A Physical Quantum Agent

TL;DR

This paper introduces a simple optical quantum agent that uses single photon pulses to probe its environment, outperforming classical agents and linking learning with thermodynamic work in a toy model.

Contribution

It presents a novel optical quantum agent model demonstrating advantages over classical agents and explores the thermodynamic implications of learning in such systems.

Findings

Quantum agent outperforms classical agent in environment probing.

Improved environmental estimation correlates with increased work done on the sensor.

Provides a toy model connecting machine learning, optics, and thermodynamics.

Abstract

The concept of an embodied intelligent agent is a key concept in modern artificial intelligence and robotics. Physically, an agent is an open system embedded in an environment that it interacts with through sensors and actuators. It contains a learning algorithm that correlates the sensor and actuator results by learning features about its environment. In this article we present a simple optical agent that uses light to probe and learn components of its environment. In our scenario, the quantum agent outperforms a classical agent: The quantum agent probes the world using single photon pulses, where its classical counterpart uses a weak coherent state with an average photon number equal to one. We analyze the thermodynamic behavior of both agents, showing that improving the agent's estimate of the world corresponds to an increase in average work done on the sensor by the actuator pulse. Thus, our model provides a useful toy model for studying the interface between machine learning, optics, and statistical thermodynamics.