Logic and AI

This workshop is part of a research project on the topic of “Logic and AI”. It will bring together international experts to explore the promising interaction of logic and modern artificial intelligence (AI). While AI struggles with explainability, interpretability, and verifiability, logic excels at this. So can logic help AI? And if so, how?

Start date: 16 July 2024
End date: 17 July 2024
Time: 09:00
Location: Oude Turfmarkt 145-147

The history of AI is often conceptualized as a pendulum: swinging back and forth between symbolic approaches based on logic and sub-symbolic approaches based on neural networks. While logic has been called the ‘Calculus of Computer Science’, it plays almost no role in modern AI after the deep learning revolution in the 2010s. However, today’s problems of AI call again for the advantages of symbolic approaches: A pressing open problem is to achieve explanation, interpretation, verification, and theory of AI—and logic historically excelled at these desiderata.

Consequentially, we see a quickly rising interest in the interaction of logic and modern AI. In this workshop, we explore how exactly a fruitful interaction can look like. The talks of distinguished experts showcase the latest work in this area, and the discussion rounds will identify promising future directions. In particular, we will focus on three topics: (1) expressive and computational power of machine learning, (2) different forms of neuro-symbolic integration, and (3) the intersection of causality, logic, and machine learning.

This workshop is organized by IAS Fellows, Levin Hornischer (LMU Munich/MCMP) and Thomas F. Icard (Stanford University), in collaboration with Johan van Benthem (University of Amsterdam/ILLC, Stanford University, Tsinghua University), Balder ten Cate (University of Amsterdam/ILLC), Frank van Harmelen (Vrije Universiteit Amsterdam), and Sara Magliacane (University of Amsterdam/AMLab) during the summer of 2024 at the Institute for Advanced Studies (IAS) at the University of Amsterdam.

Program

Tuesday, July 16 [* = streamed online]

9:00	Arrival
9:15	Welcome*
	Neuro-symbolic integration 1 — Host: Sara Magliacane
9:30	Giuseppe Marra (KU Leuven): From Statistical Relational to Neuro-Symbolic AI*
10:30	Levin Hornischer (LMU Munich): Semantics for Non-symbolic Computation: Including Neural Networks and Analog Computers*
11:30	Discussion
12:30	Lunch (provided on site)
	Expressive power of ML — Host: Balder ten Cate
14:00	Lena Strobl (Umeå University): Expressivity of Transformers: What Formal Languages Can They Represent?*
15:00	Martin Grohe (RWTH Aachen): The Logic of Graph Neural Networks*
16:00	Discussion
17:00	End

Wednesday, July 17 [* = streamed online]

9:00	Arrival
9:15	Welcome*
	Neuro-symbolic integration 2 — Host: Levin Hornischer
9:30	Sebastijan Dumančić (TU Delft): Machine learning models that provably satisfy constraints*
10:30	Herbert Jaeger (University of Groningen): What a mathematical foundation for unconventional computing should deliver and how it might look like*
11:30	Discussion
12:30	Lunch (provided on site)
	Causality, logic, and ML — Host: Johan van Benthem
14:00	Atticus Geiger (Pr(Ai)²R): Causal Abstraction as a Theoretical Foundation for Mechanistic Interpretability*
15:00	Thomas Icard (Stanford University): Causal Inference from a Logical Point of View*
16:00	Discussion
17:00	End of the workshop

Logic and AI

Program

Cookie Consent