Research map / Large language models / AI and society
Position: AI/ML Influencers Have a Place in the Academic Process
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Compiles a matched dataset of ~8,000 papers shared by two AI/ML X influencers (AK and Aran Komatsuzaki) and finds their shared papers have higher citation counts, with causal and demographic analyses, and position-style recommendations for curation.
Problem
How researchers discover and read AI/ML papers in the era of thousands of accepted papers per conference, and whether social media influencer sharing affects paper citations and the geographic and gender diversity of highlighted work.
Approach
Applies retrospective-cohort style exact and embedding-based matching of influencer-tweeted arXiv papers against Semantic Scholar controls, then estimates causal effects with a negative outcome control (review scores) and difference-in-differences framework.
Key results
- Papers tweeted by AK have median citation count 24 (95% CI 23-25) vs 14 (13-15) for matched controls; Komatsuzaki-shared papers have median 31 (27-34) vs 12 (10.5-13.5).
- Unadjusted ATET estimates are 19%, 16%, and 9% for p=50, 75, 90; after negative outcome control, effects are smaller but remain significant at 95% level.
- Gender ratio among first authors is about 80:20 (AK) and 81:19 (Komatsuzaki) male to female, close to Taulbee survey ratios of 77:23 (PhD awardees) and 76:24 (faculty).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Share a paper by AK | AI/ML papers · Matched papers | median citations: 14 -> 24 citations | Helped |
| Share a paper by Komatsuzaki | AI/ML papers · Matched papers | median citations: 12 -> 31 citations | Helped |
| Share a paper by an influencer | AI/ML papers · Highly-cited threshold p=50 | ATET: 19\% (15-24) | Helped |
| Share a paper by an influencer | AI/ML papers · Highly-cited threshold p=75 | ATET: 16\% (13-20) | Helped |
| Share a paper by an influencer | AI/ML papers · Highly-cited threshold p=90 | ATET: 9\% (6-12) | Helped |
| Use influencer sharing as the treatment instead of no influencer sharing | AI/ML papers · Highly-cited papers, after negative outcome control adjustment | ATET | Helped |
| Use influencer sharing rather than no influencer sharing | AI/ML papers with review scores · Review scores | mean review score: p-value > 0.2 | No clear effect |
Surprising findings
- Review scores of influencer-shared papers do not differ significantly from matched controls, suggesting the citation gap is not explained by quality as measured by reviews.
- Influencer-shared papers show a strong U.S. affiliation skew from 2018-2021, diverging from global AI publication trends, though the authors attribute part of this to self-reported affiliation bias.
Limitations
- Only two influencers studied, limiting generalizability; the study has no randomized control trial.
- Gender prediction is binary and less accurate for non-Western names; affiliations are self-reported and noisy.
- Causal conclusions depend on untestable assumptions such as ignorability and equi-confounding, with review scores as the negative control.
Experimental setup
Datasets: Semantic Scholar API paper metadata OpenReview review scores Stanford HAI 2023 AI Index geographic data Taulbee survey gender distribution
Metrics: median citation count ATET Epps-Singleton p-value Kolmogorov-Smirnov p-value Mann-Whitney U p-value cosine distance
Builds on: negative outcome control SPECTER2
Closest papers by meaning
- Position: The Artificial Intelligence and Machine Learning Community Should Adopt a More Transparent and Regulated Peer Review Process (ICML 2025 · similarity 0.84)
- Promoting Research Collaboration with Open Data Driven Team Recommendation in Response to Call for Proposals (AAAI 2024 · similarity 0.81)
- Citations and Trust in LLM Generated Responses (AAAI 2025 · similarity 0.81)
- AAAR-1.0: Assessing AI’s Potential to Assist Research (ICML 2025 · similarity 0.81)
- Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews (ICML 2024 · similarity 0.81)
- Fostering the Ecosystem of AI for Social Impact Requires Expanding and Strengthening Evaluation Standards (NeurIPS 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.