InTDS ArchivebyFelix Hofstätter·Oct 26, 2022How Assistance Games make AI saferAnd how they don’tA response icon1A response icon1
InTDS ArchivebyFelix Hofstätter·Jun 24, 2022Spotting Unfair or Unsafe AI using Graphical CriteriaHow to use causal influence diagrams to recognize the hidden incentives that shape an AI agent’s behavior
InTDS ArchivebyFelix Hofstätter·Apr 4, 2022How to stop your AI agents from hacking their reward functionUsing causal influence diagrams and current-RF optimisationA response icon1A response icon1
InTDS ArchivebyFelix Hofstätter·Jan 15, 2022Counterfactuals for Reinforcement Learning II: Improving Reward LearningSafer reward function learning using counterfactuals
InTDS ArchivebyFelix Hofstätter·Dec 30, 2021Counterfactuals for Reinforcement Learning I: “What if… ?”Introduction to the POMDP framework and counterfactualsA response icon1A response icon1
InTDS ArchivebyFelix Hofstätter·Nov 16, 2021How learning reward functions can go wrongAn AI-safety minded perspective on the risks of Reinforcement Learning agents learning their reward functionsA response icon2A response icon2
InTDS ArchivebyFelix Hofstätter·Nov 16, 2021Adapting Soft Actor Critic for Discrete Action SpacesHow to apply the popular algorithm to new problems by changing only two equationsA response icon3A response icon3