This is a repository that is here to explore how dreams might be necessary for the understanding of consciousness and generation of intelligent agents.
The main approach of this repository is to make a model building structure agent. The agent has the job of creating the next state that it expects to see or sense. This mainly is a problem of the agent generating an internal model of the world, where the next state is the maximum likelihood estimate given that previous state and prior experience. A dream in this sense would be to take an arbitrary initial state and follow the expected outcome of experience.
A conscious awareness would be for an agent to actively predict whatever sensory data is there for the next state in time. Now here is the controversial bit- the current conscioous experince and state would actually be a dream of the next possible second, and the state would be cleared away whenever the system increments the subsequent state. The agent would actually react to the predicted state and not the direct sensory inputs.
The basic model architecture is shown in the picture below:
