Part IV Better Together

第四部分 携手共进

AI Meets Student Mental Health Coaching

AI 与学生心理健康辅导的结合

Here’s the scenario: You’re minutes away from taking a final exam. You’ve studied, but your heart is racing and your mind has gone blank. Anxiety and fear of failure are eating up all of your confidence. A quick connection with Khanmigo and its AI-academic-coach feature becomes an emotional lifeline.

场景是这样的：你还有几分钟就要参加期末考试了。你已经复习过了，但你的心跳加速，大脑一片空白。焦虑和对失败的恐惧吞噬了你所有的自信。这时，快速连接到 Khanmigo 及其 AI 学业教练功能成为了情感救命索。

STUDENT: I have a test in ten minutes and I’m really nervous about it.

学生：我十分钟后有考试，我真的很紧张。

AI ACADEMIC COACH: Oh, I see you’re feeling a bit jittery, eh? That’s okay, it happens to the best of us! Just remember, you’ve prepared for this with hours of study and all of those quizzes you aced with me.

AI 学业教练：哦，我看到你有点紧张，对吗？没关系，这种情况最优秀的人也会发生！记住，你为此准备了好几个小时，以及你和我一起完成的所有测验。

STUDENT: I know, but I’m worried I’m going to bomb this test. So much pressure.

学生：我知道，但我担心我会考砸。压力太大了。

AI ACADEMIC COACH: It is natural to be worried. Tell me what you are most worried about if you don’t do well.

AI 学业教练：担心是很自然的。告诉我，如果你考不好，你最担心什么。

STUDENT: What if I fail the test and then fail the whole class? What if I have to retake the class next semester as a result?

学生：如果我考砸了，然后整个课程都挂科了怎么办？如果我因此不得不下学期重修这门课怎么办？

AI ACADEMIC COACH: That’s a pretty scary thought. I think that would worry almost anyone. But, if you fail the test, would you really fail the whole class?

AI 学业教练：这是一个相当可怕的想法。我认为这几乎会让任何人担心。但是，如果你考砸了，真的会挂科吗？

STUDENT: Well, realistically, probably not.

学生：嗯，实际上，可能不会。

AI ACADEMIC COACH: I agree. Odds are pretty good you’ll have a chance to raise your overall grade even if you do not do well today. One more question: I do not see any record that you’ve failed a test before in this class. What evidence is there that you’ll fail this one?

AI 学业教练：我同意。即使你今天表现不佳，你也有很大机会提高你的总体成绩。还有一个问题：我没有看到你这门课以前有考试不及格的记录。有证据表明你这次会失败吗？

STUDENT: None. But I’m still nervous.

学生：没有。但是我还是很紧张。

AI ACADEMIC COACH: Of course you are, because tests bring out anxiety in all of us. Let’s try some quick breathing exercises together that might help.

AI 学业教练：当然，因为考试让我们所有人都感到焦虑。让我们一起试试一些快速的呼吸练习，可能会有所帮助。

An AI tutor naturally guides a student through their educational journey, forming genuine connections and gaining deep understanding of their needs. Given these attributes, it’s not much of a leap to envision the AI becoming a kind of coach to the student too.

AI 导师自然地引导学生完成他们的教育旅程，形成真实的连接并深入了解他们的需求。鉴于这些特性，很容易想象 AI 也成为了学生的一种教练。

But should it? Even if it is really good at listening and talking, how competent can an AI coach or therapist really be?

但它应该这样做吗？即使它真的很擅长倾听和交谈，AI 教练或治疗师到底能有多胜任？

In the 1960s, a researcher at MIT created a surprisingly simple but notable application that helped pave the way for the development of modern chatbots and virtual assistants. Working in the Artificial Intelligence Laboratory, Joseph Weizenbaum developed ELIZA, one of the world’s first natural-language-processing computer programs that was able to simulate a conversation. Weizenbaum set out to show how artificial the communication was between a human and an AI, but instead he was shocked to learn that many people became emotionally attached to the artificial intelligence. Those using ELIZA often forgot that they were conversing with a computer. As the story goes, Weizenbaum’s secretary even asked him to leave the room occasionally so that she would be able to have a “real conversation” with the AI in confidence.

在 20 世纪 60 年代，麻省理工学院的一位研究员创造了一个出人意料的简单但显著的应用程序，这有助于为现代聊天机器人和虚拟助手的发展铺平道路。在人工智能实验室工作时，约瑟夫·魏岑鲍姆开发了 ELIZA，这是世界上第一个能够模拟对话的自然语言处理计算机程序之一。魏岑鲍姆的目的是展示人类和 AI 之间的沟通有多么人工化，但他惊讶地发现，许多人对人工智能产生了情感依赖。那些使用 ELIZA 的人经常忘记他们是在与计算机对话。据说，魏岑鲍姆的秘书甚至偶尔会要求他离开房间，以便她能够自信地与 AI 进行“真正的对话”。

One of the scripts Weizenbaum developed for ELIZA was called DOCTOR, a program that imitated a therapist. This therapist employed the psychological approach developed by Carl Rogers, known as Rogerian theory. Basically, a patient spoke to DOCTOR, and DOCTOR responded as a Rogerian therapist might, by using nondirectional questioning and reframing statements. ELIZA was not complicated or sophisticated; it simply rephrased your statements. The reason it worked was simple enough, though. Sometimes we all just crave a good listener. On one level, it seems almost fraudulent that an algorithm rephrasing whatever you said can make you feel heard. The thing was, it actually did help people. The ability to pause, reframe our questions, and dig a little bit deeper into ourselves is therapeutic.

魏岑鲍姆为 ELIZA 开发的脚本之一被称为 DOCTOR，这是一个模仿治疗师的程序。这位治疗师采用了卡尔·罗杰斯发展起来的心理方法，被称为罗杰斯理论。基本上，患者与 DOCTOR 对话，而 DOCTOR 则像罗杰斯治疗师一样，通过使用非定向提问和重构陈述来回应。ELIZA 并不复杂或精细；它只是重新措辞你的陈述。其有效的原因却非常简单。有时我们都渴望一个好的倾听者。在某种程度上，一个算法通过重新措辞让你觉得自己被听到了，这似乎几乎是欺诈性的。但实际上，它确实帮助了人们。停顿、重构我们的问题，以及更深入地挖掘我们自己的能力，都是有治疗作用的。

Once again, to his surprise, Rogers found that DOCTOR worked as well as a human therapist in key tasks. This is huge when we consider that since 2010 the United States has seen a 40 percent rise in mental health crises among school-age children. That includes one in three university students. A similar percentage leaves higher education without obtaining the degree for which they enrolled due to anxiety and depression and feelings of isolation. In 2023 the U.S. surgeon general, Vivek Murthy, called this phenomenon an epidemic of loneliness.

再一次，令他惊讶的是，罗杰斯发现 DOCTOR 在关键任务上与人类治疗师一样有效。当我们考虑到自 2010 年以来，美国在学龄儿童中精神健康危机上升了 40％，这一点是巨大的。这包括三分之一的大学生。由于焦虑、抑郁和孤独感，类似比例的学生未能获得他们注册的学位就离开了高等教育。2023 年，美国卫生局局长维维克·默西称这种现象为孤独流行病。

“I realized that behind so many of the stories I was hearing around chronic illness, around depression, around addiction were these threads of loneliness where people would often say to me, ‘You know, I feel I have to carry all these burdens in my life by myself,’ or, ‘I feel if I disappear tomorrow, nobody would even care. They wouldn’t even notice.’ People felt invisible,” Murthy tells me. “And it wasn’t just the people we stereotypically think of as lonely—perhaps the older individual who’s living at home by themselves. I was hearing this from college students who were surrounded by thousands of other students on campus but still felt profoundly alone. I was hearing this from moms and dads in neighborhoods where they were connected to the school, connected to their neighbors, but still didn’t quite feel like there was anybody that they could truly confide in, or anyone who truly got them and understood them, and so on.”

“我意识到，在我听到的关于慢性病、抑郁症、上瘾症的许多故事背后，是孤独的线索，人们常对我说，‘你知道吗，我觉得我必须自己承担生活中的所有负担，’或者，‘我觉得如果我明天消失了，没人会在乎。他们甚至不会注意到。’人们感到自己是隐形的，”默西告诉我。“这不仅仅是我们刻板印象中认为的孤独者——可能是独自生活在家里的老年人。我从大学校园里被成千上万的其他学生包围的大学生那里听到这些，但他们仍然感到非常的孤独。我从那些与学校有联系、与邻居有联系的家长那里听到了这一点，但他们仍然不觉得有任何人可以真正倾诉，或者没有人真正了解他们，等等。”

Murthy came to realize that loneliness, and this feeling of being isolated, is extraordinarily common, and it has profound effects on and implications for our physical and mental health. We also know that both mental health problems and academic underperformance have a similar cause in students: an innate sense that they have little control and purpose in their lives, he says. We have seen a much-needed turn to focus on the mental health of our kids, and for good reason. Longitudinal surveys published in JAMA Pediatricsshow that anxiety and depression have both skyrocketed among students since 2020 due to the pandemic and its associated effects, including stress, social isolation, uncertainty about the future, financial difficulties, and concerns about health and safety.

默西意识到，孤独感和这种被孤立的感觉极为普遍，并且对我们的身心健康有着深远的影响。我们还知道，学生中的心理健康问题和学业表现不佳有着相似的原因：一种他们对自己的生活几乎没有控制力和目标感的内在感觉。他说，我我们已经看到人们开始关注孩子的心理健康，这是非常必要的。发表在《美国医学会杂志·儿科学》上的长期调查显示，由于疫情及其相关影响，包括压力、社会隔离、对未来的不确定性、经济困难以及健康和安全问题，自 2020 年以来，学生的焦虑和抑郁症都飙升了。

Today, more than ever, issues of mental health have as much of a place in our schools as STEM. It was only a matter of time before people started looking to digital psychological aid, otherwise known as integrative psychological artificial intelligence. If ELIZA could help people feel heard fifty years ago, then perhaps modern AIs could go much further. To be clear, I am not espousing replacement of real therapists. No matter how attached people become to their AI therapist—or AI tutor, for that matter—the technology will never, and should never, replace the human element. However, AI can be much more accessible whenever, and wherever, someone might need it, often at times when a human counselor or therapist is unavailable or cost prohibitive. We are already seeing examples of this.

如今，比以往任何时候，心理健康问题在我们的学校中与 STEM（科学、技术、工程和数学）课程一样重要。人们开始寻求数字心理援助，也就是综合心理人工智能，这只是时间问题。如果 ELIZA 在五十年前可以帮助人们感到被倾听，那么现代 AI 或许可以走得更远。需要明确的是，我并不是主张用 AI 替代真正的治疗师。不管人们对 AI 治疗师（或 AI 导师）有多么依赖，这项技术永远不会也不应该取代人类元素。然而，AI 可以在任何时候、任何地方更容易获得，尤其是在人类顾问或治疗师无法获得或费用高昂时。我们已经看到了这样的例子。

In 2022, for instance, researchers from the South China University of Technology tested the efficacy of AI-based therapeutic interventions by having a chatbot deliver cognitive behavioral therapy, another traditional therapy for treating depression and anxiety. The therapy chatbot was able to reduce depression in people within four months and anxiety in people in as quickly as one month. It worked significantly better than self-help books.

例如，2022 年，华南理工大学的研究人员通过让聊天机器人提供认知行为疗法测试了基于 AI 的治疗干预的有效性，这是一种治疗抑郁和焦虑的传统疗法。治疗聊天机器人能够在四个月内减少人们的抑郁症状，并在一个月内迅速减少人们的焦虑症状。它的效果显著优于自助书籍。

Then GPT-4 arrived, and our team began experimenting with AI-based coaching and support. Part of this involved exploring ways to do pedagogical or mindset interventions on Khanmigo. We did this exploration with Angela Duckworth, a professor of psychology at the University of Pennsylvania who is one of a handful of leaders who studies and creates lightweight interventions, changing people’s mindset by increasing grit, perseverance, and other metacognitive capabilities. Because most of these interventions require a trained expert to deliver them, they are not generally accessible. We collaborated with Duckworth and her team to see how we might work with generative AI to scale these interventions at a high level of consistency, bringing this benefit to tens of millions of students.

然后 GPT-4 来了，我们的团队开始尝试基于 AI 的教练和支持。这其中一部分涉及探索如何在 Khanmigo 上进行教学或心态干预。我们与安吉拉·达克沃斯（Angela Duckworth）合作进行了这项探索，她是宾夕法尼亚大学的心理学教授，是研究和创建轻量干预的少数领导者之一，通过增加勇气、毅力和其他元认知能力来改变人们的心态。由于这些干预措施大多需要训练有素的专家来执行，因此通常无法普及。我们与达克沃斯及其团队合作，探索如何与生成式 AI 合作，以高水平的一致性扩展这些干预措施，将这种好处带给数千万学生。

“I think that people will always want to be with, understood by, and cared for by other people, but we are seeing a lot of promise with simulated interventions,” Duckworth says. Given the transactional nature of therapy, psychological services are not always there when you need them. Even if you do have access to proper mental health services, the AI therapist is going to fill in the gaps. It will connect with you when it is three in the morning and you are having a panic attack before class, or when you freeze before the SAT and simply cannot reach a therapist on the phone.

“我认为人们总是希望与他人在一起，被他人理解和关心，但我们在模拟干预中看到了很多希望，”达克沃斯说。鉴于治疗的交易性质，心理服务并不总是在你需要的时候出现。即使你可以获得适当的心理健康服务，AI 治疗师也会填补空白。它会在凌晨三点，你在上课前恐慌发作时与你联系，或者当你面对 SAT 考试时大脑一片空白，却无法通过电话联系到治疗师时与你联系。

Consider this test intervention:

考虑一下这个测试干预：

AI ACADEMIC COACH: Look around you. What objects do you notice?

AI 学业教练：环顾四周。你注意到了哪些物品？

STUDENT: A desk. A pencil. A cat. A cell phone.

学生：一张桌子。一支铅笔。一只猫。一部手机。

AI ACADEMIC COACH: Okay, now, which of these objects do you think is helping you with your focus, and which of these are hurting you?

AI 学业教练：好的，现在，你认为这些物品中哪些有助于你的注意力，哪些在干扰你？

STUDENT: Well . . . you know, the cell phone sometimes distracts me.

学生：嗯......你知道，手机有时会分散我的注意力。

AI ACADEMIC COACH: Well, all right, is there anything you can do about that?

AI 学业教练：嗯，好吧，那有什么你能做的事情吗？

STUDENT: I could maybe put it somewhere out of sight.

学生：我可以把它放在看不见的地方。

AI ACADEMIC COACH: There’s actually a great amount of research that shows that the farther away you place your cell phone, the higher your GPA is.

AI 学业教练：实际上，有大量研究表明，你把手机放得越远，你的 GPA 就越高。

In this example, the AI is engaging on a behavioral psychology level, offering what Duckworth calls “situational modifications,” or interventions that help young people develop metacognitive skills. “The AI enables truly interactive interventions, with students interacting with, and getting feedback, in real time,” she says.

在这个例子中，AI 在行为心理学层面上进行互动，提供达克沃斯称之为“情境修改”的干预措施，帮助年轻人发展元认知技能。“AI 使真正的互动干预成为可能，学生可以实时互动并获得反馈，”她说。

The AI is great at proposing situational modifications so that students are happier and more focused and productive and increase their growth mindset. Traditionally, only highly trained educational psychologists administered them. With the AI, you can now bring these interventions to students whenever they want or need them. “Even as we develop these self-regulation interventions, we’re also thinking about ways to make AI psychologically wiser,” Duckworth says. She and her research team hope that the world will soon be able to use an artificial intelligence that can perform such interventions in as dynamic a way as a human being can, with consistency and reliability. The problem of administering new interventions to every human being on the planet in a cost-effective way disappears when the method of administrating treatments is as pervasive as Khan Academy, and people can access it from a cell phone.

AI 非常擅长提出情境修改，使学生更快乐、更专注、更有生产力并增强他们的成长心态。传统上，这些干预措施只有经过高度培训的教育心理学家才能执行。有了 AI，你现在可以随时随地将这些干预措施带给学生。“即使我们在开发这些自我调节干预措施，我们也在考虑如何让 AI 在心理上更聪明，”达克沃斯说。她和她的研究团队希望世界很快能够使用一种能够以人类一样动态方式执行此类干预措施的人工智能，并具有一致性和可靠性。当治疗方法像可汗学院一样普及，人们可以通过手机访问时，为每个人实施新干预措施的成本效益问题就消失了。

Duckworth’s interventions are only one class of best-practice support that AI might now be able to scale. Encouraging evidence on this front is already starting to emerge. The Journal of Medical Internet Researchreports that these AIs show more promise than pre-AI general mental health applications that only offer one-size-fits-all approaches and surface-level interventions.

达克沃斯的干预措施只是 AI 现在可能能够扩展的一类最佳实践支持。在这一方面的令人鼓舞的证据已经开始出现。《医学互联网研究杂志》报告称，与仅提供一刀切方法和表面干预措施的预人工智能通用心理健康应用程序相比，这些人工智能显示出更大的希望。

People share a common assumption that AI is going to be more helpful for tasks that require IQ than tasks that require EQ, or emotional quotient. Early explorations in the effort to create AI tutors, coaches, and counselors, along with the AI-led interventions we’re working on with Angela Duckworth’s team, make me wonder whether that is true.

人们普遍认为，AI 在需要智商（IQ）的任务上比在需要情商（EQ）或情感商数的任务上更有帮助。在努力创建 AI 导师、教练和顾问的早期探索中，以及我们与安吉拉·达克沃斯团队合作的 AI 主导的干预措施中，我不禁怀疑这是否是真的。

Because AIs aren’t sentient, they can’t be truly empathetic. Empathy involves sensing and modeling other’s emotions and contexts in your own mind. They can, however, simulate empathy quite well. Even with just a chat interface, large language models can interact in ways that are hard to discern from a well-trained, caring therapist. Engineers are augmenting these models with listening, speech, and vision capabilities that can add to the AI’s “understanding” of where the user is emotionally. Perhaps we should introduce a new term, artificial empath, or AE, as a great tool in the fight against loneliness, depression, and anxiety.

由于 AI 没有感知能力，它们无法真正共情。共情包括在你的脑海中感知和模拟他人的情绪和环境。然而，它们可以很好地模拟共情。即使只是通过聊天界面，大型语言模型也可以以与训练有素的关怀治疗师难以区分的方式进行互动。工程师们正在增强这些模型的听觉、语音和视觉功能，可以增加 AI 对用户情感状态的“理解”。也许我们应该引入一个新术语，人工共情（artificial empath，AE），作为对抗孤独、抑郁和焦虑的有力工具。

“点赞有美意，赞赏是鼓励”