2nd/10_Wiki/Topics/AI_and_ML/Academic-Integrity.md

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

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
wiki-2026-0508-academic-integrity	Academic Integrity	10_Wiki/Topics	verified	self	Research Ethics Scholarly Honesty Plagiarism Policy	none	A	0.9	applied	ethics research academia ai-policy plagiarism		2026-05-10	pending	language framework N/A ICAI/COPE guidelines

Academic Integrity

매 한 줄

"매 honest attribution + reproducible claim". Academic integrity = 매 work 의 origin (idea, code, data, prose) 의 truthful disclosure + 매 method 의 reproducibility. 2026 LLM-pervasive era 의 매 redefined — 매 "who wrote it" 의 less important, 매 "what was verified" 의 central.

매 핵심

매 ICAI fundamental value

1. Honesty — 매 misrepresent 의 X.
2. Trust — 매 peer 의 work 의 build on.
3. Fairness — 매 equal standard.
4. Respect — 매 prior work 의 cite.
5. Responsibility — 매 own action 의 stand by.
6. Courage — 매 misconduct 의 report.

매 violation taxonomy

• Plagiarism: 매 attribution 없이 idea/text 의 use.
• Fabrication: 매 data 의 invent.
• Falsification: 매 result 의 manipulate (image edit, p-hack).
• Authorship abuse: ghost / gift / honorary author.
• Duplicate publication: 매 same paper 의 multiple venue.
• Peer review breach: 매 confidential manuscript 의 leak / scoop.

매 LLM era (2026) 의 new question

• AI-assisted writing: 매 disclosure required (Nature, Science, ICML 의 explicit policy 2024+).
• AI as author: 매 prohibited (모든 major venue) — 매 accountability 의 absent.
• Code generation: 매 LLM-generated code 의 review + test 의 author responsibility.
• Synthetic data: 매 disclosure + provenance log.
• AI peer review: 매 manuscript 의 LLM 의 upload 의 confidentiality breach (NeurIPS 2024 ban).

매 응용

1. Citation hygiene (DOI, BibTeX, persistent ID).
2. Pre-registration (OSF, AsPredicted) 의 p-hack 방지.
3. Code + data sharing (Zenodo, GitHub release w/ DOI).
4. Conflict-of-interest 의 disclosure.
5. AI-use statement (each paper).

💻 패턴

AI-use disclosure block (2026 standard)

## AI Tool Usage Statement
- Claude Opus 4.7 was used for: prose editing, code review, literature summarization.
- All scientific claims, experimental design, and analysis were verified by the authors.
- Generated code was reviewed line-by-line and unit-tested.
- No AI tool is listed as an author.

Reproducibility checklist (NeurIPS-style)

code: https://github.com/lab/proj  # archived to Zenodo
data: https://zenodo.org/record/XXXXX
seeds: [0, 1, 2, 3, 4]
hardware: 8x H100, 80GB
runtime_per_run: 4h
hyperparams: configs/main.yaml
preregistration: https://osf.io/XXXXX

Plagiarism / paraphrase detection

# 매 simhash + embedding 의 hybrid
from datasketch import MinHash
def fingerprint(text: str, k: int = 5) -> MinHash:
    m = MinHash(num_perm=128)
    for i in range(len(text) - k + 1):
        m.update(text[i:i+k].encode())
    return m
# 매 cosine sim of sentence embedding (>0.92) 의 secondary check

Citation graph integrity

import requests
def verify_doi(doi: str) -> dict:
    r = requests.get(f"https://api.crossref.org/works/{doi}")
    r.raise_for_status()
    return r.json()["message"]  # 매 fake DOI 의 fail 의 됨

Pre-registration diff

# 매 pre-reg vs final manuscript 의 diff — exploratory vs confirmatory 의 separate
diff prereg/hypothesis.md paper/section_3_hypothesis.md

매 결정 기준

상황	Practice
LLM 의 prose polish	Disclose, no co-author
LLM 의 idea generation	Disclose, human verify each claim
Synthetic / augmented data	Disclose generation method + seed
Reproducing prior work	Cite, share repro code
Negative result	Publish (preprint OK) — 매 file-drawer 의 anti
Reviewer 의 LLM 의 use	Generally forbidden (check venue policy)

기본값: 매 transparent disclosure + 매 verifiable artifact (code/data/preregistration).

🔗 Graph

• 부모: Research Ethics · Scientific Method

🤖 LLM 활용

언제: 매 prose editing, literature summarization, code review — 매 disclosure 와 함께. 언제 X: 매 peer review 의 manuscript upload, 매 ghostwrite 의 entire paper, 매 author listing.

❌ 안티패턴

• Hidden LLM use: 매 detection (perplexity, watermark) 의 risk + retraction.
• Citation laundering: 매 not-read source 의 cite — 매 secondary citation chain bug.
• Salami slicing: 매 one study 의 multiple paper 의 split — 매 venue policy violation.
• HARKing (Hypothesizing After Results Known): 매 exploratory 의 confirmatory 의 disguise.
• P-hacking: 매 multiple comparison 의 unreported.
• Image duplication: 매 western blot reuse — 매 detection (ImageTwin, Proofig) 의 routine 2026.

🧪 검증 / 중복

• Verified (ICAI Fundamental Values 3rd ed 2021; COPE Core Practices; Nature AI policy 2023; NeurIPS 2024 reviewer guidelines).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — 2026 LLM-era policy + reproducibility patterns