Reinforcement learning involves decision making in dynamic and uncertain environments and constitutes an important element of artificial intelligence (AI). In this work, we experimentally demonstrate that the ultrafast chaotic oscillatory dynamics of lasers efficiently solve the multi-armed bandit problem (MAB), which requires decision making concerning a class of difficult trade-offs called the exploration–exploitation dilemma. To solve the MAB, a certain degree of randomness is required for exploration purposes. However, pseudorandom numbers generated using conventional electronic circuitry encounter severe limitations in terms of their data rate and the quality of randomness due to their algorithmic foundations. We generate laser chaos signals using a semiconductor laser sampled at a maximum rate of 100 GSample/s, and combine it with a simple decision-making principle called tug of war with a variable threshold, to ensure ultrafast, adaptive, and accurate decision making at a maximum adaptation speed of 1 GHz. We found that decision-making performance was maximized with an optimal sampling interval, and we highlight the exact coincidence between the negative autocorrelation inherent in laser chaos and decision-making performance. This study paves the way for a new realm of ultrafast photonics in the age of AI, where the ultrahigh bandwidth of light wave can provide new value.

Decision making is critical in our daily lives and for society in general and is finding evermore practical applications in information and communication technologies. Herein, we demonstrate experimentally that single photons can be used to make decisions in uncertain, dynamically changing environments. Using a nitrogen-vacancy in a nanodiamond as a single-photon source, we demonstrate the decision-making capability by solving the multi-armed bandit problem. This capability is directly and immediately associated with single-photon detection in the proposed architecture, leading to adequate and adaptive autonomous decision making. This study makes it possible to create systems that benefit from the quantum nature of light to perform practical and vital intelligent functions.

Decision-making is one of the most important intellectual abilities of not only humans but also other biological organisms, helping their survival. This ability, however, may not be limited to biological systems and may be exhibited by physical systems. Here we demonstrate that any physical object, as long as its volume is conserved when coupled with suitable operations, provides a sophisticated decision-making capability. We consider the multi-armed bandit problem (MBP), the problem of finding, as accurately and quickly as possible, the most profitable option from a set of options that gives stochastic rewards. Efficient MBP solvers are useful for many practical applications, because MBP abstracts a variety of decision-making problems in real-world situations in which an efficient trial-and-error is required. These decisions are made as dictated by a physical object, which is moved in a manner similar to the fluctuations of a rigid body in a tug-of-war (TOW) game. This method, called 'TOW dynamics', exhibits higher efficiency than conventional reinforcement learning algorithms. We show analytical calculations that validate statistical reasons for TOW dynamics to produce the high performance despite its simplicity. These results imply that various physical systems in which some conservation law holds can be used to implement an efficient 'decision-making object'. The proposed scheme will provide a new perspective to open up a physics-based analog computing paradigm and to understanding the biological information-processing principles that exploit their underlying physics.

No abstract