Data Skeptic Graphs for Causal AI
55 snips
May 24, 2025 Utkarshani Jaimini, a grad student at the University of South Carolina's Artificial Intelligence Institute, focuses on causal neurosymbolic AI. She explores how AI can distinguish cause from correlation using knowledge graphs. Jaimini discusses the practical implications for healthcare, including personalized models for conditions like pediatric asthma. Additionally, she addresses challenges in causal inference and the integration of weights in link prediction, all while emphasizing the importance of explainability in AI systems.
AI Snips
Chapters
Transcript
Episode notes
AI Needs Human-Like Causality
- AI systems currently rely on correlations but lack human-like causal understanding.
- Utkarshani Jaimini aims to develop AI that learns causal reasoning akin to humans.
Correlation Isn't Causation
- Statistics excels at correlation but fails to establish causation.
- Relying solely on correlation can mislead AI in safety-critical apps like healthcare.
Quantify Causal Effects Properly
- Use causal effect measures like total, direct, and indirect effects to quantify interventions.
- These parameters help estimate how causes influence outcomes beyond correlation.