I am an Assistant Professor at the Department of Security Studies, Charles University, Prague.
I am interested in safe machine learning and the means to achieve it. Most of my works deal with inductive inference in various learning frameworks. I tend to believe that the formal study of inductive inference is important for the safety of artificial intelligence.
I am writing a book on AI alignment and (social) preference learning.
I collaborate with Vit Stritecky, my colleague from Charles, and John Symons from The University of Kansas. I am also a fellow at the Center for Cyber-Social Dynamics at KU.
My ORCID is 0000-0003-4199-645X.
You can contact me at petr [dot] spelda [at] fsv [dot] cuni [dot] cz
Learnability of State Spaces of Physical Systems is Undecidable, Journal of Computational Science (2024). DOI:10.1016/j.jocs.2024.102452, with Vit Stritecky.
Why and How to Construct an Epistemic Justification of Machine Learning?, Synthese (2024). DOI:10.1007/s11229-024-04702-z, with Vit Stritecky.
On the Need for Multiple, Independent Fact-Checking and Scoring Facilities: A Reply to Gerhard Schurz, Social Epistemology Review and Reply Collective (2024). URL, with Vit Stritecky and John Symons.
No-Regret Learning Supports Voters’ Competence, Social Epistemology (2023). DOI:10.1080/02691728.2023.2252763, with Vit Stritecky and John Symons.
Gerhard Schurz's response.
Expanding Observability via Human-Machine Cooperation, Axiomathes/Global Philosophy (2022). DOI:10.1007/s10516-022-09636-0, with Vit Stritecky.
Human Induction in Machine Learning: A Survey of the Nexus, ACM Computing Surveys (2021). DOI:10.1145/3444691, with Vit Stritecky.
What Can Artificial Intelligence Do for Scientific Realism?, Axiomathes/Global Philosophy (2020). DOI:10.1007/s10516-020-09480-0, with Vit Stritecky.
The Future of Human-Artificial Intelligence Nexus and its Environmental Costs, Futures (2020). DOI:10.1016/j.futures.2020.102531, with Vit Stritecky.
Machine learning, inductive reasoning, and reliability of generalisations, AI & SOCIETY (2020). DOI:10.1007/s00146-018-0860-6.
AI alignment safety bounds
Transformers’ in-context learning
No-regret machine learning lifecycle
Security practices in AI
Computational complexity of inductive inference
An Analysis of the Electronic Jihad’s Activity in the Social Media Environment, Czech Journal of International Relations (2017), with Vit Stritecky.
Establishing the Complexity of the Islamic State’s Visual Propaganda, Central European Journal of International and Security Studies (2017), with Vit Stritecky.
last update: 2024-09-27
𝛀