← 英作文で考える一覧へ戻る

英検準1級・CEFR B2AIと情報の見分け方

事実? 意見? それともAI?

Fact, Opinion, or AI?

Ask a chatbot to summarize today's news, and everything comes back sounding like fact — smooth, confident, and complete. That confidence is exactly the problem.

A well-written AI summary rarely signals which sentence is something that actually happened and which is an inference drawn from patterns in text the model has read before. "The policy caused public anger" sounds just as solid as "The policy took effect on Monday." Only one of those two sentences describes something a reporter could have witnessed directly.

The same blind spot shows up when a student asks a chatbot to explain a historical event for homework. The date a treaty was signed is a fact, verifiable in any textbook. The chatbot's claim about why the losing side agreed to it, though, is often just the most common interpretation among historians — stated with exactly the same even, authoritative tone as the date.

None of this makes AI summaries useless. It means an old habit — separating what happened from what someone, human or machine, thinks about it — matters more than ever. A confident tone was never proof of anything. The question worth asking is no longer "does this sound right?" It is "which part of this could I actually check myself?"

A simple habit helps: read a claim, then ask which parts describe an event that could be checked against another source, and which parts are the model's best guess dressed up in fluent prose. If a second source cannot confirm it, treat it as an inference — however confident it sounds — until proven otherwise.

257 words

Vocabulary Check

  1. In the passage, "inference" is closest in meaning to —

    1. a fact confirmed by an eyewitness
    2. a conclusion reached through reasoning, not direct observation
    3. an official government statement
    4. a question with no clear answer
    答えを見る

    正解:B.a conclusion reached through reasoning, not direct observation

  2. As used in the passage, "witnessed" means —

    1. reported secondhand
    2. predicted in advance
    3. seen directly, in person
    4. written about in detail
    答えを見る

    正解:C.seen directly, in person

  3. "...rarely signals which sentence is..." — here, "signals" is closest to —

    1. hides
    2. indicates clearly
    3. argues against
    4. repeats
    答えを見る

    正解:B.indicates clearly

  4. "The policy took effect on Monday" means the policy —

    1. had an emotional impact on Monday
    2. became legally active on Monday
    3. was cancelled on Monday
    4. was widely criticized on Monday
    答えを見る

    正解:B.became legally active on Monday

  5. This sentence is closest in meaning to —

    1. Sounding certain does not mean something is true.
    2. Confidence always indicates accuracy.
    3. Tone of voice can't be measured.
    4. AI is always more confident than humans.
    答えを見る

    正解:A.Sounding certain does not mean something is true.

  6. "...the model's best guess dressed up in fluent prose" — here, "dressed up in" is closest to —

    1. translated into another language
    2. presented in a polished, convincing form
    3. deleted from the final draft
    4. checked by a professional editor
    答えを見る

    正解:B.presented in a polished, convincing form

  7. "If a second source cannot confirm it" — here, "confirm" is closest to —

    1. explain in more detail
    2. verify that something is true
    3. strongly disagree with
    4. translate into simpler words
    答えを見る

    正解:B.verify that something is true

Comprehension

  1. According to the passage, what is the actual "problem" with AI-generated summaries?

    1. They are usually factually incorrect.
    2. They present facts and inferences in the same confident tone.
    3. They take too long to generate.
    4. They cannot summarize breaking news.
    答えを見る

    正解:B.They present facts and inferences in the same confident tone.

    本文は「AIの要約は間違っている」とは一度も言っていない。問題は正確さではなく、事実と推論が同じ調子で語られ、区別できなくなることだと明示されている(第2段落)。

  2. According to the passage, what is the difference between the treaty date and the chatbot's explanation of why the losing side agreed to it?

    1. Both are equally verifiable facts.
    2. The date is a verifiable fact, while the explanation is a common interpretation stated in the same confident tone.
    3. The date is disputed among historians, but the explanation is not.
    4. Neither can be checked against a textbook.
    答えを見る

    正解:B.The date is a verifiable fact, while the explanation is a common interpretation stated in the same confident tone.

    追加した第3段落の要点。条約が結ばれた日付は教科書で確認できる事実だが、「なぜ負けた側が合意したか」という理由づけは、歴史家の間でよくある解釈の1つに過ぎない。それでも同じ自信ありげな調子で語られる、という点が本文の指摘。

  3. What does the writer mean by "which part of this could I actually check myself?"

    1. A call to distrust AI completely.
    2. A call to verify claims against something independently confirmable.
    3. A rhetorical question with no real answer.
    4. A suggestion that only experts can judge AI output.
    答えを見る

    正解:B.A call to verify claims against something independently confirmable.

    直前の文で「confident tone was never proof」と述べたうえでの問いかけ。「もっともらしく聞こえるか」ではなく「自分で裏を取れる部分はどこか」を基準にせよ、という具体的な行動の提案として読む。

  4. Which statement would the writer most likely agree with?

    1. AI summaries should not be used at all.
    2. A calm, certain-sounding text is not automatically trustworthy.
    3. Only human-written news can be trusted.
    4. Inference is more valuable than fact.
    答えを見る

    正解:B.A calm, certain-sounding text is not automatically trustworthy.

    第3段落冒頭で「None of this makes AI summaries useless」と明言しており、(A)(C)のような全否定はしていない。一貫して主張しているのは「口調の自信と正しさは別物」という(B)。

  5. According to the final paragraph, what should a reader do when a second source cannot confirm a claim?

    1. Assume the claim is false.
    2. Treat the claim as an inference until it is proven.
    3. Ask a different AI model instead.
    4. Ignore the claim completely.
    答えを見る

    正解:B.Treat the claim as an inference until it is proven.

    最終段落が示す具体的な行動指針。「まちがいだと決めつける」(A)でも「無視する」(D)でもなく、「証明されるまでは推論として扱う」という留保つきの態度を勧めている。

Summary Writing

Summarize the passage in 60–70 words.

モデル要約と解説を見る

AI-generated summaries sound confident from start to finish, but that confidence hides an important difference: some sentences report what actually happened, while others are inferences drawn from patterns in text. Because both are stated in the same tone, readers cannot easily tell them apart. The writer suggests checking each claim against a second source, and treating anything unconfirmed as inference rather than fact.

63 words

なぜこの要約になるか

  • 元の英文は5段落=「問題提起→具体例①(政策の報道)→具体例②(歴史の宿題)→含意(結論)→実践的な確かめ方」という展開。要約では2つの具体例を1つの一般化した主張に統合し、末尾の「実践のコツ」段落は"checking each claim against a second source"という1句に圧縮している。
  • 第2段落の具体例("The policy caused..." の引用部分)はそのまま抜き出さず、「一部は事実、一部は推論」という抽象化した1文に置き換えている。固有の引用文を要約に残さないのは鉄則。
  • 63語という制限の中で、"None of this makes AI summaries useless"という譲歩の一文はあえて削り、筆者が読者に求める具体的な行動を残す取捨選択をしている。要約は「全部を薄く伝える」より「一番言いたいことを厚く伝える」方が高評価になる。

Your Turn

Some people say AI-generated summaries should always include a label showing which sentences are facts and which are inferences. Do you agree or disagree with this opinion? Write your opinion in English, using 120–150 words.

最初の1文で賛成・反対を明示する。理由は2つ、それぞれに具体例を添える。反対意見にも軽く触れてから自分の立場に戻ると(譲歩→反論)、説得力が上がる。

使える型

  • I agree with this opinion for two reasons.
  • First, ... Second, ...
  • One possible objection to this is that ..., but ...
  • For these reasons, I believe that ...

考えの道具箱

この英文が扱っているのは、既刊『新聞部、うわさの答え合わせ。』第2章で新聞部が学んだ「色分けルール」と同じ考え方です。「たしかにあったこと(事実)」と「思ったこと(意見)」を、色を分けるようにはっきり区別する——これは人から聞いた噂話だけでなく、AIが書いた文章にもそのまま当てはまります。

英文を読むときは、「confident(自信ありげ)に書かれている」ことと「正しい」ことを、いったん別の話として扱う練習をしてみてください。要約問題や意見論述で安定して得点する人ほど、実はこの色分けを英語でも自然に行っています。

「色分けルール」のページを見る →

Read the Original

The Hibari Gazette cover

この考え方が生まれた物語を、英語のまま読む。

The Hibari Gazette

Did you check that?

Read on Amazon

次のセットが公開されたら、メールでお知らせします →