Understanding and Controlling a Maze-solving Policy Network
Abstract: To understand the goals and goal representations of AI systems, we carefully study a pretrained reinforcement learning policy that solves mazes by navigating to a range of target squares, similarly to how model organisms are studied in biology. We find this network pursues multiple context-dependent goals, and we further identify circuits within the network that correspond to one of these goals. In particular, we identified eleven channels that track the location of the goal. By modifying these channels, either with hand-designed interventions or by combining forward passes, we can partially control the policy. We show that this network contains redundant, distributed, and retargetable goal representations, shedding light on the nature of goal-direction in trained policy networks.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Sebastian_Tschiatschek1
Submission Number: 3119
Loading