内容来源：https://www.sciencenews.org/article/ai-making-us-dumber-learning-skills-risk

内容总结：

AI是否让人变笨？研究显示：技能面临退化风险，但关键在于使用方式

随着人工智能工具日益普及，关于其是否削弱人类思维能力的争论愈发激烈。最新研究表明，AI确实可能导致某些技能退化，但合理使用AI作为“教练”而非“拐杖”，则能有效保持甚至提升我们的能力。

AI替代思考，技能随之退化

加拿大滑铁卢大学行为科学家特伦特·卡什及其同事在《认知科学趋势》期刊上发表的观点指出，当某项具体技能被AI接管后，人们很可能无法良好保持该技能。“证据非常明确：如果我们把某项特定技能外包给AI，就很难再熟练掌握它。”

多项研究证实了这一现象。2025年一项针对医生肠镜息肉检测的研究发现，在使用AI辅助三个月后，医生在无辅助情况下的息肉检出率明显下降，似乎“忘记了该寻找什么”。类似情况也出现在学生群体中——使用ChatGPT辅助练习数学题的高中生，在移除工具后表现反而不如从未使用过AI的同学；使用AI助手解答SAT阅读练习题的考生，在失去AI后不仅正确率下降，还更容易直接放弃答题。

保持认知参与，AI可成学习助力

然而，研究也表明并非所有AI使用都会导致能力退化。关键在于AI工具的类型和使用方式。当AI提供提示、指导和范例，而非直接给出完整答案时，反而能促进学习。

宾夕法尼亚大学沃顿商学院经济学家阿尔普·森古及其团队设计了一款带有学习防护栏的AI工具——它像ChatGPT一样工作，但不会直接给出答案，而是提供解题提示和鼓励。结果显示，使用该工具的学生与传统方式学习的学生效果相当。森古强调：“当你解题时，需要亲自动手、弄脏双手，真正学习的时刻正是在攻克新概念的过程中。”

另一项研究中，沃顿商学院行为科学家本杰明·利拉·卢特格斯发现，AI在教授求职信修改技巧方面与专业人士反馈同样有效。值得注意的是，即使AI直接重写了整封信件，参与者仍会花时间仔细琢磨并修改语言，保持了积极的认知参与。

专家建议：外包琐事，而非手艺

卡什将这一现象概括为“用进废退”，建议人们让自己保持在“认知回路”中，不要让AI替代全部思考过程。他认为，关于AI使人完全丧失思考能力的末日预言可能被夸大了。“人类具有韧性，我不认为AI会突然把我们的脑子变成一团浆糊。”

利拉·卢特格斯则给出了更为实用的建议：人们应当思考哪些技能是自己珍视并愿意保留的，无论是遣词造句的能力，还是利用现有食材创造美味佳肴的本领。“你不应该把那些让你快乐、赋予你意义的事情委托出去。外包那些琐事，而不是手艺。”

中文翻译：

人工智能是否在让我们变得更笨？未必。但我们的技能正面临风险

研究表明，最好把人工智能当作教练，而非拐杖

如今的人工智能可以替我们写大学论文、规划三餐、起草邮件。

但这类辅助带来的益处却引发了截然不同的观点。一些人将人工智能视为一种工具，可以腾出时间去做其他事情。另一些人则担心，把思考外包给电脑会削弱我们独立思考的能力。（近期不乏标题警告称，人工智能正让我们变得懒惰、愚蠢，甚至导致文明的覆灭。）

真相或许介于两者之间，加拿大滑铁卢大学行为科学家特伦特·卡什表示。他与其同事于7月9日在《认知科学趋势》期刊上发表文章指出，如果一个人习得的技能——比如如何解数学题或在做结肠镜检查时发现息肉——被人工智能接管，这些技能确实可能退化。“证据极其明确：如果我们把某项特定技能外包给人工智能，”他说，“我们很可能无法很好地保留这项技能。”

新出现的证据表明，人工智能工具的类型及其使用方式，决定了人们能否保住或习得技能。当人工智能取代了练习某项技能所需的脑力劳动时，它最有可能削弱我们的能力。相反，在学习过程中借助人工智能的反馈或示例保持思维参与，或许能保住甚至增强我们的技能。

卡什表示，归根结底，担心人工智能像某种数字折叠刀一样削磨人类智慧，这种忧虑可能被夸大了。他也不太相信人工智能会带来某些人设想的末日场景——人们彻底忘记如何思考。

“人类是有韧性的，”他说。“我不认为（人工智能）会突然把我们的脑子变成一团浆糊。”

当人工智能取代了思考过程，技能便会退化

大量新研究正在描绘我们的大脑如何应对人工智能工具提供的智力辅助。答案有时看起来并不乐观。

在一项2025年针对受过结肠镜检查息肉识别训练的医生的研究中，人工智能工具的辅助似乎削弱了这项专业技能。当医生在使用该工具三个月后回到无辅助工作状态时，他们的息肉检出率下降了。研究作者指出，医生们“有点忘了该找什么”，卡什说。

这种模式也出现在不那么小众的技能中。在高中生学习新数学概念的研究中，像ChatGPT这样的人工智能工具可以帮助解答练习题。但一旦移除该工具，这些学生的表现比从未使用过该工具的学生更差，宾夕法尼亚大学沃顿商学院经济学家阿尔普·孙古及其同事在《美国国家科学院院刊》上报告了这一结果。

研究人员在阅读理解方面也报告了类似发现。使用人工智能助手解答SAT练习题的人，在人工智能被移除后很难给出正确答案，匹兹堡卡内基梅隆大学的机器学习研究员刘格蕾丝及其同事于今年4月在arXiv.org上发布了一篇预印本论文报告了这一结果。而在失去人工智能助手后，参与者比从未使用过该工具的人更倾向于跳过问题。团队写道，他们干脆放弃了。“这些发现尤其令人担忧，因为坚持是技能习得的基础。”

总体来看，这些结果表明，当人们让人工智能接管自己的任务时——无论是识别异常、解决问题还是坚持攻克难题——他们都会陷入困境。对卡什而言，核心启示归结为一种“不用则废”的心态，他称之为让自己保持认知参与。这意味着不要让人工智能替你做所有思考。“如果你不参与认知工作，你就学不会这项技能，”他说。

人工智能能帮助人——只要它让人保持思考

使用人工智能并不意味着你的技能一定会崩塌。完全可以付出学习所需的努力，同时从我们的数字朋友那里获得一点帮助，孙古说。这意味着人工智能工具应提供指导或示例，而不是直接吐出完整的答案。

在高中生研究中，孙古和合作者设计了一款带有学习护栏的人工智能工具。它的运作方式类似ChatGPT，但被设计成不会直接给出答案。相反，人工智能为学生提供解题提示和鼓励。团队发现，这种方法帮助学生学习新数学概念的效果，与传统方式（即通过书本学习）的学生相当。

孙古说，把人工智能当作导师与把它当作答案机器是不同的。学生仍然需要为学习投入思考。攻克一个新概念的挣扎过程才是真正的学习发生之处，他说。“当你解题时，你需要亲自动手。”

对于那些愿意付出努力的人来说，人工智能可以成为好老师，同样来自沃顿商学院的行为科学家本杰明·利拉·卢特格斯说。他和同事发现，在教人如何修改求职信时，人工智能工具的效果与接受专业人士反馈相当。在借助人工智能工具或人工反馈练习之后，参与者无需任何帮助即可修改一封写得糟糕的求职信。随后，人工评审决定哪些信件可能获得面试机会。两组参与者的信件获得面试机会的可能性相当，研究人员在2026年的一份预印本论文中报告了这一结果。

在利拉·卢特格斯的实验中，人工智能工具只是直接重写了整封求职信。这听起来可能像参与者没怎么动脑，但团队发现的并非如此。人们并没有立即提交人工智能修改后的草稿，而是花了一段时间琢磨措辞。“他们仍然在干活，”利拉·卢特格斯说。结果表明，人工智能可以通过示范来教学，“前提是有愿意学习的人。”

尽管让人脑保持参与的人工智能系统或许能磨砺我们的一些技能，但科学家表示，我们应该谨慎选择外包的内容。如果你珍视某项技能，就不要把它交给人工智能，卡什建议。“我们知道技能来自练习，也通过练习得以维持。”

利拉·卢特格斯也表示赞同。他说，人们应该想清楚自己想保护哪些技能，无论是想出一句完美的措辞，还是用厨房里的食材做出一道美味的菜肴。“你不应该把那些让你快乐、赋予你意义的事情委派出去，”他说。“外包那些琐事，而不是手艺。”

英文来源：

Is AI making us dumber? Maybe not. But our skills are at risk
Research suggests it's best to use AI as a coach, not a crutch
AI today can write our college essays, plan our meals and draft our emails.
But the benefits of such assistance incite wildly different opinions. Some see AI as a tool that frees up time for other activities. Others fear that offloading our thinking to a computer will dull our capacity to think for ourselves. (See recent headlines warning about AI making us lazy, stupid and prompting civilization’s downfall.)
The truth may lie somewhere in the middle, says Trent Cash, a behavioral scientist at University of Waterloo in Canada. A person’s learned skills, such as how to work through math problems or detect polyps while performing a colonoscopy, may indeed atrophy if taken over by AI, he and his colleagues argue July 9 in Trends in Cognitive Sciences. “The evidence is extremely clear that if we offload a specific skill to AI,” he says, “we’re probably not going to retain that skill particularly well.”
Emerging evidence suggests that the type of AI tool and how it’s used makes the difference in a person’s ability to hang on to or pick up skills. AI may be most likely to weaken our skills when it replaces the mental work needed to practice that skill. Instead, staying mentally engaged using AI feedback or examples while learning may preserve, and perhaps even enhance, our skills.
Ultimately, Cash says, fears of AI whittling away human intelligence like some sort of digital pocketknife are probably overblown. And he’s hard-pressed to believe AI will bring about the doomsday scenario some have envisioned — that people will completely forget how to think.
“Humans are resilient,” he says. “I don’t think that [AI is] going to suddenly turn our brains to mush.”
When AI replaces the work, skills fade
A raft of new research is sketching out how our brains deal with the intellectual assistance AI tools offer. The answer can sometimes look grim.
In a 2025 study of doctors trained to spot polyps during colonoscopies, help from an AI tool seemed to erode this specialized skill. When doctors returned to working unassisted after three months of using the tool, their polyp detection rate dropped. The study’s authors suggest that the doctors “kind of forgot what to look for,” Cash says.
It’s a pattern that plays out in less niche skills, too. Among high school students learning a new math concept, an AI tool like ChatGPT can help solve practice problems. But take the tool away, and the students performed worse than those who never used the tool in the first place, economist Alp Sungu of the Wharton School of the University of Pennsylvania and his colleagues reported in the Proceedings of the National Academy of Sciences.
Researchers reported something similar with reading comprehension. People using an AI assistant to solve practice SAT questions had a hard time coming up with correct answers when the AI was taken away, machine learning researcher Grace Liu of Carnegie Mellon University in Pittsburgh and her colleagues reported in a preprint on arXiv.org in April. And after losing their AI helper, participants were more likely to skip questions than those who had never used the tool. They simply gave up, the team wrote. “These findings are particularly concerning because persistence is foundational to skill acquisition.”
Altogether, the results suggest that people struggled when they let AI take over their tasks, be it spotting abnormalities, solving problems or sticking with difficult questions. For Cash, the takeaway boils down to a “use it or lose it” mentality that he calls keeping yourself in the cognitive loop. That means not letting AI do all your thinking for you. “If you’re not engaging with the cognitive work, you’re not going to learn the skill,” he says.
AI can help when it keeps people thinking
Using AI doesn’t have to mean your skills will crumble. It’s quite possible to put in the effort needed for learning, while also getting a little help from our digital friends, Sungu says. That means AI tools that offer guidance or examples rather than spitting out fully formed solutions.
In the study with high school students, Sungu and collaborators designed an AI tool with some learning guardrails. It worked like ChatGPT but was designed not to give away answers. Instead, the AI offered students problem-solving hints and encouragement. That approach helped students learn a new math concept as well as students who studied the traditional way, with books, the team found.
Using AI as a tutor is different from using it as an answer machine, Sungu says. Students still had to put some thought into their studying. The struggle of working through a new concept is where real learning happens, he says. “When you’re solving a problem, you need to get your hands dirty.”
For those willing to put in some effort, AI can be a good teacher, says Benjamin Lira Luttges, a behavioral scientist also at Wharton. He and his colleagues found that when teaching people how to edit cover letters, an AI tool was as effective as receiving feedback from professionals. After practicing with either the AI tool or human feedback, participants edited a poorly written cover letter without any help. Then, human evaluators decided which letters would probably secure a job interview. Letters from both groups were equally likely to land an interview, the researchers reported in a 2026 preprint on arxiv.org.
In Lira Luttges’ experiments, the AI tool simply rewrote the entire cover letter itself. That may sound like the study participants weren’t doing much thinking, but it’s not what the team found. Instead of immediately submitting the AI-revised draft, people sat with it for a while, tinkering with the language. “They were still working,” Lira Luttges says. The result suggests AI can teach by example, “if you have people who are motivated to learn.”
Though AI systems that keep people’s brains engaged might be able to sharpen some of our skills, scientists say we should be careful what we’re outsourcing. If there’s a skill you value, don’t give it over to AI, Cash advises. “We know skills come from practice and are maintained through practice.”
Lira Luttges agrees. People should think about what skills they want to protect, he says, whether that’s coming up with the perfect way to phrase a sentence or devising a delectable dish using items in your pantry. “You shouldn’t delegate the things that you enjoy and that give you meaning,” he says. “Outsource the crap, not the craft.”