胶木化与黑箱：被隐藏的机器本质
在20世纪40年代，收音机是一项颠覆性的技术，但它并非一问世就备受追捧。20世纪初的收音机外观简陋，充斥着裸露的电线和发光的真空管，看起来就像一个搞砸了的科学展览项目。直到1929年，当收音机开始采用时尚的胶木（Bakelite：一种早期塑料外壳）包装后，它们才变得极其流行。然而，这种流行伴随着一个关键的代价——失去透明度。当技术变成一个“黑箱”，而我们再也无法看到其内部运转时，会发生什么？
快进100年，如今我们梦寐以求的科幻人工智能设备已经融入了现代生活的方方面面。但与收音机一样，这些系统也被“胶木化”了。我们把机器隐藏在无摩擦的界面、友好的声音以及儿童玩具之中。在过去的十年里，作为麻省理工学院（MIT: Massachusetts Institute of Technology）的研究员和非营利组织“AI之日”（Day of AI: 为全球数百万儿童提供学习AI运转机制机会的组织）的创始成员，我与孩子们进行了许多关于AI的交流。我发现，这些设备模拟智能和友好度的效果非常好，以至于孩子们开始信任它们——有时甚至超过信任他们自己。这导致孩子们面临对智能玩具过度依赖或产生不当情感羁绊的真实危险。然而，只要成年人做出一些有意的选择，我们就可以帮助孩子们看清技术的本质：它们只是可以用来玩耍甚至编程的对象（Objects），而不是他们寻求所有答案的实体（Entities）。
Original English
In the 1940s, the radio was a transformative technology. But it wasn't an immediate hit. That's because the radios of the early 1900s looked like this: unwieldy contraptions of exposed wires and glowing vacuum tubes, like a science-fair project gone wrong. It wasn't until 1929 when radio started to look like this, with its sleek Bakelite -- that’s a kind of plastic shell -- that they became insanely popular -- but at a critical cost. Transparency. What happens when technology becomes a black box and we stop being able to see what's going on inside? Fast forward 100 years. Today, the sci-fi artificial intelligence devices of our dreams have been woven into the very fabric of modern life. But like the radio, these systems are “bakelited.” We've hidden the machine behind frictionless interfaces, beneath friendly voices and inside children's toys. Over the past decade, I've had a lot of conversations with kids about AI as an MIT researcher and founding member of Day of AI, a nonprofit creating opportunities for millions of children around the world to learn about how AI works. And here's what I've been seeing. These devices mimic intelligence and friendliness so well that children learn to trust them -- sometimes more than they trust themselves. And there is real danger of children becoming overly reliant on or inappropriately attached to their smart toys. That said, with some intentional choices on the part of adults, we can help children see the technology for what it really is, objects that they can play with, or even program, rather than entities that they should look to for all the answers.从社交机器人到过度信任的隐患
我探索儿童与人工智能关系的旅程始于在MIT的早期工作，当时我参与研发了一款名为 Tega 的毛茸茸、可爱的社交机器人。社交机器人（Social Robot: 旨在通过社交互动帮助人们实现目标的机器人）广泛应用于医疗保健、客户服务和教育领域。在 Tega 的案例中，我们构建它的目的是观察机器人能否在幼儿学习阅读时提供帮助。在实际互动中，Tega 的职责是保持孩子对故事的参与度，它会提出问题、添加有趣的评论，并融入本周的目标词汇。
当时我的工作是介绍 Tega 并引导孩子们熟悉其使用方法。有一天，我走进研究室说：“你好，我叫 Randi，这是 Tega。今天你要和 Tega 一起读一些故事。让我来告诉你它是怎么工作的。”然而，与我合作的一个幼儿园小女孩阻止了我。她说：“不用了，谢谢。我对机器人了如指掌。我家里有一个 Alexa，她是我最好的朋友。看我的。”接着，她开始向我演示我的机器人是如何工作的。
这种反应引发了深思。首先，令人钦佩的是，这个还在学习阅读的女孩能够凭借语音技术毫不费力地浏览互联网。但其次，我感到一丝担忧。我们正在抚养成长在AI和智能玩具时代的一代人，他们倾向于将这些设备视为朋友，是有思想、有意志、有灵魂的独特生命，而非虚拟助手或技术。
为了调查这一点，我进行了一项研究，让30名3至11岁的儿童同时与各种智能玩具互动，包括玩具机器人、会说话的洋娃娃、文字聊天机器人以及 Alexa 和 Google Home 等。其中一个非常喜欢树懒的女孩，试图通过 Alexa 了解树懒。当她问道“Alexa，树懒吃什么？”时，Alexa 回答道：“对不起，我不知道该怎么帮您。”女孩失望了一秒，随后立刻振作起来，拿起了第二台 Alexa 说：“哦，我知道了，我看看这台知不知道。”
AI的局限性不仅让孩子们感到沮丧和困惑，人机交互（Human-Robot Interaction）领域的研究还表明，这会让孩子们面临更严重的风险，例如被操纵、意外接触不当内容或安全漏洞。即使是那些被宣传为教育型或免屏的设备，本质上仍然是消费品，被设计成具有无休止吸引力或成瘾性的产品。它们试图成为孩子最好的朋友并赢得信任，以便向他们投放广告，或者以每月20美元的价格将这种信任卖回给家长。智能玩具为儿童的玩耍时间、卧室和心灵打开了一扇数字之门，在决定打开这扇门之前，我们必须仔细审视门后的一切。
Original English
My journey exploring kids and their relationships with AI began early in my days at MIT, working on this fluffy, adorable social robot, Tega. So Tega being a social robot means that its goal is to help people with their goals. Social robots are used throughout healthcare, customer service and education. In Tega's case, we were building it to see if a robot could help young children as they were learning how to read. Let me show you a quick clip of Tega in action. Student: Hm, I'm lavender. Tega: What are we trying to find? Student: We are trying to find lavender-color stuff. Tega: OK. The word was lavender? Student: Yes. And what is lavender, though? Tega: Lavender is purple. Student: Yes. Just like my friend loves lavender. Tega: Hmm. Randi Williams: So as you saw, it's Tega's job to keep the child engaged in the story. Tega will ask questions, add funny commentary, and of course, weave in that target vocabulary for the week. And it was my job to introduce Tega and get children oriented to how to use it. So one day I come into the study room and I say, "Hi, my name is Randi, this is Tega. You're going to read some stories with Tega today. Let me show you how it works." And the girl I was working with, a kindergartener, stops me. "No thanks," she says. "I know all about robots. I have an Alexa at home, and she's my best friend. Watch this." And then she proceeded to show me how my robot worked. (Laughter) What do we make of a response like that? Well first, I thought it was kind of impressive that this girl who, again, was learning how to read, could effortlessly navigate the internet thanks to voice technology. And for context, I had to know how to read, write and type before I could ask Jeeves a question. But for her, not so much. But secondly, I was a little bit disappointed that this girl didn't think that my super cool robot was so super cool. And when I took a step back, I realized it wasn't just her. We are raising a generation of children who are growing up with AI and smart toys, and they tend to see these devices not as virtual assistants or technology, but as friends, distinct beings with a mind, a will and a soul. And that made me really curious about what children make of this technology. How much do they understand about what's going on? So to investigate that question, I ran a study. I had 30 children between the ages of three and 11 come together to interact with a whole bunch of different smart toys all at once. We had toy robots, talking dolls, text-based chatbots, and of course, plenty of virtual assistants like Alexa and Google Home. One of the girls in the study, her favorite animal was sloths. She loves sloths. She'd been telling us all day how much she loves sloths. And so naturally, when she got hold of an Alexa, she wanted to learn more about sloths from Alexa. So she picks up Alexa and says, "Alexa, what do sloths eat?" And then Alexa, very helpfully, replied, "I'm sorry, I don't know how to help you with that." The girl looked disappointed for a second and then all of a sudden perked right back up. "Oh, I know," she says, and she picked up a second Alexa. "I'll see if this one knows." (Laughter) The limitations of AI are not only frustrating and confusing for children, but research in child-robot interaction shows that it can make them vulnerable to even more serious risks, like manipulation or accidental exposure to inappropriate content or security breaches. And for me personally, even the devices that are marketed as educational or screen-free are still corrupt because they're consumer products. And they seem to be designed as something that can be endlessly engaging or addictive. They want to become your child's best friend and gain their trust so that they can advertise to them. Or better yet, sell that trust back to you for $20 a month. The bottom line is that smart toys open a digital door into children's playtime, bedroom and hearts. And if you're thinking about opening that door, my advice to you would be to take a very careful look at what's on the other side.用游戏拆解AI：构建反向工程的思维
幸运的是，我们可以通过顺应孩子的天性来降低AI给他们带来的风险。孩子们是天生的反向工程师（Reverse Engineers：通过拆解系统来理解其内部运作方式的人）。我多次观察到孩子们在玩AI设备时尝试破坏它，或者问一些看似荒谬的问题，例如：“Alexa，你多大了？”、“Cosmo（玩具机器人小车），你能跳吗？”或“Hey Google，我可以吃掉你吗？”这些愚蠢的问题实际上是孩子们在探究机器的本质。游戏是孩子们理解世界的基本方式，这意味着我们可以利用它来帮助他们理解正在与之互动的技术。如果孩子们能像拆解和重组物理积木那样拆解和重建智能玩具呢？
这就是我参与研发的另一款社交机器人 PopBot 背后的初衷。PopBot 的身体完全由乐高积木拼装而成。我构建它的目的是让现代AI中最复杂的概念——逻辑推理、机器学习、生成式AI——通过动手且以孩子为主导的游戏变得易于理解。我们的核心想法是，如果一个四岁的小孩能理解AI，那么任何人都可以。
例如，在学习机器学习时，孩子们会与机器人玩石头剪刀布。在开始玩游戏之前，孩子们必须先教 PopBot 游戏的规则。随后，在反复对战的过程中，PopBot 会尝试捕捉孩子出拳的规律，从而预测下一步并击败他们。对于四到六岁的孩子来说，很快就很难甚至不可能战胜 PopBot 了。
有一个小男孩因此感到非常沮丧，于是我给他出了一个“坏主意”：“如果教给 PopBot 错误的石头剪刀布规则会怎么样？这样你每次玩都能赢。”他非常喜欢这个主意。这次小小的“AI破坏行动”创造了一个独特的契机，让我们得以探讨一个关键问题：是谁教给AI规则的？答案是人类。我问他：“如果你能教 PopBot 错误的规则，那你怎么知道 Alexa 被教导的规则是正确的呢？”这引发了一场深刻的对话，他讨论了我们在多大程度上可以信任AI，以及在成人的帮助下验证AI言论以确保其规则正确的重要性。
Original English
The good news is that we can mitigate the risks of AI to children by leaning into something they already do. Kids are natural reverse engineers. Over and over again, I've watched children playing with AI devices trying to break it, just asking seemingly absurd questions like, "Alexa, how old are you?" Or “Cosmo” -- which is a toy robot car -- "Can you jump?" Or "Hey Google, is it OK if I eat you?" (Laughter) And these questions, while silly, also are an example of children probing the machine about its very nature. Play is inherent to how children understand the world, which means that maybe we can use it to help them understand what's happening with the technologies they're interacting with. What if children could take apart and reconstruct their smart toys the same way that they take apart and reconstruct physical blocks? That was the idea behind another social robot that I worked on, “PopBot.” As you can see, PopBot’s body is made completely out of LEGO bricks, and it's kind of like a typical smart toy, except I built it to make the most complex ideas of modern AI -- logical reasoning, machine learning, generative AI -- easy to understand through hands-on, child-driven play. The idea was, if a four-year-old could understand AI, then anyone can. So for example, to learn about machine learning, what children would do is they would play Rock, Paper, Scissors against the robot. Let me show you a quick clip of what that looks like with PopBots. Person: OK, go ahead. Student: Scissors... shoot. I choose rock. PopBot: Rock. You chose rock, and I choose rock. PopBot: It's a tie. Student: Rock! Student 2: It's a tie! RW: The beauties of a real-life classroom study. So before children could start playing against their PopBot, first they had to teach their PopBot the rules of Rock, Paper, Scissors. And then, as they played the game, rounds like that, over and over again, what PopBot would try to do is pick up on their pattern so that it could predict what move it could play to beat them. And if you're between the ages of four and six, pretty soon it gets hard, if not impossible, to beat your PopBot. There was one boy who was very frustrated by this, and so I planted a mischievous idea in his head. I said, "Well, what if you taught your PopBot the wrong rules of Rock, Paper, Scissors? That way as you played it, you would win every single time." (Laughter) And he really liked that idea. (Laughter) Our little act of AI sabotage created a unique opportunity to talk about who teaches AI the rules. Well, people do. And so I asked him, "Well, if you can teach your PopBot the wrong rules of Rock, Paper, Scissors, how do you know Alexa was taught the right ones?" Which led to a very insightful conversation where he discussed how much we can trust AI and the importance of verifying what it says, with an adult’s help, just to make sure that Alexa has been taught all the right rules.重写规则：培养有好奇心的AI使用者
我希望孩子们理解的核心观点是：AI并非魔法，它只是人类编写的规则集合。 我们这些在这个世界上生活得更久的人，有责任确保孩子们不被他们的玩具所塑造，而是反过来，让玩具受控于孩子。我们需要培养能够编写甚至重写AI规则的下一代。
这一工作从餐桌旁就可以开始。当全家人坐在一起，有人向最近的AI聊天机器人或智能玩具提出问题时，在得到回答后，不要只是被动接受，而是去思考：“我想知道它为什么会这样回答？”或者大声问：“我们同意还是不同意这个答案？”、“我们认为是谁编写了规则让它这样工作的？”你不需要知道所有这些问题的答案，也不需要成为某种AI专家，这并非重点。孩子们需要的是一个榜样，展示如何成为一个具有好奇心的AI使用者。他们需要有人和他们坐在一起，探索机器，探寻其极限，质疑其回答，最重要的是，敢于重写其规则。
虽然我思考这些问题大多是在儿童玩具箱的背景下，但更大的图景是，我们所有人都应该以同样的方式对待我们日常接触的AI系统。我们不仅可以帮助孩子们与智能玩具建立更安全的关系，而且我们每个人都需要在自身与日常技术之间建立更好的关系。我坚信在一个人们不会被技术遗忘在黑暗中的世界里，每个人都拥有与生俱来的权利，去参与和塑造那些影响我们生活的系统。
在回答关于如何在理想情况与现实育儿困境（如不得不借助屏幕安抚孩子）之间进行平衡时，我认为这确实体现了育儿的重重挑战。纵观历史，人类在面对图书、收音机、电视、互联网乃至今天的人工智能时，一直在与类似的挑战作斗争。历史的经验在今天依然适用：不要让孩子们独自去摸索技术。制定规则固然很好，但更有效的方法是引导孩子，让他们以符合家庭价值观及你们希望与世界接触的方式与这些技术进行互动。如果你愿意与孩子共同踏上这段探索旅程，将会带来巨大的益处。