Agency Is Frame-Dependent
David Abel, André Barreto, Michael Bowling, Will Dabney, Shi Dong, Steven Hansen, Anna Harutyunyan, Khimya Khetarpal, Clare Lyle, Razvan Pascanu, Georgios Piliouras, Doina Precup, Jonathan Richens, Mark Rowland, Tom Schaul, Satinder Singh
2025-02-10
Summary
This paper talks about how agency, which is the ability of a system to make decisions to reach a goal, depends on the perspective or 'frame' from which it is measured. It explores this idea using concepts from philosophy and reinforcement learning.
What's the problem?
Figuring out whether something has agency, like deciding if a rock, a thermostat, or a robot can make purposeful decisions, has always been a tough question. There’s no clear way to measure agency because it changes depending on how you look at it or what rules you use to judge it.
What's the solution?
The researchers argue that agency is 'frame-dependent,' meaning it can only be understood relative to the perspective or reference frame used to measure it. They use philosophical ideas and reinforcement learning concepts to show that all key features of agency depend on the frame of reference. This means there’s no universal way to measure agency—it always depends on context.
Why it matters?
This matters because understanding agency is crucial for fields like artificial intelligence, biology, and cognitive science. By showing that agency depends on perspective, this paper helps scientists and engineers develop better ways to study and design systems that can make decisions. It also opens up new ways to think about how machines and living things interact with their environments.
Abstract
Agency is a system's capacity to steer outcomes toward a goal, and is a central topic of study across biology, philosophy, cognitive science, and artificial intelligence. Determining if a system exhibits agency is a notoriously difficult question: Dennett (1989), for instance, highlights the puzzle of determining which principles can decide whether a rock, a thermostat, or a robot each possess agency. We here address this puzzle from the viewpoint of reinforcement learning by arguing that agency is fundamentally frame-dependent: Any measurement of a system's agency must be made relative to a reference frame. We support this claim by presenting a philosophical argument that each of the essential properties of agency proposed by Barandiaran et al. (2009) and Moreno (2018) are themselves frame-dependent. We conclude that any basic science of agency requires frame-dependence, and discuss the implications of this claim for reinforcement learning.