| |
Reinforcement Learning and
Artificial
Intelligence (RLAI) |
a simple plan (for achieving AI)
|
goals and affect are by reward and value
immediate decision making is by actor-critic style RL
planning is by repeated projection and sampling
knowledge is experiential
i.e., statements about future sensations, actions, and time steps
abstraction is by compositional, option-style predictions
sought and discovered so as to optimize projection and prediction
function approximation is linear
in non-linear features sought and discovered in ways yet to be fully understood