現在正是討論AI素養的適當時機。放眼所及，AI無所不在且屢上新聞頭條。有些報導強調它驚人的能力，例如產生文字、創作圖片、或撰寫程式碼；也有些關切其負面影響，例如深偽技術、決策偏誤，或對就業市場的衝擊。人們在驚嘆AI所能完成的事之餘，也對其可能對工作、隱私與信任的潛在影響感到不安。
既然我們將與AI共處並共事，「素養」就變得至關重要。這代表著要懂得如何與AI系統互動、如何與它們共同思考，以及如何負責任地使用它們。
但具體來說，這意味著什麼？為什麼現在是討論的時機？
（圖／English Career）
與機器對話的簡史
人類與機器的對話並非從ChatGPT開始。事實上，這段歷史已經持續數十年──只是過去使用的對話語言並非普通人所熟悉的語言。傳統上，要讓電腦做出有用的事，通常需要一名中介者：一個能夠撰寫程式語言的人。像Fortran、C，後來的Python等程式語言，可以協助將人類意圖轉譯為電腦可理解的指令，但大多數使用者仍需仰賴中介者撰寫程式。
這種情況在用戶逐漸自行學習一些機器語言後開始出現變化。也許你曾在Excel中寫過公式、修改過HTML標籤，或理解過「https://」與「http://」的差異。這些都是技術素養的表現——普通人開始以更直接的方式操作機器。
當然，每一波自動化浪潮總會引來同樣的老問題：我們的工作怎麼辦？但歷史顯示，科技雖然會改變職場樣貌，卻鮮少將工作徹底消滅。以試算表為例，當Excel出現時，會計師並未消失；他們反而變得更有效率、更有價值。那些學會新工具的人，不僅倖存下來，還蓬勃發展。AI時代可能也會如此。
AI的網景時刻
今日與過往的最大差異在於，我們不再需要學習機器的語言——AI開始能理解我們的語言。這種由大型語言模型（如ChatGPT背後的技術）驅動的轉變，雖然微妙，卻意義深遠。現在，任何人只要輸入一段自然語言的提示（例如簡單英文或中文），就能得到實用的結果，不論結果是圖片、段落、食譜，甚至程式碼。
我稱此為AI的「網景時刻」（Netscape Moment）。在網頁瀏覽器出現之前，網際網路早已存在，但它難以親近。網景公司的瀏覽器改變了一切。同樣的，AI研究早已默默推進多年，而ChatGPT等工具的問世，則讓AI變得可用、可親、可參與。
這是件大事，而所有重大技術轉折都伴隨著責任。
AI素養＝信任＋責任
我所說的AI素養，不只是技術知識，更重要的是懂得如何深思熟慮地使用這些系統、如何問對問題、如何判斷回應的品質。也包括懂得哪些內容不該輕信，以及何時應該啟動自己的判斷。
在我所寫的《人工智慧思維》（Think Artificial Intelligence）一書中，我提出一個公式：信任（人類＋AI）＝驚豔。
這提醒我們：最好的成果來自人類與AI的協作關係，而這種合作必須建立在信任上。素養，正是通往這種信任的途徑。
以下是幾個AI素養的面向：
提示設計
提示工程（prompt engineering）聽起來很高深，但本質上就是學會如何下達清楚、有條理的指令。愈多練習，AI的回應愈好。例如：「讓我們一步一步思考」或「給我三個建議，列出優缺點」這類語句，會形塑AI的輸出。這就像學會如何有效使用Google搜尋一樣——這是一種技能，也非常重要。
工作與學習
人們對AI取代工作感到焦慮是可以理解的，但同時，AI也開創了新角色與新工作型態。根據世界經濟論壇的數據，AI可能導致8,500萬個工作被重新分配，但也可能創造9,700萬個新職位。關鍵在於持續學習，培養那些能與AI互補、而非與之競爭的技能。
倫理與偏見
AI素養還包括對風險的認識，特別是在公平性與偏見方面。AI系統可能會反映並擴大其訓練資料中的偏誤，因此，像模型卡（model cards）與資料說明文件（data documentation）這類資訊與相關實務就非常重要。它們帶來透明度，幫助我們監督並要求AI負起責任，這是為了確保AI帶來的益處大於其害處。
為什麼我寫《人工智慧思維》
正是這些理念——信任、清晰、倫理——驅使我撰寫《人工智慧思維》一書。我希望能揭開AI的神秘面紗，說明它不是魔法，讓更多人有信心與它合作。畢竟，它只是軟體。雖然是強大的軟體，但仍可由人塑造。
今（2025）年夏天，我將與在台灣任教的老朋友暨同事林俊叡教授合作，在碁峯資訊的支持下推出本書的中文版。我非常期待能向更廣泛的讀者分享這些觀念，並持續推動與AI素養相關的對話。
下一步怎麼走？
AI素養的目的，不是要讓每個人都變成AI工程師，而是要讓大家具備足夠知識，能做出聰明的決策，以及能有效地與AI協作，並懂得在關鍵時刻提出更深入的問題。即使系統越來越聰明，也需要人類參與其中。
就像學會閱讀讓人們能參與社會活動一樣，學會提示設計、結果驗證與質疑 AI，也是一種新的數位素養。這不是科技菁英的專利，而是任何願意學習之人的機會。
人與機器之間的對話才正要開始——讓我們攜手創造一場精彩的對話。
（圖／English Career）
【英文原文】
AI Fluency
It feels like the right moment to talk about AI fluency. Everywhere you look, AI is making headlines. Some stories highlight its impressive abilities, like generating text, creating images, or writing code. Others raise concerns about deepfakes, biased decisions, or the impact on jobs. There’s a growing mix of excitement and concern. People are amazed by what this new generation of AI can accomplish, but also unsure about how it might affect jobs, privacy, or trust.
If we’re going to live and work alongside AI, fluency matters. That means knowing how to communicate with these systems, how to think alongside them, and how to use them responsibly.
But what exactly does that look like? And why now?
A Short History of Talking to Machines
The conversation between humans and machines didn’t start with ChatGPT. In fact, it’s been going on for decades—just not in a language most people speak. Historically, to get a computer to do something useful, you needed an intermediary: someone who could write code. Programming languages like Fortran, C, and later Python helped translate human intention into machine-readable instructions. But most users still depended on someone else to write the script.
That began to change when people started picking up bits of machine language on their own. Maybe you’ve written a formula in Excel, tinkered with an HTML tag, or figured out the difference between https:// and http://. These are all signs of technical fluency—small ways people have learned to operate machines more directly.
Of course, every wave of automation brings up the same old question: what happens to our jobs? But history shows that while technology does change the job landscape, it rarely wipes it out. Take spreadsheets, for example. Accountants didn’t vanish when Excel showed up. They became faster, more capable, and in many ways, more valuable. The ones who learned how to use the tools didn’t just survive; they thrived. The same will likely be true in the age of AI.
The Netscape Moment of AI
What makes today different is that we no longer need to speak the machine’s language—AI is starting to understand ours. This shift, powered by large language models like the one behind ChatGPT, marks something subtle but profound. Now, anyone can type a prompt in natural language (e.g., plain English as they say) and get something useful back, whether it’s an image, a paragraph, a recipe, or even a block of code.
This is what I like to call the “Netscape moment” for AI. The internet existed long before the web browser made it accessible to everyone. Likewise, AI research has been quietly progressing for years. But the release of user-friendly tools like ChatGPT made AI feel like something you could actually use, not just something you read about.
That’s a big deal. And like all big deals, it comes with responsibilities.
Fluency = Trust + Responsibility
When I talk about AI fluency, I don’t just mean technical know-how. I mean understanding how to use these systems thoughtfully, how to ask the right questions, and how to evaluate the answers. I also mean knowing what not to trust and when to bring in your own judgment.
In my book Think Artificial Intelligence, I introduced the formula: Trust（Human + AI）= Amazing.
It’s a reminder that the best outcomes happen when people and AI systems work together, and when that partnership is built on trust. Fluency is what gets us there.
Here are a few dimensions of that fluency:
Prompt Crafting
Prompt engineering sounds fancy, but it’s really just about learning to give clear, thoughtful instructions. The more you practice, the better the AI responds. Whether it’s “Let’s think step by step” or “Give me three ideas with pros and cons,” the language you use shapes the output. It’s not unlike learning how to search well on Google—it’s a skill, and it matters.
Work and Learning
There’s understandable anxiety about AI taking over jobs. But AI also creates space for new roles and new kinds of work. According to the World Economic Forum, AI may displace 85 million jobs by shifting how work is divided, but it could also create 97 million new ones. The key is staying open to learning, and building skills that complement AI rather than compete with it.
Ethics and Bias
Fluency also means understanding the risks, especially when it comes to fairness and bias. AI systems can reflect and amplify the flaws in their training data. That’s why practices like using model cards and data documentation matter. They bring transparency to the process and help us hold these systems accountable. It’s about making sure AI helps more than it harms.
Why I Wrote Think Artificial Intelligence
These ideas—trust, clarity, ethics—are what motivated me to write Think Artificial Intelligence. I wanted to take some of the mystery out of AI, show that it’s not magic, and help people feel more confident in working with it. It’s just software, after all. Powerful software, but still something we can shape.
This summer, I’m teaming up with Raymund Lin, a long-time friend and colleague who is a professor in Taiwan. With the support of GoTop Publishing, we’re releasing a Chinese edition of the book. I couldn’t be more excited about the chance to share these ideas with a broader audience, and to keep the conversation about fluency going.
Where We Go From Here
AI fluency isn’t about turning everyone into an engineer. It’s about knowing enough to make smart choices, to collaborate well, and to spot the moments when you need to ask deeper questions. It’s about staying human in the loop, even as the systems we use get smarter.
Just like learning to read opened up whole new ways for people to participate in society, learning to prompt, to verify, and to question AI can open up a new kind of digital literacy. One that doesn’t just benefit the tech-savvy, but includes anyone willing to learn.
The conversation between humans and machines is just getting started. Let’s make it a good one.
文／Jerry Cuomo（Former IBM Fellow, VP, and CTO）
譯／ChatGPT 編審／林俊叡 Raymund Lin