1st Workshop on Machine Ethics and Explainability - The Role of Logic Programming
The 1st edition of MEandE-LP will be co-located with ICLP2021, the 37th International Conference on Logic Programming (ICLP 2021), that will take place from the 20th of September 2021.
We are delight to announce that Luis Moniz Pereira (New University of Lisbon, Portugal) and Francesca Toni (Imperial College London, UK) have been confirmed as our invited speakers!
Description, Aim and Scope
Machine Ethics, Explainability are two recent topics that have been attracting a lot of attention and concern in the last years. This global concern has manifested in many initiatives at different levels. There is an intrinsic relation between these two topics. It is not enough for an autonomous agent to behave ethically, it should also be able to explain its behavior, i.e. there is a need for both ethical component and explanation component. Furthermore, an explainable behavior is obviously not acceptable if it is not ethical (i.e., does not follow the ethical norms of the society).
In many application domains especially when human lives are involved (and ethical decisions must be made), users need to understand well the system recommendations, so as to be able to explain the reasons for their decisions to other people.
One of the most important ultimate goals of explainable AI systems is the efficient mapping between explainability and causality. Explainability is the system ability to explain itself in natural language to average user by being able to say, "I generated this output because x,y,z". In other words, the ability of the system to state the causes behind its decision is central for explainability.
However, when critical systems (ethical decisions) are concerned, is it enough to explain system's decisions to the human user? Do we need to go beyond the boundaries of the predictive model to be able to observe a cause and effect within the system?
There exists a big corpus of research work on explainability, trying to explain the output of some blackbox model following different approaches. Some of them try to generate logical rules as explanations. However, It is worth noting that most methods for generating post-hoc explanations are themselves based on statistical tools, that are subject to uncertainty or errors. Many of the 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 with respect to the original model, as there are always trade-offs between explainability and fidelity.
On the other side, a good corpus of researchers have used inherently interpretable approaches to design and implement their ethical autonomous agents. Most of them are based on logic programming, from deontic logics to non-monotonic logics and other formalisms.
Logic Programming has a great potential in these two emerging areas of research, as logic rules are easily comprehensible by humans, and favors causality which is crucial for ethical decision making .
Anyway, in spite of the significant amount of interest that machine ethics has received over the last decade mainly from ethicists and artificial intelligence experts, the question "are artificial moral agents possible?" is still roaming around.
There have been several attempts for implementing ethical decision making into intelligent autonomous agents using different approaches. But, so far, no fully descriptive and widely acceptable model of moral judgment and decision making exists. None of the developed solutions seem to be fully convincing to provide a trusted moral behavior. The same goes for explainability, in spite of the global concern about the explainability of the autonomous agents' behaviour, existing approaches do not seem to be satisfactory enough. There are many questions that remain open in these two exciting, expanding fields.
This workshop aims to bring together researchers working in all aspects of machine ethics and explainability, including theoretical work, system implementations, and applications. The co-location of this workshop with ICLP is intended also to encourage more collaboration with researchers from different fields of logic programming. This workshop provides a forum to facilitate discussions regarding these topics and a productive exchange of ideas. Topics of interest include (but not limited to):
new approaches to programming machine ethics;
new approaches to explainability of blackbox models;
evaluation and comparison of existing approaches;
approaches to verification of ethical behavior; item logic programming applications in machine ethics;
integrating logic programing with methods for machine ethics;
integrating logic programing with methods for explainability.