SCUBA Lab
Social Computing, User Behavior, and AI
SCUBA brings together research on social computing, user behavior, and AI. We study how social platforms and AI systems shape human behavior and build tools to make these environments healthier and safer. Our research spans three connected areas:
- Digital Governance Understanding what governance interventions actually work in online spaces.
- Healthy Online Behavior Proactively cultivating desirable behavior through positive reinforcement.
- Algorithmic & AI Audits Auditing how algorithms and AI systems shape attention, information exposure, and user wellbeing.
SCUBA researchers have worked with large-scale platforms including Reddit, Facebook, Twitch, YouTube, and Discord, and our research has impacted their efforts to foster healthier online spaces. Our work informed Reddit's bans of hateful communities and is cited in 16 distinct policy documents indexed by Overton, including two National Academies reports and Spain's Integration and Citizenship Plan (2026–2030).
SCUBA is a part of the Siebel School of Computing and Data Science at the University of Illinois Urbana-Champaign.
Ongoing projects
-
Models to Foster Healthy Conversations
Measuring what drives positive feedback, surfacing community-valued behavior, and predicting prosocial conversation outcomes.
-
Context-Aware AI for Online Governance
Small and large language models for community-specific moderation, with transparent expert-style architectures for cross-community governance.
-
Safety of AI-Mediated Information
Audits of generative search and AI summaries: which sources and community voices survive, how placement shapes the influence of AI answers, and bias in retrieval-augmented generation.
Answer Bubbles · Wisdom of the Loudest · First Impressions · Masking or Mitigating
-
Tools for Curating Social Feeds
Benchmarks and models for value-aligned content curation (VASTU), expressive end-user controls for personalized feeds (Pilot on Bluesky), and moderation tooling that helps volunteers find and reward good contributions (Positive Queue).
VASTU · Pilot · Positive Queue
-
Human-AI Systems for Online Safety
Interactive visualization for threaded discussions, ML-guided panel review when moderators disagree, and AI-augmented interfaces for moderators on large-scale chat platforms.
-
Causal Evidence for Behavior Change
Randomized experiments isolating how feed rank steers attention, plus large-scale causal work on positive feedback from moderators and creators on Reddit and YouTube.
-
AI Agents and Simulated Communities
How communities of AI agents behave, and when simulations of online communities support causal policy evaluation.
-
Audits of Algorithms and AI Systems
Quasi-experimental studies of news and wellbeing on Bluesky, empirical audits of Reddit’s trending feed, and analyses of how algorithmic visibility shocks community activity.
Hidden toll of news · Trending feed audit · Algorithmic cultivation
People
Faculty
- Eshwar Chandrasekharan
Current students
Undergraduate
Alumni
SCUBA alumni have gone on to faculty roles, industry research and engineering positions, and further graduate study.