AI时代,我好像不会写东西了 | Psychology Today

正文字数1324 word         

【Para. 1】I began writing a paper the other day; laid out my plan for the "story" and set about pulling the introduction together. I hit roadblocks here and there, struggled to find the right expression for the ideas floating around in my head, paused, stood up and paced, started up again, and eventually packed it up to be dealt with in my next writing session.

【Para. 2】And at some point, it occurred to me: I like this. I find this fun. There is joy in the challenging process of writing.

【Para. 3】Now I am not anti‑AI. I have used it for a variety of purposes. (I will say I am against the theft of work to build AI models, whether that be written work or visual art, and I will be a beneficiary of the successful lawsuit against Anthropic after they stole my book co‑authored with John Eastwood, Out of My Skull.)

【Para. 4】But I do think we lose something if we give up tasks to large language models (LLMs) that we could do ourselves.

【Para. 5】Downsides of Using AI

First, we become more boring. I have written about this elsewhere. LLMs are data‑hungry, predictive algorithms. It's that predictive nature that makes them implant the most probable word in any given sentence. As Mark Bieleke and Wanja Wolff recently highlighted, this will essentially be regression to the mean; we will all begin to sound alike. Imagine if literature did this? All novels reading like they were written by Dan Brown? (You could choose an author you liked more; and to be clear, I have never read a Dan Brown book, but the idea is the same—reading would become pretty boring if all you had available were Stephen King books—and I love King’s work).

【Para. 6】Second, when we divest the task of writing to AI, we forego the joy of striving. Cognitive scientists have generally suggested that, all things being equal, we tend to want to avoid effort—even effortful thinking. That is, we are cognitive misers. This makes a lot of biological sense—the resources we need to function (e.g., glucose, oxygen) are not unlimited, and so we ought to conserve them, and avoiding more effortful tasks is one way to do that.

【Para. 7】But Michael Inzlicht has recently suggested a different account of effort—that it can feel good to do something effortful. This might be because at the end of the process, we have a sense of achievement. In essence, there can be no pleasure without some pain. I think of tattoos. We like the etchings we choose to leave on our skin not merely for their aesthetics and personal relevance, but also because we survived the self‑inflicted torture of getting them in the first place.

【Para. 8】But Inzlicht goes a step further—sometimes we choose things because they are effortful. We know beforehand this thing we’re choosing to do will be hard. But we do it anyway, because we place value on effort. Perhaps it says something about us that we can soldier through.

【Para. 9】Third, if we let AI be our only writing tool, we abdicate the responsibility for developing our own skills and our own voice. This is similar to the complaint that we all start to sound alike and become mind‑numbingly boring to read. It’s the flipside of that claim.

【Para. 10】Most of us exposed to Psych 101 classes are aware of the magical 7 ±2 number, or Miller's Law. This is supposedly the amount of information we can keep in mind over a short period of time (our short‑term, or working memory capacity). But you should read the introductory paragraph of the original paper by George A. Miller. He describes being paranoid, being constantly “persecuted” by a number, evoking possible conspiracy theories to explain why "seven" seems to be dogging him, peeking out from every corner of data he ever looks at. It’s brilliant: quirky, funny, engaging, and essentially Miller’s voice. (As an aside, I think we have become less tolerant of this kind of tongue‑in‑cheek approach in scientific writing, and we are worse off for it).

【Para. 11】If we let AI and LLMs do our writing for us, there is little chance we will find our own, unique voice. Indeed, recent work suggests that using AI to write leads us to adopt the catchphrases and buzzwords that AI tends to prefer in our own speech. We are not just beginning to sound alike; we are sounding like AI, and this should disturb us.

【Para. 12】So what are we to do? Given the prevalence of LLMs and the ease with which students and teachers alike can lean on them, it would seem ridiculous to suggest we abandon them. The genie is well and truly out of the bottle (and, according to some hyperbole, is a good chance to kill us within the decade).

【Para. 13】Some Better Uses for AI in Writing

My first suggestion would be to stop avoiding the pain. Anything worthwhile takes effort, not just writing. Our relationships, our hobbies, and even such mundanities as our gardens and looking after our pets. All take effort, and all are worthwhile. Expressing your ideas without external help is of enormous value. It forces you to crystallize the ideas in your head, and as the words unfold on the page (sometimes chaotically, sometimes painfully slowly), you are forced to challenge that crystallization. Am I getting that point quite right? Will it be understood the way I want it to be? How will others challenge it?

【Para. 14】Second, by all means use AI to help clean up your writing. My grammar (this post is likely a great example) is not exemplary. I have colleagues for whom English is their second language who have a far better grasp of the split infinitive than I do. Especially for scientists for whom English is their second language, LLMs can be particularly helpful. But I do think the process should begin with our own, human ideas, however effortful they are to develop and express.

【Para. 15】Essentially, writing is about sharing ideas. Once written and shared, your ideas no longer belong to you—they are there for the world to debate, deconstruct, and absorb into their own thinking. One recent suggested use of AI seems particularly useful to me in this context. In a recent Nature paper, Marc Schmitt suggested using LLMs as sparring partners. Present your ideas and your arguments, and ask Claude to prove you wrong. I do still believe that we get a wide range of things from person‑to‑person conversation that Chatbots can never replace. But this strikes me as a sensible use of the tool.

【Para. 16】Beyond avoiding the pain of writing, I would also say to embrace the joy. I find it exciting to engage in the hard work of putting pen to paper. But maybe that’s just me.

【声明】：本文原文摘选自 Psychology Today，原文版权归杂志所有，仅供个人学习交流使用。

如需获取外刊精读资源，加入训练营，每天都提供四份学习文档；

点链接直接查看精读训练营更多详情：

👉👉👉 外刊精读训练营