The Alberta plan for AI research

Abstract

Herein we describe our approach to artificial intelligence research, which we call the Alberta Plan. Following the Alberta Plan, we seek to understand and create long-lived computational agents that interact with a vastly more complex world and come to predict and control their sensory input signals. The agents are complex only because they interact with a complex world over a long period of time; their initial design is as simple, general, and scalable as possible. To control their input signals, the agents must take action. To adapt to change and the complexity of the world, they must continually learn. To adapt rapidly, they must plan with a learned model of the world. The purpose of this document is twofold. One is to describe our vision for AI research and its underlying intellectual commitments and priorities. The second is to describe the path along which this vision may unfold and the research problems and projects that we will pursue.