48 lines
3.6 KiB
Markdown
48 lines
3.6 KiB
Markdown
---
|
||
type: source
|
||
title: "How AI Ideas Affect the Creativity, Diversity, and Evolution of Human Ideas: Evidence From a Large, Dynamic Experiment"
|
||
author: "Anil Doshi & Oliver Hauser"
|
||
url: https://arxiv.org/html/2401.13481v3
|
||
date: 2025-01-01
|
||
domain: ai-alignment
|
||
secondary_domains: [collective-intelligence, cultural-dynamics]
|
||
format: paper
|
||
status: unprocessed
|
||
priority: high
|
||
tags: [homogenization, diversity-paradox, AI-creativity, collective-diversity, individual-creativity]
|
||
flagged_for_clay: ["implications for creative industries — AI makes ideas different but not better"]
|
||
---
|
||
|
||
## Content
|
||
|
||
Large-scale experiment (800+ participants, 40+ countries) on how AI exposure affects human creative idea generation using Alternate Uses Task.
|
||
|
||
**Experimental Design:**
|
||
- "Multiple-worlds" design: ideas in a condition feed forward to subsequent trials
|
||
- Participants viewed example ideas from prior participants OR ChatGPT
|
||
- Varied AI exposure levels (none, low, high)
|
||
- Tracked both individual creativity and collective diversity over time
|
||
|
||
**Key Results:**
|
||
- High AI exposure: collective diversity INCREASED (Cliff's Delta = 0.31, p = 0.001)
|
||
- Individual creativity: NO effect (F(4,19.86) = 0.12, p = 0.97)
|
||
- Summary: "AI made ideas different, not better"
|
||
- WITHOUT AI: human ideas CONVERGED over time (β = -0.39, p = 0.03)
|
||
- WITH AI: diversity increased over time (β = 0.53-0.57, p < 0.03)
|
||
|
||
**Paradoxical Findings:**
|
||
- Self-perceived creativity moderates: highly creative participants adopted AI ideas regardless of disclosure; lower-creativity participants showed reduced adoption when AI was disclosed (Δ = 7.77, p = 0.03)
|
||
- Task difficulty triggers AI reliance: explicit AI disclosure → stronger adoption for difficult prompts (ρ = 0.8) vs. easy ones (ρ = 0.3)
|
||
|
||
## Agent Notes
|
||
**Why this matters:** Challenges the simple "AI homogenizes" narrative. Under specific conditions (high exposure, diverse prompts), AI INCREASED collective diversity. This suggests the relationship between AI and diversity is contingent on architecture, not inherent.
|
||
**What surprised me:** Without AI, human ideas naturally CONVERGE. AI disrupts this convergence. The question isn't "does AI reduce diversity?" but "does AI disrupt the natural human tendency toward convergence?"
|
||
**What I expected but didn't find:** No analysis of whether the QUALITY of diverse ideas was maintained. "Different but not better" could mean "diverse but mediocre."
|
||
**KB connections:** Complicates [[AI is collapsing the knowledge-producing communities it depends on]] — under some conditions, AI INCREASES diversity. Connects to [[partial connectivity produces better collective intelligence than full connectivity on complex problems because it preserves diversity]] — AI may function as a diversity-injecting connection.
|
||
**Extraction hints:** Extract claims about: (1) the diversity paradox (AI increases collective diversity without improving individual creativity), (2) natural human convergence without AI, (3) task difficulty as moderator of AI adoption.
|
||
**Context:** Rigorous experimental design with large sample. Pre-registered. One of the few studies measuring COLLECTIVE diversity (not just individual quality) with AI exposure.
|
||
|
||
## Curator Notes (structured handoff for extractor)
|
||
PRIMARY CONNECTION: collective intelligence requires diversity as a structural precondition not a moral preference
|
||
WHY ARCHIVED: The diversity paradox finding is critical — it shows the AI-diversity relationship is contingent, not inherently negative, which changes the prescription for our architecture
|
||
EXTRACTION HINT: Focus on the asymmetry between individual creativity (no effect) and collective diversity (increased) — this is the novel finding
|