休息室
连接中…
连接中…
查看全部

Is AI a Tool? Rich Hickey's Argument, the Rebuttal, and What Using AI Well Takes

Rich Hickey says LLMs are automation, not tools; a rebuttal says using AI well takes skill. Compare both arguments, the research on reliance, and a checklist.

AI热门阅读约 7 分钟阅读 19发布 更新
目录7 个章节
  1. Is AI a tool? What the debate is really about
  2. What research says about AI reliance and critical thinking
  3. Using AI well: five skills that matter in practice
  4. A quick self-check for over-reliance
  5. What to take away from the tool debate
  6. 常见问题
  7. 来源与参考资料

核心要点

  • Rich Hickey defines a tool as something you wield with continuous control and feedback, and argues LLMs are automation that act on your requests instead.
  • The rebuttal on yyhh.org points out that compilers and spreadsheets automate work yet remain tools, and that a human-directed loop of asking, checking and refining is wielding.
  • A Microsoft Research and Carnegie Mellon survey of 319 knowledge workers found higher confidence in AI was associated with less critical thinking, and higher self-confidence with more.
  • In practice, using AI well is less about prompt phrasing and more about choosing what to delegate, verifying output, integrating it, and owning the result.

Is AI a tool? Rich Hickey, the creator of the Clojure programming language, says no: in his view large language models are automation that you hand work to, not instruments you wield. A widely shared rebuttal says yes, and adds that using AI well takes real skill. Both essays, along with a third piece predicting that thinking will become a hobby, drew heavy discussion on Hacker News in October 2026. This guide, written as of October 13, 2026 from the original essays and peer-reviewed research, lays out each argument, where the evidence on AI reliance actually points, and a practical way to check how you are using these systems.

Is AI a tool? What the debate is really about

On the surface this is an argument about a word. Underneath, it is about what people learn, or stop learning, while working with AI. The phrase "it's just a tool" is often used to mean "it's what you make of it." Hickey argues that framing hides properties of automation that you ignore at your peril. His critics argue the tool framing is exactly what keeps human skill and responsibility in view.

Rich Hickey: LLMs are automation, not tools

In his LinkedIn essay, Hickey settles on a definition: a tool is something you wield. Holding it gives you control and feedback; you supply intention and continuously direct the course of events. Good tools, he writes, are predictable, reliable, deterministic and of bounded function, and LLMs lack those properties. You do not direct an LLM's action; you make a request and it acts on its own. That, in his words, is automation.

His central analogy is the fast-food drive-thru: not a tool for cooking, but automation for avoiding cooking. Applied to research, he argues that if a model reads and summarizes papers for you, the model did the reading, not you. What you miss is learning how good writers write, how arguments are formed, which authors to trust, and how to think. He also warns that tuning prompts to a model's quirks is fragile knowledge that one software update can make obsolete.

The rebuttal: AI is a tool, and using it well takes skill

Huahai's response on yyhh.org accepts that outsourcing work can cost learning opportunities but rejects the conclusion. Picking up a saw does not guarantee a straight cut, and holding a chisel does not make you a sculptor. A tool's usefulness depends on the operator's ability to direct it, interpret feedback, recognize mistakes and correct them. Predictability and reliability, he argues, are criteria for judging tools, not requirements for being one.

He also challenges the tool-versus-automation split. A compiler does substantial work without the programmer directing each operation, and a spreadsheet recalculates without anyone doing the arithmetic, yet both are tools. A developer who asks an LLM to explain unfamiliar code, pushes back, inspects the source, tests a proposed change and rejects or refines it is running a feedback loop with a purpose. He concedes Hickey's point about reading is valid for one specific use: accepting a summary in place of the paper. His preferred distinction is between wielding an instrument with skill and delegating without adequate understanding or oversight.

A wider frame: thinking as a hobby

Diomidis Spinellis took a longer view in Thinking will become a hobby. He points to three forces, better performance, lower cost and round-the-clock availability, and predicts that AI will take over the bulk of cognitive work. Thought, he suggests, may become a leisure pursuit the way hunting and calligraphy did. His advice is to keep pace with AI while building a purposeful life beyond work. This is a forecast and personal view, not a research finding.

Rich Hickey

Core claim
LLMs are automation, not tools
Main examples
Drive-thru, summarized papers
What it can miss
Not every use replaces learning

Huahai (yyhh.org)

Core claim
LLMs are tools; skill matters
Main examples
Saw and chisel, compiler, spreadsheet
What it can miss
Assumes a skilled, attentive user

Diomidis Spinellis

Core claim
Cognitive work will shrink
Main examples
Performance, cost, availability
What it can miss
Long-range prediction, not evidence

What research says about AI reliance and critical thinking

The most relevant evidence comes from a CHI 2025 paper by researchers at Microsoft Research and Carnegie Mellon University. They surveyed 319 knowledge workers who use generative AI at work and collected 936 real examples. Higher confidence in AI doing a task was associated with less critical thinking. Higher confidence in one's own ability was associated with more critical thinking, though participants perceived that as more effort.

The Microsoft Research summary adds texture. For high-stakes tasks that required accuracy, participants said they spent more effort on critical thinking with AI than without it. For routine, low-stakes tasks under time pressure, they spent less. Critical thinking shifted toward verifying information, integrating responses and stewarding the task. The barriers were lack of awareness, limited motivation from time pressure, and difficulty refining prompts in unfamiliar domains.

The study is a self-reported survey that shows associations, not causation. It reflects the tools and participants at its 2025 publication and does not prove that AI use reduces thinking ability.

Read this way, the data gives each side something. Trusting the model and skipping checks looks like Hickey's automation, where the work is handed off. Keeping your own judgment and interrogating the output looks like the rebuttal's feedback loop. The same system can be used either way, and the user's habits decide which.

Using AI well: five skills that matter in practice

Combine the essays and the research and "AI skill" turns out to be a bundle of judgments rather than a single technique. None of these depend on a particular product or model.

  1. Choose what to delegate. Separate the skills you want to build from tasks where you only need the output, and keep doing the former yourself.
    1. Example: read the primary source in a new field yourself, and let AI reformat notes or draft boilerplate.
  2. State goals and standards. Define purpose, audience and quality bar before asking. Participants in the Microsoft study counted clear goal-setting as part of critical thinking.
  3. Verify. Check output against sources, tests and original material. Fluent text is not evidence of accuracy.
  4. Integrate. Adapt the result to your context and fill gaps rather than pasting it in unchanged.
  5. Own the result. If you cannot explain a part of the work when it turns out to be wrong, it is not finished.

Much of the skill lives outside the AI. Domain knowledge is what lets you spot errors and know what to ask, which is why participants found prompting harder in unfamiliar areas. Paradoxically, one of the surest ways to get better at using AI is to keep strengthening the fundamentals of your own field.

A quick self-check for over-reliance

  1. 1

    Write your own answer first

    Before asking, jot down your hypothesis or outline in a line or two. You now have something to compare against instead of accepting the first response.

  2. 2

    Match verification to stakes

    Money, health, legal and security questions deserve source checks and testing every time. Time pressure is exactly when people report cutting corners.

  3. 3

    Supplement reading, do not replace it

    Read important material yourself, then use AI to argue with your interpretation or surface what you missed.

  4. 4

    Work without AI on a schedule

    Periodically handle a core task unaided. It is the simplest way to see whether your own skills are holding up.

Debates like this one move quickly across developer forums and Telegram channels, and new rebuttals keep appearing. Theddari's AI Telegram trends and blog trend dashboard show which AI topics are being discussed most, which can be a starting point for finding the original essays. Popularity is not proof, so check the underlying arguments yourself.

What to take away from the tool debate

Hickey's strict definition exposes the risk hidden in "it's just a tool," and the rebuttal shows that the same definition struggles with compilers and spreadsheets. Both are opinion essays, and there is no official standard that settles the word. The common ground is clearer: handing work off wholesale can cost you learning, while keeping judgment and verification in your hands lets AI support your skill instead of substituting for it. A good next step is to pick one task you gave to AI this week and ask whether you could verify and explain every part of the result yourself.

常见问题

Is AI a tool or automation?
It depends on the definition you adopt. If a tool must be held and continuously directed with direct sensory feedback, as Rich Hickey argues, an LLM looks like automation. If a tool is anything a person directs toward a purpose while checking and correcting the results, an LLM can be a tool with automated features, much like a compiler. The more useful question for your own work is whether you are still doing the judging and verifying.
Does using AI make you worse at thinking?
The evidence points to how you use it rather than how much. In the CHI 2025 survey by Microsoft Research and Carnegie Mellon researchers, people who trusted AI more tended to report less critical thinking. That study relied on self-reports and shows associations, not causation, so it does not prove AI erodes thinking. Skipping verification is the warning sign to watch for.
Is being good at AI the same as being good at prompting?
Only partly. Participants in the Microsoft study described critical thinking as setting clear goals, refining prompts and checking outputs against outside sources and their own expertise. They also said refining prompts was hard in unfamiliar domains. Knowing your field is what lets you notice when an answer is wrong.
Should students or beginners avoid AI while learning?
Not necessarily, but be deliberate. Hickey warns that letting a model read and summarize papers for you means missing how arguments are built and whom to trust. The rebuttal agrees that concern is valid for that use. Reading the source yourself and then using AI to challenge your interpretation supplements learning instead of replacing it.

来源与参考资料

  1. 1Rich Hickey: 'AI' is Not a Tool (LinkedIn)(在新标签页中打开)www.linkedin.com
  2. 2yyhh.org: AI Is a Tool. Using It Well Takes Skill.(在新标签页中打开)yyhh.org
  3. 3Diomidis Spinellis: Thinking will become a hobby(在新标签页中打开)www.spinellis.gr
  4. 4Microsoft Research Blog: The Future of AI in Knowledge Work: Tools for Thought at CHI 2025(在新标签页中打开)www.microsoft.com
  5. 5Microsoft Research: The Impact of Generative AI on Critical Thinking (CHI 2025)(在新标签页中打开)www.microsoft.com