Abstract
We study formal languages which are capable of fully expressing quantitative probabilistic reasoning and do-calculus reasoning for causal effects, from a computational complexity perspective. We focus on satisfiability problems whose instance formulas allow expressing many tasks in probabilistic and causal inference. The main contribution of this work is establishing the exact computational complexity of these satisfiability problems. We introduce a new natural complexity class, named succ∃R, which can be viewed as a succinct variant of the well-studied class ∃R, and show that these problems are complete for succ∃R. Our results imply even stronger limitations on the use of algorithmic methods for reasoning about probabilities and causality than previous state-of-the-art results that rely only on the NP- or ∃R-completeness of the satisfiability problems for some restricted languages.
| Original language | English |
|---|---|
| Title of host publication | IJCAI International Joint Conference on Artificial Intelligence 2023 |
| Number of pages | 9 |
| Publication date | 2023 |
| Pages | 5730-5738 |
| ISBN (Print) | 9781956792034 |
| DOIs | |
| Publication status | Published - 2023 |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 9 Industry, Innovation, and Infrastructure
Research Areas and Centers
- Centers: Center for Artificial Intelligence Luebeck (ZKIL)
DFG Research Classification Scheme
- 4.43-01 Theoretical Computer Science
KDSF Research Field Classification Scheme
- 080 - Information technology - general
Fingerprint
Dive into the research topics of 'The Hardness of Reasoning about Probabilities and Causality'. Together they form a unique fingerprint.Prizes
-
Wissenschaftspreis der MINT-Sektionen 2025
van der Zander, B. (Award Recipient) & Ssebyatika, G. (Award Recipient), 09.12.2025
Prize: Awards of the University of Luebeck
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver