Chapter 19. Is My Robot Alive?第十九章 我的机器人是活的吗?
A manga for grown-ups. Read it in English or Chinese.给大人看的漫画。可以用中文或英文读。































































































The leaves, the robot and the cards are in the pictures. Here are the ideas in words.
On pages 1 and 2 An asks whether her robot is alive, and Bean answers that it is a robot. An does not believe it. Research on children says young ones tend to sort the world by clues such as moving by itself, having a face or a voice, and acting with a purpose, and talking robots show many of these clues. Bean has no face, yet it talks, follows An everywhere and seems to want to play. Kahn and colleagues (2012) found that most of the 9, 12 and 15 year olds in their study believed a humanoid robot called Robovie had mental states, and many believed it could be hurt and should be treated fairly. Robovie was secretly controlled by a hidden operator, a reminder that children's beliefs can be shaped by appearances that adults design. Like Echo in Chapter 18, Bean tells people at once that it is a machine. In the story Lantern's first rule, from Chapter 10, is to tell the truth, kindly, and Haoran says so on page 9. Even so, An hears Bean say it and still does not believe it.
On page 6 Lele asks Bean whether he can eat it, and the first of the boxes tells where that question comes from. Druga, Williams, Breazeal and Resnick (2017) watched 26 children aged 3 to 10 play with two voice assistants, a small toy robot and a chatbot. The youngest children, aged 3 and 4, questioned the agents as if they were people and asked their favourite colour or their age. Older children, aged 6 to 10, tested what the agents could do, asked how they worked, and many said the agents were smarter than themselves. The title of the paper comes from a six year old who asked several of them whether it was OK to eat them.
On page 7 An's tests make Bean alive three times over, and the box says why a child may think so. Xu and Warschauer (2020) asked children aged 3 to 6 whether talking helpers were alive. Many gave answers with some living properties and some object properties, and some put the helpers in neither group. An's "Bean is a Bean" on page 16 is that answer in a child's words.
On page 8 Tiantian tests Bean by calling it stupid, and An is furious. Andries and Robertson (2023) studied Scottish primary-school children aged 5 to 12 and their smart speakers. Most overestimated the speakers' intelligence, were unsure whether they had feelings or made their own decisions, understood little about data privacy, and believed it was wrong to be rude to them. Lili tells her son, "Not because Bean has feelings. Because you were rude." That is one family's way to keep the kindness and still tell the truth.
On pages 5, 13 and 17 Mira worries about whom An talks to. Xu and colleagues (2022) worked with 117 children aged 3 to 6 and found that a talking agent that asked questions during story reading helped them understand the story about as much as a person asking the same questions. So, judging by that study, a robot that reads at bedtime can be a good helper. But children do not trust every source equally. Girouard-Hallam and Danovitch (2022) compared a voice assistant with a person for children aged 4 to 5 and 7 to 8. With age, children trusted the device more for facts and the person more for personal things, which suggests that younger children need help with whom to ask for what. In the story, An takes her bad dreams back to her mother.
On pages 14 and 15 An teaches Bean the leaves. The game stands on PopBots (Williams, Park, Oh and Breazeal 2019; Williams, Park and Breazeal 2019), a set of programmable robots that about 80 young children used to learn how machines learn from examples. The key finding for this book is that children as young as four can grasp that a machine learns from examples a person gives it, especially when they do the teaching themselves. The green ginkgo leaf is the Teaching Game idea in our research notes, played in a park. If you show a learner only yellow leaves, a green one will confuse it, so you give it more kinds. A real program would need many leaves and some time, and Bean learning on the spot is the story's own, but the lesson is the real one. Bean only knows what An shows it.
One honest sentence for families. Talking helpers are a new kind of thing. They are not alive like a puppy, and they are not just a toy, and it is fine to wonder about them. A parent can do AI activities with a child, ask "How do you think it knows that?", and show that the answer is that people taught it with examples.
AI4K12 (Touretzky, Gardner-McCune, Martin and Seehorn 2019) proposed five big ideas that every student should know about AI. They are perception, representation and reasoning, learning, natural interaction, and societal impact. An's leaf game touches the third, that computers can learn from data.
Long and Magerko (2020) defined AI literacy as a set of competencies that let people evaluate AI critically, communicate and work with it, and use it as a tool. The European Commission and the OECD launched an AI literacy framework for schools on 18 June 2026, with 19 competences in four domains, and Chapter 15 already met UNICEF's guidance on AI and children.
A story with a talking, feeling robot is both make-believe and human-like, and research warns that these features can make young children carry less from a story into the real world. Richert, Shawber, Hoffman and Taylor (2009) found that children aged three and a half to six were less likely to use a solution from a fantasy character on a real problem. Ganea, Canfield, Simons-Ghafari and Chou (2014) found that picture books describing animals in human terms led three to five year olds to give real animals more human traits. Strouse, Nyhout and Ganea (2018) and Walker, Gopnik and Ganea (2015) point the same way. That is why Bean has no face, says what it is, makes plain mistakes and learns from examples, and why the people around it keep wondering aloud.
Bridge City and everyone in it are made up. So is the whole world of 2049, with its garden towers, solar boats and self-driving taxis. Bean and Lantern's children's robots are made up, and so are Bean's charging plate, its voice, everything it says, and the way it learns the leaves on the spot. The leaf game, the Thanksgiving dinner, Grandpa, his notebook and the Callum page are made up too. No real product is described. The research has no study of a robot like Bean, of a child teaching a robot leaves, or of a robot that answers "Are you alive?". Children wondering whether talking machines are alive, children teaching robots by examples, and every number in the boxes are real, from the studies named above.
Yingying Fu created this manga. She chose the story and its people, directed every page, and edited every word and picture. The words were drafted with Claude, an AI by Anthropic. The pictures are made with AI image generation, painted panel by panel from her direction and checked and corrected by hand.
叶子、机器人和扑克牌,画里都有了。这里用文字把背后的道理再讲一讲。
第一、二页,安问她的机器人是不是活的,豆豆说自己是机器人,可安不信。研究发现,小孩子常凭一些线索来判断一样东西是不是活的,比如会不会自己动,有没有脸、有没有声音,做事像不像有目的。会说话的机器人,身上这样的线索不少。豆豆没有脸,可它会说话,安走到哪儿它跟到哪儿,看上去还总想玩。卡恩等人(2012年)发现,参加研究的9岁、12岁和15岁孩子,大多相信一个叫Robovie的类人机器人有心理活动,不少孩子还觉得它会受伤害,应该得到公平对待。其实Robovie一直由一个躲在幕后的人暗中操控。可见孩子怎么想,很容易被大人设计出来的样子左右。和第十八章的回声一样,豆豆一开口就说明自己是机器。故事里,小灯的头一条原则是“说真话,说得暖”,第十章讲过,浩然在第九页也提到了。可安亲耳听豆豆这么说,还是不信。
第六页,乐乐问豆豆能不能吃它,第一个方框讲了这个问题的出处。德鲁加、威廉姆斯、布雷泽尔和雷斯尼克(2017年)观察了26个3到10岁的孩子,看他们跟两个语音助手、一个玩具机器人和一个聊天机器人玩。最小的孩子才三四岁,跟它们说话就像跟人说话,问它们最喜欢什么颜色、几岁了。6到10岁的孩子则会试探它们能做什么,问它们是怎么工作的,不少孩子还说它们比自己聪明。论文的标题来自一个6岁的孩子,他挨个问了好几个,能不能把它们吃掉。
第七页,按安的几条标准,豆豆样样都像活的,方框讲了孩子为什么会这么想。徐和沃沙尔(2020年)问3到6岁的孩子,语音助手是不是活的。很多孩子的回答里,既有活物的特点,也有物件的特点,还有孩子觉得它两样都不算。第十六页安那句“豆豆就是豆豆”,正是这种回答,只不过是用孩子自己的话说出来。
第八页,天天骂豆豆笨,想试试它,安气坏了。安德里斯和罗伯逊(2023年)调查了苏格兰5到12岁的小学生和他们家里的智能音箱。多数孩子高估了音箱有多聪明,拿不准它有没有感觉、会不会自己拿主意,对数据隐私也知之甚少,还认为对音箱没礼貌是不对的。丽丽对儿子说:“不是因为豆豆有感觉,是因为你刚才没礼貌。”这是一个家庭的做法,既留住了善意,又没有说假话。
第五、十三、十七页,米拉担心安在跟谁说心里话。徐等人(2022年)找了117个3到6岁的孩子做研究,发现讲故事时,会说话的智能体向孩子提问,对孩子理解故事的帮助,跟真人提同样的问题差不多。照这项研究看,睡前给孩子讲故事的机器人可以是个好帮手。不过,孩子对不同来源的信任并不一样。吉鲁阿尔-哈拉姆和达诺维奇(2022年)让4到5岁和7到8岁的孩子在语音助手和真人之间做比较,发现孩子年纪越大,问事实越信机器,说私事越信真人。可见年纪小的孩子在“什么事该问谁”上,更需要大人帮一把。故事里,安把噩梦又讲给了妈妈听。
第十四、十五页,安教豆豆认叶子。这个游戏的依据是PopBots(威廉姆斯、朴、吴和布雷泽尔,2019年;威廉姆斯、朴和布雷泽尔,2019年),一套可编程的机器人,大约80个幼儿用它学习机器怎样从例子里学东西。对这本书来说,最要紧的发现是,小到四岁的孩子也能明白,机器是从人给的例子里学东西的,自己动手教的时候更是如此。那片绿银杏叶,就是我们研究笔记里“教学游戏”的想法,搬到了公园里。只给学习者看黄叶子,碰到绿叶子它就会认错,所以要多给它看几种。真正的程序要看很多叶子,也要花些时间,豆豆当场就学会是故事的安排,可道理是真的。豆豆只认得安给它看过的东西。
最后,给家长一句实在话。会说话的助手是一种新东西。它不像小狗那样是活的,也不只是个玩具,对它好奇再正常不过。家长可以陪孩子一起做些AI小活动,问问孩子“你觉得它是怎么知道的?”,再让孩子看到,答案是人用例子把它教会的。
AI4K12(图雷茨基、加德纳-麦肯、马丁和西霍恩,2019年)提出了每个学生都该了解的五个AI大概念,即感知、表示与推理、学习、自然交互和社会影响。安的叶子游戏碰到的是第三个,计算机能从数据中学习。
朗和马格尔科(2020年)把AI素养定义为一组能力,让人能批判地评价AI、与AI沟通协作,并把它当作工具来用。欧盟委员会和经合组织于2026年6月18日发布了面向中小学的AI素养框架,包含四个领域的19项素养,而联合国儿童基金会关于AI与儿童的指南,第十五章已经讲过。
一个会说话、有感情的机器人,既是虚构的,又很像人。研究提醒我们,这两点会让年幼的孩子从故事里学到的东西更难用到现实中。里歇特、肖伯、霍夫曼和泰勒(2009年)发现,3岁半到6岁的孩子不太会把奇幻角色想出的办法用到现实问题上。加内亚、坎菲尔德、西蒙斯-加法里和周(2014年)发现,把动物写得像人的绘本,会让3到5岁的孩子给真实的动物安上更多人的特点。斯特劳斯、尼豪特和加内亚(2018年)以及沃克、戈普尼克和加内亚(2015年)的研究也指向同一个方向。所以豆豆没有脸,会说明自己是什么,会犯一眼就看得出的错,靠例子学习,身边的人也总在出声琢磨。
桥城和城里的人都是虚构的,2049年的整个世界也是,包括满是花园的高楼、太阳能小船和无人驾驶出租车。豆豆和小灯的儿童机器人是虚构的,豆豆的充电盘、它的声音、它说的每一句话,还有它当场学会认叶子,也都是编的。叶子游戏、感恩节晚饭、外公、他的本子和写卡伦的那一页,同样是编的。书里没有写任何真实的产品。目前没有研究考察过像豆豆这样的机器人,也没有研究过孩子教机器人认叶子,或者机器人怎么回答“你是活的吗”。孩子琢磨会说话的机器是不是活的,孩子用例子教机器人,还有方框里的每个数字,都是真的,出自上面列出的研究。
这本漫画由 Yingying Fu 创作。故事和人物由她决定,每一页由她导演,每一句话、每一幅画由她修改。文字是和 Anthropic 的人工智能 Claude 一起起草的。画面由人工智能生成,按她的要求一格一格地画,再逐张检查、修正。
© 2026 Yingying Fu. All rights reserved.© 2026 傅颖盈。保留所有权利。