The key concept that we want to convey in Lesson 3 is:
Cognitive agents must discover, learn, and exploit regularities of interaction.
Regularities of interaction (in short, regularities) are patterns of interaction that occur consistently. Regularities depend on the coupling between the agent and the environment. That is, they depend both on the structure of the environment, and on the possibilities of interaction that the agent has at its disposal.
At least since Immanuel Kant, philosophers have widely agreed on the fact that cognitive systems can never know "the world as such", but only the world as it appears to them through sensorimotor interactions. For example, in some situations, if you spread your arm repeatedly, and if you consistently experience the same sensorimotor pattern, you may infer that there is something constant out there that always makes this same sensorimotor pattern possible. Note that regularities can be experienced through arbitrarily complex instruments, which may range from a stick in your hand to complex experimental settings such as those used by physicists to interact with something out there known as the Higgs boson.
These philosophical ideas translate into AI when we acknowledge the fact that knowledge is constructed from regularities of interactions rather than recorded from input data. Designing a system that would construct complete knowledge of the world out there and exploit this model for the better is a part of AI's long-term objective.
To take this problem gradually, Lesson 3 begins with implementing an agent that can detect simple sequential regularities and exploit them to satisfy its rudimentary motivational system.
See public discussions about this page or start a new discussion by clicking on the Google+ Share button. Please type the #IDEALMOOC031 hashtag in your post: