我們?cè)儐?wèn)了專(zhuān)家們關(guān)于如何以最負(fù)責(zé)任的方式使用人工智能工具——以下是他們的觀點(diǎn)。

ChatGPT 發(fā)布三年后，已經(jīng)形成了兩大陣營(yíng)：拒絕使用它的人和每天都使用它的人。

將人工智能用作頭腦風(fēng)暴伙伴和組織者，但不要將你的判斷力外包出去。

艾麗·亨特
2026年3月18日

ChatGPT發(fā)布三年后，已經(jīng)形成了兩大陣營(yíng)：拒絕使用它的人和每天都使用它的人。

皮尤研究中心2025年的一項(xiàng)調(diào)查發(fā)現(xiàn)，三分之一的美國(guó)成年人表示他們一直在使用ChatGPT。這其中包括58%的30歲以下美國(guó)成年人——大約是兩年前的兩倍。

專(zhuān)家表示，人工智能領(lǐng)域正在出現(xiàn)分歧——一方可能已經(jīng)對(duì)人工智能投入了大量資源，而另一方則拒絕接受人工智能——這使得就如何最好地利用人工智能展開(kāi)公開(kāi)對(duì)話(huà)變得越來(lái)越重要。

那么我們應(yīng)該如何使用人工智能工具呢？以下是由專(zhuān)家提供的循序漸進(jìn)指南。

集思廣益

如果你還不熟悉人工智能的功能，那么何時(shí)使用它可能會(huì)讓你感到不知所措。你可以先從詢(xún)問(wèn)它一些你已經(jīng)想到的事情，或者待辦事項(xiàng)清單上尚未完成的任務(wù)開(kāi)始。

兩個(gè)人手牽著手漂浮在空中，周?chē)h(huán)繞著心形、玫瑰、機(jī)器人和電子元件。

他們?cè)驹诤腿斯ぶ悄馨閭H約會(huì)，卻在那時(shí)找到了真愛(ài)——彼此。

《理解人工智能》簡(jiǎn)報(bào)的作者蒂莫西·B·李 建議使用人工智能工具進(jìn)行頭腦風(fēng)暴，并將任務(wù)或項(xiàng)目分解成可實(shí)現(xiàn)的步驟。他說(shuō)：“任何時(shí)候，當(dāng)你試圖提出想法時(shí)，這都是一個(gè)很好的起點(diǎn)?！?/p>

內(nèi)容創(chuàng)作者兼人工智能教育家凱瑟琳·戈策（Catherine Goetze，TikTok賬號(hào)@askcatgpt）建議將人工智能視為“思想伙伴”，它可以幫助你碰撞出新的想法，突破創(chuàng)意瓶頸，或完善你的思路。但重要的是，在評(píng)估結(jié)果時(shí)，你要繼續(xù)運(yùn)用自己的判斷力、專(zhuān)業(yè)知識(shí)和品味，而不是讓人工智能做出最終決定。她表示，最適合人工智能的任務(wù)是那些“你知道正確答案是什么”的任務(wù)。

研究項(xiàng)目

對(duì)于更復(fù)雜或更深入的研究，人工智能工具可以提供已發(fā)表文獻(xiàn)的概覽?！澳憧梢园阉胂蟪深?lèi)似維基百科的東西，”李說(shuō)?！拔覀冎浪⒎峭昝罒o(wú)缺，我們也知道如何核查引用。”

Claude、ChatGPT 和 Perplexity 等工具都提供某種形式的“深度研究”功能。這些功能會(huì)搜索與您主題相關(guān)的文檔，并將其總結(jié)成一份幾千字的報(bào)告。機(jī)器人甚至?xí)蚰岢鲆恍┏吻鍐?wèn)題，以便收集最相關(guān)的結(jié)果。

“它真的非常棒，”李說(shuō)。它的回復(fù)會(huì)提供原始資料和鏈接，方便用戶(hù)自行查閱。李表示，深度研究功能有助于“了解概況”——確定需要閱讀的關(guān)鍵論文、需要解答的問(wèn)題以及項(xiàng)目的后續(xù)步驟。

專(zhuān)家們一致認(rèn)為，關(guān)鍵在于：在使用人工智能工具進(jìn)行初步定位之后，你仍然需要自己完成實(shí)際工作。

學(xué)習(xí)新技能或愛(ài)好

阿姆斯特丹自由大學(xué)副教授艾拉·哈弗馬爾茨（Ella Hafermalz）研究生成式人工智能對(duì)工作的影響，她說(shuō)，你也可以在人工智能“能夠拓展你的世界”時(shí)使用它。

哈弗馬爾茨一直用它來(lái)幫助種植迷你蓮花，學(xué)習(xí)股票投資的基礎(chǔ)知識(shí)，甚至還從中獲得晚餐的靈感。

她說(shuō)，它的好處在于能夠幫助你“入門(mén)”，培養(yǎng)新的興趣、活動(dòng)、愛(ài)好或技能，尤其是那些入門(mén)門(mén)檻很高的領(lǐng)域。

“對(duì)很多人來(lái)說(shuō)，嘗試新事物的障礙是尷尬、恐懼、時(shí)間、不適，甚至不知道自己需要知道什么，”她說(shuō)。

哈弗馬爾茨表示，與人工智能的互動(dòng)可以打破這些障礙，“助你邁向下一步”。但是，再次強(qiáng)調(diào)，它最好被視為一個(gè)起點(diǎn)，適用于風(fēng)險(xiǎn)較低、且你仍擁有最終決定權(quán)的任務(wù)。

組織信息

一旦你開(kāi)始進(jìn)行研究項(xiàng)目，人工智能工具可以幫助你整理研究結(jié)果，例如識(shí)別主題、回答問(wèn)題或生成時(shí)間線(xiàn)或摘要。

一幅像素風(fēng)格的插圖，描繪了一個(gè)人和一臺(tái)電腦正在制作一份文件。

我教過(guò)成千上萬(wàn)的人如何使用人工智能——以下是我總結(jié)的經(jīng)驗(yàn)。

如果你的首要任務(wù)是研究，Hafermalz 建議使用 Google 的免費(fèi)應(yīng)用程序 NotebookLM，它只使用你上傳的文檔、筆記和資料，而不是整個(gè)網(wǎng)絡(luò)。

Hafermalz表示，歷史學(xué)家現(xiàn)在正將NotebookLM用作研究助手。她繼續(xù)說(shuō)道：“它能起到組織信息的作用……如果你想整理信息、進(jìn)行綜合分析，又不想讓它偏離主題，從Reddit等網(wǎng)站上抓取內(nèi)容，NotebookLM就是一個(gè)更集中的空間?！?/p>

對(duì)于個(gè)人用戶(hù)而言，人工智能可以像家政助手或行政助理一樣工作，幫助你計(jì)劃膳食或鍛煉，制定預(yù)算，或者將下午的時(shí)間從高到低安排好。

如何獲得更好的結(jié)果

李說(shuō)，大約一年前，精心設(shè)計(jì)提示語(yǔ)對(duì)于充分發(fā)揮人工智能工具的作用非常重要，但“隨著時(shí)間的推移，它的重要性正在逐漸降低”。

但是，你仍然可以做一些事情來(lái)最大限度地利用它們。

現(xiàn)在，您可以信賴(lài)領(lǐng)先的人工智能工具，它們能夠更直觀地響應(yīng)，并能理解更隨意的語(yǔ)言和措辭——盡管上下文會(huì)有所幫助，Lee補(bǔ)充道：“您提供的信息越多，就越有可能獲得好的結(jié)果?！?/p>

戈策鼓勵(lì)大家完全放棄提示的想法。“你應(yīng)該把它當(dāng)成聊天，”戈策說(shuō)，“真正的魔力來(lái)自于一來(lái)一往的互動(dòng)?！彼a(bǔ)充道，這正是 ChatGPT 相對(duì)于谷歌搜索的優(yōu)勢(shì)所在：“我的建議是，進(jìn)去之后就隨意聊聊?！?/p>

只要你注意避免分享敏感信息，你還可以提供網(wǎng)站鏈接、上傳PDF文件和其他資料供其參考。例如，你可以分享你的新手機(jī)合同，并讓AI標(biāo)記出你應(yīng)該注意的條款或潛在的省錢(qián)機(jī)會(huì)。

如果你遇到難題，還可以嘗試一種叫做“反向提示”的方法。戈策最近在撰寫(xiě)一份文件時(shí)遇到了瓶頸，于是她讓 ChatGPT 生成了五個(gè)問(wèn)題，幫助她突破困境?！斑@些問(wèn)題引導(dǎo)我從不同的角度思考，”她說(shuō)。

永遠(yuǎn)不要盲目信任人工智能工具

檢查所有人工智能的回復(fù)至關(guān)重要。專(zhuān)家們一致認(rèn)為，雖然人工智能每天都在進(jìn)步，但仍需對(duì)照權(quán)威可靠來(lái)源進(jìn)行核查。人工智能不僅會(huì)重復(fù)虛假信息和錯(cuò)誤信息，甚至可能憑空捏造——這被稱(chēng)為“幻覺(jué)”。

戈策說(shuō)：“檢查你的消息來(lái)源，檢查那些鏈接，檢查消息來(lái)源的日期。”

如何避免濫用人工智能工具

人工智能也可能被用于縮小我們的世界、削弱我們的能力，甚至損害我們的人性。這種情況通常源于人們將其視為捷徑或流程的最后階段，而非第一步。

Goetze 使用 ChatGPT 來(lái)幫助她想出內(nèi)容創(chuàng)意，但如果她讓它編寫(xiě)腳本然后讓她逐字逐句地朗讀，她說(shuō)：“那我就完全把它當(dāng)成拐杖，這會(huì)阻礙我的創(chuàng)造力”。

三位專(zhuān)家一致認(rèn)為，未來(lái)將越來(lái)越難以分辨文本中哪些部分（或比例）是作者撰寫(xiě)的，哪些部分是人工智能生成的。戈策表示，目前“你應(yīng)該力求透明”，并注意避免抄襲和侵犯版權(quán)。

另一個(gè)潛在的危險(xiǎn)是過(guò)度依賴(lài)人工智能工具的響應(yīng)，甚至對(duì)其產(chǎn)生依賴(lài)?！澳悴幌牒腿斯ぶ悄芟萑敕答佈h(huán)——最終你會(huì)陷入困境，”哈弗馬爾茨說(shuō)道。

她建議每次使用 ChatGPT 時(shí)都設(shè)定一個(gè)明確的目標(biāo)或意圖，并逐步提高參與的程度，以便始終保持掌控。

“它不應(yīng)該成為囚禁你的牢籠，而應(yīng)該成為一塊墊腳石——讓你走出困境，去做其他事情，”哈弗馬爾茨說(shuō)。“在現(xiàn)實(shí)世界中，在你能夠親自驗(yàn)證的地方使用它?！?/p>

We asked experts about the most responsible ways to use AI tools – here’s what they said

Three years on from the release of ChatGPT, two broad camps have formed: those people who refuse to use it, and those who use it every day. Illustration: Brendan Conroy/The Guardian

Use AI as a brainstorming partner and organizer, but don’t outsource your judgment

• Sign up for AI for the people, a six-week newsletter course, here

Elle Hunt

Wed 18 Mar 2026 11.00 GMT

Share

Three years on from the release of ChatGPT, two broad camps have formed: those people who refuse to use it, and those who use it every day.

A 2025 survey by the Pew Research Center found that one-third of US adults say they have been using ChatGPT. This includes 58% of US adults under 30 – roughly double the share two years ago.

Experts say the emerging divide – between those who may already be deeply invested in AI, and those who are refusing to engage with it – makes it increasingly important to have an open conversation about how best to use it.

So how should we approach AI tools? Here is a step-by-step guide informed by experts.

Brainstorming ideas

Knowing when to use AI can be daunting if you’re not already familiar with what it’s able to do. You can start by simply asking it for help with what is already on your mind, or outstanding on your to-do list.

They were dating AI partners when they found real love – with each other

Read more

Timothy B Lee, author of the Understanding AI newsletter,recommends using AI tools to brainstorm and break down tasks or projects into achievable steps. “Any time you’re trying to come up with ideas, it’s a good starting place,” he says.

Catherine Goetze, a content creator and AI educator who goes by @askcatgpt on TikTok, suggests thinking of it as a “thought partner”, helping to bounce around ideas, break through creative blocks or refine your thinking. But it’s important that, in reviewing the results, you continue to draw on your own judgment, expertise and taste, rather than letting the AI have the final say. The best tasks for AI are those “where you know what the right answer looks like”, she says.

Researching projects

For more complex or intensive research, an AI tool can give you a rundown of what has been published. “Think about it as similar to Wikipedia,” Lee says. “We know it’s fallible, and we know how to check citations.”

Tools such as Claude, ChatGPT and Perplexity all offer some variation of a “deep research” feature. This will scour documents on your topic and summarize them in a report of a few thousand words. The bot will even ask you clarifying questions so as to assemble the most relevant results.

“It’s really astonishingly good,” Lee says. Its responses will provide primary sources and links, enabling you to refer to them yourself. Lee says the deep research function is helpful for “getting a lay of the land” – identifying key papers to read, questions to answer and next steps for your project.

That’s the important bit, experts agree: after using an AI tool for preliminary orientation, you still need to do the actual work yourself.

Learning new skills or hobbies

You can also use AI “when it allows you to expand your world”, says Ella Hafermalz, an associate professor at Vrije Universiteit Amsterdam, who studies generative AI’s impact on work.

Hafermalz has been using it for help with growing a mini lotus flower, as well as learning the basics of investing in stocks and even sparking ideas for what to make for dinner.

She says its benefit is “getting you off the ground” with a new interest, activity, hobby or skill, particularly those that have a high barrier to entry.

“For many people, the hurdle in trying a new thing is embarrassment, fear, time, discomfort, not even knowing what you need to know,” she says.

Back-and-forthing with the AI can break down those barriers, and “get you to that next step”, Hafermalz says. But, again, it’s best treated as a starting point for tasks that are perhaps lower stakes and where you remain the ultimate authority.

Organizing information

Once you’ve got started with your research project, AI tools can help you structure your findings, such as by identifying themes, answering questions, or generating timelines or summaries.

I’ve taught thousands of people how to use AI – here’s what I’ve learned

Read more

If research is your priority, Hafermalz recommends using Google’s free-to-use application NotebookLM, which draws only from documents, notes and materials you’ve uploaded, rather than the entirety of the web.

Historians are now using NotebookLM as a research assistant, Hafermalz says. She continued: “It acts as an organizer … If you’re trying to organize information and synthesize things, and you don’t want it to go off-piste and pull things from Reddit, NotebookLM is a more contained space.”

For personal use, AI can function like a domestic helper or an executive assistant, helping you to plan meals or workouts, create a budget or structure your afternoon from high- to low-priority tasks.

How to get better results

A year or so ago, crafting the prompt was important to getting the best out of an AI tool, but “it’s getting less and less important over time”, Lee says.

However, there are still certain things you can do to get the most out of them.

You can now trust the leading AI tools to respond more intuitively, and to more casual language and phrasing – though context helps, Lee adds: “The more information you give it, the more likely you are to get a good result.”

Goetze encourages people to throw out the idea of prompting altogether. “You really want to think about it as chatting,” Goetze says. “The magic actually comes from the back-and-forth.” That is ChatGPT’s advantage over Google search, after all, she adds: “What I would say is go into it and just ramble.”

So long as you’re careful to avoid sharing sensitive information, you can also link to websites and upload PDFs and other material for it to refer to in its answer. For example, you could share your new phone contract and ask the AI to flag terms you should be aware of or potential opportunities to make savings.

If you’re ever stuck on something you can also do something called a “reverse-prompt”. When Goetze recently hit a mental block with a document she was working on, she asked ChatGPT to come up with five questions that would help her push past it. “They prompted me to reflect in slightly different ways,” she says.

Never blindly trust an AI tool

It’s essential to check all of your AI responses. While they are improving every day, they still need to be checked against primary and reputable sources, the experts agree. Not only can it regurgitate falsehoods and misinformation, the model can actively make things up – what’s known as a “hallucination”.

Goetze says: “Check your sources, check those links, check the dates of the sources.”

How not to use AI tools

It is also possible to use AI in ways that shrink our world, undermine our abilities and even our humanity. Often, they stem from treating it as a shortcut, or the final stage in a process, rather than a first step.

Goetze uses ChatGPT to help her come up with ideas for content, but if she were to prompt it to write scripts that she then read verbatim, she says, “that would be me totally using it as a crutch, and it hindering my creative ability”.

All three experts agree that it will become harder to parse what parts (or proportion) of a text were written, and which were AI-generated. For now, Goetze says, “you should aim to be transparent”, and take care to avoid plagiarism and breaches of copyright.

Another potential danger is overinvesting in an AI tool’s responses and even becoming reliant. “You don’t want to stay in a feedback loop with AI – you will end up in dark places,” Hafermalz says.

She recommends setting a clear goal or intention every time you use ChatGPT, and slowly upping the stakes of your engagement so that you always remain in charge.

“It shouldn’t be a prison that holds you in, it should be a stepping stone – a way to get out, and do other things,” Hafermalz says. “Use it where you can verify it yourself, in the real world.”

The Feast app: your most useful kitchen utensil

With over 7,000 recipes to choose from, the Feast app puts world-class chefs right in your pocket. Plus smart features that make everyday cooking easier: build personalised recipe collections, create a shopping list, search smarter and cook hands-free with cook mode.

Whether you’re looking for:

?Meera Sodha’s vegan collectionfor vibrant, veggie forward meals

?Nigel Slater’s simple meals for two, perfect for quiet nights in

?Winter cooking ideas for a crowdfrom Ravneet Gill and Mattie Taiano, made for sharing and second servings

?Felicity Cloake’s perfect versions of the classicsyou'll always come back to

… you’ll find all these and more in the Feast app.Download from the App Store or Google Play to start your free trial.

Try the Feast app