Plasticity as the Mirror of Empowerment

📅 2025-05-15
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
This paper addresses the foundational question of how agents are shaped by observable information, introducing “plasticity” as a universal measure of an agent’s susceptibility to environmental influence. Method: We formally define plasticity for the first time, propose generalized directed information, and rigorously extend Massey’s directed information. Contribution/Results: We prove that plasticity and empowerment constitute a dual relationship: an agent’s plasticity equals the environment’s empowerment—mathematically equivalent and mirror-symmetric. This duality reveals an intrinsic tension between learning capacity and autonomy, fundamentally reconceptualizing agency within information theory. The result yields a computationally tractable theoretical principle for balancing perceptual sensitivity against behavioral autonomy, with foundational implications for artificial intelligence and cognitive science.

Technology Category

Application Category

📝 Abstract
Agents are minimally entities that are influenced by their past observations and act to influence future observations. This latter capacity is captured by empowerment, which has served as a vital framing concept across artificial intelligence and cognitive science. This former capacity, however, is equally foundational: In what ways, and to what extent, can an agent be influenced by what it observes? In this paper, we ground this concept in a universal agent-centric measure that we refer to as plasticity, and reveal a fundamental connection to empowerment. Following a set of desiderata on a suitable definition, we define plasticity using a new information-theoretic quantity we call the generalized directed information. We show that this new quantity strictly generalizes the directed information introduced by Massey (1990) while preserving all of its desirable properties. Our first finding is that plasticity is the mirror of empowerment: The agent's plasticity is identical to the empowerment of the environment, and vice versa. Our second finding establishes a tension between the plasticity and empowerment of an agent, suggesting that agent design needs to be mindful of both characteristics. We explore the implications of these findings, and suggest that plasticity, empowerment, and their relationship are essential to understanding agency.
Problem

Research questions and friction points this paper is trying to address.

Defining agent plasticity as a universal measure of influence from observations
Establishing plasticity as the mirror concept to empowerment in agents
Exploring the tension between plasticity and empowerment in agent design
Innovation

Methods, ideas, or system contributions that make the work stand out.

Introduces plasticity as agent-centric universal measure
Defines plasticity using generalized directed information
Reveals plasticity-empowerment mirror relationship and tension
🔎 Similar Papers
No similar papers found.
David Abel
David Abel
DeepMind / University of Edinburgh
Reinforcement LearningArtificial IntelligencePhilosophy
Michael Bowling
Michael Bowling
Amii, University of Alberta
Artificial IntelligenceMachine LearningGame TheoryReinforcement LearningComputer Games
A
Andr'e Barreto
Google DeepMind
Will Dabney
Will Dabney
DeepMind
Reinforcement LearningMachine LearningArtificial Intelligence
S
Shi Dong
Google DeepMind
Steven Hansen
Steven Hansen
Senior Research Scientist, DeepMind
Deep Reinforcement LearningMemory in Neural NetworksIntrinsic Motivation
A
A. Harutyunyan
Google DeepMind
Khimya Khetarpal
Khimya Khetarpal
Google Deepmind, Mila
AgentsArtificial IntelligenceReinforcement LearningRoboticsTemporal Abstraction
Clare Lyle
Clare Lyle
Google DeepMind
Machine Learning
Razvan Pascanu
Razvan Pascanu
Google DeepMind
deep learningreinforcement learningrecurrent neural modelsoptimizationgraph neural networks
Georgios Piliouras
Georgios Piliouras
Google DeepMind, Singapore University of Technology and Design
Algorithmic Game TheoryMachine LearningEconomicsBlockchain
Doina Precup
Doina Precup
DeepMind and McGill University
Artificial Intelligencemachine learningreinforcement learning
Jonathan Richens
Jonathan Richens
DeepMind
AgentsCausalityAI safetyThermodynamicsQuantum Physics
Mark Rowland
Mark Rowland
Research Scientist, Google DeepMind
Reinforcement LearningStatisticsProbability
Tom Schaul
Tom Schaul
Senior Staff Scientist, DeepMind
Reinforcement LearningDeep LearningDeep Reinforcement LearningGame AIExploration
Satinder Singh
Satinder Singh
Google DeepMind / U. of Michigan
Reinforcement LearningComputational Game TheoryArtificial Intelligence