MEandE-LP 2023

The 3rd edition of Workshop Dates will be co-located with ICLP2023, the 38th International Conference on Logic Programming (ICLP 2023), that will take place on July 9–15, 2023, Imperial College London, UK.

Invited Speakers

TBA

Description, Aim and Scope

Machine Ethics and Explainability are two recent topics that have been garnering significant attention and concern in recent years. This global concern has manifested itself in numerous initiatives at various levels. An intrinsic relationship exists between these two topics. It is insufficient for an autonomous agent to behave ethically; it must also be able to explain its behavior, necessitating both an ethical component and an explanation component. Moreover, explainable behavior is clearly unacceptable if it is not ethical (i.e., it does not adhere to societal ethical norms).

In many application domains, particularly those involving human lives and necessitating ethical decisions, users must comprehend the system recommendations in order to explain the reasoning behind their decisions to others. One of the most important ultimate goals of Explainable AI systems is achieving an efficient mapping between explainability and causality. Explainability refers to the system's ability to justify its actions in natural language to the average user. In other words, the system's capacity to articulate the reasons underlying its decisions is central to explainability.

However, when dealing with high-risk decision-making systems (ethical decisions), is it sufficient to merely explain the system's decisions to human users? Should we extend beyond the boundaries of the predictive model to observe cause and effect within the system?

A vast body of research on explainability attempts to clarify the output of some black-box models using various approaches. Some approaches endeavor to generate logical rules as explanations. Nonetheless, it is worth noting that most methods for generating post-hoc explanations are themselves based on statistical tools, which are subject to uncertainty or errors. Many post-hoc explainability techniques try to approximate deep-learning black-box models with simpler, interpretable models that can be inspected to explain the black-box models. However, these approximate models are not provably loyal to the original model, as there are always trade-offs between explainability and fidelity.

Conversely, a substantial number of researchers have employed inherently interpretable approaches to develop and implement their ethical autonomous agents. Many of these approaches are based on logic programming, ranging from deontic logics to non-monotonic logics and other formalisms.

Logic Programming (LP) holds significant potential in these two burgeoning research areas, as logic rules are easily understood by humans and promote causality, which is vital for ethical decision-making.

Despite the considerable interest machine ethics has received over the past decade, primarily from ethicists and AIexperts, the question "Are artificial moral agents possible?" remains unanswered. Several attempts have been made to implement ethical decision-making into intelligent autonomous agents using various approaches. However, no fully descriptive and universally acceptable model of moral judgment and decision-making exists to date. None of the developed solutions appear to be entirely convincing in providing trustworthy moral behavior. The same applies to explainability; although there is widespread concern about autonomous agents' explainability, current approaches do not seem satisfactory. Many questions remain unanswered in these two fascinating, rapidly expanding fields.

This workshop aims to convene researchers working on all aspects of machine ethics and explainability, including theoretical work, system implementations, and applications. By co-locating this workshop with ICLP, we also intend to encourage collaboration among researchers from different LP areas. This workshop offers a forum for facilitating discussions on these topics and fostering a productive exchange of ideas.

Topics of interest include, but are not limited to:

- New LP-based approaches to programming machine ethics;

- New LP-based approaches to explainability of black-box models;

- Evaluation and comparison of existing LP-based approaches;

- Approaches to verification of ethical behavior;

- LP applications in machine ethics;

- Integrating LP with methods for machine ethics;

- Integrating LP with methods for explainability;

- Neuro-symbolic AI for ethics/explainability.