CIRL model might simply not be flexible enough to represent manipulative actions. The state s is known to both agents and is supposed to represent the world, but if θ isn’t known to R then the internal state of H is not contained in s. Then there needs to be some other sH invisible to R, and an extended transition function, which is able to affect this state.
CIRL model might simply not be flexible enough to represent manipulative actions. The state s is known to both agents and is supposed to represent the world, but if θ isn’t known to R then the internal state of H is not contained in s. Then there needs to be some other sH invisible to R, and an extended transition function, which is able to affect this state.