Nandi Schoots, of the 2019 cohort of the UKRI Centre for Doctoral Training in Safe and Trusted Artificial Intelligence, presented a paper on the safety properties of inductive logic programming in February 2021.
Symbolic approaches to AI are often considered safer than neural approaches. Nandi and her co-authors investigated this by analysing how one symbolic approach, inductive logic programming (ILP), fares on specific safety properties, particularly as compared to deep learning systems. They found that ILP could satisfy many of these properties in its home domains and proposed a hybrid system using ILP as a preprocessor to generate specifications for other ML systems.
Research for this paper grew out of the AI safety research program, and the publication is co-authored with Gavin Leech and Joar Skalse.
Nandi presented at the 3rd Workshop on Artificial Intelligence Safety (SafeAI), held virtually on February 8th, 2021. The workshop was co-located with the Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI-21).