Reinforcement Learning and Artificial Intelligence (RLAI) RLAI Hypotheses

edited by Rich Sutton

The ambition of this web page is to list pages each of which describes and discusses an interesting scientific hypothesis related to reinforcement learning or artificial intelligence.  The hypotheses should be as general as possible and cleave the space of ideas in interesting ways.

Ideally, the hypotheses would be both simple and powerful, making them excellent null hypotheses -- hypotheses that should be ruled out, or accepted, before anything more complicated is considered.

• Agent hypothesis
• Reward hypothesis (Reinforcement learning hypothesis)
• Value-function hypothesis
• Empirical knowledge hypothesis
• Experiential artificial intelligence hypothesis
  
• Predictive representations hypothesis
• Generalization hypothesis
• Bootstrapping hypothesis
• Fourteen declarative principles (for an AI architecture)