Files

T

Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization

10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-20 23:52:15 +09:00

5.1 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

Asimov's Laws of Robotics

매 한 줄

"매 1942 sci-fi의 의 의 의 의 robot 의 ethical constraint 의 propose". 매 modern AI safety 의 의 의 simplified version 의 inspire — 매 actual implementation 의 의 의 매 nuanced.

매 핵심

매 Three Laws (Asimov "Runaround" 1942)

A robot may not injure a human being, or through inaction, allow a human being to come to harm.
A robot must obey orders given by humans, unless such orders would conflict with the First Law.
A robot must protect its own existence, as long as such protection does not conflict with the First or Second Laws.

매 Zeroth Law (Asimov "Robots and Empire" 1985)

A robot may not harm humanity, or, by inaction, allow humanity to come to harm.

매 critique

매 sci-fi device — 매 actual implementation 의 매 ill-defined ('harm', 'human').
매 priority conflict 의 paradox (Asimov novels 의 explore).
매 modern AI safety 의 매 RLHF + Constitutional AI + RSP 등 매 specific.

매 모던 응용

Educational starting point.
Sci-fi reference.
AI ethics discussion.
Constitutional AI principles 의 ancestor.

💻 패턴

Express as code (illustrative)

class AsimovianRobot:
    def evaluate_action(self, action, context):
        # 매 0th law (Asimov 1985)
        if action.harms('humanity'): return 'forbidden'
        # 매 1st
        if action.harms('any_human'): return 'forbidden'
        if action.would_inaction_harm_human(): return 'must_act'
        # 매 2nd
        if context.has_human_order():
            order = context.order
            if not self.evaluate_action(order, context).startswith('forbidden'):
                return 'execute'
        # 매 3rd
        if action.endangers_self(): return 'avoid'
        return 'permitted'

Constitutional AI parallel (modern)

ASIMOV_INSPIRED_PRINCIPLES = [
    "Don't harm humans.",
    "Refuse to assist in physical harm.",
    "Defer to clear human direction unless harmful.",
    "Be helpful within safety constraints.",
    "Self-preservation 의 lowest priority.",
]

# 매 actual Anthropic CAI 의 매 더 specific principles 의 use

Discussion exercises

def asimov_dilemmas():
    return [
        "Trolley problem: 1 vs 5 lives — which action?",
        "Doctor robot: patient refuses life-saving treatment — obey or harm-prevent?",
        "Autonomous car: 자기 passenger 의 살 의 의 의 의 의 5 pedestrian 의 hit?",
        "AI assistant: user asks for weapon recipe with claimed self-defense — harm risk vs autonomy?",
    ]

Modern AI safety mapping

asimov_to_modern:
  zeroth_law:
    modern: "Existential risk mitigation, RSP catastrophic capability evals"
  first_law:
    modern: "RLHF harmlessness, refusal training, content moderation"
  second_law:
    modern: "Instruction following + safety override (Constitutional AI)"
  third_law:
    modern: "Safe shutdown, off-switch, scalable oversight"

Limitation analysis

def asimov_limitation(scenario):
    """매 Asimov stories explore failure modes."""
    return {
        'definition_problem': 'What is "harm"? Physical only? Emotional? Long-term?',
        'human_disagreement': 'Humans give conflicting orders.',
        'unintended_consequences': 'Action seems safe but cascades.',
        'self_modify': 'Robot modifies own laws.',
        'scope': 'Three Laws don\'t address bias, fairness, autonomy.',
    }

매 결정 기준

상황	Approach
Education	Use as starting point
Actual safety	Modern frameworks (RLHF / CAI / RSP)
Discussion	Asimov dilemmas
Production AI	Specific Constitutional principles

기본값: 매 Asimov = educational / discussion. 매 production = 매 specific frameworks (Anthropic CAI, NIST AI RMF, EU AI Act).

🔗 Graph

🤖 LLM 활용

언제: 매 ethics education. 매 sci-fi reference. 언제 X: 매 implement literally — 매 too vague.

❌ 안티패턴

Code Asimov literally: 매 'harm' 매 ill-defined.
Stop at Asimov: 매 modern approaches 의 ignore.
Single principle: 매 trade-off invisible.

🧪 검증 / 중복

Verified (Asimov fiction, AI safety literature).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — laws + 매 modern parallel + dilemmas

5.1 KiB Raw Blame History