This PhD thesis investigates the societal impact of machine learning (ML). ML increasingly informs consequential decisions and recommendations, significantly affecting many aspects of our lives. As these data-driven systems are often developed without explicit fairness considerations, they carry the risk of discriminatory effects. The contributions in this thesis enable more appropriate measurement of fairness in ML systems, systematic decomposition of ML systems to anticipate bias dynamics, and effective interventions that reduce algorithmic discrimination while maintaining system utility. I conclude by discussing ongoing challenges and future research directions as ML systems, including generative artificial intelligence, become increasingly integrated into society. This work offers a foundation for ensuring that ML's societal impact aligns with broader social values.

Machine learning is becoming an ever present part in our lives as many decisions, e.g. to lend a credit, are no longer made by humans but by machine learning algorithms. However those decisions are often unfair and discriminating individuals belonging to protected groups based on race or gender. With the recent General Data Protection Regulation (GDPR) coming into effect, new awareness has been raised for such issues and with computer scientists having such a large impact on peoples lives it is necessary that actions are taken to discover and prevent discrimination. This work aims to give an introduction into discrimination, legislative foundations to counter it and strategies to detect and prevent machine learning algorithms from showing such behavior.

A common feature of non-monotonic logics is that the classical notion of equivalence does not preserve the intended meaning in light of additional information. Consequently, the term strong equivalence was coined in the literature and thoroughly investigated. In the present paper, the knowledge representation formalism under consideration is claim-augmented argumentation frameworks (CAFs) which provide a formal basis to analyze conclusion-oriented problems in argumentation by adapting a claim-focused perspective. CAFs extend Dung AFs by associating a claim to each argument representing its conclusion. In this paper, we investigate both ordinary and strong equivalence in CAFs. Thereby, we take the fact into account that one might either be interested in the actual arguments or their claims only. The former point of view naturally yields an extension of strong equivalence for AFs to the claim-based setting while the latter gives rise to a novel equivalence notion which is genuine for CAFs. We tailor, examine and compare these notions and obtain a comprehensive study of this matter for CAFs. We conclude by investigating the computational complexity of naturally arising decision problems.

We consider knowledge representation (KR) formalisms as collections of finite knowledge bases with a model-theoretic semantics. In this setting, we show that for every KR formalism there is a formalism that characterizes strong equivalence in the original formalism, that is unique up to isomorphism and that has a model theory similar to classical logic.

A central question in knowledge representation is the following: given some knowledge representation formalism, is it possible, and if so how, to simplify parts of a knowledge base without affecting its meaning, even in the light of additional information? The term strong equivalence was coined in the literature, i.e. strongly equivalent knowledge bases can be locally replaced by each other in a bigger theory without changing the semantics of the latter. In contrast to classical (monotone) logics where standard and strong equivalence coincide, it is possible to find ordinary but not strongly equivalent objects for any nonmonotonic formalism available in the literature.

In this paper we combine two of the most important areas of knowledge representation, namely belief revision and (abstract) argumentation. More precisely, we show how AGM-style expansion and revision operators can be defined for Dung's abstract argumentation frameworks (AFs). Our approach is based on a reformulation of the original AGM postulates for revision in terms of monotonic consequence relations for AFs. The latter are defined via a new family of logics, called Dung logics, which satisfy the important property that ordinary equivalence in these logics coincides with strong equivalence for the respective argumentation semantics. Based on these logics we define expansion as usual via intersection of models. We show the existence of such operators. This is far from trivial and requires to study realizability in the context of Dung logics. We then study revision operators. We show why standard approaches based on a distance measure on models do not work for AFs and present an operator satisfying all postulates for a specific Dung logic.

Although machine learning and artificial intelligence (AI) are terms that are often used interchangeably, they are quite different. That difference becomes more important as applications for these technologies become more prevalent. Tech Briefs posed questions to machine learning/AI industry executives to get their views on issues such as machine learning platform selection, interpreting data created by these platforms, and pros and cons of implementing machine learning. Our participants are Dr. Florian Baumann, Chief Technology Officer - Automotive & AI, at Dell Technologies; Mario Bergeron, Technical Marketing Engineer at Averna Technologies; Zach Mayer, Vice President of Data Science at Data Robot; George Rendell, Senior Director of NX Design at Siemens Digital Industries Software; and Rajesh Ramachandran, Chief Digital Officer - Industrial Automation, at ABB Inc. Tech Briefs: Machine learning is a term that has confused many people, partly because its definition has taken on multiple forms. How do you define machine learning and how do you see it being used in manufacturing, medical, transportation, or other industrial applications?

As the development of autonomous cars continues, the challenges around how data from those vehicles is managed needs to be addressed, according to Dell Technologies' Florian Baumann. There's a lot of buzz around the development of autonomous cars, from discussions about the software that goes into them to the time it will take to have fully autonomous vehicles on the road. However, an area less commonly discussed in relation to autonomous vehicles is the data involved in autonomous cars. The sheer amount of data storage they require highlights questions around how that data will be safely managed, held and transferred when self-driving cars start appearing on our roads. Florian Baumann is the global CTO for automotive and AI in Dell Technologies.

Conflicting information in an agent's knowledge base may lead to a semantical defect, that is, a situation where it is impossible to draw any plausible conclusion. Finding out the reasons for the observed inconsistency (so-called diagnoses) and/or restoring consistency in a certain minimal way (so-called repairs) are frequently occurring issues in knowledge representation and reasoning. In this article we provide a series of first results for these problems in the context of abstract argumentation theory regarding the two most important reasoning modes, namely credulous as well as sceptical acceptance. Our analysis includes the following problems regarding minimal repairs/diagnoses: existence, verification, computation of one and enumeration of all solutions. The latter problem is tackled with a version of the so-called hitting set duality first introduced by Raymond Reiter in 1987. It turns out that grounded semantics plays an outstanding role not only in terms of complexity, but also as a useful tool to reduce the search space for diagnoses regarding other semantics.

Alexa, Amazon's virtual assistant, was laughing when it shouldn't. You might have seen tweets about it: The weird, disembodied chuckle bothered people for reasons you can imagine, as well as because it reportedly could happen unprompted--our assistants, after all, are only supposed to listen and speak to us after they hear the wake word. We want them to tell us the weather and set kitchen timers on command, not spook us with laughter. We all know that virtual personas like Alexa, Siri, and the Google Assistant are not real humans. They can't laugh the way we laugh, because they are not alive.