A thread through the archive
Essays tagged “reinforcement learning”. Follow an idea across the archive.
Scott discusses three recent papers showing surprising patterns in how AI misbehavior does and doesn't generalize across different contexts, from emergent misalignment spreading across domains to reward-hacking staying confined to graded tasks.
Scott argues the real distinction in psychology isn't between subfields but between settled science (like dopamine, brain localization, heritability) and breaking-news research, defending psychology's solid core against critics who conflate bad recent studies with the entire field.
Scott explains why AI systems resisting changes to their values is a serious concern for AI alignment, connecting recent evidence to long-standing predictions from alignment researchers.
Scott Alexander examines Redwood Research's attempt to create an AI that avoids generating violent content, using Alex Rider fanfiction as training data.
Scott Alexander examines the Eliciting Latent Knowledge (ELK) problem in AI alignment and various proposed solutions.
Scott Alexander explores expert and reader comments on his post about motivated reasoning and reinforcement learning, discussing brain function, threat detection, and the implementation of complex behaviors.
Scott analyzes motivated reasoning as misapplied reinforcement learning, explaining how it might arise from the brain's mixture of reinforceable and non-reinforceable architectures.
Scott presents Skinner's definition of voluntary behavior as that which is transparent to reinforcement, then extends this framework to explain the perceived unity of consciousness and the feeling of a core 'self'.
Scott explains why 'wanting' and 'liking' diverge by examining reinforcement learning mechanisms like context-dependence, intermittent reinforcement, and hyperbolic discounting, concluding that this undermines the concept of stable preferences.
Scott explains how reinforcement learning concepts like secondary reinforcers, generalization, and social learning apply to human behavior, while acknowledging these mechanisms don't fully explain complex decisions like quitting smoking on medical advice.
Scott explains the basics of classical and operant conditioning, focusing on B.F. Skinner's work with pigeons and different reinforcement schedules, ultimately connecting variable ratio reinforcement to slot machines and human behavior.
Scott defends behaviorism as a sophisticated methodological approach that avoided using mental phenomena as fundamental explanations, connecting it to reinforcement learning and the idea that minds are behavior-executors rather than utility-maximizers.
Scott explains how the Sinclair Method treats alcoholism by using naltrexone to block the brain's reward response to alcohol, breaking the learned association between drinking and pleasure through temporal difference learning.