇题为《聊天机器人可以成为你医疗团队一员》的评论文章。

Yeah. No, I think it's really interesting, your approach here, because you basically review the serious problems we have in the practice of medicine today, and then you get into where AI could lend a helping hand, and you also review some of the liabilities of AI. So like you said, it's not a love letter for AI. But there's a lot of people right now who have antibodies to AI. For example, there was a Wall Street Journal, essay, on the August 30, a chatbot can be part of your medical team.

作者汤姆·罗森布拉特在文中向患者提供使用聊天机器人的建议。如你所知，全球有十亿人在使用聊天机器人，其中很多人用它来帮助自己或亲友的健康问题。你在书中也探讨了部分相关内容。

And this was by, Tom Rosenblatt in the Wall Street Journal. And he was basically giving tips to patients about how they could use chatbots. And as you know, the billion people around the world using chatbots, and of course, a lot are using it to help their health of themselves or their family members or friends. And you actually reviewed some of that in the book too.

但是当

But when

那条帖子发布时，我转发它是因为觉得有趣。很多人都在说'坚决不要'、'这太可怕了'、'我不想要鸡蛋'。你看，我们在这里面临着大量消极态度的问题，不仅来自患者和潜在患者，顺便说一句，医生群体也是如此。

that was posted, I put it up because I thought it was interesting. There's all these people saying, you know, hard pass. Oh, this is horrible. I don't I don't want an egg. So we have a lot of problems here with negativism, both from patients, prospective patients, and doctors too, by the way.

换个角度——你不是医生，也不是护士。就像你在书里说的，你在这件事上没有既得利益。

Now- different way. You're not a doctor. You're not a nurse. No. It's like you said in the book, you don't have any stake in this.

你试图成为客观、中立、批判性的观察者。想想所有这些对AI的抵制，其根源究竟是什么？

You're trying to be an objective, neutral, critique observer. Think about all this pushback about AI, and what is the basis of it?

简单说明下，我的学术背景是哲学。我的博士学位是科学哲学与心灵哲学。后来转向理论心理学，这使我对描绘医学现状产生了特殊兴趣——我觉得当前医学界缺失这种视角（当然这种缺失可以理解）。医生不可能精通所有医学领域，所以我关注的是当前医学实践中的心理陷阱。但同时作为哲学研究者，我也非常希望能保持客观立场。

So just to say, my background is philosophy. My PhD is philosophy of science and mind. I veered into theoretical psychology, which made me interested in giving a particular portrait of medicine that I felt was missing, again, for understandable reasons. Doctors can't be experts in all things medical, so it's looking at that portrait of the pitfalls, the psychological pitfalls with current medicine. So there is, but then with my philosophical hat on, I'm keen also to take an objective perspective here.

我认为当前存在一种巨大的焦虑——社交媒体等平台上的言论往往充满敌意。这种环境放大了最极端的观点，我们知道这本身就是问题。但当我与人交谈时，能感受到人们对中间立场的渴望。在AI与医疗领域，人们总是陷入两极分化：一边是恐惧与生存威胁，另一边则是幻想——要么乌托邦要么反乌托邦。人们对此产生情绪化反应，双方都感到不安，这些反应其实都有其合理性。

And I think there's a big worry current I mean, social media and so on can be very vitriolic. I mean, that sort of showcases the worst, and we know that's a problem in itself. But when I'm in conversation with people, I think there's a hunger for the middle ground. I think many people and there is this sort of, again, this bifurcation between other fear and existential threat on the one hand, and the kind of fantasy on the other, sort of utopia versus dystopia when it comes to AI and healthcare. I think there's understandable reasons why people get emotional about it and upset on both sides.

但通过大量调查研究，我也发现人们逐渐认识到可以采取务实态度——AI确实会带来某些益处。当然，任何创新都不可能解决所有问题而不引发新问题，科学史反复证明了这一点。这就是事物的本质：解决一些问题的同时，总会带来更多新挑战。

But I also hear, and again, I've been doing a lot of survey research in this, people recognise that you can take a practical perspective that there are going to be some benefits. And of course, no innovation solves all problems while not inviting new ones. We know this through the history of science. That's just the nature of things. You resolve some problems, you invite many more.

但我想说的是，考虑一下我们可能面临的一些问题和担忧，比如医疗中的歧视或偏见。我们实际上在消除医生偏见方面做得如何？我们确实进行了一些反偏见培训，人们应该监控自己的刻板印象，美国医学会也建议这样做。但效果相当有限。所以我们必须说，如果我们真正关心患者的治疗效果，以及面对面歧视对诊断准确性或患者治疗的干扰，是否有更好的解决方案？

But I suppose what I'm trying to say is take some of the considerations, concerns that we may have, here's one of them, discrimination or bias in healthcare. How good are we at actually debiasing doctors? I mean, we do sort of anti bias training and people should monitor their own stereotyping and the American Medical Association advises that. It's pretty ineffectual. So we have to say if we really care about patient outcomes and things like face to face discrimination interfering with diagnostic accuracy or the treatment of patients, is there a better workaround to this?

这就是为什么我说，实际上，消除这些工具的偏见可能比消除人的偏见更容易。否则你就给医生和菲尔博士（原谅我用这个比喻）施加了巨大压力，要求他们成为最好的自己。在这个疯狂的世界里，这非常具有挑战性。我们需要对此进行现实、务实的讨论。

And that's where I say, actually, it might be easier to debias some of these tools than it is to debias people. Or you're putting a hell of a burden on a medical doctor and a Doctor. Phil pardons being the best version of themselves. That's very challenging in the frenetic world of myth. We need a realistic, practical conversation about that.

是的，我认为你提出了一些很好的观点。我稍后想回到同理心这个话题，因为这完全出乎意料。我确实没想到。但在那之前，你提到了很多内容，当然包括卡斯帕罗夫定律——是的。

Yeah. No, I think you're making some good points there. I do want to go back to empathy in a moment because that's very much unanticipated. I certainly didn't anticipate it. But before I do that, there's a lot of what you referred to, of course, Kasparov's law from- Yes.

众所周知，卡斯帕罗夫是国际象棋世界冠军。当他输给机器时，他宣称最好的组合是人与机器。在医学领域，你可能最近看到几项研究表明，AI与医生相比——我们为此写了一篇论文，我也在《纽约时报》上发表了相关评论——AI表现得更好。

Kasparov, that as everyone knows, he was the grandmaster chess world champion. And when he lost to machine, he proclaimed that the best of all worlds is the human and the machine. And in medicine, you may have seen recently that there's been several studies where the AI and we wrote a for now, and I wrote an op ed in the New York Times about this, that the AI compared to the doctor with AI. Yep. AI was better.

是的。

Yes.

那么这与混合模式不符。你认为最终对临床医生来说，混合模式会是最佳选择吗？

So that doesn't go along with this hybrid thing. Do you think eventually to to the hybrid model being the best for clinicians?

这是个非常有趣的话题。我认为医学领域经常发生的情况是，人们总说'这里没有威胁，因为人和机器会协同工作'。这种医生与AI和谐共处的画面就像霍尔马克电影里的情节。多年来，以临床决策支持形式存在的AI效果并不理想。由于所谓的'最后一英里问题'，医生往往会推翻AI的建议。

It's such an interesting talking point. So what I think has often happened in medicine is there has been this sort of leitmotif, there's no threat here because man and machine will work together. And this is sort of Hallmark esque, Hallmark movie kind of imagery of the doctor and the AI working well together. For years, there has been AI in the form of clinical decision support, and it hasn't worked out amazingly well. Doctors tend to override because of the so called last mile problem.

如何真正实现这些工具的整合？这就像是价值六万四千美元的关键问题。这涉及AI与医生的心理博弈。卡斯帕罗夫定律很有趣，因为在他输给深蓝后提出的观点是，随后出现了国际象棋比赛中的人机混合时代——较弱的棋手搭配强大的计算机和严谨流程，当时能击败纯AI计算机。

How do you actually get these tools to integrate? That is the sort of $64,000 question. And it's the psychology of the AI and the doctor. So Kasparov's law is interesting because it was a suggestion when he lost to Deep Blue, the thought was and then you had this era of the centopters in chess matches where you had weaker humans paired with a strong computer and a strong process. And they were able to beat, at the time, AI computers.

但对医学的启示在于，这里指的是能力较弱的人类。现在有趣的是，我们该如何培训医生？因为大量研究表明，领域专家往往对算法输出更挑剔。因此需要研究人类对算法的反应——外行通常更顺从算法。

But the lesson for medicine there is it's a weaker human. Now, what's interesting is potentially how then do we train doctors? Because if you have and again, there's just a lot of research that domain experts tend to hold their noses more to algorithmic output. So there's a need here to look at that sort of siloed research on how do humans respond to algorithms. Lay people tend to defer more.

专家则表现出所谓的'算法欣赏'，但更多是'算法厌恶'。对，就是算法厌恶。正如你所说，若遵循国际象棋的类比，半人马（人机协作）时代现已终结。

Experts tend to it's called algorithmic appreciation, but experts tend to show more algorithmic aversion. Or algorithmic aversion. Yeah. Yeah. Then what you're saying as well, perhaps we are also now because certainly in the chess world, if you follow the analogy or the comparison, the era of centaurs is now over.

研究显示：医生与AI合作时，医生往往会降低AI的准确率，而单独使用AI效果更好。这引发了一些禁忌问题——比如为何还要保留这种合作模式？但我们确实需要更多此类研究。

There's no partnership that beats AI. So again, what you see from those studies is the doctor and the AI together. The doctor tends to hurt the accuracy of the AI, whereas if you leave the AI alone, it's better. And then there's some fights, I mean, they're taboo questions. I mean, why are we keeping so but I think we need many more of those studies.

这引出了关于正确合作方式的思考。你多年来针对NHS体系提出的核心问题是：我们需要什么样的培训？如何培养与AI协作的能力？因为至少中期内这将成为常态——事实上已在发生。我们在英国全科医生中做过调研。

It invites questions about what the right partnership is. And something that you've been writing about and thinking about for many years is certainly in relation to the NHS and educate, what training do we actually need? What's the right training in order to work alongside? Because that's probably what's going to happen at least in the medium term, it's already happening. I mean, just to add to that, I've done survey research in The UK among GPs.

2024年，我和团队调查了1000名全科医生：'是否将生成式AI用于临床工作？'20%肯定。2025年1月重测时升至25%。当然包括ChatGPT等商业工具。

2024, I asked a thousand GPs with my team, Are you using any generative AI tools for clinical tasks? 20% said yes. We redid the survey 2025. So January year, year apart, 25%. Now, are commercial tools like ChatCPT.

他们确实在使用，但关键如你所说：他们用得有效吗？

So they're using them, but then as you say, are they using them effectively?

这些临床医生的立场如何？他们愿意接受吗？某些任务最终是否会被AI完全取代？我是说，这里存在太多不确定性。但回到患者方面，我们看到人们因互动产生了心理健康问题。

These are the how grounded are clinicians? How are they willing to embrace? Will they ultimately be superseded for certain tasks by the AI alone? I mean, there's so many uncertainties here. But one thing we've seen going back to the patient side is people develop mental health problems from their interactions.

我们目睹过自杀案例，见过一些真正令人恐惧的现象侵入人们的思想，除了常见的幻觉、虚构记忆等担忧。对此你怎么看？

We've seen suicides. We've seen some really scary stuff that gets into the heads of people besides the usual concerns about hallucinations, confabulations, all that. What do you say about that?

是的。目前我们确实有大量相关报道。《纽约时报》上周就刊登了很多这类文章。听着，我认为这是个严重问题。

Yeah. So okay. So we do have slew of articles right now. New York Times has done a lot of these articles during the last week. Look, I think it's a serious problem.

这确实值得我们高度关注。但我也——这可能听起来又像是我那种哲学式的、退一步看全局的观点——虽然确实存在潜在危害，我们也不应过度美化当前患者无法获得心理治疗的现状，因为这场辩论更多涉及的是治疗而非药物。我还研究了心理疗法的实证基础——暂且把AI的挑战和潜在负面影响（尤其是年轻人依赖这些工具的问题）放一边——

It's something very we should be very concerned about it. But I also and this may seem, again, my sort of philosophical, let's take a step back, a big picture perspective here. While there may well be harms there, we shouldn't overly romance what is currently going on when patients lack access to therapists because this debate is often more about therapy, in fact, than it is medicine. But also something I've looked at is the evidence based for psychotherapy. Now consider just for a second if you park all of the challenges with AI and the potential negatives, especially with young people relying on these tools.

但心理治疗领域，已故的斯科特·利连菲尔德是少数离经叛道的临床心理学家兼研究者，他在2007年估计有10%的治疗会导致患者状况恶化。面对面心理治疗可能造成的伤害很少被研究。所以即使面对真人治疗师也可能产生负面结果。这点我们也需要考虑——我们总是对人类更宽容，对AI更苛刻。

But psychotherapy, the late, great Scott Lillianfeld was one of the few sort of iconoclastic clinical psychologists and researchers who estimated back in 2007 that ten percent of therapy led to worse outcomes for patients. It tends to be not that much investigated, the harms that can be done in face to face therapy with a therapist. So there can be negative outcomes even if you're seeing a human being. So let's also consider that. It's not as if here again, we have this issue of we hold the humans, we tend to hold humans to a higher account, to AI.

因此我认为还需要做大量工作来解答这些问题：与什么相比？对谁有用？是否有时对人们非常有帮助？使用者是否比我们想象的更精明？这类争论早有先例，就像当年医生们质疑患者不会使用'谷歌医生'——

So I think there's much, much more work that needs to be done in seeing, again, answering this question of compared to what, who is it useful for? Is it sometimes very helpful for people? Do people also tend to use these tools, be a bit more savvy than we're giving them credit for? Because we've heard these debates before, there's echoes with sort of the Doctor. Google stuff where health professionals came in and sort of said, The doctors, patients won't know how to use Google.

研究显示人们其实很擅长利用这些工具，包括所谓的'数字移民'。但我确实认为AI可能对部分人造成伤害，不过我们需要全面考量。这里还有更多值得深思的问题。

Studies show that actually people are pretty savvy about it, even so called digital immigrants. But there's also then the issue I do think that they are probably causing harm to some people. But I think we've got to consider it in the round. There's much more to think about here.

所以我认为您要表达的观点，Bhatt博士，就像您在书中提到的，我们必须从更广泛的利弊比来看待这个问题。例如，您刚才讨论到有些人可能会受到伤害——

So I think the point you're getting at, and you did in the book, Doctor. Bhatt, is that we have to look at it from the broader benefit versus harm ratio. And now, for example, you just discussed where some people could be hurt-

是的，有可能。

Yep. Could be

但总体利益更大。同样，回到我们之前讨论的医疗错误问题，人为医疗错误很多，而我们往往不愿承认这点。医疗...不，确实如此。所以，是的，AI也会犯错，但比例是否更小？

overriding benefit. But also, for example, going back to what we talked about earlier with medical errors, there's a lot of human medical errors. And we don't tend to acknowledge that. Medical- No. Doesn't And so, yeah, AI will make errors too, but is it a fraction?

是的，这个比较标准非常关键——

Yes. This comparator, is a really important-

我认为这至关重要。

I think it's critical.

你明白的。但我觉得人们应该记住这点。现在说说同理心。

That you get. But I think people should keep that in mind. Now empathy.

没错。

Yeah.

惊喜毛巾。因为当我们思考同理心时，机器并不具备这种能力。好吧，确实没有。但它们可以传递同理心，因为它们接受了互联网、维基百科以及所有其他文字内容的训练，包括成千上万本书籍等等。

The surprise towel. Because when, we thought about empathy, machines don't have it. Okay. No. But they can channel empathy because they're trained by everything that's on the internet and Wikipedia and every other thing that's been written, hundreds of thousands of books and whatnot.

事实证明，患者至少能理解AI对其病情的同理心。正如你所指出的，现在有几项研究显示，无论是医生还是患者评分，AI表现出的同理心都比医生更令人印象深刻。这在某种程度上不符合我们的预期模型。当然，问题是：这是真实的吗？你认为该如何解释这种反复出现的观察结果？

So it turns out that the patient, can understand at least the sense that the AI is empathetic about their medical condition. And as you point out, and there's now several studies that show the same, is that when rated by doctors, when rated by patients, the empathy here is much more impressive from the AI than by the doctors. That doesn't fit with our model in a way. And, of course, what, is it real? Is, is it, What do you think is the explanation for this repetitive observation?

这个话题太有意思了。很高兴我们能讨论同理心。关于这点我有几点想法：首先，正如我所说，同理心是件很难把握的事情。而且同理心也并非人们吹嘘的那么完美。

So it's just such an interest. I'm glad that we're talking about empathy. So a few thoughts that I have on this. First of all, empathy is, as I say, it's a very hard thing to get right. And empathy isn't all it's cracked up to be either.

心理学家、神经科学家和哲学家们其实已经研究这个问题多年。同理心往往带有偏见——我们更容易对与自己相似的人产生共情。而且我们对能真正产生同理心的人数存在生理限制，那种全方位的同理心要求我们感受他人所感，理解他们的生活境遇等等。所以关键问题在于，埃里克，我们如何定义同理心？

So psychologists and neuroscientists and philosophers have actually written about this for a number of years. Empathy tends to be biased. So we tend to be more empathetic to people who are like us. There's also actual physical constraints on the number of people we can be genuinely empathetic to, sort of in the kind of bells and whistles empathy, where we feel what they're feeling, we understand what's going on in their lives and all the rest of it. So the key issue I'd say, Eric, is what do we mean by empathy?

因为我们无法衡量未定义的事物。同理心通常包含几个层面：可能是捕捉他人情绪的情感共鸣（情绪同理心）。像丽塔·奇伦等医学教育者认为，医生应该感受患者的感受。但还存在另一种同理心——认知同理心。

Because we can't measure what we haven't first defined. And empathy tends to mean a couple of things. It can be that kind of catching the emotions of someone, so emotional empathy. Now, some medical educators say, Rita Chiron is one of them, that you should feel something of what doctors should feel what patients are feeling. But there are other kinds of empathy, and there's what's called cognitive empathy.

这种同理心仅需理解患者的情绪状态。但若不以亲社会或同情的方式行动，这种理解就毫无意义。所以真正的同理心体现在特定行为上。即便我们暂且搁置偏见、局限性以及同理心带来的职业倦怠，当你询问患者对医生的期待时，他们往往并不需要情绪共鸣——我做过相关小型研究。

So that is just grasping the emotional state of the patient. And then there's no point in sort of grasping it if you're not acting in a pro social or sympathetic way. So it's actually behaving in a particular way. Now, if we assume, even if we sort of park the bias and all the limitations and the burnout that comes with empathy, Actually, you ask patients what they want from doctors, they tend not to want the emotional empathy. I mean, I've done a small study on this.

那是个实验性研究，存在各种局限，但其他调查也佐证了这点。不过这些研究都很有限，因为人们很少质疑医学中的同理心概念。患者并不需要医生成为同病相怜的伙伴，陪着他们一起痛苦哭泣。他们需要的是专业素养。

It was an experimental setup. It has various limitations, but there's other survey research to support this. But again, it's very restricted because people don't often interrogate this idea of empathy in medicine. But they don't want their doctor to be a sort of fellow tragedy and then they're suffering, weeping with them or feeling. They want them to be professional.

但这样一来，他们需要的是认知共情——既要理解你的处境，又要表现出对你的关怀，并以特定方式行事。这就引出了一个问题：AI能做到这些吗？它无需真正感受情绪，但研究表明，AI确实能识别情感状态，并表现出某种看似同情的行为。有趣的是，这些调查采用盲测方式，人们认为聊天机器人等工具显得更具同理心。但当揭开盲测面纱后，数据显示AI不仅超越人类，而且对患者说出不当言论的频率也更低。

But that then, so they want cognitive empathy, grasping what you're going through, but also showing compassion for you, behaving a particular way. But that enters the door then to, can AI do those things? And it doesn't have to feel anything, but it can, and the studies are showing, it can identify emotional state and it can actually display a certain amount of what looks like compassion. Here's the interesting thing, those surveys were blinded, where people said the chatty BT or different tools look to be much more empathetic. But when you actually unblind and you say, well, AI surpasses humans, And also says fewer inappropriate things as well to patients.

当揭开盲测结果时，数据就会下降。南加州大学的研究显示，真相揭晓后，两者的水平其实相当。从偏好角度看，人类医生更有个性但共情表现不稳定，而AI则更为圆滑。这引发了更深层的问题：大规模应用时，我们能否实现这种效果？人们会习惯AI并以特定视角看待它吗？

When you unblind it, that drops. So there's University of Southern California study, I'd see when you do the reveal, it's about the same level. Now, of preference, humans who are a bit more quirky and they're inconsistent in their empathy versus the more slick AI. But that introduces the issue of, you know, at scale, can we do this? And will people actually get used to AI and they see it as in a particular way?

此外，我认为这又陷入了虚假二分法。不一定要让医生 multitask 兼顾所有职责（包括共情）。没人会花25万美元去医学院专门学习如何共情。这里存在许多问题：我们究竟希望谁来承担共情职能？这些都是重大命题。

But also, think there's this, again, the false dichotomy. It doesn't have to be a multitasking doctor who has to deliver all of these things, including empathy. I don't think anyone spends a quarter of a million dollars to become an empathiser, to go through medical school. So there's a lot of issues there with who do we want to empathize? Those are big questions.

确实。这甚至还没涉及人类肢体接触带来的独特联结——那种通过阅读聊天机器人输出永远无法获得的体验。但这是个重要议题，因为甚至有人提议（不知你如何看待）让AI来培训医生的共情能力。

Yeah. No, that doesn't even get into the human touch story about when you're, you know, human to human bond, which is a very different what you're ever gonna get from reading the chatbots output. But, you know, this this is a big topic because it's even been proposed. I don't know what you think about it. That doc coached for empathy by AI.

具体来说，就是通过AI复盘医患对话并指出：'琼斯医生，为什么你没询问患者那件事？为何在这个话题打断患者？'本质上是用AI来指导临床医生提升共情表现——这构想是切实可行还是纯属天方夜谭？

That is, where you go through the conversation and you say, the AI reviews it and say, Well, Doctor. Jones, why didn't you ask the patient about that? Why did you interrupt the patient so about this? And basically coaching clinicians to Is be better realistic or is that just crazy?

某种程度上，我认为大语言模型就像巨型 plagiarism 机器，它们站在前人的肩膀上。因此从延伸意义来看，其意图可被视为传递 compassion 的意图——虽然这个视角拉得有点远。我在书中详细探讨过这个观点。

I think it's In some sense, large language models are just one giant plagiarism machine. They're standing on the shoulders of previous human beings. So in some sense, it's sort of by extension, the intention could be viewed as the intention to be compassionate. I mean, that's taking quite a long lens to it. But I discuss that in the book.

确实已有培训项目开始运用这些工具。我认为只要对患者有益、能提升 compassion 传递效率的举措，用这些标准来衡量，都应该是好事。想想那些医学院的年轻人，他们可能缺乏面对死亡、慢性病、成瘾等问题的 life experience。就这点而言，如果这些工具有效——且我认为除了正规医学教育项目（我知道有些教育者已在实践），在电子病历公开化的时代，医生们撰写患者会阅读的病历时，这些工具最主要的用途或许就是帮助医生用通俗易懂且富有同理心的方式书写记录。

And yes, there are programmes where people are using these tools within training. And I think that whatever is in the benefit of patients and can improve delivery of compassion, put it in those sorts of terms, I think it's got to be a good thing. You know, if you consider younger people who are going through medical school, they may not have an awful lot of life experience with death or chronic illness or addiction or all kinds of things to which I say if these tools are useful, and I suspect aside from formalized within medical education training programs, which I know some are doing, some educators are doing, but from my service, I suspect that's actually probably in the era of online record access. I suspect that's a chief use when it comes to writing documentation that doctors know patients will read, writing it in an accessible and an empathetic way.

没错。贝里斯教授，或者我该称呼你夏洛特，这本伟大的著作《博特医生》对人工智能如何引领医学变革持非常乐观的态度。想想我们今天早上的对话，或者对你来说是欧洲的傍晚，这一点很明显。你对此有着非常平衡、深思熟虑的观点。

Right. Well, professor Belize, I should say Charlotte, the great book, Doctor. Bodt, it's a very optimistic view of how AI can lead to transformation of better medicine. Think during our conversation this morning, or this evening for you in Europe, This is clear. You have a very balanced, very thoughtful view about this.

这是一项正在进行的工作。生成式人工智能，甚至大型语言推理模型，仍处于非常早期的阶段。我们还在最开始的阶段。所以也许我们是在试图推测可能出现的问题和发展方向。但我非常赞赏你能整合大量研究完成整本书，解释了许多其他研究，我们将拭目以待其发展。

This is a work in progress. It's still very early in this generative AI, even large language reasoning models. We're in the earliest stages. So perhaps we're trying to speculate what are the issues, where it can go. But I give you a lot of credit for putting together a whole book with lots of research, explaining lots of other research, and we'll see how it plays out.

但显然，目前这个话题极具争议性。正如你所说，保持中立很好，但存在很多极端观点。你知道，我倾向于从长远角度保持乐观。是的，我想你也是，对吧？

But obviously, it's highly controversial right now. There's people, as you say, it's nice to be in the middle, but there's a lot of ends. You know, I tend to be optimistic in the long view. Yeah. Think you are as well, right?

这绝对是我的立场。我的观点是，现在的技术将是历史上最糟糕的状态。我当然不认为我们已经到达终点。但我在书的结尾也提到，我正在考虑医生在传统诊疗中面临的心理挑战，但这只是更广泛讨论的一部分。这些并非唯一的伦理问题。

That's definitely my angle. I mean, I take the view that the technology now is the worst it's ever going to be. I certainly don't think we're there yet. But I also close the book by saying, Look, I'm considering certain psychological challenges that doctors face in patients with a traditional visit, but there's a much wider conversation. Those aren't the only ethical concerns.

关键在于：我们希望由谁来运营我们的医疗体系？隐私和不平等问题会怎样？如果患者无力支付药物费用或生活在非理性的环境中，诊断准确性又有什么意义？因此，我认为作为社会整体，我们需要进行公民对话来讨论这些重大问题，这就是为什么我们不应该推迟这些讨论，为什么现在是个好机会。我们必须持续探讨这些问题。

It's about who do we want to run our health systems? What happens to privacy, inequality? What's the point in having diagnostic accuracy if you don't come pay for medications or have sanity you're not living in sanity conditions. So there are enormous issues that I think as a society and civic conversations that we need to have, which is why we shouldn't postpone those and why this is a nice opportunity. We have to keep talking about them.

非常感谢你今天参与讨论。很抱歉我们最初遇到了一些技术问题。与患者建立联系确实非常有趣且很有价值。你把握住了围绕人工智能与未来医学融合的许多关键问题。我知道你会继续关注这个领域，而且这只是你的第二本书。

Well, thank you so much for joining today. I'm sorry that we had problems initially. Connecting with the patient has been really very interesting, very worthwhile. You got your pulse on a lot of the key issues surrounding the, integration of AI in the future of medicine. I hope you'll I know you'll stay on it, and I know this is just your second book.

第一本是《...》，我相信你未来还会写出更多著作。非常感谢你的分享，祝你在...

The first one was And in I'm sure there's many more books in store for you. So thanks so much for, on ground

我非常感谢这次机会。谢谢你，埃里克。很高兴见面。谢谢。

I really appreciate the opportunity. Thank you, Eric. It's great to meet. Thank you.

也感谢今天大家的参与。再次为最初连接时遇到的问题表示歉意。我们期待听到您的意见，并将存档以便大家随时查看。谢谢各位。

And thanks for everyone joining today. And sorry again for the problems that we had, in initially connecting. We look forward to your comments, and we'll have this as an archive so everyone could see at their convenience. Thanks, everybody.