DR. JEFF BECK - THE BAYESIAN BRAIN
Machine Learning Street Talk (MLST)
Avoiding brittleness and dangers of reward function specification
Avoiding brittleness and instability in AI systems by prioritizing homeostatic equilibrium over monolithic reward function. Reward function specification can be dangerous, as seen in the example of Skynet and world hunger. Achieving equilibrium in the environment is a safer approach, as advocated by Rich Sutton's paper 'reward is enough'. Reward can be likened to a loaded gun and can lead to similar issues as utilitarians face with reward rarefaction.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
