Special Edition: How AI is transforming our industries
发布时间 2022-12-16 14:53:00 来源
Since 2012, when Google trained a system to recognize cats in YouTube videos, artificial intelligence has come a long way. Today, AI systems are capable of performing a wide range of tasks, from analysing medical images to translating languages and driving cars. On this special edition of short briefings of long-term thinking, we will be exploring the topic of AI in more depth. We'll be hearing from two of our investment managers who will share their insights and experiences with investing in companies using AI to drive innovation and solve complex problems.
自从谷歌在 2012 年通过训练系统在 YouTube 视频中识别猫后，人工智能已经取得了长足的进步。如今，AI 系统能够执行广泛的任务，从分析医学影像到翻译语言和驾驶汽车。在这个特别版的“长期思考简报”中，我们将更深入地探讨 AI 这个话题。我们将听取两位投资经理的见解和经验，分享他们投资于使用 AI 推动创新和解决复杂问题的公司的心得。
Do you notice anything different about that introduction? A change in voice, a change in pace maybe? A new author? Now you're getting closer. It wasn't fact written by an AI, specifically chatGPT. It's an online tool that writes all manner of text from essays to scripts to computer code. You can give it a brief description of what you want, and within seconds, it produces what might have taken you minutes or hours to craft. The platform was developed by the US Research Lab OpenAI, in which Microsoft is a major investor.
你有注意到那个介绍有什么不同吗？声音、节奏变了，或者是新的作者？现在你已经接近答案了。这段文字并不是由人类写的，而是AI——chatGPT所创作的。它是一个制作各种文本的在线工具，包括文章、剧本和计算机代码。你只需要简要描述你想要的内容，几秒钟之后，它会给你一个和你几分钟甚至几小时才能写出的结果。这个平台是由美国的研究实验室OpenAI开发的，微软是其主要投资人之一。
Over the next 25 minutes, you'll hear more about other innovations in the field from Kirstie Gibson, co-manager of Bayley Gifford US Growth Trust and the American Fund and Jilla Angelese, joint manager of the Health Innovation Fund. But before we start some important information, please remember that as with all investments, your capsules at risk and your income is not guaranteed.
在接下来的25分钟内，你将从Bayley Gifford美国增长信托和美国基金的联合经理Kirstie Gibson以及Health Innovation Fund的联合经理Jilla Angelese那里了解更多该领域的创新。但在我们开始之前，有一些重要信息需要记住，就像所有投资一样，您的资金是有风险的，您的收入不被保证。
I'm Malcolm Borthwick, managing editor at Bayley Gifford, and as mentioned earlier by my AI assistant, this is a special episode. We recorded the following conversation at our recent ideas 22 conference.
我是贝利吉福德出版社的主编马尔科姆·博思威克。正如我的AI助手先前提到的那样，这是一期特别节目。我们在最近的“思想22”会议上录制了以下对话。
I started by asking Kirstie to explain what artificial intelligence means. Artificial intelligence is a technology that enables a machine to simulate human behavior. It's like an artificial brain, but it's a super brain. And then we have terms like machine learning, which is a subset of artificial intelligence. And what that is looking to do is to automatically enable a machine to learn from past data without being explicitly programmed to do so.
我先请Kirstie解释一下人工智能的意思。人工智能是一种技术，可以让机器模拟人类的行为。它就像一个人造大脑，但是它是一个超级大脑。然后我们有诸如机器学习之类的术语，它是人工智能的一个子集。它的目的是自动让机器从过去的数据中学习，而不需要明确编程来做到这一点。
So with AI, we're looking for an intelligence system to replicate a specific task as a human would. Whereas with machine learning, what we're doing is to teach a machine using past data to perform a particular task and to do that with a high degree of accuracy. So that might be something like, can you recognize from these millions of pictures that you presented with which of these moles of cancerous and which are not?
因此，AI技术旨在模拟人类执行具体任务的智能系统。而机器学习则是利用过去的数据来训练机器执行特定任务，并在高度准确性的基础上完成。举个例子，假设我们需要识别这几百万张图片中哪些是恶性痣，哪些不是，就可以使用机器学习技术。
And then just to add confusion to all these new terms, machine learning is how an AI learns. So machine learning is how an AI develops its intelligence. And then we have subsets of machine learning, things like neural networks and deep learning. And these are about how that learning is structured. So it's about models are given more freedom than supervised less in order to learn.
然后，为了让所有这些新术语更加复杂，机器学习就是AI学习的方式。因此，机器学习是AI如何发展其智能的方式。然后，我们有机器学习的子集，例如神经网络和深度学习。这些都与学习结构有关，模型在学习过程中被赋予更多自由度，以便更好地学习，而不是当受监督的封闭式。
So I like to think of artificial intelligence as a little bit of a spectrum. At the one end you have machine learning, a kind of bare bones real data crunching end of the machine learning spectrum, is there a cat in this picture or is there not? And then at the other end of the spectrum you have the kind of general AI, which would be if you are presenting this AI with an unfamiliar task, are they able to start generating a solution to that?
我认为人工智能可以看做是一个光谱。在光谱的一端，有机器学习，这是一种基于真实数据进行简单处理的方法，例如在一幅图片中是否有猫等等。而在另一端，有通用型的人工智能，这种人工智能可以解决面对陌生任务时，是否能开始产生解决方案。
So I think, and I'm no expert in AI, but I think a really useful analogy that's helped me to bring all those pieces of that jigsaw together is to think of a toddler. And if you think about how a toddler learns, they start off learning through being taught, through practicing, and through receiving feedback. And that's like that machine learning phase. Then you have the fact that the toddlers then take that information and they start experimenting, maybe getting themselves into a little bit of trouble here and there. And that's like that neural network phase. And then finally, they start teaching themselves.
我认为，在人工智能方面并非专家，但我想分享一个能够帮助我将各种拼图组合起来的有用比喻，就是想象一下一个蹒跚学步的孩子。例如，当你观察一个孩子学习时，他们会通过被教育、练习和反馈来进行学习。这就像是机器学习的阶段。随后，孩子们会运用所学的知识进行试验，有时也会遇到一些麻烦。这就像是神经网络的阶段。最终，孩子们开始自己教育自己。
And that's because they develop the necessary algorithms in order to do so. They have the algorithms in their mind to be able to do that. And that's kind of like that general AI phase. The main difference, I guess, being between a general AI and a toddler, the sheer quantity of data that they're able to consume, and also the speed at which they learn is quite different. So I think what that serves to illustrate is that when I conceptualized AI at the beginning there, it sounds like something that's in the future. But the reality is it is here and now with those sort of machine learning and neural network phases, whilst they are still early in their opportunity, they are much more mature than that general AI, which still remains an opportunity for the future.
这是因为它们开发了所需的算法。他们心中有这些算法，以便做到这一点。这有点像一般的AI阶段。我认为，一般AI和幼儿之间的主要区别在于他们能够消耗的数据量以及学习的速度非常不同。因此，我认为这说明了当我一开始构思AI时，它听起来像是未来的事情。但事实上，它现在已经存在。那些机器学习和神经网络的阶段虽然仍处于早期机会，但它们比一般的AI更成熟，一般的AI仍然是未来的机遇。
I love the way that you describe the different ends of the spectrum with AI. It's also, I think, fair to say, there are a lot of different opinions on that spectrum about artificial intelligence. And it was interesting, chatting to friends and colleagues just in the run-up to this conference because it is something that devised opinion. But I think that's quite natural in the sense that new technology often both fascinates and frightens people. Is that just human nature, Julia?
我喜欢你用人工智能描述光谱不同端的方式，我认为可以这样说，关于人工智能的光谱有很多不同的观点。在举行这次会议之前，和朋友们、同事们聊天时，我觉得这很有趣。因为这是一种引起争议的主题。但我认为这是很自然的现象，因为新技术往往既让人着迷，又让人害怕。这就是人类的本性吗，朱莉娅？
No, absolutely. And despite that, it's so easy, the way we kind of wish explain what AI is, but it is very complex technology. And anything complex tends to frighten humankind, and it's very understandable. If you think about AI, it's probably even most frightened technology out there. First of all, it's sort of invisible. We don't even realize it's already here. And it's kind of, it works, it builds on semiconductor chips. So it's a software that needs semiconductor chips to run on. And we can almost think about it as a living organism. So they send me a cellic, a cellic on its its body and the small transistors, it's its beating hearts, many, many hearts that make it so efficient. But what is even more powerful is that it's self-learning. But also sometimes we don't really even understand how AI actually comes to certain conclusions and decisions. And that means we're losing control of that technology.
不，绝对不是。尽管如此，我们很容易希望解释AI是什么，但它是非常复杂的技术。任何复杂的东西都会让人类感到害怕，这是非常可以理解的。如果你想想AI，它可能是最令人害怕的技术。首先，它是隐形的。我们甚至不知道它已经存在了。它是通过半导体芯片来工作的软件。我们可以把它想象成一个生命体。它有一个细胞，身体和小晶体管的心脏，经过自我学习后更加强大。但更有力的是它是自我学习的。但有时我们甚至不知道AI如何得出某些结论和决策。这意味着我们失去了对这项技术的控制。
And when we're losing control, we naturally feel very uncomfortable. So in many ways, I'm trying to say, yes, it could be scary and you know, but at the same time, you know, what is actually way out of scare and being frightened is actually knowledge. It's information. It's trying to understand what technology is, what potential it is for humanity, and what potential dangers and abuse of the technology. And in that context, I believe that regulation is going to play a role there to help us normal human beings. But it's also not just a say up to regulators because they can overregulate and they actually don't understand the technology. But it's also up to the companies who are developing technologies. So we call it kind of bottom up ethical practices on the company levels because they do understand all the details and how they develop technologies there responsible of doing it in an ethical way. So it doesn't get abused. And actually, and there are the examples of the companies that actually trying to really enforce those ethical practices.
当我们失去控制时，自然而然会感到非常不舒服。因此，在许多方面上，我试图表达的是，是的，这可能是可怕的，但同时，您知道，真正摆脱恐惧和惊吓的方法实际上是知识。这是信息。尝试了解技术是什么，它对人类的潜在影响，以及技术潜在的危险和滥用。在这种情况下，我认为监管将在其中发挥作用，以帮助我们这些普通人。但这也不仅仅是监管者的说法，因为他们可能过度监管并且实际上不了解技术。但这也取决于正在开发技术的公司。因此，我们称其为公司层面的自下而上的道德实践，因为他们确实了解所有细节以及如何以道德方式开发技术。所以它不会被滥用。实际上，有一些公司正在尝试强制实施这些道德实践。
For example, I came across a company that develops emotionally AI. And immediately, you know, when you think about emotional AI, our fantasy can go as very quickly into how it can be used for the good purposes, but also can be abused in surveillance, for example. And this company has been approached by different governments for that sort of application. And I just said no. It's not what they want to develop this technology for. And actually, they're also their bonus scheme for their developers is linked to how they can actually remove the bias out of data. So it is kind of, you know, so we already see companies are trying to do it responsibly and also try to strengthen strong signals to the society out there. And we just need more of those. And it's emotionally AI, something that simulates our emotions, is it?
例如，我发现了一家开发情感人工智能技术的公司。立刻，你知道，当你思考情感人工智能时，我们的想象力可以很快地想到它可以用于好的目的，但也可能被滥用于监视等方面。这家公司已经被不同的政府联系，以进行这种应用。但我只是说不。这不是他们想开发这项技术的原因。实际上，他们的开发者的奖励计划也与他们如何消除数据偏见相关联。所以我们已经看到一些公司在尝试负责任地做这件事，同时也试图向外界发出积极信号。我们需要更多这样的公司。那么情感人工智能是模拟我们的情感吗？
Yes. So it's a machine that actually can start reading how we feel. So it's based on our language and our facial expression and our body language. So it's still a minefield, but they made really incredible progress.
是的，这是一台可以开始读取我们感受的机器。它基于我们的语言、面部表情和身体语言。虽然仍然是一个难题，但他们已经取得了非常惊人的进展。
Kirstie, a lot of AI is already here. But what's the scale of the opportunity for investors with artificial intelligence? I think that's a really interesting question, but it's a really difficult question to answer. So I think I probably approached that from kind of two two sort of directions from a kind of theoretical perspective and then hopefully we can move on and discuss some company specifics. But so I think to start with, it's the recognition that machine learning and artificial intelligence have been in existence for the best part of 60 years.
Kirstie，很多人工智能已经存在了。但是对于投资者来说，人工智能的机会规模是多少？我认为这是一个非常有趣的问题，但是很难回答。因此，我认为我可能会从两个方面来回答：从理论角度来看，然后我们可以进一步讨论某些公司的具体情况。但首先要认识到的是，机器学习和人工智能已经存在了将近60年。
But what's really changed over the past sort of decade or so has been this like massive explosion in the sheer quantity of data that we have around us. And that's been enabled by things like sensors, but it's also been enabled by computing power, the chips that we have that are able to generate data to collect that data and our ability to store it through things like cloud computing.
过去的十年左右发生了一个巨大的变化，那就是我们周围数据量的极速膨胀。这得益于传感器等器件，同时也受到了计算能力的促进，我们拥有能够生成数据、收集数据并通过云计算等方式存储数据的芯片。
And what's most interesting for me is that data explosion has not been specific to just one or two industry verticals. It's not just being e-commerce. It's not just being advertising or social media. It's across the different spectrum. It includes manufacturing companies that have sensors that are monitoring how they're manufacturing, monitoring things to try and preempt whether there's going to be a fault.
对于我来说，最有趣的是数据爆炸并不仅仅局限于某一或两个行业垂直领域。它不仅仅出现在电子商务行业，也不仅仅出现在广告或社交媒体领域，而是存在于不同领域之中。包括制造企业，它们有传感器来监测制造过程中的情况，以提前预防是否会出现故障。
So that data explosion that is vast and it's been across industries. And I think that is a good starting point on how to think about the scale of the opportunity. It is potentially vast as well.
这是一个数据爆炸的时代，其范围已经跨越了各行各业。我认为这是考虑机会规模的一个良好起点。这也可能是一个潜在的巨大机会。
I think the other way to potentially think about it is I read a book recently called Competing in the Age of AI. And it talks about this concept of an exponential system coming into contact with a saturated one. And it's this idea that the more people embrace data and machine learning and artificial intelligence, the more others are going to have to do so because you simply cannot compete with a system that is moving that fast. Any industry that doesn't sort of embrace what is currently happening is going to be left behind.
我认为另一种潜在的思考方式是最近我读了一本书，叫做《在人工智能时代竞争》。它讨论了一个指数系统与饱和系统接触的概念。这个想法是，越多的人拥抱数据、机器学习和人工智能，其他人就越必须这样做，因为你根本无法与这样快速移动的系统竞争。任何不能接受当前正在发生的事情的行业都将被落下。
And I think part of this comes from the fact that AI breaks down silos within businesses. So it used to be that you'd have marketing data and product data and sales data and HR data. And ultimately in order to generate insights that potentially humans are not aware of, we have to bring all that data together. We bring all those data sets together, store them in a sort of cloud offering. And we can then analyse them and we can run various models. So that's breaking down the silos within industries.
我认为这部分是因为人工智能打破了企业内的隔离现象。以前，营销数据、产品数据、销售数据和人力资源数据都是分开的。但最终为了产生潜在的人类无法意识到的洞见，我们必须把所有数据汇聚在一起。我们将所有数据集结在一起，存储在云平台上。然后我们可以对它们进行分析和运行各种模型。这就是打破行业隔离的表现。
But then you're also seeing the breaking down of traditional industry verticals themselves. And we saw this with the emergence of something like Alibaba, the e-commerce giant in China, moving into what has traditionally been a sort of very high barrier to entry market, which is banking. And they could do that, not just because they wanted to and they needed to, but they had the data. They had the knowledge base to move into that different industry vertical.
然而，您还可以看到传统行业垂直本身的瓦解。正如我们所看到的，像中国的电子商务巨头阿里巴巴进军传统上一直是高门槛入市的银行业市场。他们之所以能够这样做，不仅是因为他们想要和需要这样做，而且因为他们有数据和知识库，可以进入这个不同的行业垂直领域。
And I think that speaks to just a broader and more interesting point, which is when you see a new technology, if that's the right word, to describe artificial intelligence, emerge. It's not just about replacement. It's also about unlocking new opportunities and new ways of doing things. And that is also suggested to me of the potential scale of opportunity that's here.
我认为这反映了一个更广泛、更有趣的观点，那就是当你看到一项新技术，如果用人工智能来描述的话，它不仅仅是用来取代旧有技术，也是为了开启新的机会和新的做事方式。这也向我展示了这里的潜在机会规模。
Yeah, and that data is really significant, as you're saying. Jula, let's serve your area of focus in the health innovation strategy is healthcare, clearly. Give me some examples of where you're seeing artificial intelligence being used in healthcare.
是的，正如你所说，那些数据真的很重要。Jula，你关注的健康创新战略中，关注的领域显然是医疗保健。能否给我们举一些例子，说明你看到人工智能在医疗保健中的应用呢？
Yeah, no, I just agree with Scarce, so much when she talked about big data, you know, that converges of data and AI and what the difference it can make for different industries. And I think healthcare is probably one which is really ripe for massive change. And also, if you think about the human biology, it's probably the most complex system out there.
是的，我非常赞同Scarce所说的关于大数据的观点，你知道，大数据和人工智能的结合以及它对不同行业的影响。我认为医疗保健可能是非常适合大规模变革的行业之一。而且，如果你考虑人类生物学，它可能是最复杂的系统。
And the way we started in the past, it was, as you say, silers, because we haven't been able first to gather relevant data to different parts of biology, but also integrated in a thoughtful way. And first time actually in human history, we are actually in a position to do it. And the reason why we can do it first, we can gather all the relevant data to human biology like genomics, proteomics. It's microbiome, it's bugs that live in our god. So we can pull all these data, and it's just trillions of data points. It's not just, you know, so it's a massive complexity. So I call biology actually a problem of large numbers, massive numbers.
过去，我们开始研究生物学的方式，正如你所说的那样，是比较零散的，因为我们无法首先收集到不同生物学领域的相关数据，也无法以一种深思熟虑的方式将它们整合起来。而现在，人类历史上第一次，我们有了真正的能力去做到这一点。我们之所以能够首先做到这一点，是因为我们可以收集到与人类生物学相关的所有数据，如基因组学、蛋白质组学、微生物组和生活在我们体内的细菌等。因此，我们可以获取所有这些数据，这些数据点不仅是数以万亿计的，而且还具有极其复杂的性质。因此，我认为生物学实际上是一个大量数据的问题，是一种大规模的数字问题。
And then when you combine it with artificial intelligence, suddenly you have an opportunity to unlock this complexity. And there were healthcare is actually going through the massive transformation. So first, this technology is used to understand the biology. And when we start understanding the human biology, we can start studying diseases and what actually leads to the underlying sources of diseases. If we understand that, we can start developing diagnostic tools at very, very much press zones. And we can start diagnosing diseases much earlier in the evolution. But beyond that, we can also develop proper targets for diseases, so those much more precise targets. And also, we can develop medicines from bottom up, designing small molecular drugs or like messenger RNA by actually applying artificial intelligence.
当您将人工智能与此结合时，就有机会解锁这种复杂性。 事实上，医疗保健正在经历巨大变革。 首先，这种技术用于理解生物学。 当我们开始理解人类生物学时，我们可以开始研究疾病及其潜在疾病源。 如果我们理解了这一点，我们就可以开始以非常低的价格开发诊断工具。 并且我们可以在疾病演变的早期开始诊断疾病。 但是除此之外，我们还可以根据这个来为疾病制定正确的目标，因此这些目标更加精确。 此外，我们还可以通过应用人工智能设计小分子药物或信使RNA，从底层开始开发药品。
And if just to provide two examples of some of the companies we are really excited about, like Moderna. So many people are familiar with Moderna, because we know what messenger RNA is already and how that's impacted our ability to fight against COVID-19.
如果我只是想提供两个我们非常兴奋的公司的例子，比如现代。现代公司是众所周知的，因为我们已经知道什么是信使RNA以及它如何影响我们抗击COVID-19的能力。
But what many people don't know that actually know that to design the messenger and Avaxin AI also is used to actually design the sequence because the way you put the sequence together for the spike protein, it would impact how it's expressing your body. In quantities, so to make it more effective, you know, to reduce the immune response. So it's actually very powerful technology just in the design stage.
但许多人不知道的是，实际上在设计信使和疫苗时，也使用了人工智能来设计序列，因为将尖刺蛋白序列组合在一起的方式，会影响它在体内的表达量，进而也影响到免疫反应。因此，在设计阶段，运用人工智能是十分强大的技术，可以增强疫苗的有效性，减少免疫反应。
But also, Moderna is using AI to possibly predict how various con-muteate over time before it's actually mutated. And that could put us in a much stronger position actually in the pandemic preparedness or in the pandemic.
此外，莫德纳正在利用人工智能，可能预测各种突变在实际突变之前如何变化。这可以使我们在应对大流行病或在大流行病准备方面处于更强的位置。
So we can actually be an advanced virus, rather than keep chasing it. So this is entirely transformative the way we can potentially treat infectious diseases. So this is just one of the examples.
因此，我们实际上可以成为先进的病毒，而不是一直追逐它。因此，这完全改变了我们可能治疗传染病的方式。这只是其中的一个例子。
And another one, which is also, I really think is quite cool. It's company called Exigencia. They built small molecular drugs, atom by atom by using AI.
还有一个我觉得非常酷的公司，名为Exigencia。他们使用人工智能逐个原子地构建小分子药物。
And the advantage of doing that approach by using that technology is that traditionally because, you know, the way we use to design drugs is optimizing first form efficacy and then for safety.
使用这项技术进行药物设计的优点在于，传统上我们优化药物效力的方式是首先考虑安全性。
And normally drugs actually fail because of the safety. So because they're very potent, but then they attack everything else in the body that makes them unusable.
通常药物失败是因为安全性问题。因为药物非常有效，但它们会攻击身体中的其他部分，使其无法使用。
But if you could optimize in parallel for those two things, so there is trade-offs between efficacy and safety. Suddenly, we actually have high chance of those drugs coming, actually working for patients. And that's exactly what Exigencia does.
如果您可以同时优化这两件事情，即在功效和安全之间进行权衡。突然间，我们实际上有很高的机会让这些药物真正适用于患者。这正是Exigencia所做的。
They're using artificial intelligence to optimize for those two properties simultaneously. And we already see that those drugs are very, very different.
他们正在使用人工智能同时优化这两个属性。我们已经看到这些药物非常非常不同。
So it's entirely reinventing how we approach health in many ways, but as I mentioned, it's a spectrum. So it's not just one area. It's just it's so broad.
因此，它完全重新定义了我们在许多方面对待健康的方式，但正如我提到的，它是一个谱系。所以它不是一个单一的领域。它的范围很广。
And then maybe to your point, you're going across to like it's across many industries. But when you start also going deeper in one particular industry, you also see the brass of the applications, just not one thing.
也许可以理解为，在涉及多个行业时，您会发现类似的情况。但是，当你更深入地研究一个特定的行业时，你也会看到这些应用方面的实际应用，而不只是一件事情。
It's interesting. I think the benefits of our food challenge and the health care is often something which is underestimated in terms of how significant it is. Because this is a massive problem in terms of costs.
很有趣。我认为我们的食品挑战和医疗保健的好处往往被低估了，这是非常重要的。因为它在成本方面是一个巨大的问题。
So if we look at US health care spend between 1970 and 2019, health care spending in the US as a percentage of GDP has more than doubled. So it's increased from 7% to around about 80% which is massive.
如果我们看一下1970年到2019年间美国的医疗保健支出，美国医疗保健开支占国内生产总值的比例已经增长了一倍以上。从7%增加到大约80%，这是一个巨大的增长。
You talk about that convergence point between science and data. So will AI hopefully bring down health care costs over time? Well, that's a hope and probably it's not just a hope.
你谈到了科学和数据的融合点。那么人工智能能否随着时间的推移降低医疗保健成本呢？嗯，这是一种希望，而且可能不仅仅是一种希望。
We already seeing somehow companies actually, by utilizing these technologies, they increase the efficiency on their invested capital. But there is one example.
我们已经看到一些公司利用这些技术来提高其投资资本的效率。但是还有一个例子。
But yes, so traditionally, the reason why health, I mean there are many different reasons why health care expensive. But we can just maybe stay with drugs, just drugs in themselves.
但是，传统上来说，健康保健昂贵的原因有很多。但是，我们可以只讨论药品本身。
And you know, they're very difficult to develop because of the complexity of human biology. So normally it takes more than 10 years, it takes billions of dollars. And then chance of success is around 10%.
你知道吗，由于人类生物的复杂性，研发这些药物非常困难。通常需要十年以上的时间，耗资数十亿美元，而成功率只有约10%。
So it's very, very difficult industry to be in. And then of course, you know, the compensate for all this risk taking and capital, the pharma company is charging quite high prices.
所以，这是一个非常非常困难的行业。当然，为了弥补所有这些风险和资本，制药公司收取相当高的价格。
But with AI, we can potentially have an opportunity to kind of untanly reinvent the business models in drug development. And it's just because we're kind of coming out of the darkness when you talk about the light.
但是有了人工智能，我们有可能重新发明药品开发的商业模式。这是因为我们正在逐渐摆脱黑暗，迎来光明。
And I feel like AI, even though you know, light has made such a difference. But within the biology, we are still pretty much in a dark box. And AI has an opportunity to bring us out of the darkness because we start finally getting some control over the complexity, biological complexity, and the risk in the drug development.
我感觉就像是人工智能一样，尽管光明让我们有了很大的变化。但是在生物学领域里，我们仍然处在一个相对黑暗的状态。人工智能有机会让我们走出这个黑暗，因为我们终于开始掌握了生物复杂性和药物开发中的风险。
And when we change in the risk profile, suddenly the economics looks very different. And that would eventually play into the kind of overall cost of health care.
当我们改变风险配置文件时，经济形势会突然变得不同。这最终会影响到整个医疗保健成本。
Kirstie, let's look at some other sectors that you're looking at from an investment perspective that's interesting.
Kirstie，让我们从投资的角度看一些其他领域，这可能会很有趣。
Yes, I think the core point that we've sort of both touched on now is that ultimately, you know, the AI is a data problem. And the fact is, as I mentioned in my last comments, that it is that data collection is happening across industry.
我认为我们现在都触及到的核心问题是，最终AI的问题是一个数据问题。事实上，正如我在上次评论中提到的，数据收集正在各行各业中发生。
So there's a lot happening in the lots of different industries. So for example, we can take something like the advertising industry.
许多不同行业都在发生很多事情。例如，我们可以以广告业为例。
There's a company listed in the US called the Trade Desk. And what the Trade Desk is, it's a demand side programmatic advertising platform. Now that's a lot of words. But what that really means is that they use data to help their customers by advertising. So if the three of us were watching the same television program at home on a connected device, we can then be served advertisements that are relevant to each of us at the same ad break. So what that does is it moves us from a world of who is the average viewer and who do I want to attract to? Who is actually watching this and how do I advertise to them in a relevant way? That's possible because the Trade Desk can process over 12 million ads per second. That's one, I can't even begin to process how fast that is. But also that's just something that a human would never be able to achieve. They also have a dedicated AI platform called its co-op platform and that's designed for its customers. And that enables customers to means that they can automatically send advertising dollars to those areas that are being successful and remove advertising dollars from those areas that are struggling at any one point in time. That is a real time thing.
在美国上市的一家名为Trade Desk的公司。Trade Desk是一个需求端程序化广告平台。这听起来很高大上，但实际上，他们使用数据来帮助客户做广告。举例来说，如果我们三个人在家里用联网设备看同一电视节目，那么我们可以在同一广告插入时段看到与我们各自相关的广告。这将我们从“谁是平均观众，我想吸引谁”的世界中脱离出来，进入“实际上是谁在观看，我该如何以相关的方式进行广告宣传”的世界。这是可能的，因为Trade Desk可以每秒处理超过1200万个广告。这太快了，我甚至无法想象。但这也是人类永远无法实现的。他们还有一个专门的AI平台，称为co-op平台，旨在为客户提供服务。这使得客户可以在任何时候自动向成功的区域投放广告资金，并从那些陷入困境的区域撤回广告投放。这是实时的。
Then you move to an industry like transportation. Now we all know that Elon Musk and Tesla are looking to make autonomous vehicles. But there's a company in the US called Aurora that listed last year and what Aurora is looking to do is autonomous trucking. Now trucks are big, heavy pieces of machinery and having them hurtling down a highway without somebody behind the wheel is potentially a scary prospect for people. And what Aurora has been doing is training what's known as it's a Aurora driver and that driver is its autonomous software and that driver what they've been doing is training and adding features over time. And the most recent feature that they've added is the ability for that driver to navigate roadworks. And the challenge with roadworks is that many of these autonomous drivers use sensors to understand whether or not you've moved out of your lane. Are you still in your lane? Now the challenge with roadworks is that usually requires you to move out of your lane. Quite often it requires you to cross multiple lanes to go round the blockage. That's difficult to program or difficult to learn about because it's not binary. It's not as simple as saying red light means stop, green light means go. It requires experience. No matter how many times you see roadworks, you're never going to see those precise roadworks. So that is about learning and that is what the Aurora driver has been doing. It's been learning how do I use my previous experience is all roadworks and how do I apply that to the situation I'm confronted with in real life.
然后您会涉足运输行业。现在我们都知道埃隆·马斯克和特斯拉正在寻求制造自动驾驶车辆。但是美国有一家名为Aurora的公司去年上市，Aurora试图实现自动运输。现在卡车是巨大而又沉重的机器，让它们在没有司机的情况下在高速公路上飞驰对人们来说可能是一个可怕的前景。而Aurora一直在训练自己的司机，这个司机是它的自动驾驶软件，它一直在不断学习和添加功能。它们最近添加的功能是使司机可以在路面施工时行驶。而在路面施工时的挑战在于许多自动驾驶车辆使用传感器来了解您是否已经从车道中移开。你还在自己的车道上吗？而在路面施工过程中的挑战在于通常需要您从车道中移开，通常需要您跨越多条车道绕过障碍物。这对于编程或学习来说是困难的，因为它不是二进制的，不能像控制红灯停、绿灯行那么简单。这需要经验。无论您看了多少次路面施工，您都永远不会看到那些精确的路面施工。因此，这就是学习，这就是Aurora司机一直在做的事情。它一直在学习：如何利用我以前的所有路面施工经验，并将其应用于我在现实生活中面临的情况。
And then you have an industry like insurance and this is really interesting because AI has the potential to not just disrupt the business model but also the product itself. So if you think, if you boil insurance down to what it fundamentally is, it's about data. It's about monetizing statistics and it's probability theory which is ultimately what AI is looking to achieve as well. And so how can AI disrupt the insurance industry? Well, it can actually disrupt the domain over which legacy players had over the one factor of production and that's data and monetizing that data. So data and statistics because these AI models and insurance can potentially come in bringing in data that those legacy players don't even have access to or they've never collected data on and that means that you develop a model in which others just can't compete. So there's a company listed in the US called Lemonade Insurance and they are embracing the power of artificial intelligence when it comes to their insurance offering.
保险业是一个非常有趣的行业，人工智能有潜力不仅颠覆业务模式，还可能颠覆产品本身。如果你将保险原理归纳到最基础，它就是关于数据的，它也是关于赚取统计数据和概率理论，这也是人工智能所追求的。那么，人工智能如何颠覆保险业呢？它可以打破现有的保持数据和赚取数据的独特领域。因为这些人工智能模型可以引入那些旧有参与者没有的数据，或者他们从未收集过的数据，这意味着你可以建立一种他人无法匹敌的模型形式。美国上市的一个公司叫做Lemonade Insurance，他们在保险方面已经成功地应用了人工智能技术。
So they have an AI called AI Maya and AI Maya is the she she deals with premiums so policies and she asks 13 questions and collects over 1700 data points on through those questions including things like how long did you spend answering your question, how many times you visited the website, did you read all the terms and conditions etc. And as a consequence of that decides how to price your premium and can price your premium therefore based on you as an individual again as I mentioned with the trade desk rather than putting you in a kind of average pool of what's expected of somebody within your demographic.
他们有一个名为AI Maya的人工智能，她负责处理保单和保费。她会问13个问题，并通过这些问题收集超过1700个数据点，包括您回答问题的时间、访问网站的次数、是否阅读了所有条款和条件等等。
由此，她可以根据您个人的情况来决定如何定价您的保费，并且可以再次根据您的情况来定价您的保费，而不是将您放入与您人口统计学相似的平均池中。
Then they also have AI Jim and AI Jim deals with claims management. So if you need to make a claim on your insurance as long you contact Lemonade you're you put in contact with AI Jim and as long as that claim is within the guidelines of your policy then it will pay out.
他们还有一个AI吉姆，负责理赔事务。如果您需要在保险上提出索赔，只要联系Lemonade，您就会与AI Jim联系上；只要该索赔符合您的保单规定，保险公司就会进行赔付。
Now the opportunity here is is one to reduce the cost as we saw in healthcare to potentially reduce the cost of insurance because you'll you'll be able to better identify the bad actors, you're not clubbed together, the good actors and bad actors are not clubbed together and given an average price you can be priced as an individual. And the second thing is there's a potential to lower the cost for the companies operating in the industry because the administrative burden of people phoning to make claims is taken on by an AI rather than humans.
现在，这里的机会在于降低成本，正如我们在医疗保健领域所看到的那样，有潜力降低保险费用，因为你可以更好地识别出恶意的行为者，他们不会被捆绑在一起，好的行为者和坏的行为者不会被归为一个平均价格，而是可以按个人定价。其次，有可能降低运营行业公司的成本，因为通过人工智能接管理赔的繁琐工作，取代了需要人类来打电话处理的工作量。
I'm sorry I just find it fascinating what you actually talk about as a lot of personalization and the same theme actually is very powerful in healthcare space because if we think about most of the drugs are developed for other ages and when you actually assess how effective those drugs in real life most of them actually don't work because either they're not effective or they cause very severe side effects and that's also there is a paradigm shift in healthcare can we actually develop drug first for more stratified populations you know based on for example the genomic profile but also going much more personal as truly patient by patient and again bringing moderner back so they're working on the personalized cancer vaccine when each individual cancer is assessed where it's it's profile and then the vaccine is developed specifically for that profile for that patient and without AI you just can't do it because it's effective it's quick and yeah and it's yeah in this cheap so so this is kind of the same kind of personalization which was not a we are not being able to approach it in that way without that technology.
很抱歉，我对你们谈论的事情感到非常着迷，因为在医疗领域里，个性化和相同主题的讨论实际上非常强大。如果我们考虑大多数药物是为其他年龄段开发的，当你实际上评估这些药物在现实生活中的有效性时，其中大多数实际上并不起作用，因为它们要么不起作用，要么会导致非常严重的副作用。在医疗保健中，也存在着一种范式转变，即我们是否可以首先为更分层的人群开发药物，例如基于基因组学文件，但也要更加个性化，就像真正的病人一样。再回到现代，他们正在研制个性化癌症疫苗，当评估每个癌症的特点以及它的文件时，为该患者特别开发疫苗。如果没有AI的帮助，就无法实现这一点，因为它是有效的、快速的，而且价格便宜。所以，这种个性化的方式以前我们没有能力以这种方式接近它。
Kirsty you started the conversation by talking about the toddler maybe going through its growing pains in terms of growing up and that correlation with artificial intelligence how do you see that toddler developing of the next 10 years or so? yeah so I think I think you can sort of use that analogy as well for the industry itself so I think you're kind of we're probably in that kind of toddler stage and then you probably see that potential for that sort of childlike delight stage or the excitement about what's possible and then maybe you move into the kind of moody teenage phase where you know they're shot in their room you don't know what's going on it is a bit of a black box we're not entirely sure what they're thinking how they're thinking and I think that's the emergence of kind of the general AI side of things that it's hard to know exactly what's going on but I think that in itself is is less of a concern if you know as as when you're bringing up a teenager that that you've put the guidelines in place that eventually they kind of pop out as a valuable member of society and even if they went through a few growing pains at the time and I think it speaks to that idea of you know if you instill the right values and you recognize that the potential biases and externalities of what you are trying to achieve that as long as those are potentially in place before you start you will go through those periods of not understanding not knowing exactly what's going on but you know there is there is a light after the darkness of hiding in your room for five years.
Kirsty，你开始谈论幼儿可能因成长而经历生长的痛苦，以及这与人工智能的关联，你如何看待这个幼儿在未来10年左右的发展？是的，我认为您也可以将这个类比用于这个行业本身。我认为我们可能处在幼儿时期，然后你可能会看到那种兴奋于可能性的孩童期，然后可能会进入一种情绪化的青少年阶段，在那里你不知道他们在想什么，他们在想什么，这是一个黑匣子，我们不完全知道他们在想什么，如何思考。我认为这是通用人工智能涌现的一种表现形式，在这种情况下，它很难知道确切发生了什么，但我认为这本身不是太令人担忧，如果你在抚养青少年时确立了正确的价值观并认识到你所努力实现的潜在偏见和外部性，那么即使它们经历了一些成长的痛苦，它们最终仍会成为有价值的社会成员并产生影响，这也反映了这样一种观念，即如果你在开始之前确立了正确的价值观，并认识到你所努力实现的潜在偏见和外部性，那么你将经历一些不理解和不确定的时期，但你知道在躲在房间五年后的黑暗之后会有一些光明。
This idea isn't it of artificial general intelligence or AGI that deep-mind in by Google parent alphabet are trying to get to over the long term but it's something and you did talk about this earlier with Aurora this idea of edge case with roadblocks and a pigeon flying up but it's hard to underestimate quite how difficult this is in terms of understanding consciousness learning from mistake it's an extraordinarily difficult thing to do isn't it.
这个想法并不是 Google 的母公司 Alphabet 正试图在长期内实现的人工通用智能（AGI），而是另一件事情。你之前和 Aurora 谈过这个边缘情况、路障和一只鸽子飞起来的想法，但很难低估意识理解、从错误中学习这些方面的困难程度。可以说这是一件非常困难的事情，是吧。
Yeah I think I think the challenge is that we don't really truly as Julia mentioned earlier like even understand how our own brains work we don't know how we make the connections that we do you know we're potentially developing AIs that can understand emotion but do they truly understand emotion yet because what people's faces say in their body language says is that the same as what's in their head and we don't necessarily know how we arrive at the conclusions that we do but ultimately it's about making connections and over time AIs are making more and more connections and they're making those connections at a faster pace because this is an exponentially improving technology.
我认为挑战在于我们并不真正了解我们自己的大脑是如何工作的，正如Julia之前提到的那样。我们不知道我们是如何建立联系的，就算我们正在发展能够理解情感的AI，但它们真正理解情感了吗？因为人们的脸部表情和身体语言所表达的，是否和他们脑海中的想法一致，我们并不一定知道我们是如何得出结论的。但最终，这一切都是关于建立联系，随着时间的推移，AI会建立更多的联系，并且以指数级别提高的技术速度实现这些联系。
That was Kirstie Gibson co-manager of the Bayley Gifford US Growth Trust and the American Fund and Julie Angelese joint manager of our Health Innovation Fund.
这是Kirstie Gibson和Julie Angelese，分别担任Bayley Gifford美国成长信托和美国基金的联合经理，以及我们的健康创新基金的联合经理。
A video recording of the full session can be found on the Insight section of our website at bayleygifford.com forward slash insights and the longer version includes a discussion about ethics and how Bayley Gifford itself is exploring AI or artificial intelligence but for now I hope you enjoyed the special edition of short briefings on long term thinking as ever thanks for investing your time in this podcast.
我们的网站bayleygifford.com/insights的Insight栏目里可以找到完整会议的视频记录，更长的版本包括对伦理以及Bayley Gifford探索人工智能的讨论，但现在我希望您享受本期关于长期思考的特别简报。再次感谢您在本播客中投资时间。