Beyond Correlation: Causal AI Decisions Based on World Models

This ACDH lecture by Roland Fleischhacker explores an alternative to AI systems that rely primarily on statistical patterns and correlations. Using deepassist by deepsearch as an example, the lecture introduces an explicit world model based on a highly interconnected knowledge graph that represents logical and causal relationships. It discusses how combining semantic technologies with causal reasoning can make AI decisions more transparent, traceable and robust against typical errors of purely pattern-based systems, such as hallucinations.

Learning Outcomes

After completing this resource, learners will be able to:

• distinguish between correlation-based and causal approaches to AI

• explain how world models and knowledge graphs can support causal reasoning

• discuss how semantic technologies and causal logic can improve the transparency and reliability of AI decisions

  ACDH Lecture