July 22, 2025   2025 年 7 月 22 日

Why AI is making us worse thinkers (and how to avoid it)
为什么人工智能让我们的思维变得更差（以及如何避免）

Exploring how AI usage erodes our critical thinking and learning to treat it as a thinking partner for better outcomes.
探讨人工智能的使用如何侵蚀我们的批判性思维，以及如何将其视为思考伙伴以获得更好的结果。

Hi there, it's Adam from Untools.
大家好，我是 Untools 的 Adam。

With AI being so widespread these days, I think it’s important to talk about its impact on how we think as well as how much we think when we use it.
如今人工智能如此普及，我认为有必要讨论它对我们思维方式以及我们使用它时思考深度的影响。

I use AI in various forms daily and at times I’ve definitely felt the lure of just taking its outputs and using them with a sort of blind trust. Maybe you’ve felt something similar. Submitting to that is dangerous, though.
我每天以各种形式使用人工智能，有时我确实感受到仅仅接受其输出并盲目信任的诱惑。也许你也有类似的感受。但屈服于这种诱惑是危险的。

What I’ve found in my experience and in recently published research is that we’re generally better off treating AI as a thinking partner rather than a tool we delegate tasks to. In today’s post, we’ll explore why that is and how to work with AI more collaboratively.
根据我的经验和最近发表的研究发现，我们通常更适合将人工智能视为思考伙伴，而不是一个我们委托任务的工具。今天的文章中，我们将探讨原因以及如何更协作地与人工智能共事。

The AI thinking paradox
人工智能思维悖论

If you use AI regularly, you've probably noticed it gets things wrong from time to time: it can hallucinate and sometimes doesn’t produce very accurate outputs. On top of that, AI misses a lot of our context and nuances.
如果你经常使用人工智能，你可能已经注意到它有时会出错：它可能产生幻觉，有时输出的结果并不十分准确。除此之外，人工智能还会遗漏很多我们的背景信息和细微差别。

That means we need a lot of critical thinking to question and verify AI’s outputs. The problem is that research shows that frequent use of AI actually erodes critical thinking skills. That’s the paradox: Using AI a lot weakens the very skill that’s necessary for working with it well.
这意味着我们需要大量的批判性思维来质疑和验证人工智能的输出。问题是，研究表明频繁使用人工智能实际上会削弱批判性思维能力。这就是悖论：大量使用人工智能会削弱与之良好合作所必需的技能。

One study by Microsoft of 319 knowledge workers showed that using generative AI “can inhibit critical engagement with work … and diminished skill for independent problem-solving”.
微软对 319 名知识工作者进行的一项研究显示，使用生成式人工智能“可能抑制对工作的批判性参与……并削弱独立解决问题的能力”。

And this study by Michael Gerlich from SBS Swiss Business School found that “participants who reported higher usage of AI tools consistently showed lower scores on critical thinking assessments.”
而这项由瑞士商学院（SBS Swiss Business School）的 Michael Gerlich 进行的研究发现，“报告使用 AI 工具较多的参与者，在批判性思维评估中的得分持续较低。”

The main problem causing this is called “cognitive offloading”.
导致这一问题的主要原因被称为“认知卸载”。

The dangers of cognitive offloading
认知卸载的危险

Cognitive offloading means delegating your thinking to AI rather than thinking with it. Sure, it’s faster and easier but at the expense of your critical thinking skills and the potential of using AI.
认知卸载是指将你的思考任务交给 AI 处理，而不是与 AI 共同思考。当然，这样做更快更简单，但代价是你的批判性思维能力以及利用 AI 潜力的减少。

Recognising when you’ve been cognitive offloading is the first step to improving how you work with AI. Take a moment to reflect and ask yourself:
识别自己何时进行了认知卸载，是改善与 AI 协作方式的第一步。花点时间反思并问自己：

• How often do you critically evaluate AI’s output?
  你多久会对人工智能的输出进行一次批判性评估？

• How much do you question its answers and cross-check them against reliable sources?
  你会在多大程度上质疑它的答案，并将其与可靠来源进行交叉核对？

• How well could you explain the reasoning behind the solutions or decisions that AI suggests?
  你能多好地解释 AI 所建议的解决方案或决策背后的推理？

If your answers lean towards “not often” or “not very well”, you might offloading too much to AI. When you consistently offload thinking to AI, you gradually lose confidence in your own judgment and become less able to spot when AI gets things wrong.
如果你的回答倾向于“不是经常”或“不是很擅长”，那你可能把太多思考任务交给了人工智能。当你持续将思考工作交给人工智能时，你会逐渐失去对自己判断的信心，也会变得不太能发现人工智能出错的情况。

The good news is that a change in how you approach AI can help you get more out of it without compromising your thinking abilities.
好消息是，改变你使用人工智能的方式可以帮助你更好地利用它，同时不损害你的思维能力。

How to work with AI as a thinking partner
如何将 AI 作为思考伙伴一起工作

It all boils down to moving away from delegation towards what I would call ‘cautious collaboration’. AI works best (and you work best with AI) when you treat it as a thinking partner, a sounding board, a peer who gives you feedback.
归根结底，就是要从简单的委托转向我所称的“谨慎协作”。当你把 AI 当作思考伙伴、一个可以倾诉的对象、一个给予反馈的同伴时，AI 的效果最佳（你与 AI 的合作也最有效）。

Why ‘cautious’? Remember that AI doesn’t know your full context and can be prone to hallucinations and inaccuracies. It’s useful to think in terms of a generation/verification loop: whatever AI generates, a human should verify.
为什么要“谨慎”？请记住，AI 并不了解你的全部背景，且可能出现幻觉和不准确的情况。将其视为一个生成/验证循环是很有用的：无论 AI 生成什么内容，都应由人类进行核实。

There are a few principles I follow when working with AI:
我在使用人工智能时遵循几个原则：

• Be specific: Think about what context AI might need to produce a relevant output. What exactly do you want the output to be?
  要具体：考虑人工智能可能需要什么样的上下文才能生成相关的输出。你到底希望输出是什么？

• Default to suggestions, not decisions: I treat AI as an assistant and ask it for suggestions instead of final decisions or results. This naturally keeps me in control and follows the generate/verify loop.
  默认接受建议，而非决策：我将 AI 视为助手，向它寻求建议，而不是最终的决策或结果。这自然让我保持掌控，并遵循生成/验证的循环。

• Challenge your thinking: I’ll often share my existing thinking (or a draft of my work) with AI and ask it to critique it. Spot weaknesses, point out what I’m missing. Interestingly, this is the generate/verify loop in reverse.
  挑战你的思维：我经常会把自己已有的想法（或工作草稿）分享给 AI，让它进行批评。找出弱点，指出我遗漏的地方。有趣的是，这其实是生成/验证循环的反向过程。

Let’s see how this can be done in practice with a few examples.
让我们通过几个例子看看这如何在实践中实现。

Example 1: Writing a social media post
例子 1：撰写社交媒体帖子

🚫 Delegation: “Write a LinkedIn post about our Q4 results.”
🚫 委托：“写一篇关于我们第四季度业绩的 LinkedIn 帖子。”

✅ Collaboration: “Help me share our Q4 results with our LinkedIn audience in an authentic way that credits the team without being boastful. Here is my first draft – review it and suggest what can be improved.”
✅ 协作：“帮我以一种真实的方式向我们的 LinkedIn 观众分享我们第四季度的业绩，既能体现团队的贡献，又不过于自夸。这是我的初稿——请审阅并提出改进建议。”

There are a few things going in this example: we wrote our own draft first instead of mindlessly delegating it to AI. Then we’re asking it to give us feedback while providing more context on how we want the post to sound. Being specific with AI greatly helps with better output as well.
这个例子中有几件事：我们先自己写了草稿，而不是盲目地交给 AI 处理。然后，我们在提供更多关于文章语气的背景信息的同时，请它给出反馈。对 AI 提出具体要求也大大有助于获得更好的结果。

Example 2: Code review  示例 2：代码审查

🚫 Delegation: “This part of the code results in an error, fix it.”
🚫 委托：“这部分代码导致了错误，修复它。”

✅ Collaboration: “I get this error when I run this code. Analyse it step by step. Then give me possible causes and debugging approaches.”
✅ 协作：“当我运行这段代码时出现这个错误。请逐步分析它。然后给我可能的原因和调试方法。”

Here we’re getting the AI to brainstorm possible solutions to give us a chance to review and pick the best one. Also notice we’re asking it to reason step by step which can help us understand what’s going on.
这里我们让 AI 头脑风暴可能的解决方案，给我们一个审查并选择最佳方案的机会。还要注意，我们让它逐步推理，这有助于我们理解发生了什么。

Example 3: Preparing for a meeting
示例 3：准备会议

🚫 Delegation: “I want to convince our client to approve the presented strategy. Give me arguments supporting it.”
🚫 委托：“我想说服我们的客户批准所提出的策略。给我支持该策略的论据。”

✅ Collaboration: “We’ll be presenting our risk-averse client with this strategy. Review it and identify points that the client might object against. Formulate possible objections and help me rehearse the conversation.”
✅ 合作：“我们将向我们的风险厌恶型客户展示这一策略。请审查并找出客户可能反对的点。制定可能的反对意见，并帮助我练习对话。”

With the collaboration approach in this example, we ask AI to argue against our position instead of doing our work for us. In the process, we can formulate our own arguments and understand both sides well instead of just letting AI do the thinking we need to do.
在这个例子中的协作方法里，我们让人工智能反驳我们的立场，而不是替我们完成工作。在这个过程中，我们可以自己构建论点，并且充分理解双方观点，而不是仅仅让人工智能替我们思考。

How this approach avoids cognitive offloading
这种方法如何避免认知卸载

Notice the common theme in these examples: we engage our critical thinking before interacting with AI. When you create a draft first or analyse the problem before asking AI, that’s thinking you haven’t delegated.
注意这些例子中的共同主题：我们在与人工智能互动之前先进行批判性思考。当你先创建草稿或在询问人工智能之前分析问题时，那是你没有委托出去的思考。

When you develop your own thoughts first, you also have a baseline, something concrete, to compare AI’s output to. And then you can use AI to improve your work and thinking while keeping the cognitive ownership of it.
当你先发展自己的想法时，你也有了一个基准，一个具体的东西，可以用来对比人工智能的输出。然后你可以利用人工智能来改进你的工作和思考，同时保持对认知的掌控权。

Practical tip: Give AI a framework, not just a task
实用建议：给 AI 一个框架，而不仅仅是一个任务

I've found that the easiest way to avoid cognitive offloading is to structure the problem yourself before asking AI to help solve it.
我发现避免认知外包的最简单方法是在请求 AI 帮助解决问题之前，先自己构建问题的结构。

When you give AI a generic task, you're basically saying "do my thinking for me." But when you create a framework first, you outline how to think about the problem first and then use AI to work through it with you. In my experience, this makes a big difference.
当你给 AI 一个通用任务时，基本上是在说“帮我思考”。但当你先创建一个框架时，你先勾勒出如何思考这个问题，然后再用 AI 与你一起解决它。根据我的经验，这会带来很大的不同。

Here’s an example of a framework you could use for decision making or problem solving:
这里有一个你可以用来做决策或解决问题的框架示例：

Current situation: [Status quo and your context]
Goals: [What we want to achieve]
Constraints: [E.g. budget, timeline, other resources]
Options to consider:
- [Option 1]
- [Option 2] 
- [Option 3]
Success criteria: [How we'll know if this worked]

For each option, analyze pros, cons, and implementation steps.

I would encourage you to build your own framework for each specific problem you’re solving. When you build the framework, you're doing the important thinking about things like constraints, available options, and success criteria. AI becomes your thinking partner to work through each part, not your replacement for figuring out the approach.
我鼓励你为每个具体问题构建自己的框架。当你构建框架时，你正在进行重要的思考，比如考虑约束条件、可用选项和成功标准。AI 成为你思考的伙伴，帮助你处理每个部分，而不是替代你去确定解决方案的方法。

You keep ownership of the thinking because the structure comes from you. AI just helps you think through each piece more thoroughly.
你始终掌控思考，因为结构源自于你。人工智能只是帮助你更深入地思考每个部分。

---

Coming up for Vault 💎 members: AI collaboration toolkit
即将为 Vault 💎会员推出：AI 协作工具包

Next week, Vault members will receive a comprehensive guide on AI collaboration techniques, plus a printable reference toolkit. This will include:
下周，Vault 会员将收到一份关于 AI 协作技巧的全面指南，以及一份可打印的参考工具包。内容包括：

• Research-backed prompting fundamentals: How to work with examples, reasoning process and other principles that improve AI output quality.
  有研究支持的提示基础：如何利用示例、推理过程及其他原则来提升 AI 输出质量。

• AI collaboration techniques: Specific techniques for engaging with AI in a collaborative way.
  AI 协作技巧：与 AI 以协作方式互动的具体技巧。

• Thinking tools prompt templates: Practical prompt templates for popular thinking tools like Second-order thinking or Six thinking hats.
  思维工具提示模板：适用于二阶思维或六顶思考帽等流行思维工具的实用提示模板。

• Printable reference cards: A PDF toolkit you can keep handy, share with your team, or use in workshops.
  可打印参考卡：一个 PDF 工具包，方便随身携带、与团队分享或在工作坊中使用。

Along with new monthly deep dives and guides, Vault members get instant access to all previously published premium content.
除了每月的新深度解析和指南外，Vault 会员还可即时访问所有已发布的高级内容。

---

Small changes, better thinking
小改变，更好思考

The shift from delegation to collaboration can be sometimes subtle, but even small changes in how you approach AI can make a big difference. Start with one technique that feels most relevant to your work, and gradually build from there. The goal isn't to use AI perfectly but to use it in a way that makes you think better, not less.
从委托到协作的转变有时很微妙，但即使是你使用 AI 方式上的小变化，也能带来很大不同。先从一种与你工作最相关的技巧开始，逐步积累。目标不是完美使用 AI，而是以一种让你更好思考而非思考更少的方式使用它。

Coming next  即将推出

Next month, we will look at cognitive biases and how they might be impacting your thinking. We’ll explore the most common biases, the research behind them and how to avoid them skewing our thinking.
下个月，我们将探讨认知偏差及其可能对你的思维产生的影响。我们将研究最常见的偏差、相关的研究以及如何避免它们扭曲我们的思维。

Until next time,  下次见，
Adam  亚当

P.S. What's your experience been with AI so far? Are you finding it helps or hinders your thinking? Leave a comment or reply to this email.
附言：到目前为止，你使用人工智能的体验如何？你觉得它是帮助了你的思考，还是阻碍了你的思考？欢迎留言或回复此邮件。