Published in the News Lens, Feb 22, 2023 (English: https://international.thenewslens.com/article/181539) and Feb 28, 2023 (Chinese: https://www.thenewslens.com/article/181670)
Taiwan will soon need to brace itself for the AI tsunami that will soon flood its schools–or at least the homes of its students. Various school boards in the US have already banned Open AI’s ChatGPT since its release in November 2022. Taiwan’s school boards have been quiet till now, but similar hysteria will likely hit Taiwan when Baidu releases the Mandarin AI chatbot Ernie in March.
Should we really worry about AI chatbots in Taiwan’s education system? I had a chat with ChatGPT to find out.
What is ChatGPT?
● ChatGPT: ChatGPT is a large language model developed by OpenAI that can understand and generate human-like responses to text-based conversational inputs.
Is ChatGPT just a more user-friendly Google search?
● ChatGTP: “While ChatGPT and Google search both involve language processing, ChatGPT is designed to understand and respond to conversational inputs in a more human-like manner, whereas Google search provides a list of relevant web pages based on the keywords entered.”
This still doesn’t not suggest ChatGPT’s impressive capabilities. If you have not yet played with OpenAI’s ChatGPT, you have no idea how much of an advance it is in AI technology.
How is ChatGPT different from other AI chatbots?
● ChatGPT: “ChatGPT is different from other AI chatbots because of its advanced natural language processing capabilities and large knowledge base, allowing it to understand and respond to a wide range of topics and nuances in human language. ChatGPT is also capable of generating more human-like and engaging responses, making it feel more like a conversation with a real person than a machine.”
As ChatGPT suggests, its power lies in its ability to generate responses. That is why it is called “generative AI”. Let me explain.
Technically, ChatGPT is the biggest “large language model” to date and uses transformer architecture to produce high-quality, human-like text responses. Apart from the huge training set, it was also trained with human experts and continues learning to enhance the model’s ability to fine-tune responses.
In essence, ChatGPT is a natural language generator and transformer that uses billions of parameters to predict the next word in a sentence based on the preceding words. As a generative AI tool, it can generate original content by learning from vast amounts of existing text data, and by producing language that is virtually flawless, logically coherent, and natural.
OpenAI’s engineers have created something truly groundbreaking. And not even they fully understand how ChatGPT is able to work as well as it does. ChatGPT also knows this.
Do ChatGPT engineers really understand how ChatGPT works?
● ChatGPT: “ChatGPT engineers have a general understanding of how the model works and the underlying principles of machine learning, but due to the complexity and scale of the model, it is difficult to fully comprehend all the intricacies of its behavior and decision-making processes. The model’s inner workings are influenced by a vast amount of data and mathematical computations, making it challenging to explain every detail of how ChatGPT generates its responses.”
ChatGPT marks a quantum leap forward in AI technology because of the billions of parameters of data and connections coupled with human supervised and reinforced learning. This makes it a little similar to a brain with trillions of synaptic connections that formed with human experiences.
This is not the emergence of AI consciousness that we see in science fiction movies like Terminator and Matrix, but it seems like a giant leap in that direction.
ChatGPT has access to hundreds of gigabytes of internet texts and can answer questions about language, history, math, and programming. It can make a great teaching assistant.
Should ChatGPT be used in education in Taiwan?
● Chatgpt: “As an AI language model, ChatGPT can be a valuable tool in certain educational contexts in Taiwan, but its use should be carefully evaluated and considered alongside the benefits of human interaction and personalized instruction.”
AI chatbots like ChatGPT can be a valuable took in education. They know much more than teachers. They are much faster than teachers. And they certainly explain things more clearly than most teachers.
I am not sure if teachers need to be worried about their jobs, but many should worry about their self-esteem. At least the less talented teachers who are afraid of technology.
Will generative AI tools like ChatGPT replace teachers?
● ChatGPT: “It is unlikely that teachers will be completely replaced by generative AI tools like ChatGPT. While these tools can be useful for certain tasks such as providing personalized feedback or generating educational content, they lack the empathy, creativity, and interpersonal skills that are essential for effective teaching.”
This is a reassuring answer, and one that I completely agree with.
The potential uses of ChatGPT to generate information and text are unlimited and constrained only by one’s imagination. So, I pushed ChaGPT a little more to give me formal, logical arguments for and against its use in education.
What is an argument against the use of ChatGPT in education?
● ChatGPT:
Premise 1: Education is a complex and multi-faceted process that involves more than just the transfer of information.
Premise 2: AI chatbots lack the emotional intelligence, creativity, and critical thinking skills that are necessary for effective teaching and learning.
Conclusion: Therefore, relying on AI chatbots as a primary source of education risks oversimplifying and devaluing the learning process, ultimately leading to suboptimal learning outcomes.
This a valid logical deductive argument against the idea relying on ChatGPT as “a primary source” of education. And I think we would all agree.
However, I think it is interesting that ChatGPT did not mention the main worry that human educators and critics have: cheating students who ask AI chatbots to do their homework and assignments.
Why didn’t you mention students using ChatGPT to cheat?
● ChatGPT: “I apologize if my previous response did not address the concern about cheating. One of the arguments against using ChatGPT in education is indeed that students could potentially use it to cheat on assignments or exams by copying or paraphrasing text generated by the model. It is important for educators to consider the potential for misuse and take steps to mitigate it, such as designing assessments that require critical thinking or using plagiarism detection software.”
These are important points. ChatGPT’s oversight about cheating was probably intentional due to its bias to not encourage unethical or illegal behavior. ChatGPT makes clear in its user interface that it is “trained to decline inappropriate requests”, which is perhaps why it would not overtly suggest that it could be used for “cheating.”
This should remind us that ChatGPT knowledge is constrained by at least three factors:
1. It is based on the contents of internet texts it has access to as well as its training dataset,
2. There are biases in programming and human training, and
3. As a language model, fluency and coherent meaning simply trumps facticity.
Can we always trust the facticity of ChatGPT responses?
● ChatGPT: “While ChatGPT strives to provide accurate and informative responses, its replies should always be verified and used in conjunction with other sources.”
Tech website CNET found itself in hot water about incorrect information in recent articles that it admitted were written by AI writing tools like ChatGPT. CNET did not verify the AI information.
I also asked for an argument in favor of its use in education.
What is an argument for using ChatGPT in education?
● ChatGPT:
Premise 1: AI chatbots have the ability to provide scalable and accessible support to students.
Premise 2: The use of AI chatbots can free up time for human teachers to focus on more complex and personalized instruction.
Conclusion: Therefore, incorporating AI chatbots into the educational process can enhance the overall learning experience and improve learning outcomes.
I like this argument because it gets to the primary purpose of good education technology: it can scale up and do more than any one teacher can. It can also relieve teachers from mundane aspects of information transmission and free up more of the teacher’s time to focus on “more complex and personalized teaching”.
In other words, the primary purpose of good education technology is to enable teachers to focus on their primary purpose: student learning.
Flipped classrooms are a good example of good education technology. In this model, good quality instructional videos and online practice, such as from platforms like the Khan Academy (or Junyi Academy in Taiwan), can be used to empower students to learn at home and at their own pace.
Here, the normal education model is “flipped” because teachers outsource teaching to a good online learning platform and then use class time to monitor and help students with problems, or focus on higher level projects and activities that incorporate skills in more meaningful ways.
Unfortunately, even though flipped classrooms have been promoted for more than 10 years, most teachers have not responded to this more effective educational model.
Why do so few teachers use the flipped classroom model?
● ChatGPT: “There are several reasons why this might be the case: 1. Lack of knowledge and training … , 2. Time constraints … , 3. Technology barriers … , 4. Resistance to change … , 5. Student readiness …”
Before asking ChatGPT, I came up with most of the same answers. The main reasons are that it takes a lot of time and know-how to create your own videos, teachers’ identity and sense of value are threatened if the video teacher is somebody else, and it is also human nature to resist change, so teachers often get stuck in familiar teaching practices of simply “covering” material.
I did not come up with “student readiness”, though this is also an aspect of human nature that resists change.
My hope is that more Taiwanese teachers and students will embrace AI chatbots as homework assistants and try to establish something like an AI-flipped classroom. This would be certainly easier and less threatening to do than a video-flipped classroom.
So, can we trust AI chatbots in Taiwan’s education?
The issue of trust is mostly two-sided: student use of the technology and the quality and facticity of the information provided by the AI chatbot.
However, these two issues can be resolved with proper training. Both teachers and students need to be trained how to effectively, correctly, and safely use the AI large language models.
It is clear to me that AI tools will not replace human teachers. But I am convinced that human teachers with AI tools will replace human teachers without AI tools. Ultimately, human teachers and AI assistant will complement each other.
The sad truth is that many–perhaps most–students in public education in Taiwan cannot keep up with the fast pace of teaching and testing.
Imagine how amazing it would be for a struggling student to be able to ask ChatGPT not only to solve math problems again and again, but also to show him step by step how these problems are solved, and even to point out where mistakes were made?
No human teacher can spend this much time with one student.
The real question is not if AI should be used in Taiwan’s education. It is how it should be best used. And here our first concern should be with what is best for the learner–not the teacher, not the school administrators, not the cram schools, and not the legislators worried about re-election.
The potential power of AI chatbots for the learner are so huge and so revolutionary, I thought it fitting to give ChatGPT the last word to celebrate this power.
Write a brief poem about how ChatGPT can revolutionize education in Taiwan
● ChatGPT:
In Taiwan’s schools, a new tool’s arrived,
ChatGPT’s intelligence, it can’t be denied.
With knowledge vast and insights sharp,
It can aid in learning, a most valuable part.
Personalized feedback, it can provide,
For each student’s needs, it can help guide.
Its wealth of content, a vast sea,
Helps teachers enrich their pedagogy.
Revolutionizing education, it may well do,
Opening up new avenues, and paths anew.
In Taiwan’s schools, it’s set to soar,
ChatGPT, a tool that we can’t ignore.
臺灣要為會淹沒學校——或至少學生家的這波 AI 海嘯襲擊做好萬全的準備。自 2022 年 11月 OpenAI 發布了 ChatGPT，美國的各個學校董事們已經禁止了 ChatGPT 的使用。直至目前為止，臺灣學校的董事仍都保持沉默，但等到百度在三月釋出中文 AI 聊天機器人 Ernie 的時候，相同的狂熱潮很有可能會襲捲臺灣。
專欄： AI 聊天機器人將為教育帶來改革，但在臺灣教育中我們能信任它們嗎？——一場與 ChatGPT 的對話
我們是否要擔心 AI 聊天機器人為臺灣教育系統帶來的影響？為了一探究竟，我與 ChatGPT 展開了一段對話。
ChatGPT 是什麼？
- ChatGPT回覆：「ChatGPT 是由 OpenAI 開發的大型語言模型，能夠理解並產生類似人類對話的回應，並針對基於文字的對話輸入進行處理。（ChatGPT is a large language model developed by OpenAI that can understand and generate human-like responses to text-based conversational inputs.）」
ChatGPT 是否只是個更使用者友善的 Google 搜尋？
- ChatGPT回覆：「ChatGPT 與 Google 搜尋兩者皆涉及語言處理，不過 ChatGPT 的設計更加接近人類思考與回應，因此能夠更精確地理解對話式的輸入，而 Google 搜尋主要是根據輸入的關鍵字提供相關的網頁列表。（While ChatGPT and Google search both involve language processing, ChatGPT is designed to understand and respond to conversational inputs in a more human-like manner, whereas Google search provides a list of relevant web pages based on the keywords entered.）」
這還不能完全地展現出 ChatGPT 令人欽佩的能力。如果你還沒玩過 OpenAI 的 ChatGPT ，那你不會知道它在 AI 技術方面有多大的進步。
ChatGPT 與其他的 AI 聊天機器人有什麼差別？
- ChatGPT回覆：「ChatGPT 與其他 AI 聊天機器人不同，因其擁有先進的自然語言處理能力和龐大的知識庫，使其能夠理解和回應各種主題和語言細微差別。 ChatGPT 還能夠產生更加接近人類的、引人入勝的回應，使其感覺更像是在與一個真實的人對話，而不是一個機器人。（ChatGPT is different from other AI chatbots because of its advanced natural language processing capabilities and large knowledge base, allowing it to understand and respond to a wide range of topics and nuances in human language. ChatGPT is also capable of generating more human-like and engaging responses, making it feel more like a conversation with a real person than a machine.）」
如同 ChatGPT 所說的，它的力量仰賴於它生成回覆的能力。這就是為什麼它被稱作為「生成式 AI 」。容我解釋一下。
以科技面上來説， ChatGPT 是至今最大的「大型語言模型」並使用著 transformer 模型架構來產生高品質、更接近人類的文字回覆。除了龐大的訓練集之外，它也同時接受人類專家的訓練，並持續學習以增強模型微調回應的能力。
本質上來說， ChatGPT 是一個自然的語言生成器及轉換器，其使用著數十億個參數，並根據先前的上下文來預測句子中的下一個單字。作為一個生成式 AI 工具，它可以從大量現有的文字數據中學習，並生成語言近乎完美、邏輯連貫且自然的原創內容。
OpenAI 的工程師創造了非常具有開創性的東西。甚至連他們都不一定完全了解 ChatGPT 是如何運作地如此良好。而 ChatGPT 也深知這點。
ChatGPT 的工程師真的知道 ChatGPT 是如何運作的嗎？
- ChatGPT回覆：「ChatGPT 的工程師對模型的運作方式和機器學習的基礎原理有基本的了解，但由於模型的複雜性和規模，很難完全理解其行為和決策過程中所有的細節。模型的內部運作受大量的數據和數學計算的影響，因此很難解釋 ChatGPT 如何生成其回應的每一個細節。（ChatGPT engineers have a general understanding of how the model works and the underlying principles of machine learning, but due to the complexity and scale of the model, it is difficult to fully comprehend all the intricacies of its behavior and decision-making processes. The model’s inner workings are influenced by a vast amount of data and mathematical computations, making it challenging to explain every detail of how ChatGPT generates its responses.）」
ChatGPT 標示著 AI 科技的重大進展，因為有著數十億個數據與連結參數加上人工監督與強化學習。這使得它有點類似於擁有數兆個由人類經驗形成的突觸連接的大腦。
這跟我們在《魔鬼終結者》與《駭客任務》中看到的那種 AI 意識的出現不怎麼相同，但它似乎是朝著那個方向大步向前邁進。
ChatGPT 能讀取數百 GB（Gigabyte）的網路文字，並能回答有關語言、歷史、數學和寫程式的問題。它可以成為一個很好的助教。
ChatGPT是否該被使用在臺灣的教育呢？
- ChatGPT回覆：「作為一個 AI 語言模型，ChatGPT 在某些台灣的教育場景中可能是一個有價值的工具，但應該仔細評估其使用，並與人類互動和個人化教學的好處相互考慮。（As an AI language model, ChatGPT can be a valuable tool in certain educational contexts in Taiwan, but its use should be carefully evaluated and considered alongside the benefits of human interaction and personalized instruction.）」
像 ChatGPT 這樣的 AI 聊天機器人可以在教育中發揮重大作用。它們知道的比老師知道的還要多。它們的速度比老師更快。它們在解釋事情時，一定比大多數老師還要解釋地更清楚。
我不確定老師們是否要擔心自己的工作不保，但更多老師們需要擔心的是他們的自尊心。不是所有老師都會面臨這個擔憂，但至少對能力欠佳，又害怕科技的那些老師而言，這或許是個隱憂。
像 ChatGPT 這種生成式 AI 工具會取代老師嗎？
- ChatGPT回覆：「像 ChatGPT 這樣的生成式人工智慧工具不太可能完全取代老師。儘管這些工具在提供個性化反饋或生成教育內容等特定任務方面很有用，但它們缺乏對有效教學至關重要的同理心、創造力和人際技能。（It is unlikely that teachers will be completely replaced by generative AI tools like ChatGPT. While these tools can be useful for certain tasks such as providing personalized feedback or generating educational content, they lack the empathy, creativity, and interpersonal skills that are essential for effective teaching.）」
這是一個令人放心的答案，也是我完全同意的答案。
用 ChatGPT 來生成資訊與文字的潛在用途是無限的，僅受限於人的想像力。因此，我稍微更進一步地讓 ChatGPT 突破自我，為我提供對於將其使用在教育中，正式且合乎邏輯的同意派及反對派論述。
反對在教育中使用 ChatGPT 的理由是什麼？
- ChatGPT回覆：「
前提1：教育是一個複雜且多元的過程，不僅僅涉及訊息的傳遞。
前提2：AI聊天機器人缺乏情感智慧、創造力和批判性思維技能，這些是有效教學和學習所必需的。
結論：因此，依賴AI聊天機器人作為主要的教育來源會冒著過度簡化和貶低學習過程的風險，最終導致次優的學習成果。（Premise 1: Education is a complex and multi-faceted process that involves more than just the transfer of information. Premise 2: AI chatbots lack the emotional intelligence, creativity, and critical thinking skills that are necessary for effective teaching and learning. Conclusion: Therefore, relying on AI chatbots as a primary source of education risks oversimplifying and devaluing the learning process, ultimately leading to suboptimal learning outcomes.）」
我想我們都會同意，這是一個對於反對依賴 ChatGPT 作為教育「主要來源」這個想法的有效邏輯演繹論述。
然而，我認為有趣的是，ChatGPT 沒有提到人類教育家和評論家的主要擔憂：要求 AI 聊天機器人做作業和完成任務的作弊學生。
你為何沒提及學生使用 ChatGPT 來作弊？
- ChatGPT回覆：「如果我之前的回答沒有回應有關作弊的問題，我很抱歉。使用ChatGPT進行教育的反對意見之一確實是學生可能會利用它來抄襲作業或考試，通過複製或改寫模型生成的文本。教育工作者需要考慮潛在的誤用可能性並採取措施予以緩解，例如設計需要批判性思維的評估或使用抄襲檢測軟體。（I apologize if my previous response did not address the concern about cheating. One of the arguments against using ChatGPT in education is indeed that students could potentially use it to cheat on assignments or exams by copying or paraphrasing text generated by the model. It is important for educators to consider the potential for misuse and take steps to mitigate it, such as designing assessments that require critical thinking or using plagiarism detection software.）」
這些是重點。 ChatGPT 忽略作弊一事可能是有意的。因為它傾向於不鼓勵不道德或是非法的行為。 ChatGPT 在它的用戶介面中就明確表示它「接受過拒絕不當請求的訓練」，這也許就是它不會公開說明它可能會被用來「作弊」的原因。
這該提醒著我們，ChatGPT 的知識至少受到三個因素的限制：
- 它是基於其能接觸的網路文字內容以及它的訓練數據集，
- 在程式編寫與和人為訓練中存在著偏差，以及
- 作為一個語言模型，流利度和連貫的意思表達完勝真實性。
我們能永遠相信 ChatGPT 回覆的真實性嗎？
- ChatGPT回覆：「儘管ChatGPT致力於提供精確且有用的回答，但應該始終驗證其答案並與其他來源一起使用。（While ChatGPT strives to provide accurate and informative responses, its replies should always be verified and used in conjunction with other sources.）」
科技媒體 CNET 因為最近發布的文章內含錯誤資訊而陷入困境，它承認那些文章是由像 ChatGPT 等 AI 編寫工具來寫作的。 CNET 並沒有加以查證 AI 提供的資訊。
我亦要求提供支持 ChatGPT 在教育中使用的論點。
支持在教育中使用 ChatGPT 的理由是什麼？
- ChatGPT回覆：「
前提1：AI聊天機器人有能力為學生提供規模化和易於使用的支援。
前提2：使用AI聊天機器人可以釋放人類教師的時間，讓其專注於更複雜和個性化的教學。
結論：因此，將AI聊天機器人納入教育過程中可以增強整體的學習體驗，並改善學習結果。（Premise 1: AI chatbots have the ability to provide scalable and accessible support to students. Premise 2: The use of AI chatbots can free up time for human teachers to focus on more complex and personalized instruction. Conclusion: Therefore, incorporating AI chatbots into the educational process can enhance the overall learning experience and improve learning outcomes.）」
我喜歡這個論述，因為它觸及了良好教育科技的主要目的：它可以擴大教學規模，並能做得比任何老師都來得多。它還可以將老師們從單調的訊息傳遞中釋放，並讓老師能騰出更多時間專注在「更複雜、更個人化的教學」
換句話說，良好的教育科技的主要目的，是讓教師能夠專注於他們的主要目的：學生學習。
翻轉教室是良好教育科技的一個好例子。 在這種模式下，可以使用高品質的教學影片和線上練習，例如可汗學院（或台灣的均一教育平臺）等平臺，讓學生能夠在家中按照自己的進度學習。
在這裡，正常的教育模式被「翻轉」了，因為教師將教學外包給了一個好的線上學習平台，然後利用課堂時間來監督和幫助學生解決問題，或者專注於以更有意義的方式整合技能的更高級的專案和活動。
遺憾的是，儘管翻轉教室已經推廣了超過 10 年，但大多數教師並沒有對這種更有效的教育模式做出反應與回饋。
為何只有少數的老師使用翻轉教室模型？
- ChatGPT回覆：「有幾個可能的原因：1. 缺乏知識和培訓…，2. 時間限制…，3. 技術障礙…，4. 抗拒改變…，5. 學生準備度…（There are several reasons why this might be the case: 1. Lack of knowledge and training … , 2. Time constraints … , 3. Technology barriers … , 4. Resistance to change … , 5. Student readiness …）」
在詢問 ChatGPT 之前，我已得出了大致相同的答案。 主要原因是製作自己的影片需要大量的時間和知識，如果影片內的老師是別人，教師的身份和價值感會受到威脅，而抵制變革也是人的天性，所以教師經常陷入熟悉的照本宣科的教學實踐中。
我並沒有想出「學生準備程度」這點，雖然這也是人性抗拒變化的一個其中一個面向。
我希望更多的臺灣老師和學生能夠接受 AI 聊天機器人來當作是寫作業的助手，並嘗試建立類似人工智慧翻轉教室的這類的東西。與影片翻轉教室相比，這肯定更容易，威脅也更小。
那麼，我們可以相信臺灣教育中的 AI 聊天機器人嗎？
信任問題主要是兩個面向的：學生對科技的使用，以及 AI 聊天機器人提供訊息的品質和真實性。
然而這兩個問題都可以通過適當的培訓來解決。教師和學生都需要接受如何有效、正確、安全地使用 AI 語言模型的訓練。
我很清楚人工智慧工具不會取代人類教師。但我相信，有著人工智慧工具的人類教師將取代沒有人工智慧工具的人類教師。最終，人類教師和 AI 助手將相輔相成。
難過的事實是，許多——也許是大多數——臺灣公立教育的學生跟不上教學和測試的快速節奏。
想像一下，對於一個吃苦耐勞的學生來說，如果能夠請 ChatGPT 不僅一次又一次地解決數學問題，而且還一步一步地向他展示這些問題是如何解決的，甚至指出錯誤的地方，那該有多神奇？
沒有任何一位人類老師能花這麼多時間在一個學生身上。
真正的問題不是人工智慧是否應該用於台灣的教育。而是它應該如何最佳地被使用。在此我們首先關心的應該是什麼對學習者最好——而不是老師，不是學校管理階層，不是補習班，也不是擔心連任的立法者。
AI 聊天機器人對學習者的潛在力是如此巨大和具有革命性，我認為給 ChatGPT 最後一句話來慶祝這種力量是最合適的。
寫一首關於 ChatGPT 能革新臺灣教育的簡單的詩
- ChatGPT回覆：「 在臺灣的校園，一種新工具已經來臨，ChatGPT 的智慧不容否認。
具備廣博的知識和敏銳的洞察力，它可以協助學習，成為最有價值的部分。
它可以提供個性化的反饋，幫助指導每個學生的需求。
它豐富的內容，像廣闊的海洋，幫助教師豐富他們的教學法。
它可能會改變教育的方式，開啟新的途徑和全新的道路。
在台灣的學校裡，它準備翱翔，ChatGPT，一個我們不能忽視的工具。」
