先建立 AI 素養與判斷力，再學編程語法，但這不是二選一。Build AI literacy and judgement first, and learn programming syntax second. This is not an either/or choice.香港中學階段的合理次序是：先學會用 AI 學習、並核實它交出來的東西，再在任務真正需要時學習程式語言。這個次序有兩個依據：教育局的資訊素養學習框架把「新興及先進資訊科技」列為中學階段必須處理的課題[1]；而 JUPAS 的「其他經驗及成就」（OEA）要求的是可以出示官方證明的成果，不是修讀紀錄[2]。The sensible order in secondary school is to learn to study with AI and to verify what it hands back, then to learn a programming language when a task genuinely needs one. Two things support that order: Hong Kong's information-literacy framework makes emerging and advanced information technologies a required topic at secondary level[1], and JUPAS Other Experiences and Achievements (OEA) asks for achievements backed by official documentary proof, not a record of attendance[2].
兩者各自訓練甚麼What each one actually trains
把「AI 素養」與「編程」放在一起比較時，最容易犯的錯誤是把它們當成同一個範疇的兩個選項。它們訓練的是不同能力，而且會互相補足。The commonest mistake when comparing AI literacy with coding is to treat them as two options inside one category. They train different abilities, and each one strengthens the other.
| AI 素養AI literacy | 編程Programming | |
|---|---|---|
| 主要問題The guiding question | 「這個答案可信嗎？我要怎樣核實？」"Can I trust this answer, and how do I check it?" | 「這個系統怎樣運作？為甚麼會出錯？」"How does this system work, and why did it fail?" |
| 日常動作Daily work | 提問、交代條件、核對輸出、追問修正Ask, set the conditions, check the output, follow up and correct | 拆解問題、設計資料結構、除錯、測試Break down the problem, design data structures, debug, test |
| 出錯時學到甚麼What errors teach | 指令不夠清楚、假設沒有檢查That the instruction was vague, or an assumption went unchecked | 邏輯錯誤、邊界情況、效能限制Logic errors, edge cases, performance limits |
| 可轉移能力Transferable skill | 判斷力、資訊評估、對 AI 限制的認識Judgement, evaluating sources, knowing where AI fails | 抽象思維、演算法思維、系統設計Abstract thinking, algorithmic thinking, system design |
| 第一次見到成果First visible result | 第一堂就可以用在自己的功課上Usable on the student's own homework in lesson one | 通常要數十小時才有第一個可用成品Usually dozens of hours before the first usable build |
| 彼此的關係How they relate | 包含對「AI 技術與應用」的基本理解[4]Includes a basic grasp of AI techniques and applications[4] | 是實現 AI 的其中一種方式，不是唯一方式One way to build AI, not the only way |
| 兩者加起來Together | 兩者會互相加強：寫過程式的人更容易看出 AI 何時出錯；而用 AI 學習的人會更快看懂一段程式碼在做甚麼。Each strengthens the other: someone who has written code spots more easily when AI is wrong, and someone who studies with AI reads a piece of code faster. | |
香港的課程文件怎樣回答這個問題What Hong Kong's own curriculum documents say
香港的官方課程文件沒有把兩者當成二選一。教育局《香港學生資訊素養》學習框架（2024）列出九個素養範疇，第九個是「新興及先進資訊科技」；在第三、四學習階段（中一至中六），要求學生理解並辨識 AI 偏見、深偽（deepfake）、AI 聊天機械人、演算法偏見，以及 AI 機械人散播假新聞等倫理議題[1]。同一份框架的資源清單，還列出教育局的「初中人工智能課程單元」（第一至三冊）[1]。Hong Kong's own curriculum documents do not treat the two as alternatives. The Education Bureau's Information Literacy for Hong Kong Students Learning Framework (2024) sets out nine literacy areas; the ninth is emerging and advanced information technologies. At Key Stages 3 and 4 it asks students to understand and identify the ethical issues these raise, such as AI bias, deepfakes, AI chatbots, algorithmic bias and AI bots spreading fake news[1]. The same framework's resource list includes the Bureau's own junior-secondary AI course modules, Booklets 1 to 3[1].
國際層面指向同一個方向。聯合國教科文組織（UNESCO）2024 年的《AI competency framework for students》把 AI 能力分成四個維度：以人為本的心態、AI 倫理、AI 技術與應用、AI 系統設計；並分三個進階：理解、應用、創造[4]。四個維度之中只有兩個直接關於技術（AI 技術與應用、AI 系統設計），另外兩個是心態與倫理。Internationally the direction is the same. UNESCO's 2024 AI competency framework for students organises AI competence into four dimensions (a human-centred mindset, ethics of AI, AI techniques and applications, and AI system design) across three progression levels: understand, apply, create[4]. Only two of those four dimensions are directly technical (AI techniques and applications, and AI system design); the other two are mindset and ethics.
換句話說，素養是所有人都要有的部分，技術是其中一部分。這不是說編程不重要，而是它在中學階段的合理位置：在素養之後，因需要而學。In other words, literacy is the part everyone needs, and technical skill is one part of it. That is not an argument that programming does not matter. It is an argument about where it sits in secondary school: after literacy, and learned when there is a reason to learn it.
升學文件看重的是甚麼What the admissions paperwork actually asks for
JUPAS 的「其他經驗及成就」（OEA）接受的是比賽與活動的獎項或證書、由參與院校舉辦的活動、學生大使、運動員、主席／評判／領袖／裁判／導師、籌委、講者等類別；支援文件必須是官方證明，文件上的資料要與填報的內容相符，院校亦可要求申請人出示正本[2]。JUPAS Other Experiences and Achievements (OEA) accepts categories such as awards and certificates from competitions and activities, programmes run by participating institutions, student ambassador and athlete roles, chairperson, judge, leader, referee or trainer roles, committee or organising work, and presenting or speaking[2]. Supporting documents must be official proof of participation or achievement, the details on them must match what the applicant entered, and institutions may require the originals[2].
「學生學習概覽」（SLP）則由學生自行上載，用來說明全人發展、個人素質與能力；JUPAS 寫得很清楚：SLP 只作參考，院校自行決定是否及如何採用[3]。The Student Learning Profile (SLP) is uploaded by the student to show whole-person development, personal qualities and competencies. JUPAS is explicit that the SLP "is provided for reference only" and that institutions decide whether and how to use it[3].
兩份文件都不會因為你「上過 AI 課程」而加分。OEA 要的是可核實的參與或成就證明，SLP 是參考資料。所以選擇課程時，值得問的問題是：「課程結束時，學生有甚麼可以拿出來給人看？」而不是「課程名字裡有沒有 AI」。Neither document gives credit for having attended an AI course. OEA wants verifiable proof of participation or achievement; the SLP is a reference. So the question worth asking about any course is not "does the name contain AI?" but "what will the student be able to put in front of someone at the end?"
一個實際可行的兩年時間表A two-year order that is actually workable
以下是一個以兩年為單位的次序建議，起點是中一或中二、之前沒有編程經驗。如果學生起步較遲，可以直接由第二步開始，不必補回第一步的課。What follows is a suggested order over two years, starting in S1 or S2 with no prior programming experience. A student who starts later can begin at step two; there is no need to go back and make up step one.
| 階段Stage | 重點Focus | 具體做得到的事What it looks like in practice | 校內外課程（示例）Courses that fit (examples) |
|---|---|---|---|
| 第 1–2 學期Terms 1–2 | 建立判斷與核實的習慣Build the habit of checking | 為各科設定 AI 教練；每次核對 AI 的答案與課本有沒有出入；把 AI 出錯的例子記下來Set up an AI coach per subject; check the AI's answer against the textbook each time; keep a note of the cases where it was wrong | 初中 Gemini AI 工作坊（2 堂／6 小時，F.1–F.3）· 中英數·新學期一次掌握（6 堂）Gemini AI Workshop (2 lessons / 6 hours, S1–S3) · AI Subject Tutoring (6 lessons) |
| 第 3–4 學期Terms 3–4 | 第一次做出可上線的成品Ship something once | 由一個真實問題出發：定義功能、用自然語言下指令、測試與修正，最後部署並取得可分享連結Start from a real problem: define the features, instruct in natural language, test and correct, then deploy and get a shareable link | 自然語言編程與人工智能開發（6 小時）Natural-Language Programming & AI Development (6 hours) |
| 第 5–6 學期Terms 5–6 | 把成品變成可以出示的成果Turn the work into something showable | 做一個要向人解釋、要示範的作品，並整理成作品集與可核實的參與證明Build a project the student can explain and demonstrate, then organise it into a portfolio with verifiable proof of participation | 互動式數字人校園實戰課程（8 堂 × 1.5 小時）· 專屬 AI 智能護理 App 研發實戰營（30 堂）· 一人公司 AI 青年總裁培育計劃（5 模組／30 堂）Interactive AI Digital Human (8 × 1.5 hours) · AI Health & Care App (30 lessons) · One-Person Company AI (5 modules / 30 lessons) |
| 之後，視需要Later, only if needed | 真正學一門程式語言Learn a programming language properly | 當任務涉及效能、可靠性或安全判斷，才由語法、資料結構與除錯開始Once a task involves performance, reliability or a security decision, start on syntax, data structures and debugging | 學校的電腦課程或公開課程The school's own computer course, or a public one |
兩年只是長度單位，不是期限。如果一年只做到第一步，那仍然是有價值的一年；反而在第一步未做完就跳到第三步，作品通常經不起一兩條追問。Two years is a unit of length, not a deadline. A year that only completes step one is still a year well spent; skipping to step three before step one is finished usually produces a project that cannot survive a question or two.
常見反駁：不先學語法，會變成只會按按鈕？The obvious objection: without syntax first, do students just learn to press buttons?
這個擔心有實際根據。2023 年發表於 ACM CCS 的一項使用者研究發現，可以使用 AI 助手的參加者寫出的程式碼明顯較不安全，而且更傾向相信自己寫得安全[5]。AI 令寫程式變快，但核實不會自動出現。The concern is well founded. A 2023 user study published at ACM CCS found that participants with access to an AI assistant wrote significantly less secure code, and were more likely to believe their code was secure[5]. AI makes coding faster; verification does not appear on its own.
不過結論不是「先背語法」，而是「把核實變成課程的一部分」。如果課堂只教學生叫 AI 做，學生當然只學會叫。所以每一次交付，都要學生指出 AI 哪一句做錯、為甚麼錯、怎樣改，而不是一律按下「接受」。The conclusion, though, is not "memorise syntax first". It is that verification has to be part of the teaching. If a class only teaches students to tell the AI what to do, that is the only skill they leave with. So every hand-in here asks the student to point out what the AI got wrong, why, and how they fixed it, rather than accepting everything.
編程本身仍然值得學，而且有三個很具體的理由：Programming is still worth learning, for three concrete reasons:
- 要判斷 AI 的解釋是否合理，語法知識是最短的捷徑。Knowing the syntax is the shortest route to judging whether the AI's explanation is plausible.
- 當系統要顧及效能、可靠性與安全（金錢交易、醫療、個人資料），仍然需要受過訓練的工程師。When a system has to be fast, reliable and safe with money, health data or personal information, trained engineers are still required.
- 要解決 AI 沒有見過的問題，抽象與演算法思維仍然來自程式訓練。For problems the AI has never seen, abstract and algorithmic thinking still come from having written code.
所以兩者的關係不是替代，是次序。So the relationship between them is not substitution. It is sequence.
這個次序甚麼時候不適用When this order is the wrong one
一份只說「兩樣都學」的建議，對家長的實際決定沒有幫助，所以下面幾種情況應該說清楚：Advice that only says "learn both" does not help anyone make a decision, so the exceptions need stating:
- 學生已確定要在大學讀電腦科學或工程，而且喜歡由底層學起：把語法放在前面是合理的選擇。The student already intends to read computer science or engineering, and likes starting from the bottom: putting syntax first is a reasonable choice.
- 學校本身已提供正式的電腦課程：先跟學校的進度，不必另加一套。The school already teaches a formal computer course: follow the school's sequence rather than adding a second one.
- 學生未滿 13 歲：本文與我們的中學生課程都以 F.1 為起點。The student is younger than 13: this article, and our secondary courses, both start at S1.
- 學生連「把事情講清楚」都感到困難：先處理表達與閱讀。AI 不會把模糊的問題變清楚，只會把模糊的答案變快。The student struggles to describe a task clearly: work on expression and reading first. AI does not make a vague problem clear; it makes a vague answer fast.
我們怎樣教這一套How we teach it
Edcosys 於 2024 年由香港科學園的 AI 團隊創立，現時為 F.1–F.6 學生提供 7 門課程，並以到校形式與香港中、小學合作；自成立以來，我們已與 50+ 間香港中、小學合作，累計交付 1,000+ 課節。Edcosys was founded in 2024 by an AI team from Hong Kong Science Park. We run 7 courses for F.1–F.6 students and deliver them on campus with primary and secondary schools; since founding we have worked with 50+ primary and secondary schools in Hong Kong and delivered 1,000+ lessons.
對應本文的次序，最直接的兩門是：《中英數·新學期一次掌握》（6 堂）訓練第一步，即為中、英、數各設定專屬 AI 教練，用 AI 整理重點、出題與分析錯題，重點放在核對與追問；《自然語言編程與人工智能開發》（6 小時）訓練第二步，由構思、下指令、測試與修正，到部署上線，四個階段學生都要自己走一次。學校如要把同一批課程安排在到校進行，可參考「智啟學教」校本方案與課程建議書。Two of our courses map directly onto that order. AI Subject Tutoring (6 lessons) covers step one: an AI coach set up for Chinese, English and Maths, used to summarise, set practice questions and analyse mistakes, with checking and follow-up as the point of the exercise. Natural-Language Programming & AI Development (6 hours) covers step two: ideation, instructing, testing and refining, then deploying. Students walk all four stages themselves. Schools that want the same courses delivered on campus can read the school programme and proposal for EDB's 智啟學教 framework.
