Bayesian Inference of Linear Temporal Logic Specifications for Contrastive Explanations

Abstract

Temporal logics are useful for providing concise descriptions of system behavior, and have been successfully used as a language for goal definitions in task planning. Prior works on inferring temporal logic specifications have focused on "summarizing" the input dataset - i.e., finding specifications that are satisfied by all plan traces belonging to the given set. In this paper, we examine the problem of inferring specifications that describe temporal differences between two sets of plan traces. We formalize the concept of providing such contrastive explanations, then present BayesLTL - a Bayesian probabilistic model for inferring contrastive explanations as linear temporal logic (LTL) specifications. We demonstrate the robustness and scalability of our model for inferring accurate specifications from noisy data and across various benchmark planning domains.

A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence

“…Most explanation generators meant for planning-based tasks are model-specific due to the problem-and approachspecific restrictions preventing them from being used for other AI challenges. For instance, Kim et al employ a Bayesian probabilistic model for generating contrastive explanations [94]. Thus, the framework operates on a pair of plan traces defined in terms of linear temporal logic templates.…”

“…A cause of this kind is responsible for the connection between the types C and M in a system S. A contrastive explanation thus explains why a system S is claimed to be ''wired'' in such a way that an internal state of type C regularly causes a movement of type M . Similarly, Kim et al regard contrastive explanation as a constraint for a system to be satisfied by a specific set of plan traces [94]. Finally, Boulter illustrates the use of contrastive explanations to distinguish between actual and non-actual biological forms [70].…”

“…• Contfactuals for planning. Contfactual explanation generation appears highly relevant to sequential tasks in robotics such as automatic planning [94], [157], [161]. Moreover, some of the robotics-related frameworks found in reinforcement learning settings provide explanations for policies that a robot selects at a given time step [132], [156].…”

The Emerging Landscape of Explainable Automated Planning & Decision Making

“…This was investigated in [33], where both queries and answers were expressed in terms of userspecified features. [43] looked at cases where the user is not just interested in learning details of the model underlying the current decisions but rather how it differs from possible alternatives, by using LTL formulas that are true in a target set of plan traces but are not satisfied by a specified alternate set.…”

Generating justifications for norm-related agent decisions

“…Our approach operationalizes Langley (2019)'s definition of justified agency for an intelligent system as "follow [ing] society's norms and explain [ing] its activities in those terms" by providing natural language explanations which appeal to temporal logic rules which may represent moral or social norms. Further relevant recent papers in explainable planning include Vasileiou et al (2019), who formulate a logic-based approach; Krarup et al (2019), who like us employ contrastive explanations; and Kim et al (2019), who construct temporal logic specifications which demonstrate the differences between plans. While the latter work has in common with ours a focus on explainable planning and temporal logic, we are interested in justifying the agent's choice of plan, whereas they seek to succinctly describe the difference between two plans.…”