Zenith Grant Awardee

Gerard Milburn

The University of Queensland

Co-Investigators

Sally Shrapnel, The University of Queensland

Project Title

Are Quantum agents possible?

Project Summary

Agents are special physical systems that act upon, and are acted upon by, the world external to them. All agents are subject to the laws of physics. Humans and other animals appear to be the only examples of agents but technology is delivering artificial agents. We propose specific designs for simple quantum agents that exploit the laws of quantum physics to out-perform current artificial agents. Agents are described in two complementary ways: (i) an inside view, which tracks the memory records and agent actions conditioned upon these records and, (ii) the outside view in which memory records are averaged over. Agents must have memories to enable learning, creativity and contingent action. The possibility of an internal view makes a physical system an agent and enables agents to identify each other as such. By explicitly modelling simple quantum agents we will reconcile the time asymmetry of agents and the time symmetry of the laws of physics; determinism poses no challenge to agency and free will. We show that agency, quantum or other wise, is only possible in a world subject to the laws of thermodynamics.

Technical Abstract

All agents are open physical systems as they act upon, and are acted upon by, the world external to them. Using models drawn from quantum optics, we will show that quantum agents are possible, despite noise and decoherence, and can outperform classical agents at analogous tasks. Agents are described in two complementary ways: (i) an inside view which tracks the memory records and agent actions conditioned upon these records and, (ii) the outside view in which memory records are averaged over. By contrasting the internal and external view we hope to reconcile some of the puzzles of agency, for example, the apparent conflict between time asymmetry and determinism. We propose specific physical models in which the complementary internal and external views can be realised. We show that possibility of an internal view provides a way to determine if physical system is an agent. We will use the quantum theory of open systems (master equations and quantum stochastic differential equations) together with quantum information theory and classical machine learning tools to model quantum agents. The project has four components: noise and quantum agents, agent-agent interaction, quantum agents for causal discovery and agents with quantum memories.