R0029/2026-03-27¶
Research run investigating AI attribution, public sentiment, journal policies, Kurosawa-Shakespeare adaptations, and AI output misrepresentation across academic, professional, and engineering contexts.
Queries¶
Q001 — AI Attribution Frameworks — Emerging but pre-standard
Query: Has anyone proposed a mechanism, standard, or framework for attributing AI contributions to collaborative human-AI work?
Answer: Multiple structured proposals exist (IBM AI Attribution Toolkit, AIA icon system, CHI 2025 research) but no standards body has adopted an AI-specific attribution standard. The field is in active proposal phase, pre-standardization.
| Hypothesis | Status | Probability |
|---|---|---|
| H1: Multiple formal frameworks exist | Partially supported | — |
| H2: Only ad hoc disclosure norms | Eliminated | — |
| H3: Emerging but pre-standard | Supported | Likely (55-80%) |
Sources: 4 | Searches: 2
Q002 — Public Sentiment on AI Content — Mixed and context-dependent
Query: What does current research say about public and technology community sentiment toward AI-generated content?
Answer: Deeply fragmented rather than uniformly positive or negative. Global trust at 46% (KPMG, 48K+), with dramatic split: 39% advanced economies vs. 57% emerging. Trust-use paradox: 66% use AI despite majority distrust. Slow positive trend (52% to 55%, 2022-2024).
| Hypothesis | Status | Probability |
|---|---|---|
| H1: Predominantly negative | Partially supported | — |
| H2: Predominantly positive | Eliminated | — |
| H3: Mixed and context-dependent | Supported | — |
Sources: 3 | Searches: 2
Q003 — Journal AI Authorship Policies — Prohibition consensus, disclosure varies
Query: Have academic journals, conferences, or publishers issued formal policies on listing AI as a co-author or contributor?
Answer: Every major venue prohibits AI authorship — universal and absolute. Disclosure requirements form a spectrum from permissive (NeurIPS: methodology-only) through moderate (ACM/IEEE: acknowledgments) to strict (Science: full prompts in methods). All Big 5 publishers prohibit AI authorship.
| Hypothesis | Status | Probability |
|---|---|---|
| H1: Universal prohibition + consensus | Partially supported | — |
| H2: No policies or inconsistent | Eliminated | — |
| H3: Prohibition consensus, disclosure varies | Supported | Almost certain (95-99%) |
Sources: 6 | Searches: 2
Q004 — Kurosawa-Shakespeare Films — Three films confirmed
Query: Which Akira Kurosawa films were based on or inspired by Shakespeare plays?
Answer: Three films: Throne of Blood (1957, Macbeth), The Bad Sleep Well (1960, Hamlet), and Ran (1985, King Lear). Settled scholarly consensus. The Bad Sleep Well's Hamlet connection is the most contested but majority of critical sources include it.
| Hypothesis | Status | Probability |
|---|---|---|
| H1: Three films (standard list) | Supported | Almost certain (95-99%) |
| H2: Two films only | Partially supported | — |
| H3: More than three | Eliminated | — |
Sources: 3 | Searches: 1
Q005 — AI Output Misrepresentation — Documented in academic/workplace, sparse for SE
Query: Are there documented cases about people submitting AI-generated output as their own work?
Answer: Extensive data for workplace (57% hide AI use, KPMG 48K+) and academic contexts (22% student admission rate; 7,000 UK formal cases). Software engineering-specific misrepresentation data is sparse — AI code tool adoption is measured but "misrepresentation" framing does not map cleanly to engineering culture.
| Hypothesis | Status | Probability |
|---|---|---|
| H1: Widespread, all contexts | Partially supported | — |
| H2: Limited/anecdotal | Eliminated | — |
| H3: Academic/workplace documented, SE sparse | Supported | — |
Sources: 4 | Searches: 2
Collection Analysis¶
Cross-Cutting Patterns¶
| Pattern | Queries Affected | Significance |
|---|---|---|
| Principle-level consensus, implementation fragmentation | Q001, Q003 | Both attribution frameworks and journal policies show agreement on principles (AI is not an author; attribution is needed) but fragmentation on implementation details |
| Trust-use paradox | Q002, Q005 | People use AI despite not trusting it (Q002) and hide their AI use (Q005) — suggesting adoption outpaces governance and social norms |
| KPMG/Melbourne study as cross-cutting source | Q002, Q005 | The same 48K+ study provides key evidence for both public sentiment and workplace misrepresentation |
| Pre-standardization state of AI norms | Q001, Q003, Q005 | Attribution frameworks are emerging (Q001), disclosure policies vary (Q003), and misrepresentation is widespread (Q005) — a coherent picture of norms lagging adoption |
Collection Statistics¶
| Metric | Value |
|---|---|
| Queries investigated | 5 |
| Answered with high confidence | 4 (Q002, Q003, Q004, Q005) |
| Answered with medium confidence | 1 (Q001) |
| H3 (nuanced) supported | 4 of 5 queries |
| H1 (affirmative) supported | 1 of 5 queries (Q004) |
| H2 (negative) eliminated | 5 of 5 queries |
Source Independence Assessment¶
The evidence base draws from genuinely diverse sources across domains: academic research (CHI 2025, Stanford), industry (IBM Research, KPMG/Melbourne), institutional policy (Nature, Science, ACM, IEEE, NeurIPS, Big 5 publishers), survey organizations (Ipsos), and cultural institutions (BFI, Criterion Collection). The main independence concern is the KPMG/Melbourne study serving as a key source for both Q002 and Q005, creating a single-study dependency for workplace data. No other comparable workplace survey was identified.
Collection Gaps¶
| Gap | Impact | Mitigation |
|---|---|---|
| Software engineering misrepresentation data | Cannot fully answer Q005 for all three requested contexts | Targeted searches confirmed the gap; cultural framing difference noted |
| Non-Western perspectives on AI attribution | All attribution frameworks (Q001) come from US/Western institutions | Acknowledged; may limit generalizability |
| AI-generated content type-specific attitudes | Q002 measures attitudes toward "AI" generally, not "AI-generated content" specifically | Acknowledged; most relevant available data used |
| ICML-specific AI policy | Q003 could not confirm ICML has AI-specific policy beyond inherited NeurIPS guidelines | Direct page fetch attempted; limited content found |
Collection Self-Audit¶
| Domain | Rating | Notes |
|---|---|---|
| Eligibility criteria | Low risk | Consistent criteria applied across all 5 queries; defined before searching |
| Search comprehensiveness | Low risk | 15+ searches across all queries; ~150 results dispositioned |
| Evaluation consistency | Low risk | Same GRADE-adapted scoring framework applied to all 20 sources |
| Synthesis fairness | Low risk | All 15 hypotheses tested fairly; 4 of 5 H3 outcomes reflect genuine evidence nuance, not default |
Resources¶
Summary¶
| Metric | Value |
|---|---|
| Queries investigated | 5 |
| Files produced | 140 |
| Sources scored | 20 |
| Evidence extracts | 20 |
| Results dispositioned | 37 selected + 113 rejected = 150 total |
| Duration (wall clock) | 29m 6s |
| Tool uses (total) | 140 |
Tool Breakdown¶
| Tool | Uses | Purpose |
|---|---|---|
| WebSearch | 17 | Search queries across all 5 topics |
| WebFetch | 8 | Page content retrieval for key sources |
| Write | 68 | File creation |
| Read | 4 | Methodology and output format reading |
| Edit | 0 | No edits needed |
| Bash | 12 | Directory creation, file generation, validation |
Token Distribution¶
| Category | Tokens |
|---|---|
| Input (context) | ~350,000 |
| Output (generation) | ~80,000 |
| Total | ~430,000 |