David C. Hyland

Alessandro Abate, Yousif Almulla, James Fox et al.

Training reinforcement learning (RL) agents using scalar reward signals is often infeasible when an environment has sparse and non-Markovian rewards. Moreover, handcrafting these reward functions before training is prone to misspecification, especially when the environment's dynamics are only partially known. This paper proposes a novel pipeline for learning non-Markovian task specifications as succinct finite-state `task automata' from episodes of agent experience within unknown environments. We leverage two key algorithmic insights. First, we learn a product MDP, a model composed of the specification's automaton and the environment's MDP (both initially unknown), by treating the product MDP as a partially observable MDP and using the well-known Baum-Welch algorithm for learning hidden Markov models. Second, we propose a novel method for distilling the task automaton (assumed to be a deterministic finite automaton) from the learnt product MDP. Our learnt task automaton enables the decomposition of a task into its constituent sub-tasks, which improves the rate at which an RL agent can later synthesise an optimal policy. It also provides an interpretable encoding of high-level environmental and task features, so a human can readily verify that the agent has learnt coherent tasks with no misspecifications. In addition, we take steps towards ensuring that the learnt automaton is environment-agnostic, making it well-suited for use in transfer learning. Finally, we provide experimental results compared with two baselines to illustrate our algorithm's performance in different environments and tasks.

Lancelot Da Costa, Tomáš Gavenčiak, David Hyland et al.

This paper offers a roadmap for the development of scalable aligned artificial intelligence (AI) from first principle descriptions of natural intelligence. In brief, a possible path toward scalable aligned AI rests upon enabling artificial agents to learn a good model of the world that includes a good model of our preferences. For this, the main objective is creating agents that learn to represent the world and other agents' world models; a problem that falls under structure learning (a.k.a. causal representation learning or model discovery). We expose the structure learning and alignment problems with this goal in mind, as well as principles to guide us forward, synthesizing various ideas across mathematics, statistics, and cognitive science. 1) We discuss the essential role of core knowledge, information geometry and model reduction in structure learning, and suggest core structural modules to learn a wide range of naturalistic worlds. 2) We outline a way toward aligned agents through structure learning and theory of mind. As an illustrative example, we mathematically sketch Asimov's Laws of Robotics, which prescribe agents to act cautiously to minimize the ill-being of other agents. We supplement this example by proposing refined approaches to alignment. These observations may guide the development of artificial intelligence in helping to scale existing -- or design new -- aligned structure learning systems.

David Hyland, Mahault Albarracin

The human mind is capable of extraordinary achievements, yet it often appears to work against itself. It actively defends its cherished beliefs even in the face of contradictory evidence, conveniently interprets information to conform to desired narratives, and selectively searches for or avoids information to suit its various purposes. Despite these behaviours deviating from common normative standards for belief updating, we argue that such 'biases' are not inherently cognitive flaws, but rather an adaptive response to the significant pragmatic and cognitive costs associated with revising one's beliefs. This paper introduces a formal framework that aims to model the influence of these costs on our belief updating mechanisms. We treat belief updating as a motivated variational decision, where agents weigh the perceived 'utility' of a belief against the informational cost required to adopt a new belief state, quantified by the Kullback-Leibler divergence from the prior to the variational posterior. We perform computational experiments to demonstrate that simple instantiations of this resource-rational model can be used to qualitatively emulate commonplace human behaviours, including confirmation bias and attitude polarisation. In doing so, we suggest that this framework makes steps toward a more holistic account of the motivated Bayesian mechanics of belief change and provides practical insights for predicting, compensating for, and correcting deviations from desired belief updating processes.