赋予计算机嗅觉

AI 主要通过从数字世界学习而取得进步：文本、图像、音频，以及越来越多的视频。但人们希望 AI 解决的许多问题都存在于这些模态之外的物理世界中。嗅觉就是最有趣的例子之一。动物通过嗅觉检测疾病、识别食物、导航环境，并通过化学物质进行交流。然而，气味在很大程度上仍处于计算领域之外。因为与语言或图像不同，从来就没有一种实用的方法可以大规模地将其数字化。Alex Wiltschko 是 Osmo 的创始人兼 CEO，也是前 Google DeepMind 研究员，他正在努力改变这一点。他的团队正在构建所谓的“嗅觉智能”——能够建模、预测和设计气味的 AI 系统，同时创建将气味带入数字世界所需的数据集和基础设施。在这次对话中，我们探讨了如何让计算机拥有嗅觉、为什么气味是一个如此困难的 AI 问题，以及它教会了我们关于下一代基础模型的什么。以下是 Alex 的分享。

AI has advanced primarily by learning from the digital world. Text, images, audio, and increasingly video. But many of the problems people want AI to solve live outside these modalities in the physical world. Smell is one of the most interesting examples. It's how animals detect disease, identify food, navigate environments, and communicate through chemistry. Yet, scent has remained largely outside the reach of computing. Because unlike language or images, there has never been a practical way to digitize it at scale. Alex Wiltschko, founder and CEO of Osmo and a former Google DeepMind researcher, is working to change this. His team is building what they call olfactory intelligence, AI systems that can model, predict, and design scent while creating the data sets and infrastructure needed to bring smell to the digital world. In this conversation, we explore what it takes to give computers a sense of smell. Why scent is such a difficult AI problem and what it teaches us about the next generation of foundation models. Here's Alex.

这个星球上 99% 的物种只能用化学语言交流，对吧？想想细菌、真菌、植物和昆虫，它们只能用分子对话。我认为非常值得将其他这些外星形式的智能加入我们的 AI 模型。而实现这一点的方法，就是用那些其他智能体的智力输出——也就是化学——来训练它。那就是空气中的感觉。它们由生物产生，目的是相互交流。

99% of species on this planet can only speak with chemistry, right? Thinking of bacteria and fungi and plants and insects, like they just only can talk with molecules. And I think that it's really worth adding other kind of alien forms of intelligence to our AI models. And the way to do that is to train it on the intellectual output of those other intellects, which is that's chemistry. That's the sense that's in the air. They're produced by living things for reasons to talk to each other.

我是 Sam Chington，这里是 Twimmel AI 播客。十多年来，我一直在通过这样的对话探索塑造 AI 未来的想法和创新，帮助你理解什么是真实的、什么是下一步、什么才是重要的。让我们开始吧。当我想到让计算机拥有嗅觉时，有两个角度。一个是，世界上有某种气味，我希望我的计算机能像我一样识别它。另一个，我认为更接近你在 Osmo 的工作，至少最初是这样，就是让计算机深入理解气味的概念，从而能够创造新的气味。

I'm Sam Chington and this is the Twimmel AI podcast. For over a decade, I've been exploring the ideas and innovations shaping the future of AI through conversations like this one that help you understand what's real, what's next, and what matters. Let's jump in. When I think about giving computers a sense of smell, there's kind of two angles to this. One is, you know, there's some scent out in the world and I want my computer to be able to recognize it the same way I do. And the other, which is, I think, more along the lines of what you're working on at Osmo, at least initially, is to have the computer kind of grok the idea of scent so that it can create new ones.

要让计算机处理气味，大致需要三个步骤。首先，你要读取世界——也就是把原子转化为比特和信息。然后，你要映射它，理解它——能够数字化地操作、编码、发送，就像 JPEG 和 RGB 那样，对吧？最后，你还要能把它再写出来——比如打印机、显示器或扬声器。所以我们在 Google Brain 关注的是缺失的那一块，对于气味来说，就是映射。颜色已经有了映射。

Any scent that's been given to computers, there's three kind of broad steps. You got to read the world. So, like turn atoms into bits and information. You have to map it, like understand it. So, you know, be able to manipulate it digitally, encode it, send it, and that's like JPEG and RGB, right? And then you have to be able to write it back out again, right? So, a printer or a display or a speaker. And so the thing we focused on at Google Brain was the missing piece, which is for scent is the map. So, color has had a map.

某种表示，但具体是从什么到什么？

Sort of representation of well, what to what though?

完全正确。所以我们从侧面来看：视觉是怎么做的？听觉是怎么做的？对吧？我们早就有了映射。声音的映射只有一个维度——从低频到高频，说起来很简单。而颜色是三个数字，就是 RGB，或者你喜欢的任何色彩空间。但那三个数字告诉你如何处理颜色。我们的眼睛里有三个颜色信息通道。当然，你简化了很多，因为对于其他每种模态，都有很多不同的映射。

Exactly. Exactly. So let's approach it from the side like how did this work for vision? How did this work for hearing? Right? We've had maps for a long time, right? So the map for sound is just one dimension. It's low to high frequency. Really simple to say. And then for color it's three numbers. It's RGB, right? Or whatever your preferred color space is. But those three numbers tell you how to deal with color. There's three channels of color information in our eye. But of course you're simplifying a lot because for each of those other modalities, there's lots of different maps.

那些只是例子……

Those are just examples of...

它们是例子，而且某种程度上可以相互转换，但我这里是在掩盖几个世纪的心理物理学，任何了解这些的人都会冲我大喊大叫。但请原谅这些简化。我也会简化气味的东西。如果有人像我这样说话，我也会去找他们算账。所以，是的，有 CMYK、Lab、HSV，很多不同的映射。还有更复杂的映射。

They're examples and they can kind of be translated into each other, but I'm papering over like centuries of psychophysics here and anybody who knows anything about those things is going to come screaming at me. But you'll have to forgive the simplifications. I'm going to simplify scent stuff, too. And if people talked the way that I'm going to talk, I would come after them, too. So yeah, there's CMYK, there's Lab, there's HSV, there's many different maps. And then there's more complex maps.

我在研究生阶段被一门数字信号处理课伤得不轻，现在全想起来了。

I was scarred by a DSP class in grad school. It all came back in.

没错。你知道，有滤波器组、Gabor 滤波器组，各种表示图像的方法，我超级简化了，对吧？但映射确实存在，我们也知道它们。我们早就知道，能够映射颜色对于构建 CCD 和 CMOS 以及数字成像至关重要。没错。还有打印机，对吧？比如喷墨墨盒里的墨水，我们知道可以组合它们来制造数百万种颜色。眼睛里大致有 RGB 三个颜色信息通道，但鼻子里有超过 300 个嗅觉信息通道。它们到底编码什么仍然是个谜，但维度肯定高得多，至少从通道数量来看是这样。

Exactly. You know, and like there's filter sets, there's Gabor filter sets, there's all kinds of ways of representing images and I'm super simplifying it, right? But like maps, they for certainly they're there and we know them. We've known them for a while and the notion that we can map color has been instrumental in building like CCDs and CMOS and therefore like digital imaging. Exactly. And then also the printers, right? So like the ink in the inkjet cartridges, we know we can combine them to make millions of colors. There's three channels of color information roughly roughly RGB in the eye, but there's over 300 channels of olfactory information in the nose. And it's still a mystery exactly what they code for, but it's certainly much higher dimensional, at least in terms of channel count.

所以你说的鼻子里的通道数量，是指映射到物理结构吗？我可能会说错，但眼睛的话，是不是像视杆细胞和视锥细胞之类的？

So when you say channel count in the nose, is that mapping to physical structures? I'm going to butcher this, but for the eye, I'm like, is it like rods and cones and stuff like that?

完全正确。是的。所以有视杆细胞和视锥细胞，加起来大约有四个通道，大致是 RGB 加灰度。再说一次，这里有很多细节。那些是编码在你基因中的特定受体，在你的眼睛细胞中表达，对光敏感。嗅觉的对应物是那些叫做嗅觉感觉神经元的细胞。这些细胞实际上从大脑开始，穿过你的头骨，直接接触世界。它是大脑中真正离开头骨的两个部分之一。所以当你闻到东西时，你的大脑实际上在物理上接触另一个生物，它释放了一点点自己让你闻。在鼻子内部真正负责嗅觉的部分，也就是嗅觉上皮，有数百万个嗅觉感觉神经元。

Exactly. Yeah. So, there's rods and cones and together there's like four channels there, like roughly RGB grayscale. Again, there's a lot of details there. Those are specific receptors that are encoded in your genes, right? Expressed in eye cells in your eye that are sensitive to light. So, the equivalent for smell are those cells are called olfactory sensory neurons. And those cells actually start in the brain and they poke through your skull and they actually touch the world. It's one of the two parts of your brain that actually leaves the skull. And so when you smell something, your brain is physically touching another living thing that's like let off a little bit of some of itself for you to smell. So there are millions and millions of olfactory sensory neurons in the part of your nose, the inside that actually does the smelling, that's sensitive to smell called the olfactory epithelium.

这里的命名——它们是感觉神经元——是否意味着它们比视杆细胞或视锥细胞更基础？我猜想视杆细胞和视锥细胞更像是超级结构，更大的东西。

Does the nomenclature here that these are sensory neurons imply that they're more fundamental than a rod or a cone, which I'm imagining are more like super structures, like bigger things.

我可能会说错，但视杆细胞和视锥细胞是细胞类型。所以它们是细胞类型的名称。视网膜中还有许多其他类型的细胞，帮助计算这些细胞获得的原始信息。

I'm probably going to mess this up, but rods and cones are cell types. And so those are names for types of cells. And there's many other types of cells in the retina that'll help kind of compute the raw information that these cells get.

所以那些初级感觉细胞的类似物——没有这些细胞，光线就无法转化为意识，对吧？它们就是第一线。嗅觉的第一线细胞叫做 OSN，即嗅觉感觉神经元。这些细胞在顶端基本上把一部分细胞膜塞进一层小小的粘膜中，这层粘膜随后接触外界。那些小顶端充满了称为嗅觉受体的蛋白质，这些才是真正感知化学物质的家伙。

So the equivalent of those primary sensory cells—without these cells, light doesn't turn into awareness, right? So they're the front line. The frontline cells for smell are called OSNs, olfactory sensory neurons. These cells, at the tip, basically shove part of their cell membranes into a little mucous membrane that then touches the world. And those little tips are just chock full of proteins called olfactory receptors, and those are the things that actually sense the chemical.

那么有多少种嗅觉受体类型呢？

And how many of those olfactory receptor types are there?

超过 300 种。这就是那个数字的来源。每种受体类型对化学世界的敏感度都不同，对吧？就像某些视紫红质——那些是真正感知光的蛋白质——有些对蓝色更敏感，有些对绿色更敏感，有些对红色更敏感。而在嗅觉中，有数百种不同的类型。

Over 300. So that's where that number comes from. Each of those receptor types is differently sensitive to the chemical world, right? Just like some rhodopsins—those are the proteins that actually sense light—some are sensitive more to blue, some more to green, some more to red. And there are hundreds of different types in olfaction.

那么这对人类嗅觉系统的分辨率意味着什么？普通人能闻到多少种气味？

So what does that say about the resolution of the human olfactory system? Is there a number of smells that the typical person can smell?

我认为我们没有一个好的估计。有一篇非常著名的论文声称有一万亿种，但已经被彻底驳斥了。我不知道我们是否真的能估算出来。不过，当谈到灵敏度时，有几件事值得注意。想想看：我做过激光眼科手术，所以大半辈子都戴着眼镜。我看得不太清楚——无法分辨非常细微的差别——但我不是盲人。光线能进来，只是没有被正确聚焦。对于嗅觉，只要你的气道正常工作，你就能闻到并检测到东西。你可能无法察觉细微差别，但你绝对对它敏感。有些人不行，他们的气道因为解剖结构或受伤等各种原因而堵塞。Noam Sobel 做过一个惊人的实验，表明人类实际上可以进行气味追踪。如果你戴上眼罩，让人趴在地上，留下一条巧克力或肉桂的小痕迹，人们如果集中注意力，真的可以像狗一样追踪气味。他们非常慢，但能做到。所以这恰恰说明，有一个误区认为我们不擅长嗅觉。我们其实非常擅长嗅觉。对于某些分子，我们能闻到相当于奥林匹克游泳池中一滴眼泪的浓度。我们对某些东西非常敏感。这就是为什么天然气有气味——添加了极微量的称为硫醇的分子，我们能极其敏感地闻到，达到十亿分之一或万亿分之一。太不可思议了。所以人类很擅长嗅觉。

I don't think we have a good estimate for that. There is a very famous paper which claims a trillion, but that has been pretty thoroughly debunked. I don't know if we can really estimate that. There are a few things though when you talk about sensitivity. Think of it this way: I had laser eye surgery, so I had glasses for most of my life. I couldn't really see that well—I couldn't resolve really subtle differences—but I wasn't blind. Light was getting in, but it just wasn't being focused properly. With smell, as long as your airways are working, you can smell and detect things. You might not detect subtle differences, but you definitely are sensitive to it. Some people aren't, and they have their airways closed for various reasons like anatomy or injury. There's this amazing experiment by Noam Sobel who showed that people can actually do scent tracking. If you put on a blindfold and get somebody down on the ground and leave a little trail of chocolate or cinnamon, people, if they concentrate, can actually scent track just like a dog. They're super slow, but they can do it. So that just goes to show that there's this myth that we're not good at smelling. We're freaking amazing at smelling. We can smell the equivalent for some molecules of a little teardrop in an Olympic-size swimming pool. We are so sensitive to some things. That's how natural gas has a smell—the tiniest amount of molecules called mercaptans are added, and we can smell it super sensitively, like parts per billion or trillion. It's crazy. So humans are good at smelling.

所以我们试图构建这张地图。我们有从生物学中了解到的这种结构。我们如何表示这种结构？下一步是什么？

So we're trying to build this map. We have this structure that we know about from biology. How do we represent that structure? What's the next step?

让我带你了解我们是如何考虑做这件事的，或者说我们是如何做到的。我们的第一步是这个问题的非常基础的版本：假设你只闻到一种分子，你知道它的结构——就像在高中化学课上一样，你可以在黑板上画出原子和键——而且你知道它闻起来像什么。所以这组特定的碳、氮、氧组合闻起来像甜味、香草味、酚味、巧克力味。这些可能是从一百个描述符中选出的四个，适用于那个分子。于是我们收集了数千个这样的配对，然后训练一个神经网络来预测从结构到气味的关系。这个问题被称为结构-气味关系问题，一百年来一直未解，在某些情况下人们认为它是不可解的。我们在 Google Brain 的首批成果之一就是训练了一种当时相对较新的神经网络，称为图神经网络，它专门用于化学。它实际上与 Transformer 非常相关——在随后的几年里，它已经被 Transformer 所吸收。我们能够非常好地预测东西闻起来像什么。事实上，好到我们设置了一个气味图灵测试。我们找到了以前没人闻过的分子——有些从未被制造出来，自然界也没有见过。我们提前预测了它们的气味，并保密了预测结果。我们实际拿到了这些分子，送到另一个地方。我有一个很棒的合作伙伴，Monell 中心的 Joel Mainland。我们训练人们闻和描述气味——不需要大量训练，大约八小时就能做得不错，能描述出像草味、酚味、香草味或黄瓜味，大约有 50 个术语。我们进行了双盲实验：人们闻并描述全新的分子。然后我们将这个小组的表现与单个小组成员进行比较——一个人总是比小组的平均水平差。所以我们将其与我们的模型进行比较。问题是：如果你想让你的小组更好，你是愿意再加一个人，还是愿意加上模型的预测？结果发现，我们的模型预测平均优于任何一个单独的小组成员。这意味着我们通过了气味图灵测试——我们的模型预测达到了人类水平，这非常酷。

Let me walk you through how we thought about doing this, or how we have done it. Step number one for us was a really basic version of this problem: assume you're only smelling one molecule and you know its structure—you can draw it on the board like in high school chemistry class with atoms and bonds—and you know what it smells like. So this particular set of combinations of carbon, nitrogen, and oxygen smells like sweet, vanilla, phenolic, chocolate. Those are maybe four descriptors drawn from a set of a hundred that apply to that molecule. So we go get thousands of those pairs and then train a neural network to predict that relationship from structure to odor. That problem, called the structure-odor relation problem, had been unsolved for a hundred years, and in some cases people thought it was unsolvable. One of our first outputs at Google Brain was that we trained a relatively new kind of neural network at the time called a graph neural network, which was specialized for chemistry. It's actually very related to the transformer—it's kind of been subsumed by the transformer in the intervening years. And we were able to predict what things smell like very well. In fact, so well that we set up an odor Turing test. We went and found molecules that nobody had ever smelled before—some had never been made before, nature has not seen them. We predicted what they would smell like ahead of time and kept our predictions secret. We got those molecules physically, sent them to another location. I have a great collaborator, Joel Mainland at Monell. We trained people to smell and describe smell—it doesn't take a ton of training, maybe eight hours to do okay at describing something as grassy, phenolic, vanilla, or cucumber, with maybe 50 terms. We double-blinded the process: people smelled and described completely new molecules. Then we compared how well this panel of people did to individual panelists—one person is always worse than the average of the panel. So we compared that to our model. The question was: if you want to make your panel better, would you rather add another person or add the predictions of a model? It turned out our model predictions were better than any one individual panelist on average. That means we passed an odor Turing test—our model predictions were human quality, which was pretty cool.

我们对神经网络做的是把它拆开，观察所谓的嵌入层。那是神经网络的一部分，基本上将输入转化为一个向量，这个向量就是地图。这张地图我们可以切分成区域，用于分类。所以地图的这个区域是香草味，这个区域是红浆果味，等等。没有它，你实际上无法完成那个分类问题。

What we did with the neural network is we cracked it open and looked at what's called the embedding layer. That's a part of the neural network that basically turns the inputs into a vector that is the map. That map is what we can slice up into regions and use for classification. So this region of the map is vanilla, this region is red berry, etc. Without that, you actually can't do that classification problem.

所以那个嵌入向量，如果你工程做得对，它大概只需要 300 维左右就能工作得很好，这有点可疑，但你知道，只是暗示性的，对吧？我不能下任何定论。但那个嵌入向量内部有大量优美的结构，它似乎是第一个连续的嗅觉预测地图的候选，我们称之为主气味图，从那以后，它一直是我们公司所建一切的基础。

So that embedding turned out, if you do the engineering right, it just kind of needs to be around 300 dimensions to work really well, which is like suspicious but you know, just suggestive, right? I can't make any claims. But that embedding had a ton of beautiful structure in it, and that seemed to be a first candidate for a continuous predictive map of smell, and we called it the principal odor map, and that's been foundational to what we've built at the company since then.

在我们讨论那个结构之前，你提到了图神经元是这里的基本架构。图中的节点和边代表什么？

Before we get to that structure, you mentioned a graph neuron that was the fundamental architecture here. What did the nodes and the edges in the graph represent?

好问题。就像如果你在社交网络图上做机器学习，节点是人，边是人与人之间的关系，比如友谊，然后它可能是一个像 Facebook 或 Twitter 那样非常大的图。在我们的案例中，每个图都是一个分子，分子中的节点是原子。可能是碳、氮、硫或氧。图中的边是化学键，可能是单键、双键或三键。这些图不是很大，因为具有气味的分子不是很大。如果它们很大，实际上无法进入空气并飞走。而且如果它们很大，也无法放入我们之前讨论过的嗅觉受体的结合口袋中。所以它们既不太小也不太大。它们往往少于 20 个原子，但多于三四个。所以这些就是图，就是输入。图神经网络能够基本处理这些信息，传播信息，并基本上收集分子中的宏观结构。然后最终你把它变成一个固定长度的向量，描述那个分子实际可能的气味，这就是主气味图。

Yeah, great question. So like if you were doing machine learning on a social network graph, the nodes would be people and the edges would be relationships between people like friendships, and then it might be one very large graph of like Facebook or Twitter. In our case, every graph is a molecule and the nodes of the molecule are the atoms. So it might be a carbon, it might be a nitrogen, it might be a sulfur or an oxygen. And the edges in the graph are the bonds, and that might be a single bond, a double bond, a triple bond. And the graphs aren't very big because molecules that have a smell aren't very big. If they were huge, they actually wouldn't make it into the air and fly away. And if they were also huge, they wouldn't fit inside the binding pockets of the olfactory receptors that we talked about. So they're not too small, they're not too big. They tend to be less than 20 atoms, but more than like three or four. And so those are the graphs, those are the inputs. And the graph neural network is able to basically process that and propagate information and basically gather the macro structures in that molecule. And then eventually you turn that into a fixed length vector that describes how that molecule actually might smell, and that's what the principal odor map is.

现在谈谈你在嵌入空间中观察到的结构。

And now the structure that you observed in the embedding space, talk a little bit about that.

当然。所以我们取了那个几乎 300 维的向量，然后取了它的二维投影以便观察，我们使用了一种叫做主成分分析（PCA）的方法。当我们绘制这个二维地图时，我们让图上的每个点代表一个分子。所以图上大约有 5000 个点。我们在图上圈出了具有相同气味的分子区域。比如甜味区域、黄瓜味区域等等。这些区域不一定是分开的，对吧？比如所有黄瓜味的分子可能完全分散开，那样每个区域实际上可能相互重叠。但真正美妙的是，如果我们圈出花香区域，它位于左侧一个相当大的部分。我们可以把图片分享给听众，但花香区域很大，在左侧。但如果你再画出茉莉、玫瑰或紫罗兰，它们实际上成了花香内部的子区域。我们并没有告诉它存在这种自然的嵌套关系。类似地，发酵和酒精味的区域在我们第一次模型训练时实际上形状像瓶子，我们后来再没动过它，因为它太有趣了。但像发酵、酒味和朗姆酒味，所有这些发酵酒精气味都在同一个区域。这种模式基本上适用于所有气味。我觉得这非常美妙。这张地图真正让我印象深刻的一点是，我们仍在研究，如果你模糊地看，这些气味几乎像是自然界如何制造这些气味的故事，对吧？所以发酵气味，比如来自葡萄酒或朗姆酒等，是由酵母产生的，实际上是一个生物过程产生这些分子，对吧？所以有一个关于分子如何产生的故事，花朵也一样。它们在遗传上彼此之间的关系比与其他产生坚果或树皮等的物种更密切。它们都聚集在一起。所以从某种意义上说，实际生物生命和气味的故事似乎紧密交织在一起。

Sure. So we took that almost 300 dimensional vector and we took a two-dimensional shadow of it so we could look at it, and we use something called principal components analysis to do that, or PCA. And what we did is when we plotted this two-dimensional map, we made every dot on that plot be a molecule. And so there were about 5,000 dots on that plot. What we did on that plot is we circled regions of molecules that all had the same smell. So the sweet neighborhood, the cucumber neighborhood, that kind of thing. It didn't have to be neighborhoods, right? Like all these cucumber molecules could be completely spread out, in which case every region might actually be all overlapping with each other. But what was really beautiful is if we circled the floral region, it was this pretty big part on the left. We can share the images with the listeners if you'd like, but the floral region was pretty big and was on a side of the left. But then if you draw like jasmine or rose or violet, they actually ended up being sub regions inside of floral. And we didn't tell it that there was this nested relationship of nature there. And similarly, the region for fermented and alcoholic was actually shaped like a bottle during our first model train, and we haven't touched it since because it's so funny. But things like fermented and winy and rum, those all fermented alcoholic scents were all in the same region as well. And this pattern continued for basically all of scent. I thought that was very beautiful. One thing that really struck me with that map, which we're still working out frankly, is these scents actually, if you blur your eyes, it almost looks like a story of how nature makes those scents, right? So fermented scents like from wine or rum etc. Those are made by yeast that are actually a biological process producing these molecules, right? So there's a story of how the molecules are produced, and same with flowers. Those are all genetically much more related to each other than they are to say other species that produce nuts or bark or whatever. And they're all clustered together. So in a way, there's this story of actual biological life and scent that seems to be very tightly woven together.

那里有没有令人困惑的例子？我想，发酵和花香都是天然气味，但有没有其他完全非天然的气味？比如我们添加到天然气中使其有气味的东西。这些非有机气味的形状是什么样的？它们在根本上有所不同吗？

And are there like confounding examples there? I'm imagining, you know, both fermentation and florals are kind of these natural scents, but are there other scents that are like completely unnatural? Maybe the thing that we put in natural gas to give it a smell. And what are the shapes of these non-organic smells like? Do they differ in some fundamental way?

实际上，那是我们开始的第一项业务：我们能否制造出气味好、安全、可大规模生产且价格合理的新分子？事实证明，这是一个非常有趣的行业，因为如果你能制造出气味好但不安全，或者因新法规而被淘汰的分子，我们需要替代品，因为我们希望产品气味好，比如洗衣粉气味好，家里气味好。所以我们实际上使用这些多年来开发的模型来寻找自然界从未存在过的新香料分子，然后制造它们，并将它们推向市场，我们目前正在这样做。

Actually, that was our first line of business that we started: can we somehow make new molecules that smell great, and that are safe, and that we can produce at scale, and that are affordable, all that. And it turns out that's a very interesting business to be in because if you can make some molecules smell great but they're not safe, or they're being removed because of new regulatory action, we need replacements because we want our products to smell great, like our laundry to smell great, our home to smell great. So we actually use these models which we've developed over many years to actually find new fragrance molecules that have never existed in nature before, and then we make them, and we can bring them to market, which we're in the process of doing right now.

当你这样做时，你的目标是什么？我猜想在某个先前的步骤中，你有一个分类器，比如气味好、气味不好，然后你试图映射到那个？或者你可能想要它像花香、发酵味或柠檬味之类的。你是如何引导的？

And when you're doing that, is your objective like I'm imagining that somewhere in some prior step you've got a classifier of like smells great, doesn't smell great, and you're trying to map to that? Or you could be well I want it to be like floral or I want it to be fermented or I want it to be lemony or something like that. How do you guide it?

它总是非常具体，对吧？在我们的行业中，需要制造哪些分子是很清楚的。比如，我们缺少一种持久的柑橘类分子，或者我们缺少一种光学透明的香草香调，因为香草通常是棕色的，你不希望那样；人们想要透明的香料，以便可以改变产品的颜色。还有很多其他类似的情况，但我们大致知道需要制造什么。所以我们通常让团队专注于最高优先级的项目，因为最终是一个团队在做这一切，包括物理基础设施、所有这些模型以及更多的化学信息学。

It's always really specific, right? In our industry, it's pretty clear what the molecules are that need to be made. So, hey, we're missing a citrus molecule that is long-lasting, or we're missing a vanilla note that is optically clear because vanilla is typically brown and so you don't want that; people want clear fragrances so that they can change the color of the product. And there are many others like this, but we kind of know what we need to make. So we typically focus the team, because it's ultimately a team that's doing all this with physical infrastructure, all these models, and a lot more cheminformatics, on the kind of highest priority items.

当然，我们也会在过程中偶然发现一些有趣的东西，而且我们不会忽视它们。最初的分子集是那些已知的、已经记录下来的分子。你不需要去收集什么，或者通过 PCR 之类的处理；我们本来就认识它们。

And of course, we discover some interesting things by accident along the way. And we don't ignore those. The initial set of molecules were these known ones, just written down. You didn't have to collect anything or put it through PCR or something; we just knew what these are.

嗯。

Yeah.

那时候我们还没有实验室，所以收集的数据一部分是通过创意性的数据授权获得的。我们也在网上找到了一些东西。项目启动时，我已经在气味领域工作了大约 15 年，所以我知道该去哪里找这些材料。我们成功了。从那以后，我们大幅提升了数据收集能力。第一个数据集大概有 5000 个分子，而到现在我们已经数字化了 60 亿个分子。

We didn't have any labs at that point in time, so the data we collected partly came through creative data licensing. We also found some stuff on the internet. I had been in the world of scent for like 15 years when we started that project, so I knew where to go to get this stuff. And we were successful. Since then, we've dramatically scaled up our data collection abilities. We had, I think, 5,000 molecules in our first data set. We've digitized 6 billion molecules at this point.

这是什么意思？请带我们了解一下这个过程。

What does that mean? Walk us through that process.

是的。我们列举了所有可能被制造出来的分子——那些真实、物理上可实现、可能具有气味的分子。我们基本上对所有分子都运行了预测模型，并制造了大量分子。这句话的前半部分是说，有一些物理规律支配着分子如何形成。你可以根据一些标准进行筛选，比如三个原子到二十个原子，哪些键能形成、哪些不能，然后你就得到了一个潜在分子的起始列表。还有一个可制造性过滤器。所以你最终会得到这个从第一性原理推导出来的列表。

Yeah. So, we've enumerated all the molecules that could possibly be made — real, physically realizable molecules that could possibly have a smell. We've basically run predictive models on all of them and made a huge number of them. The initial part of that statement is that there's some physics governing how molecules can form. You can filter based on criteria like three atoms to 20 atoms, which bonds work and which don't, and then you get a starting list of potential molecules. There's also a manufacturability filter. So you ultimately end up with this list that you derive from first principles.

很大的列表。

Big list.

一个很大的列表，你基本上是从第一性原理推导出来的，就像这些东西如何组合在一起。是这个意思吗？

A big list that you kind of derive from first principles, like the way these things work together. Is that the idea?

没错，正是如此。所以那些是真实的分子。我们可以制造其中任何一个，而且它们很可能具有气味。至于其他方面——它们是否有用、有趣或美丽——我们有 AI 模型来预测所有这些，还有安全性。安全性非常关键。那是我们的核心数据集之一。另一件事是我们闻了很多气味。我们训练人们去闻。有很多不同的闻香方案：这个气味是好是坏，强烈还是不强烈，这个比那个好吗？有很多不同的方法，我们已经非常擅长调整这些。我们有国际上的闻香人员。我想我们已经数字化了 543 万次闻嗅——这可能是迄今为止用于训练 AI 模型的最大嗅觉数据集。我们必须从零开始构建这一切。没有 Scale AI 或 Mechanical Turk 可以用于嗅觉。我们必须在内部制作、自己消化，并生成海量的嗅觉数据用于嗅觉智能。

That's right. Exactly. So those are real molecules. We can make any of them, and they're more likely than not to be able to have a smell. Then for the rest — whether they're useful, interesting, or beautiful — we have AI models to predict all that stuff, as well as safety. Safety is really critical. That's one of our core data sets. The other thing is we've smelled a lot. We train people to smell. There are many different protocols for how to smell something: is this smell good or bad, intense or not intense, is this better than the other one? There are many different ways of doing this, and we've gotten really good at dialing that in. We have people internationally who smell. I think we've digitized 5.43 million sniffs — the largest olfactory data set for training AI models, I think, ever. We had to make all of that from scratch. There's no Scale AI or Mechanical Turk for smell. We've had to make it internally, consume it ourselves, and generate huge amounts of olfactory data for olfactory intelligence.

最终那个数据集看起来就像一个分子（无论你怎么表示它）和一组人类可能用来标记该气味的标签。

And that data set ultimately looks like a molecule, however you want to represent that, and a set of labels that the human might label that smell.

它可以更广泛。那是我们数据集的一部分——我们知道分子结构，然后闻它并标记它的气味。但也可能像你从杂货店买的黄瓜那样，我们闻它，或者是一个带有分析注释的市场产品。

It can be broader than that. That's part of our data set — we know the molecular structure and then we smell it and label what it smells like. But it might also be like a cucumber you buy from the grocery store that we smell, or a market product with analytical annotations.

好的。所以是一个东西和它的气味量。

Okay. So a thing and a smelly amount.

没错。有时我们不只是闻它；我们还会把它放进分析仪器里。这就涉及到如何获取真实世界的数据——你需要使用化学传感器。我们把大量数据通过化学传感器处理。然后我们可以将这些数据与人类感官标签对齐，从而开始将传感器与人类感知联系起来。这是我们工作的核心。

Exactly. And sometimes we don't just smell it; we put it through analytical machinery. That gets to how you get real-world data — you need to use chemical sensors. We put huge amounts of data through chemical sensors. Then we can align that with human sensory labels so we can begin to relate sensors to human perception. That's core to what we do.

当你在制造这些分子的过程中，是否有已知的毒性筛查可以……

When you're in this part of the process where you're manufacturing these molecules, are there known toxicity screens that you can...

哦，是的。在欧洲、美国和全球范围内，你必须经过一个非常严格的过程。有一整套测试需要提交，它们非常彻底且正确。这对皮肤安全吗？吸入安全吗？对眼睛安全吗？对鱼类安全吗？因为你可能会在用洗发水洗完后，把一些冲进马桶或淋浴排水管。

Oh yeah. You have to go through a very rigorous process in Europe, the US, and worldwide. There's a binder of tests you have to submit, and they're really thorough and the right tests. Is this safe on your skin? Safe to breathe in? Safe for your eyes? Safe for fish? Because you might flush some of it down the toilet or shower drain after washing with shampoo.

我好奇的问题是：你能从分子结构推导出这些吗，还是必须通过经验来做？

The question I'm curious about is: can you derive that from molecular structure, or do you have to do it empirically?

你可以预测，这对我们的效率非常重要。但之后你必须进行物理测试。这是法律要求，也是正确的事情。你只需要检查，做实验，我们对所有通过监管审查的产品都反复这样做。

You can predict it, which is super important for how efficient we are. But then you have to test it physically. It's just the law, and it's the right thing to do. You just check, do the experiment, and we do it over and over for all the products going through regulatory review.

这可能偏离了技术对话，但从商业角度：你是验证气味是否符合客户需求，然后授权给他们，他们找人制造？还是你自己大规模制造气味？

This is maybe a digression from the technical conversation, but on the business side: are you validating that the scent is what the client wants, then licensing it to them, and they find someone to manufacture it? Or are you making the scent at scale?

我们的大部分业务实际上是混合已经存在且已获批的分子和原料。如果有人来找我们，想推出空气清新剂、洗发水或高级香水，为了快速推出，我们需要混合我们能获得的分子。我们库存了许多这样的原料。我们教会了嗅觉智能——一系列不同的模型——如何将客户的请求（比如“我想要一种闻起来非常清新干净、对 Z 世代男性有用或受欢迎的气味”）转化为开始制作气味的规格。然后我们有调香大师和调香师……

Most of our business is actually blending molecules and ingredients that already exist and are already approved. If somebody comes to us wanting to launch an air freshener, shampoo, or fine fragrance, to get it out quickly, we need to blend molecules we can get. We stock many of these ingredients. We've taught olfactory intelligence — a fleet of different models — how to convert a customer's request, like 'I want a scent that smells really fresh and clean and is useful or liked by Gen Z men.' That's enough specification to begin making a scent. Then we have master perfumers and perfumers...

三分之二的斧头和……

Two-thirds axe and...

就把它们混合在一起。我的意思是，你离真相不远了。一般来说，新气味就像现有气味，因为气味是艺术，而艺术是演进的。

Just mix them together. I mean, you're not too far from the truth. Generally, new scents are like existing scents, because scent is art and art evolves.

它不会做那种疯狂的跳跃，对吧？所以，你总能分辨出香味的谱系。比如，Axe 就是其中之一。实际上，Axe 非常有名，因为它是第一次将高级香水带到大众市场的价位。所以，Axe 身体喷雾中的一些香味，普通人以前根本接触不到，因为那些……

It doesn't like take these crazy jumps, right? So, you can always tell what the lineage of a scent is and like, yeah, Axe is one of those, right? Actually, Axe is actually really famous because it was the first time that fine fragrance perfumes were actually brought to that mass market price point. So some of the scents in Axe body spray had never been like accessible to the average person cuz those

我从来没把 Axe 和“高级”这个词联系在一起。

I never associated Axe with fine anything.

对，没错。它多年来建立了自己的名声，但从营销和香味开发的角度来看，这些香味的选择和设计方式其实非常有趣。我们教会了 OI 与调香师合作，也直接与客户合作，你基本上可以提示它，说“嘿，我想设计一款香味”。我们的调香师用它作为灵感，客户有时也会为自己设计香味。事实上，今天早些时候我和一个人聊天，我发给他们一款 OI 设计的香味，他们很喜欢，就选用了，准备推出市场。就像直接从我们的软件系统里出来一样。所以，你可以把它想象成香味的 Canva 或 Figma，你和我都能做出在市场上成功的香味。这就是我们整合起来的东西。这一切的基础是你训练的那个嵌入空间，我觉得你描述的任务有点像词嵌入空间的数学，比如国王减去……我老是记不住那个例子。

So yeah, exactly. It's like it's built its own rep over the years, but like the way those scents were actually chosen and designed were super interesting from like a marketing and kind of scent development perspective. Um but yeah, we we've taught OI to collaborate with perfumers and also just work directly with our customers where you basically prompt it and you're like, "Hey, I'd like to design a scent." Our perfumers use it as inspiration and our customers sometimes design their own scents for themselves, right? In fact, I had conversation earlier today where I sent somebody an OI design scent and they loved it. They picked it. They're going to launch it, right? Like just right out of like our software system. So, you can think of that as like a Canva, right? Or or a Figma for scent where, you know, you you and me can can make scent that that that wins in the market. Um, and so that's what we've put it all together with. So the foundation of this is the you know this embedding space that you've trained and I'm envisioning the the task that you described as being akin a little bit to the word uh embedding space math like uh king minus I always get the

对，就像国王减去男人等于女王减去女人，文本空间里确实有这种算术。

Yeah, it's like king minus man is equal to queen minus woman or something there's arithmetic right in the text space yeah exactly

所以，这真的很相似。我们的核心业务是，如果你想推出一个带有气味的品牌，比如洗发水或高级香水，你需要制作香味。我们确实有工厂来生产，但核心算法流程是：我们接收你的文字、图片或音频描述，然后将其嵌入到一个感知空间中。这中间有很多步骤来确保它准确、商业上可行且可接受。然后，我们需要从那个空间解码回一个配方。这个配方是一套指令，交给工厂里的配方团队和一个校车大小的机器人，它每 100 秒就能制造出一款新香水。基本上就是这样串联起来的。算法当然由数据驱动。每次我们为客户制作新香味，他们高兴，我们也高兴，因为我们得到了报酬，他们的品牌也成功了。但我们还获得了数据，对吧？所以我们总是会闻那些香味，用化学传感器检测它们。这就形成了一个自我维持的良性循环：我们找到了一个既能服务世界、帮助人们快乐、致富并发展业务，又能推进我们使命的业务。我们最后要做的事不会是香水设计，但它肯定是第一步。它既教会了我们，也资助了我们。我觉得这非常强大，我很高兴我们找到了它，因为它对我们想做的事来说意义重大。

So no it's it's really similar right so the core business for us um is, you know, you want to launch a brand that's has some smell in it, like a shampoo or a fine fragrance. You got to get the scent made. Um, so we actually have a factory where we make it, but the the core algorithmic pipeline is we take your text or image or audio description and then we embed it into a perceptual space. And there's there's lots of steps involved in getting that right and having that be commercially like viable and acceptable. Um, but then we then have to decode from that space back to a formula. And that formula is a set of instructions for a formulation team in the factory behind me and a big robot that's the size of a school bus that can make a new fragrance every hundred seconds. Um, and uh, that's basically how all this is strung together. And the algorithms are of course driven by data. And every single time we make a new scent for a customer, like yes, they're happy, we're happy because we get paid, they're happy because their brand is working. Uh, but we also get data, right? So, we're always sniffing those those scents. We're always putting them through our chemical sensors. And so, it's creates this really nice like self-perpetuating machine where we found a business where we can serve the world and and like actually help people be happy and like become wealthy and grow their businesses. Um, but also like it helps us on our mission, right? So, we we the last thing that we do is not going to be fragrance design, but it is certainly the first. And so, it's teaching us and it's also funding us. And I think that's a really powerful like I'm glad we found it because I think it's really really powerful for for what we're trying to do.

但其中有几点我想深入探讨，主要是编码和解码部分。你提到了多模态，这非常有趣。比如，我可以描述这个东西，也可以说“我希望这个香味唤起这个图像”，或者“这个图像唤起这个香味”，方向性可以任意。所以第一个问题是：你是如何从这种分子地图过渡到多模态嵌入空间的？然后在解码端，如果你找到了某个邻域，但嵌入空间本质上还是嵌入分子表征，那么如何从一个点得到“这是这五种东西的混合物”？这也是 AI 完成的，还是更偏向确定性或基于经验的方法？

But there are a couple things in there uh that I wanted to dig into, you know, essentially the the encode and decode parts. Like you mentioned multimodal and that's like super interesting. Like so I can describe this thing but also say like I want you know I want this scent to evoke this image like or this image to evoke the scent whatever the directionality is like. Um and so the the first question is like how did you go from like this kind of molecule map thing to this multimodal embedding space right and then on the decode side. Okay, you know, you found the neighborhood of some, you know, something. If it's if we're still like in the regime where this embedding space is fundamentally embedding like representations of molecules, like how do you get from that from that like a single point to well that's a mix of these five things? Is that also AI or is that more a deterministic or experience-based thing?

好问题。这里面有几个点。首先，我们发现的这个地图——我们系统已经进化了很多。我接下来要说的只是概要，实现和工程细节已经和最简单的描述方式有所偏离。这个地图大致是一个 300 维的主序图。我们训练的第一个模型确实以单个分子作为输入，但你没有理由不能用其他输入来定位那个嵌入，对吧？比如化学传感器？你可以砍掉那个神经网络的头部，把它接到另一个处理化学传感器读数的神经网络上。答案是：当然可以。所以问题不在于如何把所有东西压缩到原始输入空间，而在于如何将新输入映射到同一个空间，我们已经做到了。然后还有一个问题：如何处理混合物？这是圣杯，我们投入了大量精力、时间和科学研究。我为我们已经做的、正在做的以及未来的工作感到非常自豪。但真正的秘诀在于如何推理这些分子之间的相互作用。

Great questions. There's like a few things buried in there, right? So the the first is um the map that we discover like we've evolved the system a lot. So I'm what I'm going to say is like the gist um but the implementation and engineering details have kind of diverged from what like the simplest way to describe it but like here goes that map that roughly 300 dimensional principal order map. The first model that we trained does in take does indeed take uh single molecules as inputs, but there's no reason that you can't target that embedding with other inputs, right? There's no reason you can't target it with a so like what about a chemical sensor? Could you just like chop off the head of that neural network and like you know put it on another neural network that took in like chemical sensor readings? And the answer is like yeah, you can for sure do that. So the the question is not so much like how do you crunch everything down to the original input space. It's more like how do you map new inputs into that same space and and we've been able to do that. Um and then there's a question of how do you deal with mixtures and like that is the holy grail and we've spent a ton of energy and time and science on how to do that. And I'm super proud of what we've done, what we're doing, what's still ahead. Um, but like the special sauce is is really there is like how do you reason about how these molecules interact? And um

这是个好问题，但我不会得到答案。

That's a good question, but I'm not going to get the answer.

不是详细答案。对，来 Osmo 工作吧，一起开拓前沿。但说真的，这确实是秘诀。最终，驱动它的数据才是真正的护城河。我认为我们拥有世界上最大的嗅觉数据集。但更重要的是，我们生成数据的速度远超其他任何人。即使是一些百年老公司，他们声称有数据，但不过是大量 Excel 表格，实际上没什么用。我们从一开始所做的一切，都是为了收集数据用于分析和模型训练。这带来了天壤之别。

Not a detailed answer. Yeah, come come work at Osmo and push back the frontiers. But no, it's like it is the special sauce. Um, ultimately like truly like the data that drives it is actually the the moat. Um, and so I think we have the largest alactory data sets in the world ever. Um, but more importantly, the rate at which we're generating data just far outstrips anybody, right? Like even the companies have been around for hundred years, like they say they have data, but they have is a lot of Excel spreadsheets that don't actually kind of add up to much. Um, and so everything that we do from the very beginning has been designed to be collected for analysis and for model training. Uh, and that makes all the difference. Yeah.

所以数据和模型训练表明，这种映射确实是学习出来的，是另一个模型，而不是……

So data and and model training suggests that that mapping is indeed a learn thing. it's another model as opposed to,

你知道，关系没那么复杂，或者是业务规则……

you know, the relationship's not that complex or it's business rules or

绝对不是业务规则。

definitely not business rules.

我的意思是，就像任何 AI/ML 项目一样，你总是先从最简单的模型开始，看看它能做到什么程度。然后，在最有趣的情况下，你会发现其实那也不算太错，对吧？只是你希望它更细致一些。而那些经得起时间考验的直觉非常棒。但这实际上是一个收集为 AI 量身定制的超大数据集，然后在其上做算法处理的过程。比如我们大约四年前创办公司时，知道如何在这种数据模态上构建特定类型的模型非常重要，这实际上成了我们的加速器。但说实话，只要你有优质数据，很多东西就会自然涌现。不过每一步你都得小心，并且要有优秀的人来思考每一步。

I mean, like, you know, as with any AI/ML project, you always start with the dumbest models first and see how far it gets you. And then, in the most interesting cases, you realize that's actually not that wrong, right? It's just you'd like to be a little bit more nuanced with it. And those are great sensibilities that stand the test of time. But it's really an exercise in collecting very large data sets tailored for AI and then what you do algorithmically on top of that. Like when we started the company almost four years ago, knowing how to build specific kinds of models on this data modality was super important and that's actually been an accelerant for us. But really, if you just have great data, a lot of stuff kind of falls out from it. But you have to be careful with every step of the way and have great people who are thinking about every step too.

那么图模型在你们的工作中仍然占重要地位，还是说你们已经在合适的应用中转向了 Transformer？

And does graph models still figure significantly into what you're doing or have you evolved to transformers in the right applications?

我们对采用的建模方法并不教条。我们真正执着的是数据规模和数据质量。

We're really not dogmatic about the modeling approach that we take. We're really dogmatic about data size and data quality.

这意味着你会为每个项目启动新模型，而不是像我设想的那样，你们有一个嗅觉基础模型，那个模型不常更新，然后直接用于新任务。

There's the implication then that you spin up new models for individual projects as opposed to what I envisioned was that you had like the olfactory foundation model and that thing changes infrequently and you just use it for a new task.

当我们说嗅觉智能时，我们指的是关于气味各个方面的预测模型套件或车队，有几十个模型。如果你去看看 Neolabs 或生命科学公司，有些公司很坦率地说，他们的核心模型或基础模型实际上是一个模型车队。自动驾驶汽车也是这么工作的。自动驾驶汽车由一组模型驱动，这些模型沿着一个主干连接在一起，协调感知、行动和规划等所有环节。所以我认为我们的思考方式更接近自动驾驶汽车，这其实也是我 CTO 的背景。Rich 负责过 Nvidia 自动驾驶汽车项目的数据架构。我们在 Twitter 工作时认识，后来他去了 Spotify 和 Brex。

When we say olfactory intelligence, we kind of mean the suite or the fleet of predictive models on all aspects of smell, and there are dozens of models. If you go look at Neolabs or life sciences companies, some of them are pretty upfront that their core model or foundation model is actually a fleet of models. That's how self-driving cars work too. Self-driving cars are driven by a fleet of models that all connect together along a spine to coordinate sensing with action with planning and all that stuff. So I think the way we think about it is closer to autonomous vehicles, which is actually where my CTO's background is. Rich ran the data architecture for Nvidia's autonomous vehicles program. We met at Twitter when we were working there, and then he went to Spotify and did Brex.

这本身就很有趣，并引出了一个问题：在自动驾驶领域，存在一种张力，一方面是那些以某种方式忠实于我们对物理世界或物理过程的理解的模型，或者经典控制方法，另一方面是端到端方法。‘我不在乎那些东西。我不会去手动调整子系统。我只会用数据进行端到端训练。’听起来你更倾向于‘我们对嗅觉系统或开发这些气味的过程有物理理解，我们会使用子组件或子模型’这一派。

That in and of itself is interesting and opens up this question: in AV there's this tension between models that are somehow faithful to the way we understand the physical world or physical process, or classical control, versus end-to-end. 'I don't care about any of that stuff. I'm not going to human tinker subsystems. I'm just going to end-to-end train on data.' It sounds like you're more in the camp of 'we've got this physical understanding of olfactory systems or some process for developing these scents, and we're going to have subcomponents or submodels.'

是的，我会说一部分对，一部分不对。不对的部分是，我们并不一定追求物理理解。我们追求预测准确性，帮助客户，并建立预测正确事物的能力以维持组织运转。有时我们会使用物理学，但那只是一个建模选择，比如使用基于物理的模型。而且我认为我们实际上无法做一个完全统一的模型，因为还有监管工作要做。所以你需要一个预测来回答‘这个东西在特定方面是否安全’。因此你至少需要——它可能是一个有 12 个头的模型——但你必须有那些特定的输出，并且需要用特定的数据集来训练它们，它们实际上在那个体制下变成了独立的模型。

Yeah, I would say one part of that's true and one's not. The part that is not true is that we don't necessarily strive for physical understanding. We strive for predictive accuracy and helping our customers and building the ability to predict the right thing to keep the organization going. Sometimes we use physics, but that's just a modeling choice, like using a physics-based model. And I would say we actually cannot do one fully unified model because there's regulatory work that has to be done. So you kind of need one prediction for 'is this safe in this specific way.' So you have to have at least, you know, it could be one model with 12 heads, but you do have to have those specific outputs, and there are specific data sets that need to be used to train those, and they effectively become independent models in that regime.

所以这些子模型——可以这么说吗，核心是预测气味，但然后你还有这些其他的头或属性也要预测，比如毒性、可制造性、制造成本、监管安全性等等？你不能跳过任何步骤，它们都是必需的。但核心是：它闻起来像什么？它好闻吗？它足够浓烈吗？这些是你需要的基本东西。然后其他一切你都可以解决。

So the submodels are like—is it fair to say the core is predicting a smell, but then you've got these other heads or attributes that you're also trying to predict, which are toxicity, maybe manufacturability, maybe cost to manufacture, maybe regulatory safety, all that stuff? And you can't really skip any steps, they're all required. But the core is: What does it smell like? And does it smell good? And does it smell strong enough? Those are kind of the core things that you need. And then everything else you can figure out.

它好闻吗？这是‘它闻起来像什么’的一个派生属性，还是你也会独立于它的气味来预测？

Does it smell good? Is that a derived property of 'what does it smell like' or do you also predict that independent of what it smells like?

这是气味元组与目标消费者的派生属性。举个例子：你可以选择帕尔马干酪、泡菜或草莓。取决于你展示给谁，他们可能喜欢也可能不喜欢。就草莓而言，实际上日本人们想到或想要的草莓种类与美国人们想到或想要的非常不同。那些是送礼用的，但光是味道，它的甜度特征就不同。在如何将其构建到产品中时，它几乎像是不同的水果。但你喜欢什么很大程度上取决于你之前接触过什么，以及接触时的感受。如果你在韩国家庭长大，接触泡菜是在温暖的家庭环境中。如果你是一个从未体验过的非韩国人，你可能会想，‘我打开了什么？’我碰巧喜欢泡菜，但那是一种后天习得的 taste。同样，在我的文化中，有各种辣根、吉拉鱼和面条 kugel 等等。我喜欢那些东西，但那是我族人的食物。不是每个人都喜欢，但我很热衷。

It's a derived property of the scent tuple with the target consumer. So, an example is: you could either choose parmesan cheese or kimchi or strawberry. Depending on who you show it to, they may or may not like it. In the case of strawberry, actually the kind of strawberries that people think of or want in Japan are pretty different from the ones that we think of or want in the US. Those are the gift ones, but just the flavor, right, it's a different sweetness profile. It's almost borderline different fruit in terms of how you construct it in a product. But what you like is heavily driven by what you've been exposed to before and how you felt when you got exposed to it. If you're in a Korean household, being exposed to kimchi is done under warm family conditions. If you're a non-Korean person who's never experienced it before, you might think, 'What did I just open?' I happen to love kimchi, but it was an acquired taste. Similarly, from my culture, there's all kinds of horseradish and gila fish and noodle kugel and all this stuff. I love those things, but that's my people's food. Not everybody likes it, but I'm into it.

这就引出了关于味觉的问题，它比……更接近嗅觉。

Which raises the question about taste, which is kind of more adjacent to smell than it is to...

从物理上讲，它们非常接近。是的。

Physically it's pretty close. Yeah.

那么这是你们涉足的领域吗？还是说它离得足够远，你们不考虑？

And so is that something that you dabble in? Is it far enough away that you don't think about it?

我们经常考虑它。但我们目前不从事味觉方面的工作，因为我们非常专注于我们的第一个垂直领域，我们必须专注。

We think about it a lot. We don't work in taste today because we're really focused on our kind of first vertical, and we've got to focus.

这方面大概有三个层面。嗅觉当然是指从鼻子吸入的气味，但还有一种反向的嗅觉，叫做鼻后嗅觉。吃东西的时候，会产生一种烟囱效应，让气味反向逸出。这就是风味的来源。如果你捏着鼻子吃一颗软糖，就尝不出是柠檬、青柠还是西柚味，只能感觉到甜。风味 90% 来自嗅觉，只有 10% 来自味觉。味觉只是舌头上发生的事：甜、鲜、酸、咸、鲜味和苦味。但从维度和丰富度来说，它没那么复杂。信息通道更少。如果闻不到，风味体验就完全被破坏了。问问那些因新冠失去嗅觉的人，或者试过捏着鼻子吃东西的人就知道了。所以风味和香气密切相关，它就在那里等着我们去探索。

So there are kind of three aspects to this. There's smell, which is for sure just what comes in your nose, but also you can smell things that go the other way. It's called retronasal action. When you're eating something, you're actually creating this chimney effect that vents scent the reverse way. This is how flavor is produced. If you ever eat a jelly bean and hold your nose, you can't tell if it's lemon or lime or grapefruit; it just tastes sweet. Flavor is 90% smell, 10% taste. Taste is just what happens on your tongue: sweet, savory, sour, salty, umami, and bitter. But in terms of dimensionality and richness, it's just not as complex. There are fewer channels of information. The experience of flavor is destroyed if you cannot smell. Just ask people who lost their sense of smell in COVID or who've tried holding their nose while eating. So flavor is super related to fragrance, and it's kind of there waiting for us when we're ready.

他们只有六种，我们有三种。

They've only got six. We've got three.

抱歉。下次你可以请一位味觉专家上播客，让他们为自己辩护，但信息通道确实更少。如果闻不到，风味体验就完全被破坏了。问问那些因新冠失去嗅觉的人，或者试过捏着鼻子吃东西的人就知道了。所以风味和香气密切相关，它就在那里等着我们去探索。

Sorry. You can have a taste person on your podcast next and they can defend themselves, but there are just fewer channels of information. The experience of flavor is destroyed if you cannot smell. Just ask people who lost their sense of smell in COVID or who've tried holding their nose while eating. So flavor is super related to fragrance, and it's kind of there waiting for us when we're ready.

我们已经聊了很多关于这一切的发展方向和可能性。还有什么补充吗？

We've talked a lot about where this is all going and what's possible. Any additional thoughts on that?

我想让计算机拥有嗅觉，这意味着读取、映射和编写气味。我们一路走来非常务实：公司前两年建立科学能力，然后找到一个很好的商业应用场景，就是香精香料行业。但我们还会走得更远。我们正在和许多组织洽谈。我认为现在需要的不仅是气味设计，还有气味检测，这正是你之前提到的：我们必须构建一个基础模型。缺失的东西，也是我们一直在构建的，就是必须收集大量数据。比如，如果你想做一项非常崇高的工作——通过闻癌症患者的气味来早期检测疾病，或者早期检测疟疾或其他主要夺走儿童生命的传染病的感染。

I want to give computers a sense of smell, and that means reading, mapping, and writing smell. We've been on a journey, super pragmatic: building the scientific capabilities in the first two years of the company, then finding a great business to put this to use in, which is the fragrance industry. But we're going to go further than that. We're talking with a number of organizations. I think the thing needed now for not just design of scent but detection of scent is exactly what you were talking about: we have to build a foundation model. The thing that's been missing, and that we've been building, is you have to collect a whole bunch of data. For instance, if you wanted to do the very noble work of smelling someone with cancer to detect their disease early, or detect early infection with malaria or another infectious disease that claims the lives of mostly kids.

这非常有趣，把这些看作数据收集问题。

That's super interesting, thinking about those as data collection problems.

问题是这样的。你可以直接去做，收集患病和未患病者的气味数据，然后尝试构建模型。但你永远找不到足够多的人来构建一个好模型。获取那么多数据实在太难了。

Here's the deal. You can go directly after that and collect scent data from people with or without those conditions and try to build models. You're never going to get enough people to build a great model. It's just hard to get that much data.

是因为信号不足，气味和疾病或其他因素之间的相关性不够强？还是仅仅因为数量太少？

Meaning because there's not enough signal and the correlation between scent and disease or some other factor? Or just low numbers?

这纯粹是一个统计学和机器学习问题。我认为基本上没有一个明显的信号说这个人有病。变化很微妙，涉及许多、可能成百上千个分子信号。我们就是不知道。但我们知道狗能做到。动物确实能检测到这些模式。所以肯定有东西存在。但计算机做不到。

It's just from a pure statistics and machine learning problem. I think basically there's not an obvious signal that says this person has a disease. There are subtle changes across many, potentially hundreds or thousands of molecular signals. We just don't know. But we know dogs can do it. Animals can actually detect these patterns. So there's something there. It's for sure there. But we can't get computers to do it.

对，你提到的时候我就在想这个。

Right, that's what I was thinking of when you raised it.

肯定存在。但计算机做不到。我们只需要获取海量数据。我们需要团结起来，投入巨大努力，去闻一万个人。我不在乎他们是健康还是生病，也不在乎年龄。当然，我们要记录所有信息，但关键是获取海量数据。我们去杂货店拿黄瓜、香蕉、鲜花、牛排，什么都行。我们去闻所有东西，然后构建一个巨大的嗅觉数据集，记录世界的气味。如果我们能做到，那么……

It's for sure there. But we can't get computers to do it. We just need to go get a ton of data. We need to band together and build a huge effort where we just go sniff 10,000 people. I don't care if they're healthy or sick, or how old they are. Let's record all that information, of course, but let's just go get a ton of data. Let's go to the grocery store and get cucumbers, bananas, flowers, steaks, and whatever. Let's just go smell everything and then build a huge olfactory dataset of what the world smells like. If we can do that, then...

听起来比再爬一次 Reddit 难多了。

It sounds a lot harder than scraping Reddit one more time.

比爬 Reddit 难多了。但值得，因为除非我们团结起来，否则根本没人会去做。这值得做。顺便说一句，我在 AI 和机器学习领域有很多朋友。我最初是作为生物学家进入这个领域的。作为生物学家，你要做实验，很难，你要去实验室自己获取数据，很艰苦。在 AI 领域，你基本假设数据就在那里，如果不在，你假设可以给某人一张信用卡来创造数据。这很棒。但你能获得的基础设施数量惊人。你必须明白，如果你想继续扩展 AI 和算力所能及的领域，可能会变得困难。你可能会遇到没有供应商能为你处理一切的情况。你可能得自己动手。我们正处于这种状态。我喜欢这样。太难了，极其艰难。但我喜欢，因为没人在做。这恰恰是我想要的生活方式：我想做那些没人做的、有意义的新奇事。这是我的选择标准。有很多人可以构建客户服务的 AI 语音助手。我认为我们需要它来改善商业，但我不是做那个的人。我想做这件新奇的事。我想让计算机拥有嗅觉。

Whole lot harder than scraping Reddit. But it's worth it because literally nobody's going to do it unless we pull ourselves together. It's worth doing. As an aside, I have a lot of friends in the world of AI and ML. I came in through this as a biologist. As a biologist, you go do experiments, it's hard, you go to the bench and get your own data, it's gritty. In the world of AI, you kind of just assume the data is out there, and if it's not, you assume you can give somebody a credit card to create it. That's amazing. But the amount of infrastructure you have access to is astoundingly huge. You just have to understand that if you want to continue to expand the realm of what AI and compute can do, it might get hard. You might reach a point where there's not a vendor that can handle everything for you. You might have to go do the work yourself. We're very much in that regime. I love it. It's so hard, beyond brutally hard. But I love it because nobody else is doing it. That's very much my preference for how I want to spend my life: I want to do weird things that nobody else is doing that matter. That's my selection criteria. There are plenty of people who can go build AI voice agents for customer service. I think we're going to need that to make commerce better, but I'm not the guy to do that. I want to do this weird thing. I want to give computers a sense of smell.

所以，我认为我们的方向是收集海量数据，但重要的是，我们拥有一个平台，可以存储并组织所有这些数据，实现数字化的规模效率，然后逐渐在物理层面，我们可以制造大量独特的嗅觉样本。我们花了几年时间辛苦搭建了所有这些基础设施，现在就去获取所有的嗅觉数据吧！让我们把化学这一现实切片融入 AI，构建一个真正的 foundation model。但数据是第一位的，对吧？我们需要先去获取所有数据。然后，每当这些模型发布时，比如 Opus 47、Mythos 或 Codex 55，它们都会在模型卡上列出各种基准测试。所以我设想的未来是，我们收集所有这些基础数据，疟疾检测会成为一个基准，癌症检测会成为一个基准，在马来西亚为新品洗发水打造出色的市场产品也会成为一个基准。但更进一步，通用 foundation model——我指的是通用目的的模型——不仅限于文本和常见模态，现在还要有关于嗅觉和味觉的基准或模型卡声明。

And so, you know, I think where we're going is we are collecting huge amounts of data, but importantly, we have this platform that allows us to store all this data, organize it, like massive efficiencies of scale digitally, and then increasingly physically, we can make a huge number of unique scents. Like we have all this infrastructure that we've built painstakingly over a few years, and now let's go get all the smell data, right? And let's build this chemical slice of reality into AI, which let's build a true foundation model. But the data comes first, right? We need to go get all the data. And then, you know, whenever these models come out, like Opus 47 or when Mythos comes out or Codex 55, they talk about the model card in terms of all these benchmarks. So what I imagine for the future is that we'll collect all this foundational data, and malaria detection will be a benchmark, all right, and cancer detection will be a benchmark, and you know, building a great market product for a new shampoo launch in Malaysia will be a benchmark. But even the idea that the generic foundation model—by generic I mean kind of general purpose, for lack of a better term—is not just text and the common modalities, but now has some kind of benchmark or model card statement around smell and taste.

是的，就像我们要添加这些奇怪的新基准，比如我们能在多大程度上预测某物的气味，或者我们在世界中检测到的某种气味——它只是分子的组合——如何预测我们关心的事情。

Yeah, it's like let's add these weird new benchmarks, which is how well can we predict what something smells like, or how some smell that we detect in the world—which is just a combination of molecules—is predictive of something that we care about.

我喜欢用另一种方式来看待这项工作：文本和图像的 foundation model 正在接近，有时甚至超越人类智能——在某些领域肯定超越了我。我一直在推动它，然后心想，天哪，这比我还聪明了。这是人类智能，因为它是在人类智力产出上训练的。但这个星球上 99% 的物种只能用化学语言交流，对吧？想想细菌、真菌、植物和昆虫——它们只能用分子交谈。我认为非常值得将其他外星形式的智能加入我们的 AI 模型。方法是在那些其他智能的智力产出上训练模型，也就是化学。那是空气中的感觉，是生物为了相互交流而产生的。你知道，著名数学家陶哲轩在他刚写的一篇论文中有一段话，他反思这些新形式的 AI 以及它们能做的奇怪数学。他说，我认为我们越来越需要一种哥白尼式的智能观。意思是，在天文学中，很长一段时间我们以地球为中心，但后来不得不推翻它来面对事实。地球是个好地方，但它不是宇宙的中心。同样，我们把人类智能放在了关于人工智能的辩论中心。做人类很酷，很棒，我喜欢。但为什么我们应该在中心？还有哪些其他形式的智能？当我读到他的文字时，立刻产生了共鸣，我非常认同这种观点：我们应该有哥白尼式的智能观。智能有如此多的形式，有如此丰富的智能事物，有些我们甚至可能不认为是智能。我认为，如果我们为人类和地球的未来构建 AI，就应该把这些形式的智能内置进去。

One other way I like to think about this work is like foundation models for text and for images—they are approaching, in some cases exceeding, human intelligence, certainly mine in certain regimes. I've been pushing it and I'm like, holy crap, this is smarter than me now. It's human intelligence, and that's because it's trained on human intellectual output. But 99% of species on this planet can only speak with chemistry, right? Thinking of bacteria, fungi, plants, and insects—they only can talk with molecules. I think that it's really worth adding other kinds of alien forms of intelligence to our AI models. The way to do that is to train it on the intellectual output of those other intellects, which is chemistry. That's the sense that's in the air. They're produced by living things for reasons to talk to each other. And you know, Terry Tao, this famous mathematician, he has this one paragraph in a paper he just wrote where he's reflecting on these new forms of AI and the kind of weird mathematics that they can do. And he says, I think increasingly we need to have a Copernican view of intelligence. Meaning for a long time in astronomy, we had the Earth at the center, but then we had to dislodge it to actually face facts. Earth's a great place to be, but it's not the center of the universe. Similarly, we've placed human intelligence at the center of the debate about artificial intelligence. It's pretty cool to be human, it's awesome, I love it. But why should we be at the center? What other forms of intellect are there? So when I read what he wrote, it instantly resonated with me, and I've very much taken on this view that we should have a Copernican view of intelligence. There are so many forms of it, and there's so much richness out there of things that are intelligence, some of which we might not even recognize as such. I think if we're building AI for our species and for the planet going forward, we should have those forms of intelligence built in.

想想我们在广义的物理 AI 方面取得的进展——机器人、自动驾驶汽车等等——再想想我们的传感器在近似视觉刺激方面有多好。你了解嗅觉摄像头或芯片上的鼻子发展到什么程度了吗？

Thinking about all the progress we've been making in broadly physical AI—robots, autonomous vehicles, all that—and thinking as well about how good our sensors are for this approximation of visual stimuli. Do you have a sense for where we are with olfactory cameras or a nose on a chip?

我想说，我们用来检测化学物质的设备确实非常出色，但它们一直非常专门化，只能待在实验室里。所以想想 1970 年代的计算——大型主机困在后台办公室或数据中心。所以我们在化学传感和嗅觉智能方面仍然处于数据中心时代。我们的气味打印机有一辆校车那么大。它根本不会离开，是固定在地上的。

I would say the devices that we use to detect chemistry are truly phenomenal, but they've been very specialized to live in a laboratory. So think about computation in the 1970s—big mainframes stuck in a back office or a data center. So we're still in the data center era of chemical sensing and of olfactory intelligence. Our scent printer is the size of a school bus. It's just not going to leave. That's bolted into the concrete.

嗯。

Yeah.

但最终所有这些都应该能放进我们的手机里，对吧？随着时间的推移。这就是技术的故事：某样东西有效，有价值，然后我们努力把它做小。所以这将会实现。

But eventually all this should fit in our phone, right? Over time. And that's the story of technology: something that works, it's valuable, and now we just work really hard to make it small. So that will come.

癌症的例子很好。如果我们手机上有辅助嗅觉装置芯片，它还能让我们做到哪些鼻子做不到的事情？

The cancer example is a good one. What are other examples of if we had the auxiliary olfactory apparatus chip on the phone, what would it let us do that our noses don't do for us?

我非常兴奋的是，如果我们向所有人开放，我认为会带来的创造力。我们怎么可能预测到 App Store 上的所有应用、摄像头手机的所有用途？这太不可思议了。我认为肯定有一些事情我们可以考虑：狗鼻子能做的一切。

I'm super excited by the creativity that I think would come if we opened that up to everybody. How could we have predicted all the apps on the App Store, all the uses of camera phones? It's incredible. I think there are definitely things we can contemplate: everything that a dog's nose can do.

对。我在想那个你下载后能告诉你谁放了屁的应用。

Right. I'm thinking of the app that you download that tells you who actually dealt it.

没错。好吧，今天的气味指纹肯定是 Sam 的。我有数据证明。所以未来肯定会有放屁应用。我只是——这是人性。我告诉过你我在 Twitter 工作过，我们清理过。我就是了解人性。你知道，我们对很多事情着迷，包括粪便学。但我认为狗鼻子能做的任何事情——大概有 80 种左右你可以训练狗做的事情，比如检测。

Exactly. All right, well, the old fingerprint is definitely Sam today. And I've got data to prove it. So there'll definitely be fart apps for sure in the future. I just—it's human nature. I've told you I worked at Twitter and we clean up. I just know human nature. It's just, you know, we've got a fascination with many things including the scatological. But I think that anything a dog's nose can do—and there are like 80 things more or less that you can train a dog to do, like detect.

比如在手机上追踪。

So tracking, for example, on your phone.

没错。

Exactly.

太棒了。

That's amazing.

变质的农产品、变质的肉。这些是你想在冰箱里知道的事情，对吧？你想知道，我能吃这个吗？我应该吃这个吗？它好了吗？煮熟了吗？这对机器人厨房很有用。人类鼻子和狗鼻子能做的事情太多了。我认为，到目前为止，我们看到的这些例子中，没有一个单独能成为巨大的业务或一定可访问的业务。

Spoiled produce, spoiled meat. These are things you want in your fridge, right? You want to know, can I eat this? Should I eat this? Is it done? Is it cooked? That's useful for robot kitchens. There are so many things that human noses and dogs' noses can do. None of the examples alone, I think, are so far that we've seen huge businesses or necessarily accessible businesses.

闻出癌症很重要，但商业上真的很艰难，对吧？医疗诊断不幸地不是一个好做的生意。问题不仅在于如何造出来，还在于如何让它自我维持。这是我努力应对的难题之一，坦白说我没有答案。

Smelling cancer is important but the business is really rough, right? Medical diagnostics are unfortunately not a great business to be in. It's figuring out how to not only build it but then make it self-sustainable. That's one of the harder questions I wrestle with, and I don't have the answer frankly.

你觉得这个问题的难度是否在抑制技术的发展？如果商业模式到位了，我们能轻松搞定芯片上的鼻子吗，还是说它本身就非常难，经济门槛非常高？

Do you feel like the difficulty of that question is suppressing the development of the technology? If the business model was there, could we easily figure out the nose on a chip, or is it just really hard and the economic bar is really high?

我觉得是混合的。如果商业模式解决了，我想过去四年我们会全力冲刺这个，而不是在看似无关的香氛行业建立很好的业务——但深入科学和技术后，它们高度相关。我们有设备可以让人来体验或推着走，能完成这个任务。问题只是如何投入适当资源来规模化和小型化。非常可能。我很容易想象一个像 80 年代手机大小的盒子，比手机大一点，但能像人鼻一样闻。事实上，我们很接近了。我们的设备大约两个鞋盒大小，有一个可部署的传感器，性能和你我鼻子一样好，而且它工作。把它做小是直接的工程问题，不容易但直接。再小到手机大小，然后集成到芯片，就需要一些创造力和工作。我无法确切告诉你如何实现。我们有证据表明它存在于我们的眼睛和鼻子之间——我们就是活证据，但要让它负担得起且普及，还有很多工作要做。

I think it's a blend. If the business model was figured out, I think we would have been sprinting at this for the last four years exclusively, as opposed to building a really great business in the fragrance industry, which seems unrelated, but when you get down to the science and technology, it's super related. We have devices that you can bring people to or wheel around that can do this. It's just a matter of how to throw an appropriate amount of resources behind scaling and miniaturizing that. Very possible. I can easily envision a device the size of an 80s cell phone, a bigger box, that can smell as well as a human nose. In fact, we're pretty close. Ours is about the size of two shoe boxes. We have a deployable sensor that's as good as our nose, and it's the size of two shoe boxes. It works. Getting that smaller is straightforward engineering, not easy but straightforward. Going smaller than that, putting it into something the size of this phone, and then into a chip, that's going to require some creativity and work. I couldn't tell you exactly how that's going to happen. We have proof it lives between our eyes and our nose, we're walking proof, but there's work to be done to make it affordable and pervasive.

在 ML/AI 的应用中是否存在某种固有局限？重要的不是映射到单一标签的气味，比如黄瓜或肉桂，而是映射到情感，这是一个更丰富的东西。你肯定已经在做了，这无论如何都是人们试图达到的目标。

Is there some inherent limit in the application of ML/AI? What's important isn't really mapping to single label scents like cucumber or cinnamon, but mapping to emotion, which is a much richer thing. You must be already doing that, that's what people are trying to get at anyway.

标签不止肉桂和黄瓜。感官信息的丰富性，当你开始不用人类语言而用数值尺度——这可以做到，只需要一些微调——你实际上会更仔细地标记事物，从而学到很多关于化学世界的知识。

There's more labels than just cinnamon and cucumber. The richness of sensory information, when you start to not use human language and use numeric scales—which you can do, it just takes some fine-tuning—you actually learn a lot about the chemical world when you label things more carefully that way.

从某种意义上说，这是 Google 嵌入工作的最初观察：接近度实际上意味着感知接近度，所以地图上靠近的东西闻起来相似。

In some sense, that's the original observation from the embedding work at Google: proximity actually means perceptual proximity, so things that are nearby in the map smell similar.

对，没错。

Right, exactly.

我在设想将情感效价与嗅觉信息联合嵌入。我不知道你怎么做，但这似乎是可能的。

I'm envisioning jointly embedding emotional valence with olfactory information. I don't know how you would do this, but it seems possible.

我认为这非常可能。挑战在于现在有一个趋势叫神经感知，坦白说，很多人用 EEG。我过去在训练中也用过 EEG，我认识经营 EEG 公司的人。他们很坦诚：EEG 对于判断你是否睡着或癫痫发作很有用，量化睡眠和癫痫很有价值，但它无法告诉你感觉好还是不好。那里没有信号。所以有很多垃圾。我们还没有涉足那个领域，因为我必须做对。如果我们做不对，我无法原谅自己。但有很多例子表明感官确实能以某种方式解锁情感。我们会做到的，而且我认为我们会做对。但我们必须非常小心地量化情感，这本身就是一个棘手的问题。我以前也研究过这个；我喜欢这个话题。

I think that's very much possible. The challenge is there's a trend now called neurosense, which largely, if I may be frank, people use EEGs. I used to use EEGs in my training, and I know people that run EEG companies. They're very upfront: EEGs are useful for telling if you're asleep or having a seizure, and it's valuable to quantify sleep and seizures, but it's not going to tell you if you're feeling up or low. There's no signal there. So there's a lot of crap out there. We've not waded into that space yet because I have to do it right. I can't live with myself if we don't do this right. But there are so many examples of sense really being able to unlock emotions in one way or another. We will get there, and I think we will do it right. But we're going to have to be very careful about how we quantify emotion, which is its own thorny problem. I used to work on this too; I love the topic.

我在想象房产中介在开放日喷新鲜烤巧克力曲奇的气味。

I'm imagining the real estate agent spraying fresh baked chocolate chip cookie scent in the open house.

他们已经这么做了，对吧？

They already do that, right?

我知道。他们烤了饼干。有专门的蜡烛吗？

I know. They baked them. Is there an actual candle?

哦，是的。肯定有蜡烛。新车气味是发明出来的，对吧？这是真实存在的。但很多也是联想。就像泡面的例子：它可能好也可能坏，就像壁炉。如果你曾经在生火的家里，它闻起来是像燃烧的建筑还是像舒适、棉花糖和热巧克力？这取决于你的经历。但可以肯定，证据很清楚，有些气味能积极影响你的情绪，几乎是生理性的，能提高专注力、警觉性或减少焦虑。这对我来说很清楚，但我觉得我们需要非常小心。

Oh, yeah. There's candles for sure. New car smell is invented, right? It's a thing. But a lot of this is also association. Same with the kimchi example: it might be good or it might be the hearth. If you've ever been in a home where a fire was lit, does it smell like a burning building or like comfort, marshmallows, and hot chocolate? It depends on what you're exposed to. But for sure, the evidence is pretty clear that there are some scents that do positive things to your mood, almost physiologically, and can increase your focus, awareness, or reduce anxiety. It's pretty clear to me, but I think we need to be really careful.

就像芳香疗法那种？

Like aromatherapy kind of stuff?

是的，我认为芳香疗法中有非常深刻的真理元素，阿育吠陀也一样。我认为这些都是非常古老的传统。它们很吸引人，体验起来很美；只是闻起来好闻。但在阅读研究时，没有什么是非常明确的。我坚信其中有深刻的真理内核。就像针灸现在有西方对应物叫干针，实际上是治疗性的，针灸更早达到；他们只是用不同的方式谈论。本质是一样的。

Yeah, I think aromatherapy has elements of really deep truth in it, and same with Ayurveda. I think these are really old traditions. They're appealing and beautiful to experience; it just smells nice. But in reading the research, there's nothing super clear-cut. I am of the conviction that there's some kernel of deep truth there. In the same way that acupuncture now has this western equivalent called dry needling that's actually therapeutic, acupuncture got there way earlier; they just talk about it differently. It's the same stuff.

也许在某些方面，西方改编中仍有一些未被充分认识的东西。但我认为，这种对应关系也会发生在阿育吠陀、芳香疗法，以及中医的芳香部分。植物就像是我们的药物和毒药工厂，它们制造所有有益或有害的分子。我们所有的药物，很多药物都有某种天然来源，或者受到天然来源的启发。所以一定存在某种东西——我就是相信这一点。我认为气味确实有强大、可控的影响，可以提升我们的情绪，让我们感觉更好，或表现更好，或实现我们任何可能达成的愿望。我们只需要认真对待，并且要彻底，仅此而已。

And maybe some ways there's still things that haven't even been appreciated in the western adaptations of it. But I think that that correspondence will also happen for Ayurveda and for aromatherapy and then also probably for the aromatic aspects of herbal Chinese medicine as well. Plants are like our medicine and poison factories. They make all the molecules that do good and bad things. And all of our drugs, many of our drugs, have some natural origin to them or they were inspired by a natural origin. So there's got to be some — I just believe it. I think there's got to be something there where scent has real, powerful, harnessable impact to uplift our mood and to make us feel better or perform better or whatever our desire might be that's actually realizable. We just have to be serious about it and we have to be thorough, that's all.

嗯，Alex，很高兴和你聊天，了解了嗅觉智能的最新进展。

Well Alex, it's been great catching up with you and getting the download on olfactory intelligence.

聊这个非常有趣。这显然涉及很广，但我们正在拓展前沿。所以所有你通常认为理所当然的事情，我们都需要从头到尾思考一遍。我一直都很兴奋能谈论这些，因为我全身心投入其中。但和你分享也很有趣，Sam。谢谢你邀请我。

Super fun to talk about it. It's obviously wide ranging, but it's like we're pushing back the frontiers. And so all the things you normally take for granted, like we have to think all the way through. So I'm just thrilled to talk about this stuff all the time because I live it and breathe it. But it's fun to share with you, Sam. Thanks for having me on.