AI的本质：与Drew Purves共同解决地球数据缺口

有时候你会认为，真正的改变可能来自于这些觉醒时刻，在长远来看，人们几乎可以在一夜之间改变他们与自然的关系。如果人工智能能够帮助赋能这一点，那么长远来看，这可能是人工智能最强大的作用。

Sometimes you think that the real change can come in the long run from these moments of awakening where people almost overnight can change their relationship with nature. If AI can help to empower that, that might in the long run be the most powerful role of AI.

欢迎回到谷歌DeepMind播客。我是汉娜·弗莱教授。本播客中我们重点关注人工智能及其与人类的互动。但还有另一个故事正在展开，人工智能可能帮助保护我们的星球。想想海洋、森林、沙漠、脆弱的生态系统，以及地球上爬行、游泳和飞行的数百万种动物物种。

Welcome back to Google DeepMind, the podcast. I'm Professor Hannah Fry. Now we focus a lot in this podcast on AI and its interactions with humans. But there is another story that's unfolding, one in which AI could help protect our planet. Think oceans and forests and deserts and fragile ecosystems and the millions of animal species that are crawling, swimming and flying across this earth.

人工智能有潜力应对我们这个时代最大的挑战之一——对自然和生态系统的破坏。但面对如此庞大的问题，该从哪里入手呢？谷歌DeepMind的自然领域负责人德鲁·珀维斯将作为我的向导，带我探索人工智能助力自然的广阔领域。他拥有二十年的生态研究经验，并在谷歌DeepMind工作了近十年。德鲁，非常感谢你的加入。

AI has the potential to tackle one of the biggest challenges of our time the damage to nature and ecosystems. But with a problem that's this vast where do you even begin? Well Drew Purvis Nature Lead at Google DeepMind is my guide through the rich terrain of AI for Nature. He has got two decades of experience in ecological research and has been at Google DeepMind for almost ten years now. Drew, thank you so much for joining me.

哦，谢谢。很高兴来到这里。

Oh, thanks. It's great to be here.

我想大多数人现在可能都同意，环境是我们未来的一个重要方面，值得保护和呵护。在这个领域，是什么阻碍了我们？是什么让它成为一个难题？

I think most people probably agree by now that the the environment is an important aspect of our future, something that deserves preserving and looking after. What's holding us back in this area? What's making it a difficult problem?

嗯，通常的简短答案是缺乏信息。你说得对，现在有一种共识正在兴起，认为生物多样性、生态系统和自然非常重要，然后我们在不同部门，包括私营部门和政府部门，看到了行动的迹象。你知道，如果你看一些数字，例如，全球有189个国家签署了“30x30”计划，目标是在2030年前保护30%的陆地和海洋生态系统。有很多值得乐观的地方，但通常当涉及到实际采取行动来保护或恢复生态系统中的生物多样性时，往往只是缺乏基本信息。

Well, the short answer to that often is lack of information. So you're right, know, there's this groundswell now of agreement about the importance of biodiversity, ecosystems and nature, and then we have signs of action in different sectors, in the private sector and the government sector. You know, if you look at some of the numbers, for example, I mean, are 189 countries around the world that have signed up to the 30 by 30 plan, which is to protect 30% of ecosystems on land and in the oceans by the year 02/1930. There's so much to feel good about, but often when it comes down to actually taking action on the ground to either protect or restore biodiversity in ecosystems, it's just a lack of basic information.

那么，人工智能在这个领域可以帮助回答哪些重大问题呢？

So what are the big questions that AI could help answer in this space then?

我的意思是，显然有很多种情况。但举个例子，如果你考虑的是保护场景，保护生物多样性，你需要知道生物多样性在哪里。可能是关键物种，也可能是生物多样性热点地区等等。可能是特定的特有物种。如果是恢复场景，你会想找到世界上具有最高生态恢复潜力的特定地点。

I mean, are a number of them, obviously. But for example, if you're thinking about a protection scenario, protecting biodiversity, you need to know where the biodiversity is. It might be focal species, or it might be biodiversity hotspots and so on. It might be particular endemic species. If it's a restoration scenario, you want to find particular places around the world that have the highest potential for ecological restoration.

但例如，你可能需要知道种植哪些树种或重新引入哪些动物物种。所以在地方层面，因为生物多样性具有很强的地方特异性，你需要这些与当地相关的信息来指导社区或其他地方行动，而这些信息往往缺失。

But for example, you might need to know which species of trees to plant or which species of animals to reintroduce. So down at this local level, it's because biodiversity is so place specific, you need that locally relevant information for communities or whatever local action is happening to guide the action down there, and often it's just missing.

那么，谷歌DeepMind在自然界应用AI的主要目标，是否就是像我们在人类世界所做的那样，真正填补信息空白？

Is that the big goal then of Google DeepMind using AI in nature, to really fill in the information gaps in the same way that we have done for the human world?

嗯，在谷歌DeepMind，我们正在发展一个围绕AI用于自然的项目组合，至少有三个关键的AI用于自然的类别需要探索。第一个是用于数据的AI，这可能意味着从实地获取数据，比如通过摄像头、麦克风等设备，或者从文献中识别和获取数据，因为那里有海量数据。第二类则是将所有数据与卫星数据等其他数据源结合，生成决策者保护或改善自然所需的衍生信息。然后是第三类，这一点容易被忽视，即世界上所有的信息都很好，但人类可能会被如此大量的数据淹没。因此，AI在积极部署以辅助决策、帮助人们理解所有这些信息方面扮演着关键角色。

Well, here at Google DeepMind, we're growing a portfolio of work around AI for nature, And there are at least three key categories of AI for nature to explore. So the first of those is AI for data, and that can mean bringing in data from the field, from things like cameras, microphones, and so on, or identifying and bringing in data from the literature, because there's huge amounts of data there. Second category though is taking all of that data and combining it with lots of other source of data like satellite data and so on to create that derived information that decision makers need to protect or enhance nature. And then the third category, which is easy to miss, is that all the information in the world is fine, but human beings can get overwhelmed by that amount of data. So then there's this key role for active deployment of AI to help decision making, to help people make sense of all that

数据并采取行动。我想这部分是因为，你知道，这算是一个相当新的领域，对吧？我的意思是，它不像AI的其他一些应用存在那么久。解决生物多样性问题的方式不像解决，比如说疾病那样明显。

data and act. And that's, I guess, in part because, you know, this is sort of quite a new area, right? I mean, this hasn't been around for as long as some of the other applications of AI. It's not as obvious of how to sort of solve biodiversity as it might be to solve, you know, disease, for instance.

我认为确实如此。它确实感觉像是AI中的一个增长领域。实际上很有趣的是，生态学显然有很长的历史，但有趣的是，数学生态学、统计生态学甚至机器学习生态学也有很长的历史。例如，费希尔，著名的费希尔统计学，很多都是在生态学背景下发展起来的。我认为这是因为生态学本身相当具有挑战性。

I think that's true. It does feel like a growth area in AI. It's kind of interesting, actually, that ecology obviously has a long history, but interestingly, mathematical ecology and statistical ecology and and even machine learning ecology has has a long history too. Fisher, for instance, the famous Fisherian statistics, a lot that was developed in an ecological context. I think that's because ecology is inherently quite challenging.

就像，没有明显的方法来记录和构建生态学。你有各种各样的物种。过程在不同的尺度上运作。时间信号总是非常嘈杂。嗯。

Like, there's no obvious way to sort of record and frame ecology. You've got a vast variety of species. You've got processes operating at different scales. The temporal signals are always really noisy. Mhmm.

所以这很有趣。这一直是一个有点挑战性的领域，它推动了统计学和机器学习的创新。因此，我们非常期待在我们现在的工作中尝试应用深度学习，比如最近的人工智能革命到生态学中，这也将是一种创新的方式。

And so it's it's interesting. It's always been a little bit of a challenging domain that has led to innovation in statistics and machine learning. So we very much expect that in the work we're doing now trying to apply this deep learning, for instance, the recent AI revolution to ecology, that also it will be a way to innovate.

好的，那我们来深入探讨一些这方面的问题。特别是中间那一层，就是接收数据并从中推导出某些信息的部分，因为，我的意思是，这其中很多涉及制图，对吧？就像那个基础阶段。给我们讲讲绘制生物圈地图的一些情况吧。

Okay, well let's dig into some of this stuff then. In particular, in that middle layer, right, of like taking in data and sort of deriving something from it, because, I mean, lot of this involves mapping, right? Like sort of that fundamental stage. Tell us a little bit about mapping the biosphere.

环境和生态学中的许多决策都归结为地点，这是因为在生态学和生物多样性中，每个地方都如此不同，不同的物种、不同的栖息地，以及往往需要处理的不同问题，你知道，无论是农业还是野火等等。所以我之前提到的缺失信息，大多数时候是缺失地理信息。所以很简单，你可以把它想象成地图。你说得对。因此我们需要绘制栖息地地图。

So many of the decisions in environment and ecology come down to place, and that's because in ecology and biodiversity, each place is so different, different species, different habitats, and often different issues that you need to deal with, you know, whether it's agriculture or wildfires, etc. So that missing information I mentioned before, most of the time it's missing geo information. And so very simply, you can think of that as like maps. You're right. And so we we need to map habitats.

这有点像生态学上演的基底，因为基底主要由植物定义。所以栖息地和植物密切相关。所以，你知道，无论是森林、草原等等。但当然，理想情况下我们还想绘制所有物种的地图。这是一个非常具有挑战性的问题，仅仅因为物种有数百万种，而且很多非常小，看不见等等。

That's kind of the substrate over which ecology plays, because substrate is largely defined by plants. So habitats and plants are closely related. So, you know, whether that's forest, grasslands, etc. But of course, then we want to map all the species ideally. That's a very challenging problem just because there are millions of them and a lot of them are very small and can't be seen and so on.

当然，现在有地图，但我们可能还想了解历史变化。那些变化的基线非常重要，当然还有能够将地图投射到未来，这完全是另一个挑战。但在生态学中，所有道路都通向地图。

Of course, there's maps for now, but also we might want to understand historical change. Those baselines of change are very important and of course being able to project maps into the future, which is a whole different challenge. But all roads lead to maps in ecology.

所有道路都通向地图。好的，这很有趣。不过我很惊讶这竟然还不存在。我的意思是，你不能直接从谷歌地球获取这些吗？

All roads lead to maps. Okay, that's interesting. I'm surprised that this doesn't exist already though. Mean, you not just get this from Google Earth?

这是一个非常有趣的问题，因为我们已经太习惯了。当然，通过像谷歌地球和谷歌地图这样的工具，我们拥有的地理信息已经非常惊人。你知道，将所有这些整合在一起并使其可用是人类的一项真正成就。但到目前为止，这些数据非常以人类为中心，所以我们当然有道路，有每个购物中心、每个邮局，诸如此类的一切，实际上在绘制自然世界地图方面还很不发达。

It's a really interesting question, because we've got so used to it. Of course, it's been amazing the geographic information we have through things like Google Earth and Google Maps. You know, real human achievement to bring all that together and make it available. But it's very human centric so far, that data, so of course we've got roads and you've every shopping centre, every post office, everything like that, actually it's much less developed in terms of mapping the natural world.

它更侧重于人而非环境。

It's focused more on people than the environment.

是的，没错，而且我们需要的许多关于自然世界的信息从未以纸质形式记录过，比如所有不同树种的位置、栖息地的分布，或者粗草地与湿地之间的边界。因此我们实际上需要首次创建这些信息。

Yeah, that's right, and lot a of the information that we want about the natural world has never been recorded like that on paper necessarily, where all the different species of trees are or where the habitats are or where the boundaries between the rough grassland and wetland are. So we need to actually create that information for the first time.

我觉得有点不可思议，现在都2025年了，这件事居然还没完成。我的意思是，之前确实有过尝试。比如Landsat项目在1970年代就开发了，对吧？就是通过卫星图像对不同区域进行分类。在那之前也有一些基于卫星图像的军事用途项目。

It sort of seems a bit bonkers to me that we're in what, 2025 and this hasn't been done already. I mean, are attempts at this that exist. I mean, Landsat was developed in the 1970s, right? Of like taking satellite images and categorising different regions. There I was some stuff before that for military purposes using satellite images.

但这只是规模上的不同吗？

But is this just a different scale to those things?

这也让我感到惊讶。有几件事让我惊叹不已。一是卫星本身的存在就很不可思议，对吧？这简直让人难以置信，而且它们已经存在了几十年。几十年来我们基本上一直有这些漂浮在高空的高清数码相机俯视着地球，这绝对令人惊叹。同样令人惊叹的是我们实际拥有的地理信息量——存在大量与环境相关的惊人数据集和地图，但有时你仍然会遇到最基础的问题，比如：森林在哪里？

It amazes me too. Several things amaze me. One thing that amazes me is that the satellites even exist. I mean that's incredible, right? It just blows my mind and they've been actually for many decades We've basically had these floating high res digital cameras looking down at the earth for decades That's absolutely incredible It's also incredible the amount of geographic information we actually do have There are amazing data sets or amazing maps of all kinds of things that are relevant to the environment, and yet at the same time sometimes you come in with the most basic questions like where are the forests?

我们难道不知道森林在哪里吗？

Do we not know where the forests are?

没错，说实话这也让我吃惊。对于我们需要理解的大多数栖息地，包括森林，我们并没有公认的黄金标准全球地图，所以并不存在绝对公认的森林/非森林黄金标准。话虽如此，如今确实出现了一些相当不错的地图，但仍有改进空间。另一方面，我们关注的是，即使有了这样的地图，为了指导决策，通常还需要区分不同种类的森林。比如最基本的区分：天然林与人造林。

That's right, and honestly it surprises me too. We don't have gold standard accepted global maps for most of the habitats that we need to understand, including forests, so there is no absolutely universally accepted gold standard for forest non forest. Now having said that, there are some pretty good ones nowadays appearing, but there's still room for improvement. On the other hand, what we're looking at is, even if you had that, to guide decision making, you often need to distinguish different kinds of forests. For example, the most basic split: natural forest from planted forest.

而且对于这一点，绝对没有一个公认的黄金标准地图。我们目前正在做的一些工作，就是尽力提供迄今为止可用的最佳地图。我们称之为世界天然林项目。

And there definitely isn't a gold standard accepted map for that. Some of the work we're doing at the moment is trying to do our best to provide the best yet available map of that. We call it the Natural Forests of the World project.

森林地图为什么有用？你能用它们做什么

Why are forest maps useful? What can you use them

？森林极其重要，蕴藏着生物多样性和碳汇。因此，有许多政府政策和国际法规围绕森林制定，森林也是保护组织关注的主要焦点。在所有情况下，他们都需要最好的地图，以便知道如果要保护森林应该保护哪里，可能在哪里恢复森林，在哪里监测森林病害等问题，诸如此类。

for? Forests are incredibly important, harbouring biodiversity and carbon. And for that reason, there's a lot of government policy and a lot of international regulation around forests, and forests are a major focus of concern for conservation groups. And in all cases, they need the best maps they can so they can know where to protect if they're looking to protect forests, where to potentially restore forests, where to monitor for problems like forest diseases and on and on.

我看到了吗？你拿到了吗？

I see it? Have you got it?

是的，拿到了。对，就在我笔记本电脑上。首先，我们制作的是全球10米分辨率的地图，因此每个10米像素都被分类，我们给出了它是天然林的概率，这就是我们在这里看到的。

Yes, have. Yeah, I've got it right here on the laptop. Firstly, what we produced is a map for the whole globe at 10 metre resolution, so each 10 metre pixel is classified, and we give it a probability that it's natural forest, and that's what we're viewing here.

我可能应该为只听不看的人做一点音频描述。我们在浏览器中基本上有一个看起来表面上很像谷歌地球的界面，但覆盖在上面的是这些蓝绿色的像素，它们与卫星图像中可见的树木大致对齐但并不完全一致。

I should probably do a little audio description for the people that are listening rather than watching. We have essentially in a browser what looks superficially well, exactly like Google Earth, but overlaid on top are these teal colored pixels that align but not exactly with what you can see to be trees from the satellite images.

没错。所以你可以举例说明，比如——这一点很重要，但也很难区分天然林和人工林，因为天然林通常具有更高的生物多样性，如果是一些古老森林的话，碳储量也往往更高。因此，这类区域你往往会更倾向于保护。同样地，如果你检测到森林减少，那么天然林的减少通常比人工林的减少更令人担忧，许多人工林本来就是为砍伐而种植的，比如木材林，它们实际上可以是可持续的。但在这个案例中，例如在美国东南部，就有很多火炬松人工林。

That's right. So you can illustrate that, for instance, in and this this is important, but also difficult to tell the difference between natural forest and planted forest because the natural forest will typically be much higher in biodiversity and often carbon if it's somewhat old growth. So this will be an area that you would tend to want to protect more. And equally, if you detect forest loss, then the loss of natural forest would typically be much more concerning than, for instance, the loss of planted forests, many of which are planted in order to cut them down, like a timber so they can actually be sustainable. But in this case, in the Southeast US, for example, there are a lot of loblolly pine plantations, for example.

所以这是一种人工林，种植后大约五年就能快速提供木材。好的。你可以看到这里，我们有被我们归类为天然林的区域和归类为人工林的区域混合在一起。

So these are a type of planted forest which is planted to give timber very quickly after about five years. Okay. So you can see there we've got a mix of areas that we classify as natural and areas that we classify as planted forest.

是的。而且

Yeah. And

如果你放大缩小查看，你知道，大多数时候当然会有不准确之处。但我们在测试时准确率超过90%。所以你可以看到这里，当我们逐渐显示预测结果时，然后如果我们查看那些剩余部分并深入观察，你会发现这些整齐的行列，

if and if you zoom in and out, you know, most of the time, of course, they're inaccuracies. We're but we're we're over 90% accurate when we test it. So you can see here, as we fade up our predictions there, then if we look at those remaining parts and you go in, you'll find these neat rows,

那就说明

and that's telling

它是人工林。能够大规模做到这一点，没错正是这样。

us it's planted forest. To be able to do this at scale, yeah exactly.

那么好的，我注意到在这个例子中，它有点像是天然林又不是天然林，但我也注意到你们那里有一个置信度阈值。是的。所以你们是在用概率进行分类，而不是简单地说‘是天然林’或

So okay, I noticed in this example, it's sort of natural forest, not natural forest, but I also noticed that you've got a confidence threshold there. Yes. So are you sort of classifying this with a probability rather than just saying yes it's natural or

‘不是’？是的，我们确实如此，而且这是我们相当热衷的一点，因为如果看到一张以非黑即白方式分类的地图，人们往往会相信它。即使大多数时候相对准确，意识到不准确性非常重要，这就是为什么我们更倾向于以这种不确定性地图的形式呈现。当然，你通常会做的一件事是，好吧，我会选择一个阈值，然后用于我的目的来区分森林和非森林，但根据你的目标，你可能会选择不同的阈值。所以如果你真的想要确保保护一个地区所有剩余的天然林，你会设置你的置信度阈值。

it's not? Yes, we are and it's something we're quite passionate about really because there is a tendency if you see a map that is classified in a black and white light way to believe that. And even if it's relatively accurate most of the time, it's really important to be aware of the inaccuracies, and so that's why we much prefer to present things as this uncertainty map. Now of course, one thing you'd often do is to think, Okay, I'll choose some threshold then, which I will use for my purposes to distinguish forest and non forest, but depending what you're after, you may choose different thresholds. So if you were really looking to make sure you're protecting all of the remaining natural forest in an area, you would set your confidence threshold.

实际上，设置低阈值是为了让你能捕捉到所有这些信息，但当然你也会收集到大量的人工林数据，不过你在过程中可能不太在意。另一方面，如果你资源非常有限，比如说，你想验证天然林的损失情况，并且只能负担得起访问某些地点，你可能会设置一个非常高的阈值，以确保你真正访问的是你最有把握的地方等等。我们以这种不确定性的形式向下游用户提供这些信息。我们开源了所有地图，实际上也开源了数据以及我们开发的新模型，向社区开放。对下游用户来说，这意味着他们可以用比以前更少的计算量、更低的数据要求，最重要的是，更低的技能要求，获得自己的高质量遥感地图。所以如果我们做对了，类似这样的东西真的可以帮助遥感技术在外界实现民主化。

It's actually setting it low so that you pick all of that up, but of course you'll also pick up a whole load of planted forest, but you sort of don't mind along the way. Whereas on the other hand, if you had very limited resources, let's say, and you wanted to verify that the loss of natural forest had occurred and you can only afford to visit certain places, you may put a very high threshold on it to make sure that you're really visiting the places where you're most confident about and so on. We're sort of providing that to downstream users in this uncertainty form. We're open sourcing all the maps, we're open sourcing actually the data and also the new models we developed open sourcing out to the community. What it means for downstream users is that they can get to their own high quality remote sense maps with much less compute than before, lower data requirements, and all importantly, lower skills requirements as So if we do it right, something like that can really help to democratise remote sensing on the outside.

在幕后技术上，你们使用的是视觉变换器（vision transformer），对吧？是的。它是如何工作的？

In terms of what's going on behind the scenes, you're using a vision transformer, right? Yeah. How does that work?

总的来说，我们正在引入这些海量的卫星信息。首先，这些图像非常庞大，像素数量巨大。然后对于每个像素，它不仅仅是像普通图像那样的RGB，通常还有许多不同的波段，比如红外线等等。而且这些波段是随时间变化的。然后你还有多个卫星。

Overall, we're doing is bringing in this massive satellite information. So these are enormous images for a start, the sheer number of pixels. And then for each pixel, it's not just RGB like it would be with an image, but it's often many different bands, infrared, etcetera. And those bands are going up through time. Then you've got multiple satellites.

每个卫星每隔几天就会提供记录。你有很多缺失的数据，比如云层遮挡。所以你需要处理这些海量数据，并以某种方式将其压缩，提取出你真正想知道的信息，比如，这是森林吗？而我们中间使用的模型就是这个视觉变换器模型。

Each satellite is giving you records every few days. You have a lot of missing data from things like clouds. So you need to take this enormous amount of data and somehow crunch it all down to pull out the thing that you actually want to know, like, is it a forest? And the model in the middle that we're using is this vision transformer model.

但我的意思是，变换器通常人们会将其与大型语言模型联系起来，对吧？就像关注句子的不同部分比其他部分更重要。在地图方面，它是如何应用的？

But I mean, transformers are sort of people usually associate them with large language models, right? Like paying attention to different bits of sentences is more important than others. How does it feed in when it comes to maps?

没错。变换器最初主要是为处理语言而开发的，包括注意力头等概念。然后它们被适配到视觉变换器中，在图像上做类似的事情，关注图像的不同部分。我们在这里所做的是将其扩展为一个特殊的视觉变换器，专门设计来处理卫星数据的挑战。所以它现在实际上是一个多模态时空视觉变换器，但它仍然是视觉变换器的一种形式。

That's right. So Transformers were mostly developed to work on language and this idea of these attention heads and so on. They were then adapted into vision transformers to do similar things on images, tending to different parts of images. What we've done here is then expand that out to a special vision transformer that's set up to deal with the challenges of satellite data. So it's actually now a multimodal temporal spatial vision transformer, but it's still a form of vision transformer.

这是一个很棒的故事，展示了方法如何在不同领域之间交流，从语言建模到视觉，再到遥感，并最终应用到环境政策中。

It's a great story about how you can get this exchange of methods between different areas from language modelling into vision, and in this case, into remote sensing and out there into environmental policy.

所以它是在工作时进行支付，无论是森林还是非森林，它都会关注——有时是红外数据，有时是图像，有时是该区域的地形之类的东西。

So it's paying when it's working out, whether it's forest or not forest, it's paying attention as it were to sometimes the infrared data, sometimes the image, sometimes the sort of the topology of the area, that kind of thing.

没错。Transformer架构是一种极具表现力的架构，它赋予模型极大的自由度来选择关注什么以及如何在下游将这些信息整合到预测中。这在遥感等领域尤其有价值，因为那里有各种各样的模态输入，而且所有数据集都非常庞大。因此，这个领域确实受益于这种额外的模型表现力和灵活性。

That's right. The transformer architecture is extremely expressive architecture that gives the model a huge amount of freedom to choose what it attends to and how it then combines that information downstream into its prediction. And that's particularly valuable in an area like remote sensing, where you've got such a wide variety of modalities coming in and all the data sets are so huge. And so it's an area that really benefits from this kind of extra sort of expressivity and flexibility of the model.

但那种表现力最终归结起来就是森林，非森林。

But that expressivity in the end comes down to forest, not forest.

这就是有趣的地方，对吧？所有那些最终都被压缩成了这个，在这个案例中，就是那张单一的地图。

That's what's funny, right? All of that gets crunched down into just that in this case, that single map.

嗯，好吧，我只是接着你刚才回答中提到的一点来说，因为你谈到了你们拥有这些随时间变化的卫星图像。那么，这不仅仅是绘制森林存在与否的地图，是否意味着你们可以观察森林是如何变化的？

Well, Okay, I'm just picking up on something that you said in your answer there because you were talking about how you have these satellite images over time. So then rather than just mapping where the forests are or aren't, does that mean that you can look at how the forests are changing?

是的，这是一个非常重要的点。确实如此。所以遥感卫星数据的有趣之处在于，由于卫星已经运行多年，如果你能进行这种绘图，比如说你取一年的卫星数据，比如今年，我们可以看到今年森林在哪里，我们可以自动地重新处理过去的数据来估算变化情况。这非常重要。有一个名为全球森林观察（Global Forest Watch）的项目，谷歌之前支持过，它就在做这件事，所以每年最新的地图都会发布，你可以看到森林覆盖随时间的变化。

Yes, that's a very important point. Yeah, indeed. So the interesting thing with remote sensing and satellite data is because the satellites have been up for years, if you could do this mapping, you know, say you take one year's worth of satellite data, say for this year, and we can see where the forests are this year, we can automatically redo that from the past to estimate the picture of change. And this is really important. There's a project called Global Forest Watch, which Google has supported before, that does this, and so each year the latest map comes out and you can look at the changes in forest cover through time.

最近我们创建了一系列我们称之为毁林驱动因素的地图。这实际上是观察每个森林损失单元，并根据采伐、农业扩张等原因对损失原因进行分类。但我们将此应用到了过去二十年。我们与世界资源研究所（WRI）的同事合作，绘制了从2000年至今每年的全球毁林原因地图。

More recently we've created this stack of maps that we call deforestation drivers, What this is actually looking at is it's taking each unit of forest loss and categorising the cause of that loss in terms of logging, agricultural expansion and so on and so on. But we've applied that for the last twenty years. We have, in collaboration with our colleagues at the World Resources Institute at WRI, we've worked up this global map of the causes of deforestation for each year from the year 2000 to today.

当你这么做的时候，你发现了什么？

And what do you find when you do that?

观察这些模式真的很有趣。

It's really interesting to look at these patterns.

继续。给我看看。给我看看。我知道你找到了。我知道你找到了。

Go on. Show me. Show me. I know you've got it. I know you've got it.

想看看它。

Wanna see it.

所以当你放眼全球时，首先能看到的是森林损失无处不在。

So the first thing you can see when you look across the world is that there's forest loss everywhere.

是的。北欧地区相当多。

Yeah. Quite a lot in Northern Europe.

没错。举个例子。所以很容易认为所有这些都只发生在热带地区，或者全球南方，对吧。

Yes. For example. So it's so easy to think that all of these problems are just occurring in the tropics or in Yeah. In the global South. Right.

没错。你知道吗？而且第二点是，不同地区的损失原因确实会有所不同。所以如果你看一下巴西和南美洲的放大图，可以看到，我们可以进一步放大观察。

Exactly. You know? And and the second thing though is the causes of the loss does change between different regions. So if you have a look at zooming on Brazil here and South America, can see, again, can go right in.

那留下的是永久性农业用地。

That leaves a permanent agriculture.

说得对。这种深黄色区域。当你逐渐放大缩小时，经常会发现这与林地清理等有关，所以你能看到它是如何运作的。

That's right. So this sort of darker yellow colour. You'll find when you fade in and out, you'll often find that this is associated with clearings and so on, so you can see how that's working.

你几乎可以清晰地看到它是如何有效识别出农田边缘的。

You can literally see how it picks out the edges of fields effectively.

没错，所有这些精细数据实际上都包含在卫星信号中。之前的地图分辨率只有现在的十分之一。正是这种高分辨率让你能够提取出这些局部模式。而很多决策正是在局部尺度上做出的，所以理解局部尺度的情况非常重要。

Right, all that fine grained data is actually in the satellite signals. There have been previous maps, this one is 10 times finer resolution. And that's what really enables you to pull out these local patterns. And that's where a lot of the decision making happens, is at local scales. So understanding things at local scale is really important.

把这些综合起来看，如果我们展望未来，这个系统有实时功能吗？能否在非法砍伐发生之前就进行阻止？

Putting all of this together, if you're sort of long into the future, right, is there like a real time aspect to this? Could you prevent illegal logging from happening before it did?

这是个很好的问题，因为我们到目前为止讨论的工作主要是利用长期卫星数据来评估当前情况或预测未来趋势

It's a really good question because the work that we've been talking about up to now tends to be taking satellite data from an extended period to estimate how things were either right now or looking into the

过去。回顾性地。

past. Retrospectively.

没错。而实时警报以及更长期的未来预测这些想法，正是我们真正的增长领域之一。实际上已经有一些组织正在使用卫星尝试实时捕捉森林砍伐警报，所以这确实存在。然而，其中一个挑战是，为了捕捉到大部分的真正森林砍伐，他们不得不采用一种方法，但这种方法同时也会产生很多误报。

Right. Whereas this idea of real time alerts and then also longer term future forecasting is some of the real growth areas for us. There are already actually existing organisations that are using satellites to try to pick up deforestation alerts live, so that that actually exists. One of the challenges, though, is that in order to pick up most of the real deforestation, they've had to use a methodology that picks up a lot of false positives at the same time.

我明白了。

I see.

所以这实际上非常具有挑战性。所谓的森林砍伐警报，虽然是一项了不起的成就，但同时也感觉还有改进的空间。显然，如果我们能够将警报的准确性大大提高，那么人们根据这些警报采取行动就会容易得多。

So it's actually really challenging. So there's so called deforestation alerts. It's still an incredible achievement, but at the same time, it feels like that could be improved. So obviously, if if we could get it down to the point where the alerts were much more accurate, then it would be much more easy for people to act on them.

我猜这就不只是卫星数据那么简单了。我的意思是，你们必须开始处理其他类型的数据了，对吧？

I guess this goes beyond just satellite data then. I mean, you you have to start working in other types of data, wouldn't you?

是的。我，嗯，我想是的。我的意思是，有趣的是，你可以建立一个变化模型，用过去的数据来训练它，你可以非常科学地做到这一点，进行所有的保留数据验证以及所有那些聪明的机器学习工作。但问题是，另一方面，围绕气候变化和环境变化的整个情况是，未来与过去不同。所以我们能确信可以将那个模型重新应用到未来吗？

Yes. I I well, I think so. I mean, the interesting thing is that you can produce a model of change, and you can train that on past data, and you can do that very scientifically, and you do all of your held out data and your validation and all that clever machine learning stuff. The problem is, on the other hand, that the whole thing around climate change and environmental change is the future is different to the past. So can we be confident that we can reapply that model into the future?

这不同于预测一周后的天气。整个基础模型都不同了。

This isn't predicting the weather a week in advance. The whole underlying model is different.

没错。当我们审视，比如说，我们一直在讨论的森林砍伐过程时，这是一个人为驱动的过程，而人类可以非常迅速地改变行为以应对市场变化等因素。也许大豆价格上涨会促使更快的森林砍伐。另一方面，政府法规和政策或地方法规政策等也是如此。因此，传统的机器学习方法将很难应对这种情况，因为它们只能处理数字，并且只能基于过去的数据。

That's right. And when you look at, say, this process of deforestation that we've been talking about, that is a human driven process, and humans can change behaviour very, very quickly in responses to things like changes in the market. Perhaps the price of soybeans goes up and then that encourages more rapid deforestation. On the other hand, government regulation and policies or local regulation and policies and so on. So the traditional machine learning approaches are going to really struggle to deal with that because they can only deal with numbers and they can only deal with the past.

我认为这就是为什么我们对于将模拟建模的科学传统与机器学习、数据驱动的预测方法进行某种程度的融合感到非常兴奋的原因之一，尤其是像Gemini这样的大型语言模型，它们可能能够捕捉到政府政策的变化、新闻、媒体信息，或许还能整合农产品价格等因素，帮助我们至少稍微调整并识别出这些未来预测周围的一些不确定性情景。

And I think that's one the reasons why we're really excited about hybridising, in a sense, the best of the scientific tradition of simulation modelling with the machine learning, data driven approach to forecasting, with actually things like the large language models like Gemini, which potentially could pick up on, you know, changes in government policy, news, media, perhaps integrate together things like agricultural prices and so on to help us at least to slightly adjust and identify some of the sort of uncertainty scenarios around those future predictions.

如果那是植被，比如树木、树篱和灌木，那么我们也很关心生活在其中的生物。你们如何将这一点纳入考虑呢？

If that's vegetation, though, if that's like trees and hedgerows and shrubs, what about I mean, we also care quite a lot about the creatures that are living inside them. How do you bring that in?

栖息地是一回事，但大多数物种无法从太空看到，它们也不构成那种意义上的栖息地。相反，我们关注的是昆虫、真菌、鸟类等等。我们对生物多样性的关注以及保护行动的很大一部分实际上都集中在这些物种上。所以你之前听说没有公认的森林地图时看起来很惊讶，这是事实。但世界上大多数物种的地图也是如此。

So habitats is one thing, but then most species are not visible from space, and they don't form a a habitat in that sense. Instead, it's all the insects, the fungi, the birds, etcetera. So much of our concern around biodiversity and so much of the action around conservation is actually around those species. And so you looked surprised earlier when you said there isn't an agreed map of forests, and that's true. But that's also true for most of the species in the world.

再次强调，我想让大家真正认识到，我们在生态学和生物多样性方面所做的一切都建立在过去组织和辛勤工作的巨大基础之上。有一个很棒的组织叫IUCN，它为全球14万个物种提供了分布图，这非常了不起。但另一方面，这些地图非常粗略，它们是基于专家意见绘制的。

Now, again, I want to it's really important to realize everything we're doing in ecology and biodiversity builds on this huge foundation of organizations and hard work from the past. So there's an amazing organization called the IUCN that does provide maps for 140,000 species worldwide. It's amazing. On the other hand, those maps are very coarse. They're driven by expert opinion.

这些专家确实非常专业，但这些地图必然相当粗糙。比如大约50公里见方的网格单元，而且大约每十年才更新一次。

Those experts, they they really know what they're talking about, but necessarily, these are quite coarse maps. So 50 kilometer or so cells, and they're only refreshed about every ten years.

哦，哇。

Oh, wow.

是的。所以理论上，利用最新的人工智能和遥感技术等，显然存在一个潜在的作用，可以制作出好得多的地图。因此，我们也在探索这方面。

Yeah. So there's an obvious potential role there for producing, in theory, much better maps using the latest in AI and remote sensing and etcetera. So we're sort of exploring that as well.

好的。但具体怎么做呢？我的意思是，因为你没有相当于卫星的东西，你知道，在太空中拍摄昆虫的照片。

Okay. But then how do you do it, though? I mean, because you're not you haven't got the equivalent of satellites, you know, in space photographing insects.

没错。所以你可以把它看作一种预测问题。你的意思是，我确实拥有的信号，比如卫星信号和其他输入数据，可以帮助我产生一种概率性估计，预测某物存在的可能性，而这通常总是存在相当大的残余不确定性。那么，你如何着手做这件事呢？有趣的是，在最基础的层面上，从某种意义上说，它还是那种老一套的机器学习方法。

That's right. So you could view it as a sort of prediction problem. What you're saying is the signals I do have, like the satellite signals and other input data, can help me produce a sort of probabilistic estimate of how likely something would be there, and there's probably always quite a large residual uncertainty on that. So how do you go about doing that? Interestingly, at the absolutely courses level, it's kind of the same old, in a sense, machine learning approach.

换句话说，你获取你的训练数据，你获取像卫星这样的输入数据，然后引入一个特定的模型，拟合模型，得到输出，再评估它。明白吗？就是经典的那一套。但在这个案例中，这真的很有挑战性，因为我们引入的数据大部分是公民科学数据。嗯。

In other words, you take your training data you take your input data like the satellites, and you you bring in a special model, you fit the model, you get the output, you evaluate it. You know? It's the classic stuff. But the in this case, it's really challenging because the data we're bringing in mostly is citizen science data. Mhmm.

所以这是人们在野外注意到某些东西。A 是的。没错。有一个不可思议的平台叫iNaturalist。它现在就在你的手机上，如果你看到了什么，可以拍张快照，然后上传到iNaturalist。

So this is people out in the field noticing something. A Yeah. That's right. There's an incredible platform called iNaturalist. It's on your phone now, and if you see something, you can take a snapshot of it, and then you can upload that to iNaturalist.

而iNaturalist实际上有自己的机器学习模型，所以他们可以将图像分类到物种级别，这太棒了。并且，现在全球这些实地物种观察的主要来源就是这个名为iNaturalist的神奇手机应用。所以这非常棒，但实际上这些数据仍然非常有偏见且零散。

And iNaturalist actually have their own machine learning models, so they can classify that image to species, which fantastic. Is And this is now the main source of these on the ground observations of species globally is coming from this amazing phone app called iNaturalist. So that's fantastic, but actually that data is still very biased and patchy.

为什么？

Why?

例如，人们通常在离家较近的地方拍照，所以人们去的地方、去的时间通常是在好天气，以及他们选择观察和拍摄的对象。因此你会发现，你会得到很多在晴朗天气下城市附近色彩鲜艳鸟类的照片。而你得不到的是在偏远地方下雨时许多小棕色蘑菇的照片。虽然我是在开玩笑，但这实际上还有一个更严肃的方面，那就是当你从全球范围看公民科学数据的分布时，世界上有很多地区数据非常少，比如撒哈拉以南非洲，因为实际上iNaturalist需要人们有一定量的闲暇时间才能真正进行这类活动。

So for example people typically are close to their home when they take a picture, so where people go, when they go, they're typically on a nice day, and also what they choose to look at and take pictures of. So what you'll find is you'll get lots of pictures of brightly coloured birds close to cities on sunny days. And what you don't get is many pictures of small brown mushrooms in remote places when it's raining. Although I'm joking about that, there's actually a much more serious element to this too, which is when you look globally at the distribution of citizen science data, there are major areas of the world with very little, Sub Saharan Africa for example, because actually iNaturally depends on having a certain amount of leisure time really to do these sorts of activities.

是的，还有网络连接速度和智能手机摄像头的质量。

Yeah, also speed of internet connection and quality of your smartphone camera.

没错，这在某种程度上具有双重讽刺意味。一是从基础生态学角度，如果你不小心，如果你对此天真，你会得出结论认为所有物种都喜欢住在城市边缘的中产阶级社区附近，对吧？因为那是大部分数据所在的地方。而在全球范围内，还有一个真正的讽刺，那就是我们在许多生物多样性最丰富的地方、在当前生物多样性威胁最大的地方、以及人们生计最依赖生物多样性的地方缺乏数据，这有点糟糕，对吧？所以，在某种意义上，我们在这个领域寻求深度学习的魔力，就是看我们是否能以某种方式从这些非常零散的有偏见的数据中泛化，推断出全球所有物种在生态上合理的分布。

That's right, and this is a real irony then for two reasons in a way. One is from the fundamental ecology, if you weren't careful, if you were naive about it, you would conclude that all species like to live near to middle class neighbourhoods on the edge of cities, right? Because that's where most of data is. And then globally, there's a real irony there, which is we're lacking data in many of the places that have the most biodiversity, in many of the places where the current threats to biodiversity are the greatest, and where people's livelihoods depend most on biodiversity, which is just kind of awful, right? And so the magic, in a sense, that we're looking for from deep learning in this space is to see if we can somehow generalize from this very patchy bias data that we do have to ecologically plausible distributions for for all of the species worldwide.

这是一个巨大的挑战。目前，我认为你会说这是否可能尚无定论，但我们看到了一些证据。

And and there's a big challenge. And at the moment, I think you would say the jury's out as to whether it's possible, but we are seeing some evidence.

但是，你们所学到的、现在在生态商业（注：可能指生态学应用或相关领域）中大致了解的关于植被的知识，是否会反馈到你们预期物种所在的地方？我的意思是，就像一种知识帮助另一种？

But does the things that you've learned and that you now sort of know in adverbial commerce about the vegetation then feed into where you expect the species to be? I mean, just like one sort of help the other?

这是一个非常好的观点。我们对生态学的了解足以知道，像气候、海拔和栖息地这样的因素是非常强的物种预测指标。所以你完全正确。我们已经讨论过，我们可以从太空对栖息地进行精细分类。因此，我们预计这本身就能高度预测哪些物种在那里。然后，引入所有这些观测数据，我们就可以在有数据的区域观察这些因素之间的重叠，并希望能获得足够普遍的理解，从而能够在全球范围内重新应用。

That's a really good point that we know enough about ecology to know that actually things like climate and elevation and habitat are very strong predictors of species. So you're absolutely right. We've already talked about the fact that we can classify habitats from space down to a fine grain. So that itself we would expect to be highly predictive of which species are there. So then bringing in all this observation data, then we can look at the overlap between these in the areas where we do have data and hopefully pick up a general enough understanding to then be able to reapply that globally.

这里还有另一个挑战，那就是生态学中到处都是这种长尾分布。你有少数常见物种，然后有很多非常稀有的物种。所以我们需要在生命分类树上做同样的事情。我们能否从数据丰富的更常见物种中学习关键关系，从而能够为实际上数据相当匮乏的物种做出我们所能做的最佳预测。例如，如果我们引入物种性状数据，我们可能会发现，在生命树的这个部分，比如说，是体型较大的物种倾向于生活在较冷的环境中。

There is another challenge here, which is you have these long tail distributions in ecology all over the place. You have a few common species and then lots of very rare ones. So we need to do the same thing across the taxonomic tree of life. Can we learn the key relationships with the more common species that are data rich, enabling us to make the best predictions we can for species actually fairly data poor. So for example, if we bring in species trait data, we might find out that in this part of the tree of life, let's say, it's the larger bodied species that tend to live in the colder environments.

我们或许能够在数据丰富的地方，比如欧洲和美国，发现类似这样的规律，有朝一日，我们可以将这些规律以一定的置信度应用于数据稀缺的地区，比如全球南方。这并不是说我们会盲目猜测，而是实际上这会降低这些地区的数据需求。你知道，一旦我们掌握了一些更普遍的规则，就意味着相对较少的观测数据可能就足以大致确定该物种的分布范围。这真是一个令人兴奋的领域。

And we might be able to discover laws like that in places where we do have lots of data, like Europe and The US, in a way one day where we can apply that with a certain level of confidence in areas where we have much less data, like the global south. Not that we'd ever have a blind guess, but what it would really do is lower the data requirements in those areas. You know, once we've learned some of the more general rules, it would mean that actually a relatively small amount of observational data may be enough to sort of anchor the distribution of that species. Such an exciting space.

所以几年前，我想是这个播客的第二季，我们采访了你的一些同事，他们正在处理塞伦盖蒂的一些数据。表面上看，他们的一些目标与你相似，即了解栖息在该特定区域的物种类型。实际上，我有一段梅雷迪思·帕尔默的录音片段，她是一位保护科学家，谈到了这个项目。请听一下。

So a few years ago, I think series two of this podcast, we spoke to some of your colleagues who are working with some data from Serengeti. On a surface level, some of their ambitions were similar to yours, which was to understand the types of species that were inhabiting that particular area. In fact, I have a little clip of Meredith Palmer, who's a conservation scientist, talking about this project. Have a listen to this.

人工智能在解决这类问题上的能力让我感到震惊。我有过这样的时刻：我看着一张图像，计算机识别出了一个物种，而我第一眼甚至都没注意到它的存在。

It blows my mind how incredible AI can be for solving these kinds of problems. I've had moments where I've looked at an image and had a computer identify a species I, you know, on first glance didn't even recognize was there.

我确实想知道这里是否还有其他类型的数据。我的意思是，你谈到了关于地形、海拔、植被的数据，但是否还有其他数据元素，比如照片，也可以纳入其中？

I do wonder about other types of data here. I mean, you're talking about data about the landscape, about elevation, about the vegetation, but are there other elements of data like photographs, for instance, that you can also bring in here?

是的，有的。总的来说，可以将其视为多模态模型和多模态人工智能的趋势。我认为这在生态学和生物多样性科学领域尤其引人注目，因为你有如此多种类的生物体，它们尺度各异，因此不同的模态适合监测不同种类的物种。他们谈到的是人类有意用相机拍摄的图像。这些是iNaturalist专门处理的类型。

Yes, there are. And in general, could think of that as this trend towards multimodal models and multimodal AI. And that particularly lights up I think in the world of ecology and biodiversity science because you've got such a wide variety of organisms at different scales and so different modalities suit monitoring different kinds of species. They're talking there you've got images that, let's say, humans take deliberately with a camera. They're the ones that say iNaturalist specialises in.

通常，比如在塞伦盖蒂项目和其他项目中，我们谈论的是所谓的相机陷阱。这些是运动激活的相机。很多时候，它们上面有红外灯，所以实际上在夜间拍摄的是红外图像。所以，你知道，可以拍摄不同类型的图像。然后你还有航空影像，比如来自无人机或飞机的。

Often, say in the Serengeti projects and others, we're talking about so called camera traps. These are cameras that are motion activated cameras. And a lot of the time they've got infrared lights on them, so they're actually taking infrared images at night. So even images of you know, there are different kinds of images you can take. You've then got aerial imagery, let's say, from drones or from planes.

我们谈到的还有遥感卫星影像。然后，还有生物声学。生物声学是指你在野外部署麦克风。这可能很多人熟悉，比如那些识别鸟鸣的应用程序，如Merlin和eBird等。

You've got the remote sensing satellite imagery we're talking about. Then, also, you've got bioacoustics. They call bioacoustics is where you're deploying microphones in the field. That might be familiar to a lot a lot of people, for example, around the apps that identify birdsong, Merlin and eBird and similar.

因为那些已经存在了。我的意思是，我稍微玩过它们。对吧？比如，你在外面，你打开它。

Because those those already exist. I mean, I've I've played with them a little bit. Right? Like, you're outside. You turn it on.

它会告诉你正在听的是什么鸟。

It tells you what birds you're listening to.

没错，就是这样。

That that's right.

但那是监督学习，对吧？它能做到这一点。

But that's supervised learning, is it? That manages to do that.

是的，没错。所以你需要一种我们称之为带标签的数据集来开始。你需要一张带有标签的图片，比如这是一只帝王蝶，或者这是一只乌鸦，

Yeah. That that that's right. So you need a sort of what we call a corpus of labelled data to start with. So you need a picture with a label along with it. You know, this is a monarch butterfly or this is a A crow,

乌鸦在走路。

crows walking.

是的，完全正确。所有这类东西。我的意思是，有了足够的数据，你就可以训练一个能够泛化的模型。这就是所谓的监督学习。没错。

Yeah, exactly. All of this sort of stuff. I mean, enough of that, you can then train a model that generalises. So that's what's called supervised learning. That's right.

正是这种技术推动了各个独立模态中的这些发展，无论是图像、遥感等等。

And that's powered most of these developments in each of the individual modalities, whether that's images, remote sensing, etc.

不过你们有一个名为PERCH的项目，对吧？给我讲讲PERCH。

You've got a project though called PERCH, right? Tell me about PERCH.

PERCH属于生物声学领域，在野外部署麦克风的想法非常吸引人，因为几乎总有生物在发出声音。是的。实际上，我们能从声音中提取的信息极其丰富。你可以把麦克风放在野外连续几天，记录所有这些数据。而且声音不仅关乎物种，还可能涉及行为。

So PERCH is in this space of bioacoustics, and the idea of deploying microphones in the field is really attractive because there's almost always something making a sound. Yeah. And it's incredibly rich, the information we can extract from sound, actually. And you can leave the microphones out for days on end to record all that data. And because sound is not necessarily just species, but it can be behavior.

我们或许能区分幼体和成体。你可能捕捉到警告信号。这些昆虫会发声，爬行动物会发声，鸟类会发声，两栖动物也会发声，等等。你可以捕捉到所有这些，而且不需要直接看到生物体。它还能在夜间工作。

We might be able to distinguish juveniles from adults. You might be able to pick up warning signals. So these insects make sounds, reptiles make sounds, know, birds make sound, amphibians make sounds, you know, etcetera. And you pick all of that up, and you don't have to have line of sight to the organism either. It can work at night.

因此，这是一种极具前景的技术，用于从野外提取数据。

So it's an incredibly promising technology for pulling out data from the field.

但如果这已经实现了，我是说如果已经有应用能告诉你听到的是乌鸦，那还有什么需要做的呢？

But if it's already been done, I mean if you do already have these apps that can tell you if you're listening to a crow, then what else is there to do?

PERCH采取了一种略有不同的方法，它是一个自然声音的基础模型。与其他基础模型类似，它不是开箱即用的，不仅仅是提供识别多种鸟类的能力——虽然它确实能做到。但它真正设计的目标是让人们能快速创建自己的检测器，用于检测尚未被纳入现有检测器的内容。例如，如果你研究的是特别稀有的物种，它可能尚未出现在这些应用中，因为它太罕见了。

Perch has taken a slightly different approach, it's a foundational model for natural sound, And what that does is like other foundational models, it's not out of the box, it's not just providing you with the ability to identify lots of different birds, for example. I mean, it does do that. But what it's really designed to do is to allow people to rapidly create their own detectors for things that haven't been in the detector already. So for instance, if you're working on particularly rare species, it won't be in one of these apps already. It's too rare.

他们还没有把它放进去。但你可能也在研究一个更熟悉的物种，不过你想把幼鸟和成鸟区分开，或者比如，一首歌的地方口音。

They haven't put it in there. But you might also be working on a more familiar species, but you'd like to divide the juveniles from the adults or a local accent, for instance, of a of a a song.

鸟类确实有口音，是吗？

Birds do have accents, do they?

哦，没错。嗯，是的。是的。我我觉得是这样。它们的歌声确实有地方性的变体。

Oh, that's right. Well, yeah. Yeah. I I think that's right. They certainly have local variants on their on their songs.

对吧？所以你可以追踪这类事情。

Right? So you'd be able to track these sorts of things.

那么生物声学建模实际上是如何工作的呢？

So how does bioacoustic modelling actually work then?

PERCH的工作方式是引入这些音频数据文件，它们非常庞大。它将所有这些数据压缩成更小量的数据，即所谓的嵌入，然后这个嵌入可以有所有这些下游用途，其中最明显的是开发新的检测器。所以你可以把它看作是一个大型处理过程，输入大量数据，压缩成少量数据，保留所有你想要用于理解生态和识别物种的显著信号，并尽可能丢弃噪音。然而，还有至少另一个重要的用例，叫做搜索。所以如果你取不同的音频片段在这里压缩它们，也可以轻松找到相似的音频片段。

The way PERCH works is it brings in these audio data files, and they're huge. And it crunches all that down into a much smaller amount of data which is this so called embedding and then that embedding then can have all these downstream uses, the most obvious of which is developing new detectors. So you can think of it as a big kind of process that takes huge amounts of data in and crunches it down to a small amount of data that retains all of the salient signal that you'd want for understanding ecology and identifying species and has thrown away as much of the noise as possible. However, there's also, for example, at least one other important use case, which is called search. So if you take different audio clips and crunch them down here, can also then easily go and find similar audio clips.

因为通常一个主要障碍是，假设你想监测一些稀有物种，你把麦克风放置几个小时，那么，啁啾声在哪里？你得听一整天的录音。如果你能在这里找到一两个，那么你可以说，好了，现在我可以找到相似的，因为我在寻找那些有类似PERCH嵌入的片段。甚至还有第三个，我不介意推测一下，例如，我和很多同事一直在讨论的一个想法是，如果你想在实地验证生态恢复，你可以想象，如果我在一个好的地方比如自然保护区放置麦克风，在另一个地方比如我的花园，我一直在重新野化我的花园，对吧？那么随着时间的推移，你可以看到，比如说，我花园中的PERCH嵌入是否变得更类似于当地的自然保护区。

Because what often one of the main barriers is, let's say you're looking to monitor some rare species and you put the microphone out for hours, well, where's the chirp? You've got to listen to the thing for today. If you can just find one or two here, then you can say, right, now I can find the similar ones because I'm looking for the ones that have got a similar perch embedding to these. And even a third one, which I don't mind speculating around, for instance, is an idea I've been talking with a lot of colleagues about is if you're looking to do something like verify ecological restoration in the field, you can imagine if I put microphones in a good place like a nature reserve and in some other place like my garden, I've been rewilding my garden, right? Then through time you can see whether the perch embeddings, say, in my garden become more similar to the local nature reserve.

好的，让我来理解一下。假设你正在进行一个项目，想要增加某种特定物种的栖息数量，你可以在那里放置一个麦克风，通过分析声音来判断你的干预措施是否增加了它们的种群数量，无论具体是什么物种。

Okay, so let me understand that then. So let's say you're doing a project where you want to increase the occupancy of a particular type of species, you can put a microphone in there and work out whether your intervention is increasing the population of them, whatever it might be.

完全正确。至少有两种情况需要考虑。一种是，如果你专注于一个或多个特定物种，那么perch会非常有帮助，因为你可以为这些物种开发检测器。而且我们知道这不仅仅是物种的有无问题，还包括叫声的频率甚至叫声的某些变化。所以在某些情况下，perch甚至能区分个体，顺便说一下。

It's absolutely right. So there's at least two scenarios to think about. One is yes, if you're focused on one or more particular species, then perch can be massively helpful because you could develop detectors for those species. And we also know it's not just a species yes, no, but the frequency of the calls and even some of the variation in the calls. So in some cases, can tell individuals apart, by the way, here.

这样我们就能知道，是一只鸟整天在唱歌，还是有四五只鸟在唱歌？perch甚至能在这方面提供帮助。

So we can say, is it one bird singing all day long, or do we have four or five birds singing? And perch can even help with that.

是那只特定的鸟吗？

What that specific bird?

特定的鸟。哇。好的。没错。是的。

Specific bird. Wow. Okay. Exactly. Yeah.

这真的很重要，对吧，要知道，比如听到了很多黑鸟的歌声，但到底是一只黑鸟还是十只黑鸟？因此，通过这种方式，你可以将生物声学与perch结合使用，提高你测量物种存在与否的能力，还能测量丰度以及随时间的变化趋势，看是否变得更加丰富。这是一种情况，一种非常以物种为中心的视角，这也是目前大多数人对待生物多样性和生态系统的方式，通过这种物种透镜，这非常重要。但对我来说也很有趣，我只是在这里分享一个想法，我和很多同事都分享过，就是在世界上一些生物多样性非常高但数据和理解却少得多的地区，你可以做的是，即使无法识别所有物种，也几乎可以将perch的提取视为对一个地方音频生态的提炼。然后如果我选择一个好地方，比如一个本地自然保护区，再选择一个我想要恢复的地方，我就可以总体上判断，perch的嵌入是否变得更加相似？

It's really important, right, to know, you know, got lots of blackbird song, but is it one blackbird or is it 10 blackbirds? So So in this way, you can use bioacoustics in combination with perch to increase your ability to measure not just the presence or absence of species, but the abundance and then the change in that through time that is getting more abundant. So that's one scenario, a very species focused view, and this is how most people currently approach biodiversity and ecosystems is through this species lens, which is very important. But it's also interesting to me, and I'm just I'm sharing an idea here, right, I've shared with lots of colleagues, which is in some areas of the world where you've got a really high biodiversity and much less data, much less understanding, what you could do instead is almost view the perch distillation as a distillation of the audio ecology of a place, even if you can't identify all the species. Then if I pick a good place, like a local nature reserve, and I pick a place that I'm looking to restore, then I could just say, on aggregate, are the perch embeddings becoming more similar?

我可能不知道涉及哪些物种。有所有这些声音在发生，你知道，噼啪声、咔嗒声、口哨声等等。但在某种意义上，perch是一种深度学习的数值答案，来回答这个问题：这个地方的自然声音听起来是什么样的？

And I may not know which species are involved. I've got all these things going, you know, pops and clicks and and whistles and and etcetera. But in a sense, what perch is is a kind of deep learning numerical answer to the question, what does the nature sound like in this place?

嗯，不过我想问一下，因为，好吧，perch这个名字听起来很像主要是针对鸟类的，但你能把它应用到其他环境中吗？比如水下？它在那里能用吗？

Well, me ask though, because, okay, so perch, I mean, given the name, sounds quite a lot like it's mainly focused on birds, but can you put this into other environments? Like what about underwater? Does it work there?

你说得对，perch最初只是用来分类大量鸟类物种的，但令人惊讶的是，当他们在水下用这些水听器测试时，它似乎效果非常好。

So you're right, perch started by just classifying very large numbers of bird species, but amazingly, it turned out that then when they tried it underwater on these hydrophones, it seemed to work really well.

什么？同一个

What? The same

模型？完全相同的模型。而且确实有效。我认为这告诉我们的是，自然声音及其进化方式存在某种特性，当然不会完美，但在水下具有惊人的高度可迁移性。你知道，这涉及到整个声学生态学领域，有证据表明物种在音频频谱的使用上已经相互进化分离。因为很容易忘记，你在生物声学中捕捉到的大部分声音都是生物有意产生的，对吧？

model? Exactly the same model. Worked And really I think what this teaches us is that there's something around natural sound, the way that that's evolved, that actually has now, of course, won't be perfect, but has a a surprisingly high degree of transfer underwater. And that's you know, this is whole area, right, of acoustic ecology, and what we think, there's evidence for the fact that species have evolved away from each other in terms of their use of the audio spectrum. You know, because easy to forget that most of the sound you're picking up in bioacoustics is deliberately produced by organisms, right?

这是交流。这才是奇妙之处。它们在彼此交流。就像我们一样，它们需要考虑频谱带

It's communication. That's what's amazing. They're communicating with each other. And just like us, they need to think about the spectral bands in

哦，它们希望被听到。

Oh, they want to be heard.

是的。音频频谱带。当然，它们是分化的。所以看起来

Yeah. Audio spectral bands. Of course, they're differentiated. So it looks like

所以有些使用低频声音，有些使用高频声音，对吧。因为你想确保你的沟通是成功的。

So some use low sounds, some use high sounds Right. Because you wanna make sure that your your communication is successful.

是的，完全正确。所以总的来说有高有低，然后你考虑时间因素，知道吗，是像这样还是那样？所以你可以想象这某种程度上是作为一种互动演化而来的，不仅仅是物种内的交流，而是所有物种之间的。它们都以某种方式相互响应，并最终形成了这些不同的模式。

Yes, exactly right. So overall high and low and then you think about the temporal thing, know, is it like or is it oh oh thing? So you can imagine this has somehow evolved as an interaction, not just within species communication, between all the species. They've all responded to each other somehow and settled down on these different patterns.

当你在水下放置麦克风时，你发现了什么？从这些水听器中你得到了什么？

And what do you find when you put microphones underwater? What do you get from these hydrophones?

所以我们实际上已经准备了几段来自珊瑚礁的录音，我想听听看。

So we've queued up actually a couple of sound recordings from coral reefs that I'd like to listen to.

好的，开始吧。好的。我是说，哦，有一些咕噜声，吱吱声。还有噼啪声。

Okay. Go for it. Okay. I mean, oh, some grumbles going on, squeaks. Crackly as well.

然后这是另一段。

And then here's another one.

安静多了。没有咕噜声。我是说，这对这个珊瑚礁来说意味着什么呢？

Much quieter. No grumbles. I mean, what does that say about this coral reef though?

所以现在听到第一个是比第二个健康得多的珊瑚礁，可能不会让你感到惊讶。哦，真的吗？那种生态声音的多样性和丰富性，动物交流的丰富性，是生态系统健康的有力指标。即使作为人类，我们也能听出这种差异。

So maybe now it won't surprise you to hear that the first one of those was a much healthier coral reef than the second. Oh really? So that diversity and richness of ecological sound, animal communication, is a strong indication of a healthy ecosystem. Even as humans we can hear that difference.

哇，所以你确实能从音频信号中如此丰富地理解环境本身。

Wow, so you really have that richness of understanding the environment itself just from the audio signal.

没错。这说明了即使只是整体信号，我们也能以某种方式嵌入，从而大致了解生态系统的健康状况。当然，有了更具体的物种数据，我们实际上可以开始梳理出单个物种。所以如果我们特别关注濒危物种，甚至可能是入侵物种，我们可以提取出动物行为的季节性和日常时间模式。我们可以开始了解物种栖息地的地理差异，所有这些只需将这些麦克风——在这个案例中是放入海洋——就能实现。

That's right. And that's an illustration of how even just that overall we can embed that overall signal in a way that can just give us a general indication of the health of ecosystems. Of course, with more specific species data, we can actually start to tease out individual species. So if we're particularly concerned with endangered species, maybe even invasive species, we can pull out seasonal and daily temporal patterns in animal behaviour. We can start to understand geographic differences in where species live, all just by dropping these microphones, in this case, into the ocean.

但这带来了可能性的爆炸式增长，因为如果以前你把麦克风放在水下，你必须听无数小时的录音，而且没有真正的方法，我的意思是，基本上是人类大脑试图识别模式。但现在如果你能像识别出单个物种那样，甚至识别出特定个体的独特声音。是的，我的意思是，这潜力巨大。

But this is an explosion in possibilities, because if previously you stuck a microphone underwater, you have to listen through hours and hours and hours of tapes and have no real way of like, I mean, it's basically the human brain trying to pick up on patterns. But I mean, now if you can like single out, not just individual species, but individual voices as it were from particular individuals. Yeah. I mean, that's gigantic potential.

是的，绝对巨大。这就是为什么我们这么多人对于生物声学与深度学习这样结合的力量感到如此兴奋，尤其是Perch带来的这种生物声学的基础性方法。它至少有两个你已经暗示的优势。一个是它能极大地加速人类所能做的事情。

Yeah. It's absolutely huge. That's why so many of us are so excited by the power of sort of bioacoustics combined with deep learning in this way, and especially with this, like, foundational approach to bioacoustics that Perch is bringing. And it has at least two advantages that you've hinted at there. One is that it can just massively accelerate what humans could do.

所以你可以处理人类能解决的问题。只是你需要成百上千的人听数千小时的音频。所以，你知道，这非常非常低效。但你说得对。还有一些领域你可能超越人类的能力。

So you could take a problem that humans could do. It's just that you'd have to have hundreds of people listening to thousands of hours of audio. So, you know, it's very, very inefficient. But you're right. There are also areas where you may go beyond human capabilities.

你知道，人类不一定擅长分辨自然声音或通过声音识别物种。这是一项相当难学的技能，可能有些情况比如区分不同个体，最终可能对人类来说过于微妙而无法做到。

You know, humans are not necessarily all that good at passing out natural sounds or identifying species from sound. It's quite a hard skill to learn, and there may be cases like pulling out different individuals. It might eventually be even too subtle for humans to do.

在水下待一会儿。我是说，有一些物种，比如鲸鱼和海豚，有证据表明它们的交流方式相当复杂。对吧？你能用这些想法来研究它们吗？能不能，我不知道，学会说海豚语？

Staying underwater for a moment. I mean, there are some species, I'm thinking about whales and dolphins here, where there is some evidence that their communication is sophisticated. Right? Can you use these ideas for that? Can you sort of, I don't know, learn to speak dolphin?

我愿意相信，将来我们能够借助人工智能与动物交流，像PERCs这样的嵌入基础建模方法在这方面会非常重要。当然，你还需要其他一些元素。有趣的是，我的一些同事参与开发了一个名为Dolphin Gemma的项目，这是一个大型语言模型，经过调整后适用于解码海豚的交流。它接收声音，将其分离、标记化，基本上将其带入大型语言建模的世界。所以这是一个早期的例子，但这是人工智能被积极用于研究动物交流的一个实例，其水平是我们以前无法达到的。

I like to think that we will be able to talk to animals using AI at some point and that these embedding foundational modelling approaches like PERCs would be really important in that. But of course you'd need some other elements there. Interestingly, some colleagues of mine have been involved in producing this thing called Dolphin Gemma, which is a large language model that's been adapted to be suitable for beginning to decode dolphin communication. It takes the sounds, separates them out, tokenises them and basically brings it into the world of large language modelling. So that's an example of it's early days but that's an example of AI actively being used to study animal communication at a level that we really couldn't do before.

问题是，我能看到这在科学兴趣方面的潜力，但我确实想知道，假设我们达到了能理解海豚在说什么、与地球上更高级的动物交流的程度，这是否会改变我们如何看待自己以及我们在世界上的位置？

The thing is, can see the potential for this in terms of scientific interest, but I do wonder whether, let's say, we get to a point where you can understand what dolphins are saying, communicate with the higher animals on the planet. Does it change how we view ourselves and our place in the world?

我认为，是的，我认为它绝对有这种潜力。正如我提到的，我们现在做的大部分工作是在填补已知过程中的信息空白等等。但有时你会想，真正的改变可能来自那些觉醒的时刻，人们几乎一夜之间改变与自然的关系，而这可能会在长期带来影响。我认为过去至少有两个例子。一个是第一张从月球拍摄的地球照片，历史学家可能会对此有争议，但他们经常将现代保护运动的一些近期进展追溯到那张照片，你知道，人们第一次看到地球，意识到在这个黑暗的宇宙中只有一个地球，我们所有人都共享它，等等。

I think, yeah, I think it absolutely has that potential. I mean, most of work we're doing at the moment, as I mentioned, sort of filling known information gaps in known processes, etcetera. But sometimes you think that the real change can come in the long run from these moments of awakening where people almost overnight can change their relationship with nature. And I think we've seen at least two examples in the past anyway. One was the first picture of the Earth on the moon, and I guess historians can debate this, but they often trace some of the more recent increases in the modern conservation movement to that picture, you know, where people looked at the Earth for the first time and realized there was just one in this dark universe and all of us shared it together, etcetera.

另一个实际上是鲸鱼的歌声，即使我们不知道它们在说什么，仅仅是聆听鲸鱼的声音，就真正改变了人们对鲸鱼的看法。所以，能够用像Perch这样的工具为更多物种做到这一点，然后能够解码它们实际在说什么，也许有一天甚至能进行某种对话。如果人工智能能够帮助实现这一点，那可能在长期来看是人工智能最强大的作用。

Another one is actually whale song, and this is just listening to whales, even if we can't tell what they're saying, that really changed the way people thought about whales. So so the idea of being able to do that for more species with things like Perch, but then being able to decode that to say what they're actually saying and maybe one day even have some kind of conversation. And if AI can help to empower that, that might in the long run be the most powerful role of AI.

好吧，我们在这一集中涵盖了很多内容，所以也许我会以关于未来的问题来结束。人工智能将如何改变生态学家能够提出的问题类型？

Well, okay, we've covered a lot of ground in this episode, so maybe I'll finish by asking about the future. How will AI change the kinds of questions that ecologists can ask?

我认为到目前为止，正如我们所讨论的，我们在监测和从田野及文献中提取数据的能力方面取得了惊人的进步，在测绘和地理方面也取得了惊人的进步，而且只会变得更好。我认为未来有几个方向值得探索，一个是如果我们真的能够 confidently 预测生态系统的未来，包括所有植物、所有昆虫、土壤、土壤生物、真菌在一系列不同情景下的情况。如果我这样做，生态系统在十年后会是什么样子。如果我那样做，会发生什么。并且还要考虑到未来的气候变化等等。

I think up to now, as we've discussed, amazing progress in our ability to monitor and extract data from the field and from the literature, amazing progress, I think, in mapping and geo, which is only going get better. I think a couple of the future directions to explore, one is what if we really could confidently predict the future of ecosystems, all the plants, all the insects, the soil, the soil organisms, the fungi under a range of different scenarios. If I do this, this is what the ecosystem will look like in ten years. If I do this, this will happen. And also in a way that took into account future climate change, etcetera.

所以，这些对未来生态系统进行精细尺度、稳健且具有前瞻性的条件预测。如果我们能做到这一点，那将彻底改变我们与自然的关系。我们可以利用这种模拟能力，找到我们应采取的最佳行为方式，识别关键权衡点，决定哪些物种需要保护，构建新型再生农业系统，发展混种农林复合系统，同时推进野化恢复和太阳能碳回收。如果我们能做到这些，那么每一个影响自然或做出我认为会影响自然的决策的人，都将充分了解这些行动的未来后果。在我看来，这具有惊人的潜力。

So these sort of future proof, robust, conditional predictions of the future of ecosystems down at the fine grain. If we could do that, because that could unlock an entirely different relationship between us and nature. We could use that simulation ability to then be able to find all the optimal ways that we should behave, identify the key trade offs, which species to plan, build new kinds of regenerative agriculture systems, agroforestry with mixed species where we've also got rewilding and we're also bringing the solar carbon back. And if we could do that, then every person that's affecting nature or taking a decision that I think affects nature would be fully informed as to the future consequences of those actions. And that feels to me like an amazing potential.

当然，这也取决于价值观，但我希望大多数人平均而言会利用这种能力为自然造福。

Of course, it would depend on values as well, but I would hope that on average, most people would use that ability to do good for nature.

太棒了。德鲁，非常感谢你参加节目。

Amazing. Drew, thank you so much for joining me.

谢谢邀请。真的很棒。

Thanks for the invite. It's been great, really.

我认为人们很容易认为生态学是人工智能的一个小众应用，是一个值得做的项目，排在视频生成和药物发现等重大突破之后。但实际上，它只是比其他领域落后几步。它们仍处于数据收集和数据同化的阶段。但这里的路线图同样雄心勃勃，长远来看，我认为这里真正的潜力不仅在于利用人工智能保护现有的一切，而在于彻底改变我们与自然界本身的关系。您刚刚收听的是由汉娜·弗莱教授主持的《谷歌深度心智》播客。

I think it can be tempting to imagine that ecology is a bit of a niche application for AI, a nice worthy project that sits behind all of the big flashy advances like video generation and drug discovery. But in reality, this is just a few steps behind the others. They are still at the data gathering and data assimilation phase of the process. But the roadmap here is just as ambitious and in the long run, I think there is real potential here not to just use AI to conserve what we have already, but to completely change our relationship with the natural world itself. You have been listening to Google Deep Mind the podcast with me Professor Hannah Fry.

如果您喜欢本期节目，请订阅我们的YouTube频道或在您喜欢的播客平台上留下评论。当然，我们还有更多涵盖各种主题的节目即将推出，请务必关注。下次再见。

If you enjoyed this episode then do subscribe to our YouTube channel or leave a review on your favourite podcast platform. And of course, we have plenty more episodes on a whole range of topics to come, so do check those out. See you next time.