背景知識
Reading Day — 月曜
今日は Academic Reading — Technology & Ethics（医療AI） の練習です。IELTS Academic Reading は3パッセージ60分ですが、今日は1パッセージ（約750語）20分の集中練習です。
| 設問形式 | 今日の出題数 | 攻略ポイント |
|---|---|---|
| True / False / Not Given | 6問（Q1〜6） | 「矛盾」と「情報なし」を区別する |
| Matching Paragraphs | 3問（Q7〜9） | 各段落のトピックをつかむ |
| Multiple Choice | 3問（Q10〜12） | ディストラクターに注意し本文根拠を確認する |
今日のテーマ: 医療AI（Artificial Intelligence in Healthcare）
IELTS頻出分野。algorithmic bias（アルゴリズムの偏り）・regulatory frameworks（規制の枠組み）・ethical responsibility（倫理的責任） などのアカデミック語彙が頻出。エンジニアとして技術的背景の理解を活かして読むと定着しやすい。
IELTS頻出分野。algorithmic bias（アルゴリズムの偏り）・regulatory frameworks（規制の枠組み）・ethical responsibility（倫理的責任） などのアカデミック語彙が頻出。エンジニアとして技術的背景の理解を活かして読むと定着しやすい。
T/F/NG 採点基準
| 判定 | 条件 | 判定方法 |
|---|---|---|
| TRUE | 本文の内容と一致する | 本文に同じ意味の記述がある |
| FALSE | 本文の内容と矛盾する | 本文に逆の意味の記述がある |
| NOT GIVEN | 本文に情報がない | 本文から判断できない（あいまいも NG） |
最頻出ミス: NOT GIVEN と FALSE の混同
NOT GIVEN = 「本文に書かれていない」。FALSE = 「本文と正反対のことが書かれている」。
自分で推論・常識で判断した結果は必ず NOT GIVEN。
NOT GIVEN = 「本文に書かれていない」。FALSE = 「本文と正反対のことが書かれている」。
自分で推論・常識で判断した結果は必ず NOT GIVEN。
パッセージ
⏰ 目安: パッセージ読解 8分 / 設問 12分
READING PASSAGE
ARTIFICIAL INTELLIGENCE IN HEALTHCARE: PROMISE AND PERIL
A
The integration of artificial intelligence into healthcare systems has accelerated dramatically over the past decade, driven by advances in machine learning, the availability of large medical datasets, and growing computational power. Proponents argue that AI has the potential to transform diagnostics, drug discovery, and patient monitoring in ways that far exceed the capabilities of human practitioners working alone. Early studies suggest that certain AI systems can detect conditions such as diabetic retinopathy and some forms of cancer from medical imaging with accuracy rates comparable to, and in some cases surpassing, those of experienced clinicians.
B
Despite these promising results, the pathway from laboratory demonstration to clinical deployment remains fraught with challenges. One recurring concern is the issue of algorithmic bias. Many AI systems are trained on datasets that do not adequately represent the full diversity of the patient population. When the training data skews toward patients of a particular demographic—whether defined by age, ethnicity, or socioeconomic status—the resulting model may perform less reliably for underrepresented groups. A landmark 2019 study published in Science demonstrated that a widely used commercial algorithm systematically underestimated the health needs of Black patients compared to White patients with equivalent medical conditions, not because of explicit racial coding, but because the algorithm used healthcare costs as a proxy for health need, and Black patients historically received less healthcare spending.
C
Regulatory frameworks have struggled to keep pace with the rapid development of AI medical tools. In the United States, the Food and Drug Administration (FDA) has approved hundreds of AI-based medical devices since 2017, yet critics argue that the approval process was not originally designed with adaptive algorithms in mind. Unlike traditional medical devices that remain static after approval, many AI systems continue to learn and evolve from new data after deployment—a characteristic sometimes called "continuous learning." This means a system that was safe and effective at the time of approval may behave differently months or years later, raising questions about post-market surveillance and accountability.
D
The question of legal and ethical responsibility when an AI system errs is particularly vexing. If an AI-assisted diagnostic tool fails to detect a tumour that a human radiologist might have caught, who bears responsibility—the hospital, the software developer, the clinician who relied on the tool, or the AI system itself? Current legal frameworks in most countries are ill-equipped to answer this question. Some legal scholars advocate for the creation of a new category of "electronic personhood" that would allow AI systems to hold a form of legal identity, though this proposal remains highly controversial and has attracted more academic debate than legislative action.
E
Patient trust represents another critical factor in the successful adoption of AI in clinical settings. Research conducted in the United Kingdom found that patients generally expressed more willingness to accept AI-assisted diagnoses when they were offered a clear explanation of how the technology works and when the final decision remained with a human clinician. Interestingly, patients over the age of 65 showed slightly higher acceptance rates for AI tools in the context of chronic disease monitoring compared to younger age groups, a finding that contradicts common assumptions about older individuals and technology adoption. However, acceptance varied considerably by cultural background and prior healthcare experiences.
F
Looking ahead, many experts believe that the most productive model for AI in healthcare is not one in which machines replace clinicians, but one in which AI serves as a sophisticated decision-support tool—amplifying the capabilities of human practitioners rather than supplanting them. This "augmentation" paradigm, as it is sometimes called, requires not only technical refinement of AI systems but also investment in training clinicians to critically evaluate AI outputs, understand their limitations, and integrate them effectively into clinical workflows. Medical schools and residency programmes around the world are only beginning to incorporate AI literacy into their curricula.
設問
Questions 1–6: True / False / Not Given
Do the following statements agree with the information given in the reading passage?
Write TRUE if the statement agrees with the information
Write FALSE if the statement contradicts the information
Write NOT GIVEN if there is no information on this topic
Write TRUE if the statement agrees with the information
Write FALSE if the statement contradicts the information
Write NOT GIVEN if there is no information on this topic
1
Some AI systems have demonstrated diagnostic accuracy for certain cancers that matches or exceeds that of experienced doctors.
2
The 2019 Science study found that the commercial algorithm contained explicit racial data fields that caused biased outputs.
3
The FDA has approved more than one hundred AI-based medical devices since 2017.
4
All AI medical devices approved by the FDA are required to remain unchanged after deployment.
5
In the UK study, patients over 65 were less willing to accept AI diagnoses than patients in younger age groups.
6
Medical schools in the United States currently offer mandatory AI literacy courses to all medical students.
Questions 7–9: Matching Paragraphs
Which paragraph contains the following information?
Write the correct letter (A–F) in the boxes.
Write the correct letter (A–F) in the boxes.
7
A description of why AI systems may function differently after receiving official approval.
8
An example of a study that revealed unintended discrimination in a commercial health algorithm.
9
The suggestion that legal systems may need entirely new categories to address AI accountability.
Questions 10–12: Multiple Choice
Choose the correct letter, A, B, C, or D.
10
According to the passage, what is the primary reason some AI systems perform less accurately for certain patient groups?
- A The algorithms are deliberately programmed to favour certain demographics.
- B Training data does not sufficiently represent all segments of the patient population.
- C Regulatory bodies impose restrictions on the types of data that can be collected.
- D Continuous learning causes the algorithm to drift away from its original purpose.
11
The "augmentation" paradigm referred to in Paragraph F implies that:
- A AI will eventually replace most clinical decision-making roles.
- B Clinicians should avoid relying on AI tools until regulations improve.
- C The most effective use of AI enhances rather than substitutes human expertise.
- D Medical schools should prioritise technical AI development over clinical training.
12
The passage suggests that patient acceptance of AI in healthcare is influenced by:
- A The cost of the AI system relative to traditional care.
- B Whether explanations are provided and human clinicians retain final authority.
- C The age of the patient, with younger groups showing consistently higher acceptance.
- D The specific type of disease being monitored.
有用語彙・表現
| 語彙・表現 | 意味 | 使用例 |
|---|---|---|
| diagnostic accuracy | 診断精度 | AI can achieve high diagnostic accuracy in radiology. |
| algorithmic bias | アルゴリズムの偏り | Algorithmic bias may disadvantage minority groups. |
| proxy for | 〜の代替指標 | Cost was used as a proxy for health need. |
| post-market surveillance | 市販後監視 | Post-market surveillance ensures device safety after approval. |
| augmentation paradigm | 拡張パラダイム | The augmentation paradigm sees AI as a tool to assist clinicians. |
| supplant | 取って代わる | AI is unlikely to supplant human judgment entirely. |
| fraught with | 〜に満ちた（困難など） | The process is fraught with ethical challenges. |
| ill-equipped | 準備不足の | Current laws are ill-equipped to handle AI liability. |
| electronic personhood | 電子的人格 | Scholars propose a category of electronic personhood for AI. |
| continuous learning | 継続的学習（ML文脈） | Continuous learning algorithms update after deployment. |
ヒント（解いた後に読む）
各段落のテーマ: A=AIの効果、B=アルゴリズムの偏り、C=規制の課題、D=法的責任、E=患者の信頼、F=未来の方向性。設問を解く前に各段落の主題を把握してからパッセージに戻ること。
Q2: Paragraph Bに「not because of explicit racial coding」という否定表現がある。これを見落とすとTRUEと誤答する。
Q5: UK研究で65歳以上は「slightly higher acceptance rates（やや高い受容率）」を示した。設問は「less willing（より低い）」と逆なのでFALSE。
Paragraph Fに「only beginning to incorporate AI literacy into their curricula」とあるが、これは「義務化（mandatory）」を意味しない。設問にはmandatoryとあり、それはパッセージに書かれていない情報 → NOT GIVEN。
モデル解答
| Q | 答え | 根拠となるパッセージの箇所 |
|---|---|---|
| 1 | TRUE | "accuracy rates comparable to, and in some cases surpassing, those of experienced clinicians" (Para. A) |
| 2 | FALSE | "not because of explicit racial coding" (Para. B) — 明示的な人種コーディングが原因ではないと書かれている |
| 3 | TRUE | "the FDA has approved hundreds of AI-based medical devices since 2017" (Para. C) |
| 4 | FALSE | "many AI systems continue to learn and evolve from new data after deployment" (Para. C) — 静止しないと書かれている |
| 5 | FALSE | "patients over the age of 65 showed slightly higher acceptance rates" (Para. E) — 65歳以上の方が受容率が高い |
| 6 | NOT GIVEN | 義務化（mandatory）については記述なし。Para. Fは「取り入れ始めている」とのみ述べる |
| 7 | C | "a system that was safe and effective at the time of approval may behave differently months or years later" (Para. C) |
| 8 | B | "A landmark 2019 study published in Science demonstrated that a widely used commercial algorithm..." (Para. B) |
| 9 | D | "Some legal scholars advocate for the creation of a new category of 'electronic personhood'" (Para. D) |
| 10 | B | "Many AI systems are trained on datasets that do not adequately represent the full diversity" (Para. B) |
| 11 | C | "amplifying the capabilities of human practitioners rather than supplanting them" (Para. F) |
| 12 | B | "offered a clear explanation...and when the final decision remained with a human clinician" (Para. E) |
解答分析
Paragraph Bの核心は「not because of explicit racial coding（明示的な人種コーディングが原因ではない）」という否定表現。設問は「explicit racial data fields が偏りの原因だった」と言っているので、パッセージと正反対 → FALSE。否定語を含む文は特に丁寧に読むこと。
「remain static（静止したまま）」と書かれているのは traditional medical devices の特性。AI systemsについては「continue to learn and evolve（継続的に学習・進化する）」と正反対のことが書かれている。設問は「All AI medical devices...remain unchanged」なので FALSE。
「continuous learning」についての記述はParagraph Cにあるが、それはQ4の話題。Q10の根拠はParagraph Bの「training data skews toward patients of a particular demographic」。選択肢Dはパッセージに書かれた情報を使っているが、Q10の問いへの答えではない典型的なディストラクター。
よくある日本人のミス
| ミス | 原因 | 正しい判断 |
|---|---|---|
| Q2をNOT GIVENと回答 | "not because of" の否定を見落とす | FALSE（パッセージで明確に否定されている） |
| Q5をTRUEと回答 | "higher acceptance"を設問と同方向に誤読 | FALSE（65歳以上の方が受容率が高い） |
| Q6をFALSEと回答 | "only beginning"をnot doingと解釈して「やっていない」と判断 | NOT GIVEN（義務化の有無は書かれていない） |
| Q10でDを選択 | パッセージで読んだ"continuous learning"という語が印象に残る | B（根拠はParagraph Bのtraining data） |
自己評価
正解数:
/ 12問
所要時間:
分
気づき・メモ: