トップ > ChatGPTとの対話シリーズ　「本質的な問い」を考えるヒントを得る

ChatGPTとの対話シリーズ　「本質的な問い」を考えるヒントを得る

またもやChatGPTシリーズ。発問を考えるのは大変なので、またもやChatGPTに手伝ってもらおうシリーズ！　先日研修会に出たらその際に、

「生徒が興味をひきそうな活動を考えるのが難しいといつも思うんですけど、どうやって作っているんですか？」

と聞かれた。個人的には、SELHiやSGHで、カリキュラムも教科書もへったくれもないゼロから構築する授業をずっと経験していて、その度に面白くない活動を作ったら生徒が寝るなどの超シビアで分かりやすいFeedbackをもらっていたので、知らず知らずのうちに鍛えられたのではないかと思う。最終的には、教科書の内容をもとに「本質的な問い（真正な問い）」を考えて、一緒に生徒と考えて答えを作り、発表すると言うのがみんな楽しいかなと今では感じている。

では、そんな活動を作るためのヒントはChatGPTから得られるのか？　答えは「得られる」だが、あくまでもヒントである。今日はそのヒントを得るためのChatGPTとの対話を晒して（！）、自分自身の忘備録としたい。

最初に英文を読ませる。

教科書は著作権の問題があるので、Public DomainのVOA learning Englishの英文を使用して考えてみる。今回はこれ。GoogleのMusicLMについて書かれた英文で、ChatGPTのことも出ている英文。
learningenglish.voanews.com

（ChatGPTを始めるには、ニュースサイトのZDnetの記事がわかりやすいです（クリック））。

まずは英文を読んでもらわないといけないので、コピー＆ペースト。とりあえず、キーワードリストとそれを使った要約を作ってもらう。

Please choose 10 keywords you will definitely use to summarise the following passage. After listing 10 keywords, please write short and easy summary of the passage for 2nd-year students in elementary school, using all the 10 keywords in the summary: （ここでShiftを押しながらEnterキーを押して改行して、VOAの英文をペースト）

ChatGPT:

  1. Keywords: Google, AI, MusicLM, tool, music, descriptive words, ChatGPT, OpenAI, huge amounts of data, high-quality music, biases.
  2. Summary: Google has created a new AI tool called MusicLM that can create music based on a limited set of descriptive words entered by a user. MusicLM operates similarly to the machine-learning system used by ChatGPT, another AI tool created by OpenAI. Both MusicLM and ChatGPT are trained on huge amounts of data to create realistic works. The Google team has not released MusicLM for public use yet, but they have released a dataset of music-writing pairs for other AI music generators. The team believes MusicLM will help anyone create high-quality music, but they also recognize the risks of biases in the training data and plan to study and limit these biases. In the future, they plan to improve MusicLM to include lyrics generation and better voice and music quality.

上記のように、キーワードと要約を作ってくれる。小学校２年生向けとお願いしたが、実際に使うにはやや難しそうなので、後で手を加える必要があるなあと思いながら次に進む。

Bloomの分類で質問づくり。

次は、Bloomの分類で質問を作ってもらう。まずはいつも通りジャブで、Bloomの分類を知っているかどうか確認する。

Do you know bloom's revised taxonomy?

ChatGPT:
Yes, I am aware of Bloom's Revised Taxonomy. It is a framework for organizing and classifying educational learning objectives into different levels of complexity and specificity. The taxonomy was created by Benjamin Bloom in the 1950s and revised in the 1990s. It is widely used in education and provides a structure for designing lesson plans, assessments, and educational activities that promote higher-order thinking skills such as analysis, synthesis, and evaluation. The taxonomy is organized into six levels: Remembering, Understanding, Applying, Analyzing, Evaluating, and Creating.

「はいはい、その通りですよ！」と思いながら（あんまり、ちゃんと読んでない（笑））、 Bloomの分類に従って、レベルごとの質問を５つずつ作ってもらうことにする。

About this passage, please make five questions for each of the six levels of Bloom's Revised Taxonomy

ChatGPT:
Here are five questions for each of the six levels of Bloom's Revised Taxonomy for the passage:

  1. Remembering:
    1. What is the name of the AI tool that Google researchers built to create music?
    2. Who developed ChatGPT, the AI language model mentioned in the passage?
    3. What type of content can computer-generated systems like ChatGPT produce?
    4. How does MusicLM operate?
    5. What did Google announce about the release of the MusicLM models?
  2. Understanding:
    1. How does MusicLM work?
    2. What does the training process for AI systems like MusicLM and ChatGPT involve?
    3. How does the quality of MusicLM's generated music relate to the level of detail in the description provided by the user?
    4. What is the purpose of MusicCaps, the "high-quality dataset" released by Google?
    5. What are the risks that the MusicLM team recognizes with regards to the machine-learning process?
  3. Applying:
    1. Can you give an example of a description that a user might enter into MusicLM to produce music?
    2. How would you describe the difference between the release of MusicLM models and the release of ChatGPT?
    3. How would you compare the results produced by MusicLM with the work of a human composer?
    4. What might be the implications of cultural biases in the training data for MusicLM's generated music?
    5. How might the team's plans for future development of MusicLM address any potential cultural appropriation in the system's results?
  4. Analyzing:
    1. How does MusicLM differ from other AI music tools that have been developed in the past?
    2. What are the advantages and disadvantages of using AI systems like MusicLM and ChatGPT to produce creative content?
    3. What do the examples of MusicLM-generated music demonstrate about the tool's musical creative ability?
    4. How does the use of AI in music creation compare with traditional methods of composing music?
    5. How might the development of AI music tools impact the future of the music industry?
  5. Evaluating:
    1. How effective do you think MusicLM is at producing high-quality music?
    2. What criteria would you use to determine the value or quality of MusicLM-generated music?
    3. How would you assess the potential risks associated with using AI systems to generate creative content?
    4. How would you compare the results produced by MusicLM with the results produced by ChatGPT or other AI content generators?
    5. In what ways do you think the development of AI music tools will influence the future of music composition and production?
  6. Creating:
    1. How would you modify the training process for MusicLM or ChatGPT to address any potential biases in the training data?
    2. How could you use MusicLM or ChatGPT to create a new piece of music that incorporates elements from different cultures or musical styles?
    3. How might you use MusicLM or ChatGPT in combination with other AI tools or technologies to create new forms of musical expression?
    4. How would you design a new AI system for music creation that builds on the strengths of MusicLM and ChatGPT?
    5. What type of music would you generate using MusicLM or ChatGPT, and why?

だいぶ、ChatGPTとの対話をブログに書くのも慣れてきた感じだが（いいのか！？笑）、割と面白そうな質問を考えてくれる。ただし、この答えを求めても、ChatGPTが満足する答えを出すとは限らないので、質問生成でやめておいた方が今は無難だと思う。なので、ここでやめる。Bloomの分類については、それぞれ10個作ってとするとさらにヒントとなる問題を作ることもあるので、数は必要に応じて加減すればいいかと思う。

さらにCLIL的要素を加えた活動を作ってもらうのもいい。これは前回も記事にしたので、そちらを参考にしてもらいたい（クリック）。今回は、それとは別に、Pros and consについて尋ねるのも良いと思われるので、書くようにお願いする。

Pros and Consをリストアップする。

記事についての賛否両論をリストアップすると、授業は楽しくなるかもしれない。授業を楽しくするコツはGAPを作ることだとLongの十戒にも述べてあると思ったら、なかった笑。でもPrabhuの３つのGap（information, opinion, reasoning）を作ると、コミュニケーションを行う意味が出てくるので、意識するに越したことはない。てなわけで、リストアップである。

What will be the pros and cons about the advent of this new AI? Please list 5 pros and 5 cons.

ChatGPT:

  1. Pros:
    • Convenient and quick music creation: With the use of MusicLM, users can create music quickly and easily using just a few descriptive words.
    • High-quality music production: The system is designed to produce high-quality music that is faithful to the text instructions provided by the user.
    • Personalized music generation: The system can also create examples based on a person's own singing, humming, whistling, or playing an instrument, making the music creation process more personal and unique.
    • Diversified music production: The tool can generate music from a variety of genres and styles, allowing for a diverse range of musical creations.
    • Potential for musical innovation: The system's ability to generate music from written descriptions and personal musical inputs may lead to new and innovative forms of musical expression.
  2. Cons:
    • Lack of human creativity: With the use of AI, there is the risk of losing the human touch and creativity in music production.
    • Potential for biases in the training data: The researchers have identified the presence of biases in the training data, which could raise questions about the appropriateness of music generation for cultures underrepresented in the training data.
    • Limited musical expression: The system may be limited in its ability to generate unique and diverse forms of musical expression.
    • Not yet publicly available: At this time, the Google team has not released the MusicLM models for public use, so only a limited number of people will have access to the tool.
    • Risks associated with machine learning: The team recognizes some risks linked to the machine learning process, such as the potential for cultural appropriation and biases in the system's results.
それぞれの意見に、個人的な意見を交えた英文を作らせる。

それぞれについて、個人的なエピソードを交えた短い英文を作り、どれに最も共感するかなどの活動を入れたら面白いかもしれない。個人的な思いを入れた短い英文を作るようお願いする。

Please write easy and short passage to claim the 1st pros idea you provided with a personal anecdote. Please write the claim first.

ChatGPT:
Claim: The advent of AI music tools will make music creation easier and more accessible to people who previously lacked the technical skill.

  • With the rise of AI music tools like MusicLM, creating music is becoming easier and more accessible to people who previously lacked the technical skill. I remember when I was younger, I always dreamed of making my own music but felt discouraged because I didn't know how to play any instruments. But now, with AI tools like MusicLM, I can simply input a description of the music I want to create and let the AI do the rest. It's amazing how technology is breaking down barriers and making our dreams a reality.

How about the second pros idea?

ChatGPT:
Claim: The use of AI in music creation can lead to increased efficiency and productivity.

  • As a music lover, I often spend hours trying to come up with the perfect melody or beat for a new song. With AI tools like MusicLM, this process can be significantly faster and more efficient. Instead of struggling to come up with new ideas, I can simply input a description of the music I want and let the AI do the rest. This frees up more time for me to focus on other aspects of my music creation, such as refining lyrics or practicing my instrument. With increased efficiency and productivity, I can produce more music in less time, allowing me to take my passion for music to new heights.

Consの方も以下のようにして、聞いてみる。

HPlease write easy and short passage to claim the 1st cons idea you provided with a personal anecdote. Please write the claim first.

ChatGPT:
Claim: The use of AI in music creation raises concerns about the loss of human creativity and originality in the music industry.

  • As a music lover, I have always appreciated the unique sound and style that each artist brings to the table. However, with the advent of AI music creation tools like MusicLM, I worry that the music industry will begin to rely too heavily on technology to produce songs. This could lead to a loss of originality and creativity in the industry, as songs become formulaic and lack the personal touch that comes from human artists. I fear that the music we listen to in the future will all start to sound the same, and the heart and soul that make music so special will be lost.

読むとわかる通り、割とよくあるステレオタイプな答えを書いてくるので、英文を理解をするのに生徒が困ることはあまりないのではないかと思う。

自分の意見も書かせる？

これを参考に自分の意見を書かせる活動も考えられるだろう。その際は、きっと上山先生のトリディス（クリック）を取り入れて、書かせることになる。意見は何度もいうことによって作られていくものだし、精緻化する。だから、何度もいろんな人に話して（意見を英語で全部書くより、意見を述べるときのキーワードを考えさせて、話させた方が早い。）、まず考えを固めた上で書かせる方が個人的には好きだ。ただし、先日noteをさまよっていたら、書くことの指導についての鋭い指摘があった。

note.com

指摘は以下の通りである。

元の英文の正確性が著しく低いと適切な添削はしてくれない

確かに！と思う意見である。実例も述べてあるので、分かりやすい。これについては、AI-poweredな学びとしては、以下の学習手順が考えられる。

  1. 日本語でまず意見を書かせる。その時に主語と述語を必ず入れるようにということと、あとで英語に直しやすい日本語にするように指示する。
  2. 自分でその日本語を英語に直させる。
  3. DeepL Writeに自分の書いた英語を入力して、どう違うのか、なぜ違うのか考えさせる。
  4. DeepLに、日本語を入力させて出力された英語と、最初に自分で書いた英語を比較させて、違いを考えさせる。
  5. Peer feedbackを行うために、ペアやグループで交換させて、意見をもらう。自分が予想する感想と、友達の感想がちぐはぐな場合は、英語が間違っていることが多いので、訂正を続けさせる。

さまざまなアプリや指導を見ていて、この手順が良いのではないかというのが現在感じているところである。まあ、これについては、また別記事で（←また！？）。

今日はこんなところです。