Zenith Grant Awardee

Jayne Thompson

National University of Singapore, Centre for Quantum Technologies

Co-Investigators

Mile Gu, National University of Singapore, Centre for Quantum Technologies

Project Title

Are quantum agents more energetically efficient at making predictions?

Project Summary

Information processing is one of the most essential parts of being alive, and one of the more costly. Our brains consume 40% of our body's energy budget, yet are clearly indispensable. One critical aspect of survival is the ability to anticipate future events. Yet there are intrinsic energetic and memory costs associated with making a prediction. Predicting the future involves carefully tracking relevant pieces of information about the past, which are key indicators of future circumstances. Surprisingly even the memory optimal predictive devices are forced to store past events that are irrelevant to their future predictions – this information is simply retained in memory and then discarded, resulting in overhead memory costs and extraneous energy expenditure. Yet these results all assumed the memory was classical. Can having access to a quantum memory challenge this status quo? Quantum devices are known to save memory over their optimal classical counterparts. To what extent can they also shave away the energetic cost of making a prediction? What are the signatures and implications of such a quantum enhancement in energetic efficiency? This grant will explore these possibilities, to understand to what extent intelligent agents can leverage quantum mechanics in lieu of a crystal ball.

Technical Abstract

Looking ahead to anticipate the future behaviour of environmental systems is a basic imperative for survival. Yet as the systems we seek to predict become increasingly more complex, forecasting their future behaviour requires tracking vast amounts of information about the past. This incurs both memory and energetic costs that scale with the complexity of the system. Recent work has demonstrated that quantum processing can reduce the memory cost of generating statistically accurate future predictions beyond classical limits. However, in many situations energetic efficiency is a much more stringent bottom line for a process's viability. In this grant, we ask to what extent quantum mechanisms for generating future predictions exceed classical bounds on energetic efficiency. We adopt a novel confluence of tools from computational mechanics, information thermodynamics, and quantum information processing. The former to put limits on the ultimate thermodynamic efficiency of classical mechanisms for prediction, the latter to design more efficient quantum mechanisms. Our results could carry implications across a variety of areas from uncovering regimes where the energy savings strongly incentivize prediction through quantum processing, to identifying whether a quantum agent will perceive extra energetic costs for tracking a process in one temporal direction over the other.