# CausalKG: Causal Knowledge Graph Explainability using interventional and counterfactual reasoning

**Authors:** Utkarshani Jaimini, Amit Sheth  
**arXiv:** [2201.03647](https://arxiv.org/abs/2201.03647)  
**Date:** 2022-01-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2201.03647v1/

## Key points

- Humans rely on causality and hypothetical retrospection in decision-making, planning, and understanding life events
- Counterfactual scenarios play a crucial role in developing a causal model of the world
- Knowledge graphs (KGs) are used to represent time, space, and facts in AI algorithms
- Existing KGs struggle with supporting counterfactual reasoning due to reliance on linguistic patterns for extracting causal relationships
- The Causal Knowledge Graph (CausalKG) framework leverages advancements in causality and KG for enhanced explainability in AI systems
- CausalKG utilizes a hyper-relational graph representation to effectively capture complex causal relations
- It offers interventional and counterfactual reasoning capabilities to enhance domain explainability for AI systems
- CausalKG fills the gap in domain-adaptable causal modeling and provides a comprehensive approach to defining context, causal information, and effects within a higher representation framework

## Summary

Humans rely on causality and hypothetical retrospection in their daily decision-making, planning, and understanding of life events. This involves developing a causal model of the world to make inferences and consider counterfactual scenarios. These unseen possibilities are known as counterfactuals. To represent concepts of time, space, and facts in AI algorithms, knowledge graphs (KG) are utilized. KGs are graphical data models that capture semantic relationships between entities like events, objects, or concepts. However, existing KGs often struggle with supporting counterfactual reasoning due to their reliance on linguistic patterns for extracting causal relationships from texts. To address this limitation and enhance explainability in AI systems, a richer representation of causality is needed. Introducing the Causal Knowledge Graph (CausalKG) framework: leveraging recent advancements in causality and KG towards explainability. By utilizing a hyper-relational graph representation within the KG, CausalKG aims to effectively capture complex causal relations. Its interventional and counterfactual reasoning capabilities have the potential to greatly enhance domain explainability for AI systems. Not only does this framework fill the gap in domain-adaptable causal modeling but it also provides a comprehensive approach to defining context, causal information, and causal effects within a higher representation framework. Authored by Utkarshani Jaimini and Amit Sheth from the Artificial Intelligence Institute at the University of South Carolina, this research article delves into how CausalKG offers a promising solution for improving understanding and interpretability within AI systems through its advanced approach to representing causality.