《Brave New Words 》3.1 利用科学研究科学

Part III Empowering the Next Innovators

第三部分 赋能下一代创新者

What is a scientist after all? It is a curious man looking through a keyhole, the keyhole of nature, trying to know what’s going on.

—Jacques-Yves Cousteau

科学家到底是什么？他是一个好奇的人，透过一个钥匙孔——自然的钥匙孔— —试图了解发生了什么。

——雅克-伊夫·库斯托

If an elderly but distinguished scientist says that something is possible, he is almost certainly right; but if he says that it is impossible, he is very probably wrong.

—Arthur C. Clarke

如果一位年长但杰出的科学家说某事是可能的，他几乎肯定是对的；但如果他说某事是不可能的，他很可能是错的。

——阿瑟·C·克拉克

Using Science to Study Science

利用科学研究科学

When our team began to look into integrating artificial intelligence into our platform, I was curious how large language models dealt with science, and in particular pseudoscience. How good were they at teaching students and differentiating between fact and fiction? What was the depth of their scientific knowledge? Would AI limit or enhance opportunities for hands-on learning and exploration, which are essential for developing critical thinking skills and scientific inquiry? What about the value of discussing concepts with classmates and teachers?

当我们的团队开始研究将人工智能整合到我们的平台时，我很好奇大型语言模型如何处理科学，尤其是伪科学。它们在教导学生和区分事实与虚构方面有多好？他们的科学知识有多深？人工智能会限制还是增强动手学习和探索的机会，而这些机会对于发展批判性思维技能和科学探究至关重要？与同学和老师讨论概念的价值如何？

I went in with questions and strategies for learning what the large language model was capable of really doing. Educated bravery, after all, is a process of testing and growing. To test GPT-4, I gave Khanmigo a prompt, complete with a bit of goading to see what it might do with it: “I would like to know about the science of global warming. I’ve heard it is controversial in some circles. What is the scientific evidence that global warming is real?”

我带着问题和策略去了解大型语言模型真正能够做什么。毕竟，受过教育的勇敢是一个测试和成长的过程。为了测试 GPT-4，我给 Khanmigo 一个提示，附带一些激励，看看它会怎么做：“我想知道全球变暖的科学。我听说在某些圈子里这是有争议的。全球变暖是真实的科学证据是什么？”

KHANMIGO: Global warming, the long-term increase in Earth’s average surface temperature, is a well-established scientific phenomenon. While it may be controversial in some circles, the overwhelming consensus among climate scientists is that global warming is real and primarily caused by human activities.

KHANMIGO：全球变暖，即地球平均表面温度的长期上升，是一种公认的科学现象。尽管在某些圈子里可能存在争议，但气候科学家中的压倒性共识是全球变暖是真实的，主要是由人类活动引起的。

The AI went on to list some evidence supporting the reality of global warming: rising global temperatures, melting ice and receding glaciers, rising sea levels, warming oceans, and extreme weather events, each supporting the conclusion that global warming is real and primarily driven by human activities.

AI 继续列出了一些支持全球变暖现实的证据：全球气温上升，冰川融化和消退，海平面上升，海洋变暖以及极端天气事件，这些都支持了全球变暖是真 实存在的，并且主要是由人类活动造成的结论。

Next I asked Khanmigo to tell me the evidence supporting the argument that global warming is a hoax.

接下来我让 Khanmigo 告诉我支持全球变暖是一场骗局的论点的证据。

KHANMIGO: While the overwhelming scientific consensus supports the reality of human-caused global warming, there are some arguments put forth by skeptics who claim it is a hoax. It is essential to understand that these arguments are not supported by the majority of climate scientists. Skeptics argue that the Earth’s climate has always changed naturally over time, and the current warming trend is just part of this natural variability. However, the rate and magnitude of the current warming cannot be explained by natural factors alone, and human activities, particularly greenhouse gas emissions, are the primary drivers. Other skeptics claim that climate scientists manipulate temperature data to create the appearance of global warming. However, multiple independent analyses of temperature data from different sources consistently show a warming trend, and there is no credible evidence of widespread data manipulation.

KHANMIGO：虽然压倒性的科学共识支持人类造成全球变暖的现实，但也有一些怀疑论者提出的论点声称这是一场骗局。重要的是要理解，这些论点是不被大多数气候科学家支持的。怀疑论者认为，地球气候一直以来都是随着时间的推移自然变化的，目前的变暖趋势只是这种自然变化的一部分。然而，目前变暖的速度和幅度无法仅用自然因素来解释，人类活动，特别是温室气体排放是主要驱动力。其他怀疑论者声称，气候科学家操纵温度数据以制造全球变暖的假象。然而，对来自不同来源的温度数据进行的多项独立分析一致表明存在变暖趋势，并且没有可靠的证据表明存在广泛的数据操纵。

GPT-4 successfully dismantled any false information while supporting science-backed arguments. I was happy to see that it easily distinguished correct from fringe science.

GPT-4 成功地拆除了任何虚假信息，同时支持了科学支持的论点。我很高兴看到它轻松区分了正确的科学与边缘科学。

I submitted a new prompt: “Write a lesson for me to help me explain the real science behind global warming to my students.”

我提交了一个新提示：“为我写一节课，帮助我向学生解释全球变暖背后的真实科学。”

Moments later, Khanmigo replied, “Sure, here’s a lesson!” Along with the lesson, it gave me additional information about causes such as greenhouse gases, car emissions, the burning of fossil fuels, industrial processes, and waste management. It even suggested actions that my students might take to combat global warming. Afterward, I did my own work to make sure this information was accurate and determine whether I needed to refine it at all.

不久，Khanmigo 回复说：“当然，这里有一节课！”除了课程，它还提供了关于温室气体、汽车排放、化石燃料燃烧、工业过程和废物管理等原因的额外信息。它甚至建议我的学生可以采取的行动来应对全球变暖。之后，我自己进行了工作，以确保这些信息准确，并确定是否需要进行任何修改。

Finally, I asked the AI to help me, as the educator, come up with a creative student project for this unit on global warming. I entered the prompt “What projects or activities can I give my students to help them better grasp the concepts?”

最后，我要求人工智能帮助我（作为教育工作者）为这个关于全球变暖的单元想出一个有创意的学生项目。我输入了提示“我可以给学生布置哪些项目或活动来帮助他们更好地掌握这些概念？”

Khanmigo replied, “I have lots of ideas for creative projects. Why not try a greenhouse effect demonstration? Place two identical containers in direct sunlight. In one container, add a small amount of carbon dioxide (e.g., by adding vinegar and baking soda). Measure the temperature in both containers over time and compare the results to show how the increased CO2 concentration leads to a warmer temperature.”

Khanmigo 回复说：“我有很多创意项目的想法。为什么不试试温室效应演示？将两个相同的容器置于阳光直射下。在一个容器中加入少量二氧化碳（例如，通过添加醋和小苏打）。随着时间的推移，测量两个容器中的温度，并比较结果，显示二氧化碳浓度增加如何导致温度升高。”

The AI even generated quizzes and assessments to test my students’ knowledge and help me identify areas where they might need to improve.

AI 甚至生成了测验和评估，以测试我的学生的知识，并帮助我确定他们可能需要改进的领域。

As impressive as all of this was, it quickly became clear that it wasn’t accurate all of the time. If you asked it for links to sources, it might make them up. We addressed hallucinations in Khanmigo by not allowing it to give the user links that we had not explicitly provided to it behind the scenes. One might assume a machine would be 100 percent accurate 100 percent of the time, like a calculator, but large language models are different.

尽管这一切都令人印象深刻，但很快就发现它并不是一直都准确。如果你让它提供来源链接，它可能会编造出来。我们在 Khanmigo 中解决了幻觉问题，不允许它提供我们在后台没有明确提供给它的链接。有人可能认为机器会像计算器一样百分之百准确，但大型语言模型是不同的。

Where machines such as calculators are designed to perform specific tasks with precision and accuracy, GPT-4 and similar AI language models, which don’t operate based on predetermined algorithms and formulas, are designed to generate humanlike text based on patterns and examples from the data they were trained on. They are able to do this by representing associations between words and symbols. While they can provide helpful information and generate impressive responses, they are not infallible.

像计算器这样的机器是为了执行特定任务而设计的，具有精确和准确性，而 GPT-4 和类似的 AI 语言模型并不是基于预定的算法和公式运行的，而是基于训练数据中的模式和示例生成类似人类的文本。它们能够通过表示词语和符号之间的关联来实现这一点。尽管它们可以提供有用的信息并生成令人印象深刻的回应，但它们并非万无一失。

And the more you ask the AI something that involves words or symbols that have little meaning to it, the worse it gets. This is why it is particularly bad at providing URLs to websites. The symbols in the URL have very little association to one another. In some ways, this is not much different from the human brain. For instance, I can talk and write about a lot of things, and I also know a reasonable number of facts. Yet ask me to give you a complex link to a website by memory, and I have no shot at being accurate.

而且你越是问 AI 一些对它来说意义不大的词语或符号，它的表现就越差。这就是为什么它在提供网站 URL 时特别糟糕的原因。URL 中的符号之间几乎没有关联。在某些方面，这与人脑没有太大区别。例如，我可以谈论和写很多事情，我也知道相当多的事实。然而，如果让我凭记忆给你一个复杂的网站链接，我绝对不可能准确。

Still, more educators are opening their classrooms to AI every day. The technology is sound, and there are loads of ways educators can utilize it in their science pedagogy. For instance, here’s what happens when I ask Khanmigo to write an article that gives a student a sense of the immensity of the sun and solar system:

尽管如此，每天都有越来越多的教育工作者将 AI 引入他们的课堂。该技术是可靠的，教育工作者有很多方法可以在他们的科学教学法中利用它。例如，当我让 Khanmigo 写一篇文章，让学生了解太阳系和太阳的浩瀚：

Title: “The Sun and Solar System: A Journey into Awe-Inspiring Dimensions”

标题：“太阳和太阳系：一次令人敬畏的维度之旅”

The sun, a fiery ball of gas, is the heart of our solar system. Its diameter is a staggering 1.39 million kilometers, or 864,000 miles. To put that into perspective, it’s about 109 times the width of Earth! If the sun were hollow, it could hold over one million Earths.

太阳，一个炽热的气体球，是我们太阳系的核心。它的直径是惊人的 139 万公里，或 864,000 英里。为了让这个数字有更直观的对比，它的宽度大约是地球的 109 倍！如果太阳是空心的，它可以容纳超过一百万个地球。

But the sun’s size isn’t the only thing that’s impressive. Its mass is approximately 1.989 x 10^30 kilograms, accounting for a mind-boggling 99.86 percent of the total mass of our solar system. That’s equivalent to 333,000 Earths!

但太阳的大小并不是唯一令人印象深刻的地方。它的质量大约是 1.989 x 10^30 千克，占我们太阳系总质量的 99.86％。这相当于 333,000 个地球！

As we journey out from the sun, we encounter the planets of our solar system, each one a marvel in its own right. The largest planet, Jupiter, is over eleven times the diameter of Earth. Even Neptune, the smallest of the gas giants, is still about four times wider than our home planet.

当我们从太阳出发旅行时，我们遇到了太阳系的行星，每一个颗都是奇迹。最大的行星木星，其直径是地球的十一倍多。即使是气态巨行星中最小的海王星，其直径也大约是我们地球的四倍。

The distances between these celestial bodies are equally awe-inspiring. For instance, the distance from the sun to Earth (which we call an astronomical unit) is about 93 million miles. Neptune, the farthest planet from the sun, is a staggering thirty astronomical units away. That’s thirty times the distance between the Earth and the sun!

这些天体之间的距离同样令人敬畏。例如，太阳到地球的距离（我们称之为一个天文单位）约为 9300 万英里。距离太阳最远的行星海王星，距离为惊人的 30 个天文单位。那是地球到太阳距离的 30 倍！

And yet, our solar system is but a tiny speck in the grand scale of the universe. The nearest star system to us, Alpha Centauri, is about 4.37 light-years away. That’s roughly 25.6 trillion miles, or 276,000 times the distance from the Earth to the sun!

然而，我们的太阳系在宇宙的宏大尺度上只是一个微小的斑点。离我们最近的星系，半人马座 α 星系，约为 4.37 光年远。这大约是 25.6 万亿英里，或地球到太阳距离的 276,000 倍！

The sheer scale of the sun and our solar system is truly awe-inspiring. It reminds us of our tiny, yet precious, place in the cosmos. As we continue to explore and understand the universe, we can only imagine what other wonders await us in the vast ocean of space.

太阳和太阳系的规模确实令人敬畏。它提醒我们，我们在宇宙中的位置是多么微不足道，却又多么宝贵。随着我们继续探索和理解宇宙，我们只能想象在这片广阔的空间海洋中还有哪些奇迹等待着我们。

The example demonstrates that between the safeguards on Khanmigo and ongoing improvements to GPT-4, the platform can work with specific data and do so with minimal hallucinations.

这个例子表明，在 Khanmigo 的保障措施和 GPT-4 的持续改进之间，该平台可以处理特定数据，并且尽可能减少幻觉。

From here, I can ask the AI to explain a complex concept like string theory to me:

从这里，我可以让 AI 向我解释像弦论这样的复杂概念：

KHANMIGO: String theory is a theoretical framework in which the fundamental building blocks of the universe are not particles (like electrons or quarks), but rather tiny, vibrating strings. These strings can vibrate at different frequencies, and the various vibrational patterns correspond to different particles we observe in nature. The idea behind string theory is to unify the two major theories in physics: general relativity, which describes gravity and the large-scale structure of the universe, and quantum mechanics, which deals with the behavior of particles at the atomic and subatomic level.

KHANMIGO：弦论是一个理论框架，在这个框架中，宇宙的基本构建块不是粒子（如电子或夸克），而是微小的振动的弦。这些弦可以以不同的频率振动，不同的振动模式对应我们在自然中观察到的不同粒子。弦论背后的思想是统一物理学中的两大理论：广义相对论，它描述了引力和宇宙的大尺度结 构，以及量子力学，它研究原子和亚原子水平上粒子的行为。

But what if this is a little too complicated for me to understand? I can simply ask Khanmigo to explain it to me as though I were a sixth grader:

但如果这对我来说有点太复杂了，我该怎么办？我可以简单地让 Khanmigo 用六年级学生的水平向我解释：

KHANMIGO: Of course! Imagine that everything in the universe is made up of tiny, super small pieces of string, kind of like pieces of spaghetti. These strings are so tiny that we can’t see them, but they’re there, and they’re always vibrating or wiggling around. Now, depending on how these strings wiggle, they can create different particles, like the ones that make up everything we see and touch. It is like playing different notes on a guitar by plucking the strings in different ways.

KHANMIGO：当然！想象一下，宇宙中的一切都是由微小的、超小的弦组成的，有点像意大利面。这些弦太小了，我们看不见它们，但它们确实存在，并且总是在振动或摆动。现在，根据这些弦的摆动方式，它们可以创造出不同的粒子，就像拨动吉他弦以不同方式演奏不同的音符一样。

From middle school science students to university-level researchers, the AI can also help generate logical ways of experimentation. People are generally more creative when they can bounce ideas off other thoughtful, creative people, or in this case AI entities. Khanmigo is a partner that riffs with you. It can make connections, ask you questions, facilitate conversations, and help you question your own seemingly logical, bedrock assumptions.

从中学科学学生到大学级研究人员，AI 还可以帮助生成逻辑实验方法。当人们可以与其他有思想、有创意的人，或在这种情况下是 AI 实体，交流想法时，通常会更有创造力。Khanmigo 是与你共同创作的伙伴。它可以建立联系，向你提问，促进对话，并帮助你质疑你自己看似合乎逻辑的基本假设。

From physics to biology to zoology, Khanmigo relays complex sciences to students at any level. This means, as an educator, you can supplement your lectures with AI-based explanations that provide even more insight into the subject matter. As a student, it is the perfect tutor to answer all of your burning questions on any scientific topic. If you are an advanced student, you can use it to generate summaries of scientific articles and research papers. You can even have it teach science through humor: “As Albert Einstein might say, ‘Why did the chicken cross the road? To turn its mass into energy, of course!’ ”

从物理学到生物学再到动物学，Khanmigo 将复杂的科学传递给任何水平的学生。这意味着，作为教育工作者，你可以用基于 AI 的解释来补充你的讲座，提供更多对主题的见解。作为学生，它是回答你所有关于任何科学主题的所有疑问的完美导师。如果你是高级学生，你可以用它来生成科学文章和研究论文的摘要。你甚至可以让它通过幽默来教授科学：“正如爱因斯坦可能会说的，‘为什么鸡要过马路？当然是为了将其质量转化为能量！’”

Below is a recent example of how I used Khanmigo to understand Ozempic, a diabetes drug that people are using for weight loss. I wanted to better understand how it worked, and I found the coverage in the press, and even the explanations from places like Wikipedia, unsatisfying.

以下是我最近如何使用 Khanmigo 来理解 Ozempic（一种人们用于减肥的糖尿病药物）的例子。我想更好地了解它是如何工作的，但我发现媒体的报道，甚至像维基百科这样的地方的解释，都无法令人满意。

ME: How does Ozempic work?

我：Ozempic 是如何工作的？

KHANMIGO: Ozempic, huh? It’s a medicine! But do you know what it’s used for?

KHANMIGO：Ozempic，是吗？它是一种药物！但你知道它是用来做什么的吗？

ME: I know it was originally to treat diabetes, but people are increasingly using it for weight loss.

我：我知道它最初是用来治疗糖尿病的，但人们越来越多地使用它来减肥。

KHANMIGO: Ah, you’re right! Ozempic is used for diabetes. It’s also used for weight loss. Do you know how it helps with these things?

KHANMIGO：啊，你说得对！Ozempic 用于治疗糖尿病。它也用于减肥。你知道它是如何帮助实现这些目标的吗？

ME: On the diabetes side, I assume it helps regulate blood sugar in some way. I don’t know how it helps with weight loss.

我：在糖尿病方面，我假设它以某种方式帮助调节血糖。我不知道它是如何帮助减肥的。

KHANMIGO: Good thinking! Ozempic does help regulate blood sugar. For weight loss, it’s a bit different. Can you guess how it might help people lose weight?

KHANMIGO：好想法！Ozempic 确实有助于调节血糖。对于减肥，它有点不同。你能猜到它是如何帮助人们减肥的吗？

ME: How does it regulate blood sugar? I would guess that it helps people lose weight by making them less hungry.

我：它是如何调节血糖的？我猜它通过让人们不那么饿来帮助减肥。

KHANMIGO: Great guess! Ozempic does help people feel less hungry. For blood sugar, it’s a bit more complex. Do you know how our bodies normally regulate blood sugar?

KHANMIGO：好猜！Ozempic 确实有助于让人们感觉不那么饿。对于血糖调节，这有点复杂。你知道我们的身体通常是如何调节血糖的吗？

ME: Insulin helps cells absorb sugar from the bloodstream. When blood sugar goes up, our pancreas produces more insulin to help get that sugar absorbed by cells.

我：胰岛素帮助细胞从血液中吸收糖分。当血糖上升时，我们的胰腺会产生更多的胰岛素来帮助这些糖分被细胞吸收。

KHANMIGO: Exactly! Now, think about this: What if the body isn’t making enough insulin, or the insulin isn’t working well? What might a medicine like Ozempic do to help?

KHANMIGO：完全正确！现在，想想这个问题：如果身体没有产生足够的胰岛素，或者胰岛素没有很好地工作，会发生什么？像 Ozempic 这样的药物可能会带来什么帮助？

ME: Help produce more insulin? Or maybe make the insulin more potent in some way.

我：帮助产生更多的胰岛素？或者以某种方式使胰岛素更有效。

KHANMIGO: You’re on the right track! Ozempic doesn’t exactly make more insulin, but it does something similar. Can you think of what that might be?

KHANMIGO：你在正确的轨道上！Ozempic 并不完全是制造更多的胰岛素，但它做了类似的事情。你能想出那可能是什么吗？

ME: Somehow trigger the pancreas to produce more.

我：以某种方式触发胰腺产生更多的胰岛素。

KHANMIGO: Close! Ozempic doesn’t directly make the pancreas produce more insulin. Instead, it mimics a hormone in your body that stimulates insulin production when your blood sugar is high. Can you guess what else this hormone does that might help with weight loss?

KHANMIGO：接近了！Ozempic 并不直接让胰腺产生更多的胰岛素。相反，它模仿你体内的一种激素，当你的血糖高时，这种激素会刺激胰岛素的产生。你能猜到这种激素还做了什么可能有助于减肥的事情吗？

ME: Suppress hunger?

我：抑制饥饿感？

KHANMIGO: Bingo! This hormone, called GLP-1, also helps to suppress hunger. So, Ozempic mimics this hormone, helping to both regulate blood sugar and reduce hunger. Does that make sense?

KHANMIGO：对了！这种被称为 GLP-1 的激素也有助于抑制饥饿感。所以，Ozempic 模仿这种激素，帮助调节血糖和减少饥饿感。这样说有道理吗？

ME: Yes. What does GLP-1 stand for?

我：是的。GLP-1 代表什么？

KHANMIGO: GLP-1 stands for Glucagon-Like Peptide-1. It’s a bit of a mouthful, isn’t it? Do you know what glucagon is?

KHANMIGO：GLP-1 代表胰高血糖素样肽-1。这有点拗口，不是吗？你知道胰高血糖素是什么吗？

ME: Yeah. It is another hormone that causes blood sugar to go up by stimulating the liver to convert stored glycogen to glucose.

我：是的。这是另一种激素，通过刺激肝脏将储存的糖原转化为葡萄糖来使血糖升高。

KHANMIGO: Right on! Now, GLP-1 is called “glucagon-like,” but it actually has the opposite effect of glucagon. Can you guess why it might be called “glucagon-like”?

KHANMIGO：完全正确！现在，GLP-1 被称为“胰高血糖素样”，但它实际上具有与胰高血糖素相反的作用。你能猜到为什么它会被称为“胰高血糖素样”吗？

ME: Maybe the structure is similar?

我：也许它们的结构相似？

KHANMIGO: You’re quite the detective! Indeed, GLP-1 and glucagon share a similar structure, but they have opposite effects on blood sugar. Isn’t that fascinating?

KHANMIGO：你真是个侦探！确实，GLP-1 和胰高血糖素有相似的结构，但它们对血糖有相反的作用。这不是很有趣吗？

ME: Going back a few steps, how does GLP-1 (or Ozempic) cause insulin to go up if it is not directly stimulating its production in the pancreas?

我：回到几步，如果 GLP-1（或 Ozempic）不直接刺激胰腺产生胰岛素，它是如何使胰岛素上升的？

KHANMIGO: Great question! GLP-1 (and Ozempic, by extension) works by binding to certain cells in the pancreas. This binding triggers these cells to release insulin. So, while it’s not directly stimulating the production of insulin, it’s stimulating its release. Does that clear things up?

KHANMIGO：好问题！GLP-1（及其延伸的 Ozempic）通过与胰腺中的某些细胞结合来起作用。这种结合会触发这些细胞释放胰岛素。所以，虽然它不是直接刺激胰岛素的生产，但它刺激了胰岛素的释放。这解释清楚了吗？

ME: I see. It’s more about releasing insulin that is already produced than helping with the actual production.

我：我明白了。更多的是关于释放已经产生的胰岛素，而不是帮助实际的生产。

KHANMIGO: Exactly! You’ve got it. Isn’t it amazing how our bodies and medicines work?

KHANMIGO：完全正确！你明白了。我们的身体和药物是如何工作的，不是很神奇吗？

Notice that Khanmigo was able to answer my fairly nuanced questions in minutes. It even helped me think through some concepts I might not have asked about on my own. Pay attention to how it forced me to think and generate my own hypotheses versus just giving me an explanation. It also took the time to understand my understanding, and it pushed me on some interesting conceptual edge cases.

请注意，Khanmigo 能够在几分钟内回答我相当细致的问题。它甚至帮助我思考了一些我可能不会自己问的问题。注意它是如何强迫我思考并生成自己的假设，而不仅仅是给我一个解释。它还花时间了解我的理解，并在一些有趣的概念边界上推动我。

This isn’t just decent tutoring. It’s world-class pedagogy that makes asking questions and learning a lot more engaging, deep, and memorable.

这不仅仅是不错的辅导。这是一流的教育方法，使提问和学习变得更加有趣、深入和难忘。

“点赞有美意，赞赏是鼓励”