IELTS Reading: What are the social implications of increasing reliance on AI in governance? – Đề thi mẫu có đáp án chi tiết

Mở bài

Trong bối cảnh chuyển đổi số khu vực công, câu hỏi “What Are The Social Implications Of Increasing Reliance On AI In Governance?” đang trở thành tâm điểm bàn luận toàn cầu. Chủ đề này xuất hiện ngày càng thường xuyên trong IELTS Reading nhờ tính thời sự, đa chiều và phù hợp với các dạng câu hỏi kiểm tra kỹ năng đọc hiểu sâu. Trong bài viết này, bạn sẽ được luyện một bộ đề IELTS Reading “như thi thật” gồm 3 passages (Easy → Medium → Hard), xoay quanh tác động xã hội khi cơ quan công quyền tăng cường sử dụng AI trong quản trị. Bạn sẽ học cách xử lý nhiều dạng câu hỏi phổ biến, có đáp án chi tiết, giải thích rõ ràng, cùng từ vựng trọng tâm và kỹ thuật làm bài thực chiến. Bài phù hợp với học viên từ band 5.0 trở lên muốn nâng trình kỹ năng skimming, scanning, suy luận và nhận diện paraphrase. Đây là một tài liệu luyện tập có tính ứng dụng cao, giúp bạn tự tin hơn với các chủ đề công nghệ – xã hội, thường gặp trong đề thi chính thức.

1. Hướng dẫn làm bài IELTS Reading

Tổng Quan Về IELTS Reading Test

• Thời gian: 60 phút cho 3 passages
• Tổng số câu hỏi: 40 câu
• Phân bổ thời gian khuyến nghị:
  • Passage 1: 15-17 phút
  • Passage 2: 18-20 phút
  • Passage 3: 23-25 phút

Chiến Lược Làm Bài Hiệu Quả

• Đọc câu hỏi trước, sau đó đọc passage (skimming để lấy ý chính, scanning để tìm chi tiết)
• Chú ý từ khóa và paraphrase (đặc biệt là synonyms, cấu trúc bị động/chủ động)
• Quản lý thời gian chặt chẽ, đánh dấu câu khó để quay lại
• Không bỏ trống câu nào; đoán có cơ sở khi cần

Các Dạng Câu Hỏi Trong Đề Này

• Multiple Choice
• True/False/Not Given
• Sentence Completion
• Yes/No/Not Given
• Matching Headings
• Summary/Note Completion
• Matching Features
• Short-answer Questions

Hình minh họa IELTS Reading chủ đề AI trong quản trị, tips và chiến lược làm bài

---

2. IELTS Reading Practice Test

PASSAGE 1 – Everyday Governance Meets Algorithms

Độ khó: Easy (Band 5.0-6.5)

Thời gian đề xuất: 15-17 phút

In many public services, artificial intelligence (AI) is no longer a distant concept but a tool quietly helping officials work faster. When a city receives thousands of requests to repair streetlights or remove waste, an AI system can triage cases, grouping them by urgency and location. This kind of predictive and prioritization technology is attractive because it promises to reduce backlogs, cut waiting times, and make services more equitable across different neighborhoods.

However, the social implications of increasing reliance on AI in governance are more complex than simple efficiency. Citizens want decisions that are not only accurate but also fair, and they expect processes to be transparent. If an algorithm decides which school a child can attend or who qualifies for a housing benefit, people naturally ask: who designed the rules, and can these decisions be appealed?

One common misconception is that AI is a neutral machine that automatically removes human bias. In reality, algorithms learn from data that may reflect historical inequalities. If past investment favored certain districts, a model trained on that data might recommend continuing that pattern. Without careful auditing and retraining, an AI could unintentionally reinforce the very biases it was expected to fix.

Another social concern is explainability. People tend to trust decisions when they understand the reasons behind them. But many AI models are like black boxes, producing outputs without clear justifications. To build public trust, agencies can require plain-language explanations that show the key factors behind a decision. They can also publish impact assessments that outline risks, benefits, and measures taken to prevent harm, especially to vulnerable communities.

Importantly, AI does not act alone. It is shaped by policy choices, budgets, and values. A helpful way to think about this is to view AI as a tool that can be set to different goals. If the goal is only cost savings, social outcomes may be mixed. If the goal includes equity, public health, or environmental metrics, then the model can be designed to balance these priorities. In practice, trade-offs are inevitable, but making them visible allows citizens to discuss what matters most.

Public engagement can also improve AI in governance. Some cities have created citizen panels to review how algorithms are used in areas like traffic enforcement or welfare eligibility. These panels might not write code, but they can set guiding principles: avoid disproportionate impacts, allow human review, and provide accessible appeals. Moreover, publishing datasets, within privacy limits, invites independent researchers to identify errors. This collaborative approach can catch problems early and improve outcomes.

None of this means AI should be abandoned. It means AI should be used wisely. When procurement teams buy AI systems from vendors, they can include requirements for fairness testing, data quality checks, and ongoing monitoring. When managers deploy new tools, they can conduct pilots and adjust based on feedback. When frontline staff use algorithms, they can apply human judgment rather than following the model blindly. In short, responsible use of AI depends on design, oversight, and culture.

In the end, AI in governance will be judged by lived experience. If residents see faster repairs, more consistent decisions, and clear explanations, trust will grow. If they encounter errors without recourse, or if certain groups are systematically disadvantaged, skepticism will deepen. The social implications of AI are not a fixed outcome but a reflection of how we choose to build and govern these systems. That is why transparency, accountability, and participation matter just as much as technical accuracy.

Key takeaway: AI can help governments serve people better, but only when it is combined with clear goals, strong safeguards, and meaningful engagement. Otherwise, even the most advanced system may fail the communities it is meant to support.

Lưu ý: Trong bài đọc, các cụm như predictive, auditing, impact assessments, và cấu trúc nhấn mạnh như In practice, trade-offs are inevitable được làm đậm để bạn nhận diện từ vựng và cấu trúc quan trọng.

Questions 1-13

Questions 1-5
Choose the correct letter, A, B, C or D.

1. What is a main appeal of AI tools in public services mentioned in the passage?
   A. They eliminate the need for human workers entirely
   B. They can triage cases to improve service speed
   C. They guarantee perfect fairness in all decisions
   D. They reduce the number of citizen complaints

2. According to the passage, a common misconception about AI is that it
   A. is too expensive for most cities
   B. always needs real-time data
   C. automatically removes human bias
   D. can explain its decisions clearly

3. Publishing impact assessments primarily aims to
   A. protect vendors from legal action
   B. show the public the risks and benefits
   C. simplify the algorithmic code
   D. speed up procurement processes

4. The passage suggests that setting goals for AI systems should
   A. focus only on cost savings
   B. include multiple social priorities
   C. be delegated to vendors
   D. avoid environmental metrics

5. Citizen panels are described as useful mainly because they
   A. can directly rewrite algorithmic code
   B. decide the annual AI procurement budget
   C. establish guiding principles and review use
   D. guarantee zero errors in implementation

Questions 6-9
Do the following statements agree with the information in the 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

6. AI always reduces costs for governments in the short term.
7. Public trust depends only on the accuracy of AI decisions.
8. Some cities involve citizens in reviewing how algorithms are used.
9. International standards have already solved algorithmic bias.

Questions 10-13
Complete the sentences below.
Choose NO MORE THAN TWO WORDS from the passage for each answer.

10. Without careful auditing and retraining, AI may unintentionally __ existing biases.
11. To build trust, agencies should provide __ explanations for decisions.
12. Responsible use of AI relies on design, oversight, and __.
13. The social implications of AI reflect __ we build and govern these systems.

---

PASSAGE 2 – Governing Algorithms: Trust, Accountability, and the Transparency Paradox

Độ khó: Medium (Band 6.0-7.5)

Thời gian đề xuất: 18-20 phút

[Paragraph A]
As AI systems move from pilots to core public infrastructure, governments face a transparency paradox: the more complex a model becomes, the less explainable it often is to non-specialists, yet the greater the demand for accountability. Citizens affected by automated eligibility decisions may accept the use of advanced analytics if they receive clear notices, meaningful appeal rights, and evidence of independent oversight. However, disclosing too much about a system’s decision logic can enable gaming or compromise security. Striking a balance between public scrutiny and operational integrity is therefore a central challenge.

[Paragraph B]
Accountability mechanisms increasingly include algorithmic impact assessments, independent audits, and red-teaming exercises designed to probe failure modes. Still, accountability can become performative if reports are infrequent, methodologically shallow, or not acted upon. Effective oversight demands continuous monitoring, transparent datasets (with privacy-preserving safeguards), and a feedback loop that demonstrably leads to system improvements. Without these, accountability risks devolving into a checkbox ritual.

[Paragraph C]
Fairness is equally complex. Statistical definitions of equity—such as demographic parity or equalized odds—can conflict, making it impossible for a single model to satisfy all fairness criteria simultaneously. Policymakers must therefore specify the relevant context and justify trade-offs. For example, in public health allocation, minimizing false negatives may be prioritized to prevent harm, whereas in permit approvals, reducing false positives could matter more for resource allocation. In short, fairness is a policy choice operationalized in code, not a purely technical property of the model.

[Paragraph D]
Many agencies rely on “human-in-the-loop” assurances, assuming that a human will correct algorithmic mistakes. But automation bias can lead staff to over-weight model outputs, particularly under time pressure or when performance metrics reward throughput over deliberation. To mitigate this, training must include scenarios where officials are expected to document disagreements with the model, and governance must reward well-reasoned dissent, not just speed.

[Paragraph E]
Procurement also shapes outcomes. When AI solutions are purchased as opaque products with minimal auditability, the public sector may become locked into vendor dependencies that limit adaptation. Contracts should require model cards, data documentation, and recourse pathways for affected individuals. Public agencies might also pilot open models where feasible, enabling community review and reducing long-term costs, while keeping sensitive components protected.

Lưu ý: Các cụm như transparency paradox, performative, equalized odds, automation bias và cấu trúc nhấn mạnh như In short, fairness is a policy choice operationalized in code được làm đậm để giúp bạn định vị ý chính.

Questions 14-26

Questions 14-18
Do the following statements agree with the views of the writer?
Write YES if the statement agrees with the writer’s views
Write NO if the statement contradicts the writer’s views
Write NOT GIVEN if it is impossible to say what the writer thinks about this

14. Revealing too much about a system’s decision logic can create risks.
15. Independent audits are sufficient even if they occur infrequently.
16. The writer believes fairness criteria can conflict and require policy judgment.
17. A human-in-the-loop always eliminates automation bias.
18. Open models can sometimes reduce costs while improving scrutiny.

Questions 19-23
Matching Headings
Choose the correct heading for each paragraph from the list of headings below.
Write the correct number, i-ix, next to the paragraphs A-E.

List of Headings
i. The limits of transparency and the need for balance
ii. When accountability becomes ritual
iii. Defining fairness through metrics and context
iv. The inevitability of public backlash
v. Training and incentives to counter automation bias
vi. Legal reforms that standardize procurement
vii. Contracts that enable auditability and flexibility
viii. The ethics of replacing humans with machines
ix. Why open-source is always safer

19. Paragraph A
20. Paragraph B
21. Paragraph C
22. Paragraph D
23. Paragraph E

Questions 24-26
Complete the summary below.
Choose NO MORE THAN TWO WORDS from the passage for each answer.

Summary
Effective accountability requires more than occasional reports; it needs continuous monitoring, transparent datasets with privacy safeguards, and a __ that leads to improvements. While “human-in-the-loop” can help, it is vulnerable to __, especially when speed is rewarded. To avoid vendor lock-in, contracts should include model documentation and ensure __ pathways for affected individuals.

24. ---

25. ---

26. ---

---

PASSAGE 3 – Power, Design, and the Future of AI-Enabled Governance

Độ khó: Hard (Band 7.0-9.0)

Thời gian đề xuất: 23-25 phút

AI in governance sits at the intersection of institutional design and public values. Beyond the instrumental question of “what works,” there is a normative inquiry: which arrangements are legitimate, and for whom? A useful starting point distinguishes between the “governance of AI” (rules that constrain AI) and “governance with AI” (using AI to govern). These interlock: weak ex ante rules can produce systems that later constrain citizens in ways that are hard to contest, while robust procedural safeguards can empower citizens to contest and reshape those systems.

Consider polycentric governance, where multiple authorities—local, regional, and sectoral—coordinate oversight. Such multiplicity can reduce single-point failure and encourage experimentation, but it may also create institutional isomorphism, as agencies mimic one another’s practices without adequate evaluation. When procurement cycles reward speed, a path dependency emerges: early vendor choices shape later technical architectures and even how problems are conceptualized. Over time, sociotechnical imaginaries—shared visions of what AI “should” do—stabilize policy preferences and narrow the range of acceptable alternatives.

At the front line, the psychology of decision-making matters. Automation bias and authority gradients can make human reviewers reluctant to contradict model outputs, especially when metrics emphasize throughput. To counter this, agencies can realign incentives toward procedural justice: ensuring that processes are consistent, transparent, and allow voice. Research shows that people often accept adverse outcomes when they believe the process was fair, particularly if they receive coherent explanations and a chance to present additional evidence.

Another pressure arises from information asymmetries. Vendors and specialized units possess epistemic advantages—knowledge about model design, limitations, and data provenance. In principal-agent terms, public managers (principals) must design contracts that induce agents (vendors) to reveal information credibly. Mechanisms include model registries, tiered access for auditors, and algorithmic impact assessments with standardized metrics. Some propose a data fiduciary model, creating duties of loyalty and care for data handlers, while others advocate algorithmic sandboxes where systems are tested under constraint before wide deployment.

Equally important is the question of power redistribution. When eligibility or enforcement is partly automated, the location of discretion shifts. A street-level bureaucrat once decided case by case; now, much discretion is embedded in feature engineering, threshold setting, and error tolerances. The politics of defaults—what happens when data are missing, when uncertainty is high, or when conflicts arise between metrics—become central. The more opaque the pipeline, the harder it is for affected communities to mobilize, contest, or seek remedies.

Design, then, is political. Choices about objective functions, loss weighting, or fairness criteria entail moral commitments, even if expressed in code. Institutional design must therefore pair technical controls with contestability: avenues to challenge, revise, and recalibrate systems. That might mean mandated explanations, plain-language notices, and timelines for appeals; it could also mean periodic sunset clauses that force reevaluation, preventing inertia from hardening into governance by default.

In this framing, the social implications of increasing reliance on AI in governance are not merely side effects but the main event. They include shifts in administrative capacity, public trust, and the distribution of burdens and benefits. Well-governed AI can enable targeted services, faster responses, and early risk detection. Poorly governed AI can amplify historical inequities, erode accountability, and entrench private power within public functions. Between these poles lies a landscape of design choices, incentives, and oversight regimes that determine whose values are ultimately encoded.

A final point concerns learning. Institutions evolve through feedback. Algorithmic impact assessments that are filed and forgotten achieve little; those coupled to remedy mechanisms, public dashboards, and independent evaluation can recalibrate systems in light of empirical harms. Incentive-compatible oversight—where agencies gain recognition for surfacing and fixing problems—can help overcome blame avoidance. If the goal is legitimate governance with AI, then design for learning is not optional; it is the backbone of a system capable of earning and sustaining public trust.

Lưu ý: Các thuật ngữ như polycentric governance, procedural justice, epistemic, principal-agent, data fiduciary, contestability, và cấu trúc nhấn mạnh như Design, then, is political được làm đậm để phản ánh độ khó Passage 3.

Questions 27-40

Questions 27-31
Choose the correct letter, A, B, C or D.

27. The passage distinguishes between
    A. private and public AI funding
    B. governance of AI and governance with AI
    C. centralized and decentralized procurement
    D. model accuracy and model speed

28. Polycentric governance may lead to
    A. the elimination of vendor dependence
    B. guaranteed evaluation quality
    C. institutional isomorphism
    D. reduced need for transparency

29. In principal-agent terms, vendors act as
    A. principals who set public objectives
    B. agents who possess epistemic advantages
    C. neutral third parties without incentives
    D. auditors responsible for enforcement

30. The passage suggests that discretion in automated systems often resides in
    A. street-level decisions exclusively
    B. feature engineering and thresholds
    C. citizen panels and protests
    D. model registries and sandboxes

31. Incentive-compatible oversight is intended to
    A. punish agencies for reporting issues
    B. reward the concealment of failures
    C. encourage learning by recognizing fixes
    D. reduce independent evaluation

Questions 32-36
Matching Features
Match each concept with its description.
Choose the correct letter, A-G, and write the letter next to questions 32-36.

Features/Descriptions
A. Duties of loyalty and care in handling data
B. Multiple centers sharing oversight responsibility
C. Bias toward accepting automated recommendations
D. Mechanism for pre-deployment testing under constraints
E. Ensuring processes are fair, consistent, and allow voice
F. A register that records technical and governance details
G. Analytical method that guarantees all fairness metrics at once

Concepts
32. Polycentric governance
33. Automation bias
34. Procedural justice
35. Algorithmic sandbox
36. Data fiduciary

Questions 37-40
Answer the questions below.
Choose NO MORE THAN THREE WORDS for each answer.

37. What kind of clauses can prevent inertia by forcing reevaluation?
38. Which artifacts record standardized metrics to inform oversight?
39. What shifts when eligibility decisions are partly automated?
40. Which advantage do vendors hold due to specialized knowledge?

---

3. Answer Keys – Đáp Án

PASSAGE 1: Questions 1-13

1. B
2. C
3. B
4. B
5. C
6. NOT GIVEN
7. FALSE
8. TRUE
9. FALSE
10. reinforce
11. plain-language
12. culture
13. how

PASSAGE 2: Questions 14-26

14. YES
15. NO
16. YES
17. NO
18. YES
19. i
20. ii
21. iii
22. v
23. vii
24. feedback loop
25. automation bias
26. recourse

PASSAGE 3: Questions 27-40

27. B
28. C
29. B
30. B
31. C
32. B
33. C
34. E
35. D
36. A
37. sunset clauses
38. algorithmic impact assessments
39. discretion
40. epistemic advantages

---

4. Giải Thích Đáp Án Chi Tiết

Passage 1 – Giải Thích

• Câu 1 (B): Đoạn 1 nêu AI có thể “triage cases,” “reduce backlogs,” “cut waiting times” → cải thiện tốc độ phục vụ. A, C, D không được nêu hoặc sai.
• Câu 2 (C): Đoạn 3: “One common misconception is that AI… removes human bias.” → hiểu lầm rằng AI tự động loại bỏ thiên kiến.
• Câu 4 (B): Đoạn 5: đặt nhiều mục tiêu xã hội (equity, public health, environment) để cân bằng ưu tiên → B đúng. A, C, D sai hoặc mâu thuẫn.
• Câu 6 (NG): Bài không nói “luôn” giảm chi phí trong ngắn hạn.
• Câu 7 (F): Đoạn 2, 4: niềm tin phụ thuộc công bằng, minh bạch, khả năng kháng nghị; không chỉ độ chính xác.
• Câu 10 (reinforce): Đoạn 3: “reinforce the very biases…”
• Câu 12 (culture): Đoạn 7: “design, oversight, and culture.”

Passage 2 – Giải Thích

• Câu 14 (YES): Đoạn A: tiết lộ quá nhiều có thể dẫn đến “gaming” hoặc rủi ro an ninh.
• Câu 15 (NO): Đoạn B: kiểm toán không thường xuyên, nông → “performative” → không đủ.
• Câu 16 (YES): Đoạn C: fairness metrics xung đột; cần quyết định chính sách và trade-offs.
• Câu 17 (NO): Đoạn D: automation bias vẫn tồn tại; human-in-the-loop không tự động loại bỏ.
• Câu 18 (YES): Đoạn E: open models có thể giảm chi phí và tăng “community review”.
• Matching Headings: A-i (giới hạn minh bạch), B-ii (accountability ritual), C-iii (định nghĩa công bằng), D-v (huấn luyện và khuyến khích), E-vii (hợp đồng, auditability).
• Summary: feedback loop (B), automation bias (D), recourse (E).

Passage 3 – Giải Thích

• Câu 27 (B): Đoạn mở đầu nêu “governance of AI” vs “governance with AI”.
• Câu 28 (C): Đoạn 2: polycentric → nguy cơ “institutional isomorphism”.
• Câu 29 (B): Đoạn 4: vendors có “epistemic advantages”; trong khung principal-agent, họ là “agents”.
• Câu 30 (B): Đoạn 5: quyền tùy nghi nằm trong “feature engineering, thresholds, error tolerances”.
• Câu 31 (C): Đoạn cuối: “Incentive-compatible oversight… recognizing fixes.”
• Matching Features: 32-B, 33-C, 34-E, 35-D, 36-A phù hợp định nghĩa trong đoạn 2-4.
• SAQ: 37 “sunset clauses” (Đoạn 6), 38 “algorithmic impact assessments” (Đoạn 4, 7), 39 “discretion” (Đoạn 5), 40 “epistemic advantages” (Đoạn 4).

---

5. Từ Vựng Quan Trọng Theo Passage

Passage 1 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
triage	v/n	/ˈtriːɑːʒ/	phân loại theo mức độ ưu tiên	AI can triage cases	triage requests/cases
predictive	adj	/prɪˈdɪktɪv/	dự đoán	predictive technology	predictive model/analytics
backlog	n	/ˈbæk.lɒɡ/	tồn đọng	reduce backlogs	clear a backlog
equitable	adj	/ˈekwɪtəbəl/	công bằng	more equitable services	equitable access/outcomes
appeal	n/v	/əˈpiːl/	kháng nghị	decisions can be appealed	appeal process/right
inequality	n	/ˌɪn.ɪˈkwɒl.ə.ti/	bất bình đẳng	reflect historical inequalities	reduce inequality
audit	v/n	/ˈɔː.dɪt/	kiểm toán/kiểm định	careful auditing	independent audit
explainability	n	/ɪkˌspleɪnəˈbɪləti/	khả năng giải thích	lack of explainability	model explainability
black box	n	/ˌblæk ˈbɒks/	hộp đen	like black boxes	black-box model
impact assessment	n	/ɪmˈpækt əˌses.mənt/	đánh giá tác động	publish impact assessments	environmental/algorithmic impact assessment
procurement	n	/prəˈkjʊə.mənt/	mua sắm công	procurement teams buy	public procurement
pilot	n/v	/ˈpaɪ.lət/	thí điểm	conduct pilots	pilot program/study

Passage 2 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa	Ví dụ	Collocation
transparency paradox	n	/trænˈspærənsi ˈpærədɒks/	nghịch lý minh bạch	face a transparency paradox	paradox of transparency
accountability	n	/əˌkaʊn.təˈbɪl.ɪ.ti/	trách nhiệm giải trình	demand accountability	accountability mechanism
gaming	n	/ˈɡeɪ.mɪŋ/	lách luật	enable gaming	game the system
red-teaming	n	/ˈred ˌtiːmɪŋ/	kiểm thử phản công	conduct red-teaming	red-team exercise
performative	adj	/pəˈfɔːmə.tɪv/	mang tính hình thức	accountability becomes performative	performative transparency
privacy-preserving	adj	/ˈpraɪvəsi prɪˈzɜːvɪŋ/	bảo toàn quyền riêng tư	privacy-preserving safeguards	privacy-preserving tech
demographic parity	n	/ˌdeməˈɡræfɪk ˈpærɪti/	công bằng theo nhóm	achieve demographic parity	parity constraint
equalized odds	n	/ˈiːkwəlaɪzd ɒdz/	xác suất cân bằng	optimize for equalized odds	odds constraint
automation bias	n	/ˌɔːtəˈmeɪʃn ˈbaɪəs/	thiên kiến tự động	vulnerable to automation bias	mitigate automation bias
throughput	n	/ˈθruːˌpʊt/	thông lượng	reward throughput	high throughput
auditability	n	/ˌɔːdɪtəˈbɪləti/	khả năng kiểm toán	minimal auditability	ensure auditability
recourse	n	/rɪˈkɔːs/	biện pháp cứu vãn/kháng nghị	ensure recourse pathways	legal recourse
model card	n	/ˈmɒd.əl kɑːd/	thẻ mô hình (tài liệu mô tả)	require model cards	model card documentation
dependency	n	/dɪˈpen.dən.si/	phụ thuộc	vendor dependencies	dependency risk

Passage 3 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa	Ví dụ	Collocation
instrumental	adj	/ˌɪn.strʊˈmen.təl/	công cụ, thực dụng	instrumental question of “what works”	instrumental value
normative	adj	/ˈnɔː.mə.tɪv/	chuẩn tắc	normative inquiry	normative framework
ex ante	adj/adv	/ˌeks ˈænti/	trước khi xảy ra	weak ex ante rules	ex ante safeguards
polycentric governance	n	/ˌpɒliˈsentrɪk ˈɡʌvənəns/	quản trị đa trung tâm	consider polycentric governance	polycentric oversight
institutional isomorphism	n	/ˌɪnstɪˈtjuːʃənl aɪˈsɒməˌfɪz(ə)m/	đồng hình thể chế	create institutional isomorphism	coercive isomorphism
path dependency	n	/ˌpɑːθ dɪˈpendənsi/	phụ thuộc đường mòn	a path dependency emerges	lock-in/path dependency
sociotechnical imaginaries	n	/ˌsəʊsɪəʊˈtɛknɪkəl ɪˈmædʒɪnəriz/	hình dung xã hội-kỹ thuật	stabilize sociotechnical imaginaries	shared imaginaries
authority gradient	n	/ɔːˈθɒrɪti ˈɡreɪdiənt/	dốc quyền lực	steep authority gradients	flatten authority gradients
procedural justice	n	/prəˈsiː.dʒər.əl ˈdʒʌs.tɪs/	công bằng thủ tục	emphasize procedural justice	procedural fairness
epistemic	adj	/ˌepɪˈstiːmɪk/	tri thức, nhận thức	epistemic advantages	epistemic community
principal-agent	adj	/ˈprɪnsɪp(ə)l ˈeɪdʒənt/	ủy thác (chủ-thừa hành)	principal-agent terms	principal-agent problem
model registry	n	/ˈmɒd.əl ˈredʒɪstri/	sổ đăng ký mô hình	create model registries	registry entry
data fiduciary	n	/ˈdeɪtə fɪˈdjuːʃəri/	ủy thác dữ liệu	adopt a data fiduciary model	fiduciary duty
contestability	n	/kənˌtɛstəˈbɪləti/	khả năng tranh biện	ensure contestability	contestable decision
sunset clause	n	/ˈsʌn.set klɔːz/	điều khoản hết hiệu lực	mandate sunset clauses	sunset provision
remedy	n	/ˈrem.ə.di/	biện pháp khắc phục	couple with remedy mechanisms	effective remedy
incentive-compatible	adj	/ɪnˈsɛntɪv kəmˈpætəbl/	tương thích động lực	incentive-compatible oversight	incentive-compatible design

---

6. Kỹ Thuật Làm Bài Theo Từng Dạng Câu Hỏi

Multiple Choice

• Cách làm:
  • Đọc kỹ câu hỏi, gạch chân từ khóa (names, numbers, distinctive terms).
  • Xác định đoạn chứa thông tin bằng scanning.
  • Loại trừ đáp án chứa thông tin sai hoặc không được nhắc.
  • Chọn đáp án khớp nghĩa (paraphrase) chứ không dựa vào từ giống hệt.
• Lỗi thường gặp:
  • Bị đánh lừa bởi từ trùng lặp (word traps).
  • Không đối chiếu toàn câu dẫn.
• Ví dụ: P1 Q4 – “include multiple social priorities” paraphrase từ “designed to balance priorities… equity, public health, environmental metrics.”

True/False/Not Given

• Cách phân biệt:
  • True: Khớp nội dung (có thể paraphrase).
  • False: Trái ngược với bài.
  • Not Given: Bài không nói rõ để kết luận.
• Lỗi thường gặp:
  • Suy đoán theo hiểu biết ngoài bài.
  • Nhầm False với Not Given.
• Ví dụ: P1 Q6 – “AI always reduces costs in the short term.” Bài không khẳng định → Not Given.

Yes/No/Not Given

• Khác với T/F/NG: Hỏi về quan điểm tác giả (opinion/claim).
• Mẹo:
  • Tìm từ chỉ quan điểm: the writer suggests/believes/argues.
  • Kiểm tra câu khẳng định tuyệt đối (always, never) để nghi ngờ NO/NG.
• Ví dụ: P2 Q15 – Tác giả cho rằng kiểm toán không thường xuyên là performative → NO.

Matching Headings

• Cách làm:
  • Đọc nhanh câu mở đầu, câu kết đoạn để nhận diện main idea.
  • So khớp heading mang ý khái quát, không dính chi tiết nhỏ.
  • Loại trừ headings đã dùng.
• Tip: Chú trọng từ khóa trừu tượng (paradox, performative, trade-offs).
• Ví dụ: P2 Paragraph B → “performative accountability” → heading ii.

Summary/Note Completion

• Cách làm:
  • Xác định đoạn văn tóm tắt đề cập.
  • Tìm từ đồng nghĩa, paraphrase quanh chỗ trống.
  • Tuân thủ giới hạn từ (NO MORE THAN TWO WORDS).
• Ví dụ: P2 Q24 “feedback loop” bám sát cụm trong đoạn B.

Matching Features

• Cách làm:
  • Đọc kỹ định nghĩa mô tả (A-G).
  • Tìm thuật ngữ trong bài và đối chiếu meaning.
  • Cẩn trọng với các cặp gần nghĩa (registry vs sandbox).
• Ví dụ: P3 – “Data fiduciary” → duties of loyalty and care.

Short-answer Questions

• Cách làm:
  • Xác định từ khóa và tìm câu trả lời chính xác.
  • Giữ giới hạn từ.
  • Viết danh từ riêng/thuật ngữ đúng chính tả.
• Ví dụ: P3 Q37 – “sunset clauses”.

Chiến lược IELTS Reading với chủ đề AI governance, tips skimming scanning paraphrase

---

7. Chiến Lược Đạt Band Cao

Band 6.0-6.5: Nền Tảng

• Nắm ý chính từng đoạn bằng skimming.
• Hoàn thành đúng 23-26/40 câu.
• Đầu tư Passage 1-2, hạn chế mất điểm dễ.
• Luyện nhận diện từ đồng nghĩa phổ biến.

Band 7.0-7.5: Trung Cấp Cao

• Nhận diện paraphrase phức tạp, cấu trúc bị động.
• Làm đúng 30-32/40 câu, kiểm soát thời gian Passage 3.
• Vững tất cả dạng câu hỏi, đặc biệt Matching Headings.

Band 8.0-9.0: Nâng Cao

• Đọc hiểu sâu, suy luận ý ngầm và quan điểm tác giả.
• Làm 35-40/40, phân bổ thời gian linh hoạt.
• Rà soát bẫy logic (absolute claims, exceptions).

---

Kết bài

Tóm Tắt

Chủ đề “What are the social implications of increasing reliance on AI in governance?” không chỉ thời sự mà còn giàu tính học thuật, rất phù hợp để luyện IELTS Reading test. Bộ đề hôm nay cung cấp 3 passages theo độ khó tăng dần, bám sát cấu trúc Cambridge và khung câu hỏi thi thật. Đáp án kèm giải thích giúp bạn tự đánh giá, nhận diện paraphrase và bẫy thường gặp. Bảng từ vựng và kỹ thuật làm bài mang tính thực chiến sẽ hỗ trợ nâng band điểm IELTS Reading hiệu quả. Hãy lưu lại tài liệu này để ôn tập có hệ thống, và tiếp tục mở rộng kiến thức với các chủ đề công nghệ – xã hội tương tự tại [internal_link: chủ đề liên quan].