Zenith Grant Awardee

Eric Cavalcanti

Griffith University

Project Title

Causal reasoning in quantum agents

Project Summary

One of the abilities that distinguishes humans from other animals is our capacity to represent the world in terms of cause-and-effect relationships. It has been argued that human-level artificial intelligence will only be achieved if we endow machines with similar abilities to reason about the effects of their interventions in the world, and to imagine the potential outcomes of counterfactual courses of action. In current research in artificial intelligence, the mathematical machinery for this kind of representation is provided by the classical theory of causal models. Unfortunately, this theory is in direct conflict with quantum mechanics, as demonstrated through a number of theorems and experiments, such as those involving the phenomena of quantum nonlocality and contextuality. In recent years, a new research program is attempting to resolve this conflict by developing a quantum generalisation of the theory of causal models. In this project, I will aim to investigate how quantum causality differs from its classical counterpart in the representation and analysis of statements involving interventions and counterfactuals. This could form a fundamental part of the toolbox of automated reasoning in quantum artificial agents, and further our understanding of the nature of intelligence and agency in the physical world.

Technical Abstract

It has been argued that human-level AI cannot be achieved by analysis of observational data alone, however large and complex, but requires agents endowed with the ability to express and evaluate statements about interventions and counterfactuals. Judea Pearl identifies these actions of “seeing”, “doing” and “imagining” with a 3-Level Causal Hierarchy of models. This classical framework, however, fails to describe quantum correlations, such as those violating Bell or Kochen-Specker inequalities. The program of quantum causal models aims to resolve that incompatibility by extending the framework of causal models to the quantum case. These frameworks however, remain largely at the Level 1 of the causal hierarchy, lacking a fully developed machinery for the calculus of interventions and counterfactuals. In this project I will aim to systematically address Levels 2 and 3 of Pearl’s hierarchy, to determine the extent of their incompatibility with quantum theory, and search for a theoretical framework for the representation and analysis of interventions and counterfactuals in quantum mechanics. Following Pearl’s thesis, this could form a fundamental part of the toolbox of automated reasoning in quantum artificial agents and help further our understanding of the nature of intelligence and agency in the physical world.