TL;DR ๐ฆ
- ๐ค AI ethics is the vocabulary family for asking, โShould this system do that, and who is responsible if it harms someone?โ
- โ๏ธ The core terms are ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ (AI ethics), ์ฑ ์ (responsibility), ํธํฅ ๋ฐฉ์ง (bias prevention), and ์ธ๊ฐ์ฑ ๋ณดํธ (protection of humanity).
- ๐งญ Advanced Korean discussions often pair these nouns with verbs like ๋ณด์ฅํ๋ค (to guarantee), ์๋ฐฉํ๋ค (to prevent), ๊ท์ ํ๋ค (to regulate), and ํผ์ํ๋ค (to damage).
๐ง Iru Helper Note โ Use Korean Helper Widgets whenever you want to hear these advanced phrases aloud. Try typing full expressions like ์๊ณ ๋ฆฌ์ฆ์ ํธํฅ์ ๋ฐฉ์งํด์ผ ํ๋ค (algorithmic bias must be prevented) so your ears can catch the rhythm. ๐ง
ย
๐ The Big Picture
Talking about artificial intelligence in Korean is not only about shiny robots and dramatic sci-fi music. It is also about public responsibility, data fairness, human dignity, and institutional accountability. In Korean, this topic usually sounds formal, policy-oriented, and debate-friendly. ๐๏ธ
AI ethics vocabulary often uses compact noun phrases. For example, ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ means โAI ethics,โ while ์ฑ
์ ์์ฌ means โwhere responsibility liesโ or โwho is accountable.โ These are high-value phrases for essays, presentations, policy discussions, and advanced reading. ๐
The key move is to connect abstract nouns with action verbs: ํธํฅ์ ๋ฐฉ์งํ๋ค (to prevent bias), ํฌ๋ช
์ฑ์ ํ๋ณดํ๋ค (to secure transparency), ์ธ๊ฐ์ ์กด์์ฑ์ ๋ณดํธํ๋ค (to protect human dignity), and ์ฑ
์์ ๋ช
ํํ ํ๋ค (to clarify responsibility). ๐งฉ
๐จ The Core AI Ethics Family
- ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ โ AI ethics; the standards and principles used to decide whether AI development and use are socially responsible. ๐ค
- ์ฑ ์ โ responsibility; use it for moral, legal, or institutional duty. In AI contexts, ์ฑ ์ ์์ฌ means โthe locus of responsibility,โ or who should be held accountable. โ๏ธ
- ํธํฅ โ bias; especially unfair distortion in data, algorithms, or judgment. ์๊ณ ๋ฆฌ์ฆ ํธํฅ means โalgorithmic bias.โ ๐งฎ
- ํธํฅ ๋ฐฉ์ง โ bias prevention; a policy or design goal that tries to stop unfair outcomes before they spread. ๐ก๏ธ
- ์ธ๊ฐ์ฑ ๋ณดํธ โ protection of humanity; a broader phrase for preserving human dignity, autonomy, empathy, and moral judgment. ๐ซถ
- ํฌ๋ช ์ฑ โ transparency; the quality of being explainable, inspectable, or open to review. ๐
- ๊ณต์ ์ฑ โ fairness; a key value when AI affects hiring, credit, education, medicine, or public services. ๐งโโ๏ธ
- ๊ฐ์ธ์ ๋ณด ๋ณดํธ โ personal information protection; privacy protection, especially around data collection and use. ๐
- ๊ท์ โ regulation; rules or controls set by institutions or governments. ๐๏ธ
- ๊ฐ๋ โ oversight or supervision; human or institutional monitoring of AI systems. ๐
- ์ค๋จ์ฉ โ misuse and abuse; a compact noun used for harmful or improper use. ๐จ
- ์์จ์ฑ โ autonomy; in ethics, the ability of humans to make meaningful choices. ๐งญ
- ์กด์์ฑ โ dignity; often appears as ์ธ๊ฐ์ ์กด์์ฑ (human dignity). ๐ฑ
๐ ๏ธ How They Work in a Sentence
- ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ๋ ๊ธฐ์ ๋ฐ์ ๊ณผ ์ธ๊ฐ์ ์กด์์ฑ ์ฌ์ด์ ๊ท ํ์ ์๊ตฌํ๋ค. (AI ethics requires balance between technological development and human dignity.) โ๏ธ
Here, ์ฌ์ด์ ๊ท ํ means โbalance between,โ and ์๊ตฌํ๋ค means โto require.โ
- ๊ธฐ์ ์ ์๊ณ ๋ฆฌ์ฆ ํธํฅ์ ๋ฐฉ์งํ๊ธฐ ์ํด ํ์ต ๋ฐ์ดํฐ๋ฅผ ์ ๊ฒํด์ผ ํ๋ค. (Companies must inspect training data in order to prevent algorithmic bias.) ๐งฎ
๋ฐฉ์งํ๊ธฐ ์ํด means โin order to prevent,โ and ์ ๊ฒํ๋ค means โto inspect/check.โ
- ์๋ํ๋ ๊ฒฐ์ ์ ์ฑ ์ ์์ฌ๋ฅผ ๋ช ํํ ํ์ง ์์ผ๋ฉด ํผํด ๊ตฌ์ ๊ฐ ์ด๋ ค์์ง๋ค. (If the locus of responsibility for automated decisions is not clarified, remedying harm becomes difficult.) ๐งพ
๋ช
ํํ ํ๋ค means โto clarify,โ and ํผํด ๊ตฌ์ means โremedy/relief for harm.โ
๐งฉ General Formation Rule
Advanced AI ethics phrases often form as abstract noun + policy/action noun:
- ํธํฅ (bias) + ๋ฐฉ์ง (prevention) โ ํธํฅ ๋ฐฉ์ง (bias prevention)
- ๊ฐ์ธ์ ๋ณด (personal information) + ๋ณดํธ (protection) โ ๊ฐ์ธ์ ๋ณด ๋ณดํธ (privacy protection)
- ์ฑ ์ (responsibility) + ์์ฌ (where something lies) โ ์ฑ ์ ์์ฌ (where responsibility lies)
- ์ธ๊ฐ์ฑ (humanity) + ๋ณดํธ (protection) โ ์ธ๊ฐ์ฑ ๋ณดํธ (protection of humanity)
These phrases become natural when paired with verbs:
- ๋ณด์ฅํ๋ค โ to guarantee
- ํ๋ณดํ๋ค โ to secure
- ๊ฐํํ๋ค โ to strengthen
- ํผ์ํ๋ค โ to damage/undermine
- ์ด๋ํ๋ค โ to cause/bring about
- ๊ท์ ํ๋ค โ to regulate
๐ Cross-check note
Key Korean vocabulary and usage patterns were checked against ๊ตญ๋ฆฝ๊ตญ์ด์ / korean.go.kr, including public corpus and Korean-learning vocabulary resources that list terms such as ์ธ๊ฐ์ฑ (humanity) and research materials discussing ์ธ๊ณต์ง๋ฅ (artificial intelligence), ํธํฅ์ฑ (bias), and ethical language in Korean corpora. โ
๐ช Synonym Alert
- ์ธ๊ณต์ง๋ฅ / AI โ both mean artificial intelligence; ์ธ๊ณต์ง๋ฅ is more Korean and formal, while AI is common in media and business. ๐ค
- ํธํฅ / ํธ๊ฒฌ โ ํธํฅ is bias or skew in systems/data/judgment; ํธ๊ฒฌ is prejudice, often human/social. ๐ง
- ์ฑ ์ / ์ฑ ๋ฌด โ ์ฑ ์ is general responsibility; ์ฑ ๋ฌด is more formal, duty-bound responsibility. ๐
- ๊ฐ๋ / ๊ด๋ฆฌ โ ๊ฐ๋ emphasizes oversight; ๊ด๋ฆฌ emphasizes management or administration. ๐
- ๋ณดํธ / ๋ณด์ฅ โ ๋ณดํธ means protection from harm; ๋ณด์ฅ means guaranteeing that a right or condition is secured. ๐ก๏ธ
โ ๏ธ Pitfalls to Dodge
- โ AI๊ฐ ์ฑ ์์ ๊ฐ์ง๋ค โ โ AI ๊ฐ๋ฐ์์ ์ด์์๊ฐ ์ฑ ์์ ์ง๋ค (In Korean ethics discussions, responsibility is usually assigned to people or institutions, not the tool itself.) โ๏ธ
- โ ํธ๊ฒฌ ๋ฐฉ์ง for algorithmic fairness โ โ ํธํฅ ๋ฐฉ์ง (Use ํธํฅ for data/system bias; ํธ๊ฒฌ is more like prejudice.) ๐งฎ
- โ ์ธ๊ฐ์ฑ ๋ณด์ฅ as a default phrase โ โ ์ธ๊ฐ์ฑ ๋ณดํธ or ์ธ๊ฐ์ ์กด์์ฑ ๋ณด์ฅ (Protect humanity; guarantee dignity/rights.) ๐ฑ
- โ ํฌ๋ช ํ AI only โ โ ์ค๋ช ๊ฐ๋ฅํ ์ธ๊ณต์ง๋ฅ / ํฌ๋ช ํ ์์ฌ๊ฒฐ์ ๊ณผ์ (For systems, โexplainable AIโ or transparent decision-making often sounds more precise.) ๐
- โ ๊ท์ ๋ ํ์ ์ ๋ฌด์กฐ๊ฑด ๋ง๋๋ค โ โ ์ ์ ํ ๊ท์ ๋ ํ์ ๊ณผ ์์ ์ ๊ท ํ์ ๋๋ชจํ๋ค (Advanced discussions need balanced framing, not cartoon-villain regulation.) ๐๏ธ
โจ If you can connect ์ฑ
์, ํธํฅ, ํฌ๋ช
์ฑ, and ์กด์์ฑ, you can discuss AI ethics in Korean with nuance instead of just pointing at a robot and whispering โscary.โ ๐ค๐ซถ
ย
Try this exercise
Practice choosing the strongest advanced Korean term for each AI ethics sentence. Read the full sentence, fill the blank, then open the Answer toggle to check the nuance.
1. ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ๋ ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๋ณดํธํ๋ ๋ฐฉํฅ์ผ๋ก ๋
ผ์๋์ด์ผ ํ๋ค.
AI ethics should be discussed in a direction that protects human dignity. ๐ฑ
Answer
์กด์์ฑ โ ์ธ๊ฐ์ ์กด์์ฑ is the natural formal phrase for human dignity.
2. ์ฑ์ฉ ์๊ณ ๋ฆฌ์ฆ์ ์ฑ๋ณ์ด๋ ๋์ด์ ๋ฐ๋ฅธ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์ค์ฌ์ผ ํ๋ค.
A hiring algorithm should reduce bias based on gender or age. ๐งโ๐ผ
Answer
ํธํฅ โ Use ํธํฅ for systemic or data-driven bias.
3. ์๋ํ๋ ๊ฒฐ์ ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋ฅผ ๋ช
ํํ ํด์ผ ํผํด์๋ฅผ ๋ณดํธํ ์ ์๋ค.
The locus of responsibility for automated decisions must be clarified to protect victims. ๐งพ
Answer
์ฑ
์ ์์ฌ โ This asks where responsibility lies.
4. ๊ฐ์ธ์ ๋ณด ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์์ด๋ ์ ๋ขฐํ ์ ์๋ AI ์๋น์ค๋ฅผ ๋ง๋ค๊ธฐ ์ด๋ ต๋ค.
Without personal information protection, it is difficult to create trustworthy AI services. ๐
Answer
๋ณดํธ โ ๊ฐ์ธ์ ๋ณด ๋ณดํธ means privacy/personal data protection.
5. ์ค๋ช
๊ฐ๋ฅํ ์ธ๊ณต์ง๋ฅ์ ์์ฌ๊ฒฐ์ ๊ณผ์ ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๋์ธ๋ค.
Explainable AI increases transparency in the decision-making process. ๐
Answer
ํฌ๋ช
์ฑ โ ํฌ๋ช
์ฑ is transparency or openness to review.
6. ์ ๋ถ๋ ๊ณ ์ํ AI์ ์ค๋จ์ฉ์ ๋ง๊ธฐ ์ํด ํฉ๋ฆฌ์ ์ธ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋ฅผ ๋ง๋ จํด์ผ ํ๋ค.
The government should prepare reasonable regulation to prevent misuse and abuse of high-risk AI. ๐๏ธ
Answer
๊ท์ โ ๊ท์ means regulation or control.
7. ์๋ฃ AI๋ ์ง๋จ ๊ฒฐ๊ณผ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๋ณด์ฅํด์ผ ํ๋ค.
Medical AI must guarantee fairness in diagnostic results. ๐ฅ
Answer
๊ณต์ ์ฑ โ ๊ณต์ ์ฑ is fairness.
8. AI ์์คํ
์๋ ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด ๋ฐ๋์ ํ์ํ๋ค.
AI systems absolutely need human oversight. ๐
Answer
๊ฐ๋
โ ๊ฐ๋
means oversight/supervision.
9. ๋ฌด๋ถ๋ณํ ์๋ํ๋ ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์ฝํ์ํฌ ์ ์๋ค.
Indiscriminate automation may weaken human autonomy. ๐งญ
Answer
์์จ์ฑ โ ์์จ์ฑ means autonomy.
10. ๊ธฐ์
์ AI ์๋น์ค์ ์ํ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ์ง ๋ง๊ณ ๊ณต๊ฐํด์ผ ํ๋ค.
Companies should disclose the risks of AI services rather than concealing them. ๐จ
Answer
์ํํ๋ค โ ์ํํ๋ค means to conceal; disclosure builds trust.
Vocabulary Box
- ์กด์์ฑ โ dignity
- ํธํฅ โ bias, skew
- ์ฑ ์ ์์ฌ โ locus of responsibility
- ๊ฐ์ธ์ ๋ณด ๋ณดํธ โ personal data protection
- ํฌ๋ช ์ฑ โ transparency
- ๊ท์ โ regulation
- ๊ณต์ ์ฑ โ fairness
- ๊ฐ๋ โ oversight
- ์์จ์ฑ โ autonomy
- ์ํํ๋ค โ to conceal
Tip ๐ก Most AI ethics vocabulary becomes powerful when you attach it to verbs like ๋ณดํธํ๋ค (protect), ๋ณด์ฅํ๋ค (guarantee), and ๋ช
ํํ ํ๋ค (clarify).
Want more reps? Head into the Practice Hub below and train this vocabulary until your Korean sounds policy-panel ready. ๐๏ธ
ย
๐ฏ Practice Hub โ The Ethics of Artificial Intelligence
ย
Welcome to your AI ethics practice zone, where robots behave politely and vocabulary carries legal consequences. ๐ค Youโll practice the core language of ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ (AI ethics), ์ฑ
์ (responsibility), ํธํฅ ๋ฐฉ์ง (bias prevention), and ์ธ๊ฐ์ฑ ๋ณดํธ (protection of humanity) through advanced, policy-style Korean sentences. โ๏ธ
ย
๐ Audio reminder: Use Korean Helper Widgets to hear any full sentence you want to shadow aloud.
ย
๐ Instructions: Fill each blank with the Korean term or verb that best fits the sentence. Say the completed sentence aloud, check the Answer toggle, and notice whether the sentence is about fairness, responsibility, transparency, protection, or regulation.
ย
Reference Hub โ Core nouns ๐ง
- ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ โ AI ethics
- ์ฑ ์ โ responsibility
- ์ฑ ์ ์์ฌ โ where responsibility lies
- ํธํฅ โ bias
- ํธํฅ ๋ฐฉ์ง โ bias prevention
- ์ธ๊ฐ์ฑ ๋ณดํธ โ protection of humanity
- ์กด์์ฑ โ dignity
- ์์จ์ฑ โ autonomy
- ํฌ๋ช ์ฑ โ transparency
- ๊ณต์ ์ฑ โ fairness
ย
Reference Hub โ Action verbs โ๏ธ
- ๋ณดํธํ๋ค โ to protect
- ๋ณด์ฅํ๋ค โ to guarantee
- ๋ฐฉ์งํ๋ค โ to prevent
- ํ๋ณดํ๋ค โ to secure
- ๊ท์ ํ๋ค โ to regulate
- ๊ฐ๋ ํ๋ค โ to oversee/supervise
- ํผ์ํ๋ค โ to damage/undermine
- ์ด๋ํ๋ค โ to cause/bring about
- ์ํํ๋ค โ to conceal
- ๋ช ํํ ํ๋ค โ to clarify
ย
Exercise 1 โ Core Ethics Nouns โ๏ธ
A. ์ฌํ๋ ์ธ๊ณต์ง๋ฅ ๊ธฐ์ ์ ๋ฐ์ ๊ณผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์ฌ์ด์ ๊ท ํ์ ๊ณ ๋ฏผํด์ผ ํ๋ค.
Society must consider the balance between AI development and ethics. ๐ค
Answer
์ค๋ฆฌ โ ๊ธฐ์ ๋ฐ์ ๊ณผ ์ค๋ฆฌ ์ฌ์ด์ ๊ท ํ is a natural policy-discussion frame.
B. ์๋ ์ถ์ฒ ์์คํ
์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ํน์ ์ง๋จ์๊ฒ ๋ถ๋ฆฌํ ๊ฒฐ๊ณผ๋ฅผ ๋ณ์ ์ ์๋ค.
The bias of an automated recommendation system may produce unfavorable results for a specific group. ๐ฑ
Answer
ํธํฅ โ Use ํธํฅ for skew or bias in systems and data.
C. ์ธ๊ณต์ง๋ฅ์ ์ค๋ฅ๋ก ํผํด๊ฐ ๋ฐ์ํ์ ๋ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์์ฌ๊ฐ ๋ถ๋ถ๋ช
ํ๋ฉด ๋ฌธ์ ๊ฐ ์ปค์ง๋ค.
When harm occurs because of an AI error, the problem grows if responsibility is unclear. ๐งพ
Answer
์ฑ
์ โ Together, ์ฑ
์ ์์ฌ means where responsibility lies.
D. ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์งํค๋ ๊ฒ์ ๊ธฐ์ ๊ฐ๋ฐ์ ๊ธฐ๋ณธ ์ ์ ์ด๋ค.
Protecting human dignity is a basic premise of technological development. ๐ฑ
Answer
์กด์์ฑ โ ์ธ๊ฐ์ ์กด์์ฑ is the standard formal phrase.
E. AI๊ฐ ๋ชจ๋ ๊ฒฐ์ ์ ๋์ ํ๋ฉด ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด ์ฝํด์ง ์ ์๋ค.
If AI makes every decision instead, human autonomy may weaken. ๐งญ
Answer
์์จ์ฑ โ ์์จ์ฑ is autonomy or self-directed choice.
F. ์์ฌ๊ฒฐ์ ๊ณผ์ ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด ๋ฎ์ผ๋ฉด ์๋ฏผ๋ค์ ๊ฒฐ๊ณผ๋ฅผ ์ ๋ขฐํ๊ธฐ ์ด๋ ต๋ค.
If transparency in the decision-making process is low, citizens find it difficult to trust the result. ๐
Answer
ํฌ๋ช
์ฑ โ ํฌ๋ช
์ฑ is transparency.
G. AI ์ฑ์ ์์คํ
์ ํ์๋ค์๊ฒ ๊ฐ์ ๊ธฐ์ค์ ์ ์ฉํ๋ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ํ๋ณดํด์ผ ํ๋ค.
An AI grading system must secure fairness by applying the same standards to students. ๐
Answer
๊ณต์ ์ฑ โ ๊ณต์ ์ฑ means fairness.
H. ๋ฏผ๊ฐํ ์๋ฃ๋ฅผ ์ฐ๋ ์๋น์ค์์๋ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ๋ณดํธ๊ฐ ํ์์ ์ด๋ค.
In services that use sensitive materials, personal information protection is essential. ๐
Answer
๊ฐ์ธ์ ๋ณด โ ๊ฐ์ธ์ ๋ณด ๋ณดํธ means personal data/privacy protection.
I. ๊ณ ์ํ ์ธ๊ณต์ง๋ฅ์๋ ๋ฒ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์ง์์ ์ธ ๊ฐ๋
์ด ํ์ํ๋ค.
High-risk AI needs legal regulation and continuous oversight. ๐๏ธ
Answer
๊ท์ โ ๋ฒ์ ๊ท์ means legal regulation.
J. AI์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ํธ๋ฆฌํจ๋ณด๋ค ์ฌ๋์ ์ฐ์ ํด์ผ ํ๋ค๋ ๊ด์ ์ ๋ด๊ณ ์๋ค.
The protection of humanity in AI contains the view that people should come before convenience. ๐ซถ
Answer
์ธ๊ฐ์ฑ ๋ณดํธ โ This phrase focuses on preserving human-centered values.
Vocabulary Box
- ์ค๋ฆฌ โ ethics
- ํธํฅ โ bias
- ์ฑ ์ โ responsibility
- ์กด์์ฑ โ dignity
- ์์จ์ฑ โ autonomy
- ํฌ๋ช ์ฑ โ transparency
- ๊ณต์ ์ฑ โ fairness
- ๊ฐ์ธ์ ๋ณด โ personal information
- ๊ท์ โ regulation
- ์ธ๊ฐ์ฑ ๋ณดํธ โ protection of humanity
- ์์ฌ๊ฒฐ์ ๊ณผ์ โ decision-making process
- ๊ณ ์ํ ์ธ๊ณต์ง๋ฅ โ high-risk AI
Tip ๐ก Abstract nouns like ํฌ๋ช
์ฑ, ๊ณต์ ์ฑ, and ์์จ์ฑ often pair with ํ๋ณดํ๋ค (secure) or ๋ณด์ฅํ๋ค (guarantee).
ย
Exercise 2 โ Responsibility and Accountability ๐งพ
A. ๊ฐ๋ฐ์๋ AI ๋ชจ๋ธ์ ํ๊ณ๋ฅผ ์ฌ์ฉ์์๊ฒ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
Developers must disclose the limitations of an AI model to users. ๐ฃ
Answer
๊ณต๊ฐํ๋ค โ ๊ณต๊ฐํ๋ค means to disclose or make public.
B. ์ด์์๋ ์๋น์ค ์ฅ์ ๋ก ์ธํ ํผํด์ ๋ํด ์ฑ
์์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ํ๋ค.
Operators must take responsibility for damage caused by service failure. ๐ ๏ธ
Answer
์ ธ์ผ โ ์ฑ
์์ ์ง๋ค means to take responsibility; ์ ธ์ผ ํ๋ค means must take.
C. ์๋ํ๋ ์ฌ์ฌ์์ ์ค๋ฅ๊ฐ ๋๋ฉด ์ฑ
์ ์์ฌ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ํด์ผ ํ๋ค.
If an error occurs in automated screening, the locus of responsibility must be clarified. ๐
Answer
๋ช
ํํ โ ๋ช
ํํ ํ๋ค means to clarify.
D. ๊ธฐ์
์ด ์ํ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ๋ฉด ์๋น์์ ์ ๋ขฐ๋ ๋น ๋ฅด๊ฒ ๋ฌด๋์ง๋ค.
If a company conceals risk, consumer trust collapses quickly. ๐ณ๏ธ
Answer
์ํํ๋ค โ ์ํํ๋ค means to conceal, often with a negative tone.
E. ๊ณต๊ณต๊ธฐ๊ด์ AI ๋์
์ ์ ์ํฅ ํ๊ฐ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
Public institutions must conduct an impact assessment before introducing AI. ๐ข
Answer
์ค์ํ๋ค โ ํ๊ฐ๋ฅผ ์ค์ํ๋ค means to conduct an evaluation.
F. ํผํด์๊ฐ ๊ตฌ์ ๋ฅผ ๋ฐ์ ์ ์๋๋ก ์ ์ฐจ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
Procedures must be prepared so victims can receive remedy. ๐งโโ๏ธ
Answer
๋ง๋ จํ๋ค โ ์ ์ฐจ๋ฅผ ๋ง๋ จํ๋ค means to prepare or establish procedures.
G. AI ์ ์ฑ
์ ๊ฐ๋ฐ ์๋๋ณด๋ค ์ฌํ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์ฐ์ ํด์ผ ํ๋ค.
AI policy should prioritize social responsibility over speed of development. ๐๏ธ
Answer
์ฑ
์ โ ์ฌํ์ ์ฑ
์ means social responsibility.
H. ์ค๋ฅ๊ฐ ๋ฐ๋ณต๋๋ฉด ๊ฐ๋
๊ธฐ๊ด์ด ์กฐ์ฌ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์ ์๋ค.
If errors repeat, a supervisory agency may launch an investigation. ๐
Answer
์ฐฉ์ํ โ ์กฐ์ฌ์ ์ฐฉ์ํ๋ค means to begin/launch an investigation.
I. ์ฑ
๋ฌด๋ฅผ ๋คํ์ง ์์ ๊ธฐ์
์ ๋ฒ์ ์ ์ฌ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์ ์๋ค.
A company that fails to fulfill its duty may face legal sanctions. โ๏ธ
Answer
๋ฐ์ โ ์ ์ฌ๋ฅผ ๋ฐ๋ค means to receive/be subject to sanctions.
J. AI ์ค๋ฆฌ๋ ์ฑ
์์ ๊ธฐ์ ๋ค์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ์๋๋ก ๋ง๋ ๋ค.
AI ethics prevents responsibility from being hidden behind technology. ๐ญ
Answer
์จ๊ธฐ์ง โ ์จ๊ธฐ๋ค means to hide; this sentence criticizes responsibility-dodging.
Vocabulary Box
- ๊ฐ๋ฐ์ โ developer
- ์ด์์ โ operator
- ํ๊ณ โ limitation
- ๊ณต๊ฐํ๋ค โ to disclose
- ์ฑ ์์ ์ง๋ค โ to take responsibility
- ์ฑ ์ ์์ฌ โ locus of responsibility
- ๋ช ํํ ํ๋ค โ to clarify
- ์ํํ๋ค โ to conceal
- ์ํฅ ํ๊ฐ โ impact assessment
- ์ค์ํ๋ค โ to conduct
- ์ ์ฐจ๋ฅผ ๋ง๋ จํ๋ค โ to establish procedures
- ์กฐ์ฌ์ ์ฐฉ์ํ๋ค โ to launch an investigation
- ์ ์ฌ๋ฅผ ๋ฐ๋ค โ to face sanctions
- ์จ๊ธฐ๋ค โ to hide
Tip ๐ก Responsibility vocabulary often uses set phrases: ์ฑ
์์ ์ง๋ค, ์ฑ
์ ์์ฌ๋ฅผ ๋ฐํ๋ค, and ์ฑ
๋ฌด๋ฅผ ๋คํ๋ค.
ย
Exercise 3 โ Bias Prevention and Fairness ๐ก๏ธ
A. ํ์ต ๋ฐ์ดํฐ๊ฐ ํ์ชฝ์ผ๋ก ์น์ฐ์น๋ฉด ์๊ณ ๋ฆฌ์ฆ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด ๋ฐ์ํ ์ ์๋ค.
If training data leans to one side, algorithmic bias may occur. ๐งฎ
Answer
ํธํฅ โ ์๊ณ ๋ฆฌ์ฆ ํธํฅ is algorithmic bias.
B. ํธํฅ ๋ฐฉ์ง๋ฅผ ์ํด ๋ฐ์ดํฐ์ ๋ํ์ฑ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
To prevent bias, the representativeness of data must be secured. ๐
Answer
ํ๋ณดํ๋ค โ ๋ํ์ฑ์ ํ๋ณดํ๋ค means to secure representativeness.
C. ๋ชจ๋ธ ํ๊ฐ์๋ ๋ค์ํ ์ง๋จ์ ๊ฒฐ๊ณผ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ๋ ๊ณผ์ ์ด ํ์ํ๋ค.
Model evaluation needs a process of comparing results from diverse groups. ๐ฅ
Answer
๋น๊ตํ๋ค โ ๋น๊ตํ๋ค means to compare.
D. ๋ถ๊ณต์ ํ ๊ฒฐ๊ณผ๊ฐ ๋ฐ๊ฒฌ๋๋ฉด ๊ธฐ์ค์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
If unfair results are found, the standards must be adjusted. โ๏ธ
Answer
์กฐ์ ํ๋ค โ ๊ธฐ์ค์ ์กฐ์ ํ๋ค means to adjust standards.
E. ์ฑ์ฉ AI๋ ์ง์์์ ๋ฐฐ๊ฒฝ๋ณด๋ค ์ง๋ฌด ๋ฅ๋ ฅ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
Hiring AI should evaluate job ability rather than an applicantโs background. ๐ผ
Answer
ํ๊ฐํ๋ค โ ๋ฅ๋ ฅ์ ํ๊ฐํ๋ค means to evaluate ability.
F. ์ฌํ์ ์ฝ์์๊ฒ ๋ถ๋ฆฌํ ๊ฒฐ๊ณผ๋ฅผ ๋ด๋ ์์คํ
์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์๋๋ค.
A system that produces unfavorable results for socially vulnerable groups loses fairness. ๐งโโ๏ธ
Answer
๊ณต์ ์ฑ โ This sentence centers on fairness.
G. ํธ๊ฒฌ์ ์ฌ๋์ ํ๋์ ๊ฐ๊น๊ณ , ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์๋ฃ๋ ํ๋จ์ ๊ธฐ์ธ์ด์ง์ ๊ฐ๊น๋ค.
Prejudice is closer to a personโs attitude, while bias is closer to skew in data or judgment. ๐ง
Answer
ํธํฅ โ This is the key distinction: ํธ๊ฒฌ = prejudice, ํธํฅ = bias/skew.
H. ์๋ฃ AI๋ ํน์ ์ง์ญ ํ์๋ฅผ ๊ณผ์ ๋ํํ์ง ์๋๋ก ๋ฐ์ดํฐ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
Medical AI must supplement data so patients from certain regions are not underrepresented. ๐ฅ
Answer
๋ณด์ํ๋ค โ ๋ฐ์ดํฐ๋ฅผ ๋ณด์ํ๋ค means to supplement or improve data.
I. ์ฌ์ฌ ๊ธฐ์ค์ ๊ณต๊ฐํ๋ฉด ๊ฒฐ๊ณผ์ ๋ํ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๊ฐ ๋์์ง๋ค.
If screening standards are disclosed, trust in the results increases. ๐ค
Answer
์ ๋ขฐ โ ์ ๋ขฐ๊ฐ ๋์์ง๋ค means trust increases.
J. ํธํฅ ๋ฐฉ์ง๋ ๊ธฐ์ ๋ฌธ์ ๊ฐ ์๋๋ผ ์ฌํ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๋ฌธ์ ์ด๊ธฐ๋ ํ๋ค.
Bias prevention is not only a technical issue but also an issue of social fairness. โ๏ธ
Answer
๊ณต์ ์ฑ โ This frames bias prevention as a fairness issue.
Vocabulary Box
- ํ์ต ๋ฐ์ดํฐ โ training data
- ์น์ฐ์น๋ค โ to lean/skew
- ์๊ณ ๋ฆฌ์ฆ ํธํฅ โ algorithmic bias
- ๋ํ์ฑ โ representativeness
- ํ๋ณดํ๋ค โ to secure
- ๋น๊ตํ๋ค โ to compare
- ์กฐ์ ํ๋ค โ to adjust
- ํ๊ฐํ๋ค โ to evaluate
- ๊ณต์ ์ฑ โ fairness
- ํธ๊ฒฌ โ prejudice
- ํธํฅ โ bias/skew
- ๋ณด์ํ๋ค โ to supplement/improve
- ์ ๋ขฐ โ trust
Tip ๐ก Use ํธํฅ for algorithmic or data bias; reserve ํธ๊ฒฌ for prejudice in human perception or attitude.
ย
Exercise 4 โ Humanity, Privacy, and Oversight ๐ซถ
A. ๊ฐ์ ๋
ธ๋์ ๋์ ํ๋ AI๋ ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ํผ์ํด์๋ ์ ๋๋ค.
Even AI that replaces emotional labor must not damage human dignity. ๐ซถ
Answer
์กด์์ฑ โ The natural phrase is ์ธ๊ฐ์ ์กด์์ฑ.
B. ๋ฌด๋ถ๋ณํ ๊ฐ์๋ ๊ฐ์ธ์ ๋ณด ๋ณดํธ ์์น์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ ์ ์๋ค.
Indiscriminate surveillance can damage the principle of personal information protection. ๐
Answer
ํผ์ํ๋ค โ ์์น์ ํผ์ํ๋ค means to undermine/damage a principle.
C. AI๊ฐ ์กฐ์ธ์ ์ ๊ณตํ๋๋ผ๋ ์ต์ข
ํ๋จ์๋ ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด ๋จ์ ์์ด์ผ ํ๋ค.
Even if AI provides advice, human judgment must remain in the final decision. ๐งโโ๏ธ
Answer
ํ๋จ โ This protects human judgment in decision-making.
D. ํ์์ ์ ๋ณด๋ ์ฐ๊ตฌ ๋ชฉ์ ์ด๋ผ๋ ์๊ฒฉํ๊ฒ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋์ด์ผ ํ๋ค.
Patient information must be strictly protected even for research purposes. ๐ฅ
Answer
๋ณดํธ๋๋ค โ Passive ๋ณดํธ๋๋ค means to be protected.
E. ์ธ๊ฐ์ฑ ๋ณดํธ๋ ํจ์จ์ฑ๋ณด๋ค ์ฌ๋์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๋จผ์ ์๊ฐํ๋ค.
The protection of humanity considers peopleโs value before efficiency. ๐ฑ
Answer
๊ฐ์น โ ์ฌ๋์ ๊ฐ์น is a broad, human-centered phrase.
F. ์๋ํ๊ฐ ์ปค์ง์๋ก ์ธ๊ฐ์ ๊ฐ๋
์ฒด๊ณ๋ฅผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํด์ผ ํ๋ค.
As automation grows, human oversight systems must be strengthened. ๐
Answer
๊ฐํํ๋ค โ ์ฒด๊ณ๋ฅผ ๊ฐํํ๋ค means to strengthen a system.
G. ์ค๋ช
์์ด ์ ์๋ง ์ ์ํ๋ AI๋ ์ฌ์ฉ์์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์ ํํ ์ ์๋ค.
AI that presents only scores without explanation can limit the userโs autonomy. ๐
Answer
์์จ์ฑ โ A lack of explanation can weaken meaningful choice.
H. ๋๋ด ๋ก๋ด์ ํธ๋ฆฌํจ๋ฟ ์๋๋ผ ๊ด๊ณ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋ ๊ณ ๋ คํด์ผ ํ๋ค.
Care robots must consider not only convenience but also the humanity of relationships. ๐ค
Answer
์ธ๊ฐ์ฑ โ Here ๊ด๊ณ์ ์ธ๊ฐ์ฑ means the humane quality of relationships.
I. ๋ฐ์ดํฐ ์์ง์ ํ์ํ ๋ฒ์ ์์์๋ง ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋์ด์ผ ํ๋ค.
Data collection should be allowed only within the necessary scope. ๐
Answer
ํ์ฉ๋๋ค โ ํ์ฉ๋๋ค means to be allowed.
J. AI ์ค๋ฆฌ๋ ๊ธฐ์ ์ด ์ฌ๋์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ์ง ์๋๋ก ๊ฒฝ๊ณํ๋ค.
AI ethics guards against technology alienating people. ๐ซ๏ธ
Answer
์์ธ์ํค๋ค โ ์์ธ์ํค๋ค means to alienate or marginalize.
Vocabulary Box
- ๊ฐ์ ๋ ธ๋ โ emotional labor
- ์กด์์ฑ โ dignity
- ๋ฌด๋ถ๋ณํ ๊ฐ์ โ indiscriminate surveillance
- ๊ฐ์ธ์ ๋ณด ๋ณดํธ โ personal information protection
- ํผ์ํ๋ค โ to damage/undermine
- ํ๋จ โ judgment
- ๋ณดํธ๋๋ค โ to be protected
- ๊ฐ์น โ value
- ๊ฐ๋ ์ฒด๊ณ โ oversight system
- ๊ฐํํ๋ค โ to strengthen
- ์์จ์ฑ โ autonomy
- ์ธ๊ฐ์ฑ โ humanity
- ํ์ฉ๋๋ค โ to be allowed
- ์์ธ์ํค๋ค โ to alienate
Tip ๐ก ๋ณดํธ prevents harm; ๋ณด์ฅ guarantees a right or condition. Choose the verb based on the ethical relationship.
ย
Exercise 5 โ Mixed Mastery + Pitfall Patrol ๐ก๏ธ
A. AI๊ฐ ์ค์ค๋ก ๋ฒ์ ์ฑ
์์ ์ง๋ค๋ ํํ๋ณด๋ค ๊ฐ๋ฐ์์ ์ด์์์ ์ฑ
์์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ํ๋ค๊ฐ ๋ ์์ฐ์ค๋ฝ๋ค.
Rather than saying AI itself bears legal responsibility, it is more natural to say responsibility of developers and operators must be clarified. โ๏ธ
Answer
๋ช
ํํ โ โ assigning responsibility to AI itself โ โ
clarifying human/institutional responsibility.
B. ์๊ณ ๋ฆฌ์ฆ ๋ฌธ์ ์๋ ํธ๊ฒฌ๋ณด๋ค ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ด๋ผ๋ ๋ง์ด ๋ ์ ํํ ๊ฒฝ์ฐ๊ฐ ๋ง๋ค.
For algorithm problems, โbiasโ is often more accurate than โprejudice.โ ๐งฎ
Answer
ํธํฅ โ โ ํธ๊ฒฌ ๋ฐฉ์ง for data fairness โ โ
ํธํฅ ๋ฐฉ์ง.
C. ์ธ๊ฐ์ฑ์ ๋ณดํต ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ๊ณ , ๊ถ๋ฆฌ๋ ์กด์์ฑ์ ๋ณด์ฅํ๋ค๊ณ ๋งํ๋ค.
We usually say humanity is protected, while rights or dignity are guaranteed. ๐ซถ
Answer
๋ณดํธํ๋ค โ โ defaulting to ์ธ๊ฐ์ฑ ๋ณด์ฅ โ โ
์ธ๊ฐ์ฑ ๋ณดํธ.
D. ์ค๋ช
๊ฐ๋ฅํ ์ธ๊ณต์ง๋ฅ์ ๋จ์ํ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ AI๋ณด๋ค ๋ ๊ตฌ์ฒด์ ์ธ ํํ์ด๋ค.
Explainable AI is more specific than simply โtransparent AI.โ ๐
Answer
ํฌ๋ช
ํ โ โ only saying ํฌ๋ช
ํ AI โ โ
using ์ค๋ช
๊ฐ๋ฅํ ์ธ๊ณต์ง๋ฅ when explainability is meant.
E. ์ ์ ํ ๊ท์ ๋ ํ์ ์ ๋ง๊ธฐ๋ง ํ๋ ๊ฒ์ด ์๋๋ผ ์์ ๊ณผ ํ์ ์ ๊ท ํ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟํ๋ค.
Appropriate regulation does not merely block innovation; it promotes balance between safety and innovation. ๐๏ธ
Answer
๋๋ชจํ๋ค โ โ regulation as only anti-innovation โ โ
regulation as balance-seeking.
F. ์ํ์ ์จ๊ธฐ๋ ํ์๋ ๊ณต๊ฐ๊ฐ ์๋๋ผ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ๊ฐ๊น๋ค.
Hiding risk is closer to concealment than disclosure. ๐จ
Answer
์ํ โ ์ํ is concealment; it carries a critical tone.
G. AI๊ฐ ์ฌ๋์ ๊ฒฐ์ ์ ๋์ ํ ์๋ก ์ธ๊ฐ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ์ ์งํค๋ ์ฅ์น๊ฐ ์ค์ํด์ง๋ค.
As AI replaces human decisions, mechanisms that protect human autonomy become more important. ๐งญ
Answer
์์จ์ฑ โ The ethical issue is meaningful human choice, not just convenience.
H. ๋ถํฌ๋ช
ํ ๋ชจ๋ธ์ ๊ฒฐ๊ณผ์ ๋ํ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ๋ฅผ ๋จ์ด๋จ๋ฆด ์ ์๋ค.
An opaque model may lower trust in the result. ๐ซ๏ธ
Answer
์ ๋ขฐ โ Opaque decisions weaken trust.
I. ์ฌํ์ ์ฝ์๋ฅผ ๋ถ๋ฆฌํ๊ฒ ๋ง๋๋ ๊ฒฐ๊ณผ๋ AI์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ๋ฌธ์ ๋ก ์ด์ด์ง๋ค.
Results that disadvantage socially vulnerable groups lead to a fairness problem in AI. ๐งโโ๏ธ
Answer
๊ณต์ ์ฑ โ This directly concerns fairness.
J. ์ธ๊ณต์ง๋ฅ ์ค๋ฆฌ์ ํต์ฌ์ ๊ธฐ์ ์ ๋ฉ์ถ๋ ๊ฒ์ด ์๋๋ผ ์ฌ๋์ ์ค์ฌ์ ๏ผฟ๏ผฟ๏ผฟ๏ผฟ ๊ฒ์ด๋ค.
The core of AI ethics is not stopping technology, but placing people at the center. ๐ฏ
Answer
๋๋ โ ์ค์ฌ์ ๋๋ค means to place at the center. Very handy phrase; tiny grammar, big ethical cape.
Vocabulary Box
- ๋ช ํํ ํ๋ค โ to clarify
- ํธ๊ฒฌ โ prejudice
- ํธํฅ โ bias/skew
- ๋ณดํธํ๋ค โ to protect
- ํฌ๋ช ํ โ transparent
- ์ค๋ช ๊ฐ๋ฅํ ์ธ๊ณต์ง๋ฅ โ explainable AI
- ๋๋ชจํ๋ค โ to promote/seek
- ์ํ โ concealment
- ์์จ์ฑ โ autonomy
- ์ ๋ขฐ โ trust
- ๊ณต์ ์ฑ โ fairness
- ์ค์ฌ์ ๋๋ค โ to place at the center
- ์ฌํ์ ์ฝ์ โ socially vulnerable group
Tip ๐ก For pitfall patrol, ask: Is this about a human attitude, a data skew, an institutional duty, or a right that must be guaranteed?
ย
๐ Excellent work. You can now discuss AI ethics in Korean with words that point to real-world accountability: ์ฑ
์ ์์ฌ, ํธํฅ ๋ฐฉ์ง, ํฌ๋ช
์ฑ, ๊ณต์ ์ฑ, and ์ธ๊ฐ์ฑ ๋ณดํธ. The robots may still be mysterious, but your vocabulary is no longer hiding in the server room. ๐ค๐