Wintersemester 2021/22 Diversity and Fairness in Artificial Intelligence

English language translation will be provided.

Panelists: 

• Prof. Dr. Florian Lemmerich, University of Passau;
• Miriam Rateike, Max Planck Institute for Intelligent Systems;
• Theresa Tran, Lufthansa Industry Solutions;

Moderation: Isabella Graßl, University of Passau

The kick-off event is organized by the MINT Women's Network of the University of Passau in cooperation with Femtec Alumnae e.V..

The unequal treatment of equal facts as well as the equal treatment of unequal facts often happens unconsciously on an individual level - and remains unrecognized. Self-learning algorithms that evaluate corresponding decisions, however, will quickly recognize the underlying patterns and carry the corresponding schematization into the masses where they become obvious. In the case of trivial discriminatory features, this can be detected and corrected - but there are also cases of indirect and covert discrimination, where the consequences may not be immediately apparent even to those reviewing or monitoring the algorithm. For the users of corresponding algorithms, but also for those who create specifications or have to monitor or subsequently enforce them, the question therefore arises as to when and how unequal treatment can or must be avoided in advance or, conversely, whether state sanctions are imposed or redress must be provided. The underlying trade-off is by no means trivial: if even a non-learning algorithm can have unrecognized (indirect) discriminatory consequences, one would hardly impose due diligence obligations on developers or operators.

Speaker: Prof. Dr. Michael Beurskens, University of Passau

Artificially intelligent systems have become an integral part of our everyday lives. They influence us more than many people realize. Novel machine learning methods, especially multi-layered artificial neural networks, have helped new product categories such as voice assistants, self-driving cars or chatbots to become widespread in recent years. Many companies, but also end users, are not aware that these systems are not free of biases and susceptible to specific manipulation attempts, this is called bias effects. In my lecture, I critically question the hype around Artificial Intelligence (AI) as the savior of a digital and automated society. Based on recent studies, relevant expert statements and a detailed practical example, I illustrate that besides the performance of artificially intelligent systems, other quality characteristics such as robustness against discrimination tendencies and unintentional misbehavior will play an important role.

Speaker: Claudia Pohlink, Telekom Innovation Laboratories

This lecture will be held in German will take place.

Natural Language Processing is a branch of computer science that deals with the automated processing of human language, in text or speech data. Typical tasks include, for example, performing automatic spelling and grammar checking, automatically extracting information from large amounts of data (text mining), or performing linguistic communication with a user (e.g., voice control). Machine learning is often used to efficiently overcome such challenges and to provide the computer with the best possible understanding of human language. What happens when social stereotypes are hidden in language models is what this talk deals with. There will also be a brief background on machine learning and bias in AI systems at the beginning.

Speaker: Prof. Dr. Mascha Kurpicz-Briki, Bern University of Applied Sciences

This lecture will be held in English.

Comprehensive technologisation is changing the many areas of human life. With the use of technology, our understanding of the human being and the body is changing at the same time. How does technology influence how we understand human beings? The lecture puts a special focus on the topic of 'diversity' in technologisation and treats it from an anthropological and ethical perspective.
Where is diversity lacking in technologisation and how can it be promoted? On the one hand, it shows how a lack of diversity represents a challenge in technological processes, and on the other hand, it highlights how technology can also be an opportunity for more diversity. It argues for a more inclusive, relational understanding of people and bodies. In addition to techno-feminist approaches and perspectives on gender and intersectionality, the focus is also on the human-animal-world relationship within the framework of a critique of anthropocentrism. Following critical posthumanism and Donna Haraway's figure of the "cyborg", it is shown that in the course of technological developments, traditional categories such as 'woman'-'man', 'human'-'animal'-'machine' or 'nature'-'culture' become blurred.

Speaker: Anna Puzio, University of Münster, Munich School of Philosophy

This lecture will be held in German.

The lecture will focus on the types of harms brought upon by the development or deployment of narrow AI systems as well as the way those harms are taken into account, both by existing laws and stakeholders (notably businesses). The adopted perspective will mix ethics of AI systems (ethics and philosophy), applied ethics, business and human rights as well as European law.

Speaker: Imane Bello, Institut d'Etudes Politiques de Paris

This lecture will be held in English.

Discourses around artificial intelligence are closely linked to the projection of ethnic and gender characteristics onto digital technologies. We are familiar with such discourses from classics of AI films as well as from current discourses on assistance systems like Alexa and co., and they also form a central point of reference for feminist theory. Within these discourses, dichotomies such as 'nature vs. culture', 'emotionality vs. rationality', or 'power vs. powerlessness' are on the one hand reproduced in technical contexts, but on the other hand also subverted, which provides opportunities for their cultural renegotiation. Accordingly, the lecture will start with a look at corresponding topoi in AI film, contrast them with poststructuralist theorizing, and on this basis critically discuss the medial interfaces of AI in everyday contexts (work, care, family) as well as their marketing. Special attention will be given to the question of what reflective competence might mean in this context.

Speaker: Dr. Martin Hennig, University of Tübingen

This lecture will be held in German.

At the MINT Women's Network Meeting on January 24, 2022, we will discuss together with Die Juristinnen* and external legal experts, interfaces between legal tech and AI, the influence of AIin the legal industry, and AI and ethics. A special focus will also be on the topic of discrimination by AI, with particular attention to discrimination against women by AI. In addition, we want to talk about possibilities to design a feminist AI and exchange ideas with all participants.  You are welcome to send questions to the MINT Women's Network in advance, which we will try to answer during the virtual event.

Registration for participants of the University of Passau via StudIP, external registrations please send to: mint-frauen@uni-passau.de.

This lecture will be held in German.

Data-driven technologies are shaping our everyday lives. Big data analytics and artificial intelligence play a key role here. Can algorithms contribute to more fairness or discrimination? And what role does diversity in artificial intelligence play in developing inclusive technologies? Mina Saidze, Forbes 30 under 30 founder of Inclusive Tech, will answer these and other questions as well as present practical examples.

Speaker: Mina Saidze, Founder of Inclusive Tech

Artificial intelligence applications and products are already influencing the everyday lives of millions of people, for example through the use of voice assistants or by making suggestions when shopping online. AI tools and services recommend medical treatments, translate documents into hundreds of languages, decide on loans, make recommendations when recruiting employees, reintegrating the unemployed into the labor market, or make predictions about the recidivism of offenders, to name just a few. Many of these systems aim for greater objectivity than could be expected from human decision makers in the past. Some of these systems do serve their purpose. However, it is now known that several AI systems discriminate or have discriminated in the past against people with dark skin or on the basis of gender, for example.

Often, the problem lies in incorrect or missing training data, inadequate testing, or lack of quality control. A 2019 study by the New York-based AI Now Institute also concludes that the AI industry is facing a diversity crisis. The study worries that AI system developers are unconsciously perpetuating bias. Discrimination against minorities and misogyny are found not only in the composition of developer teams or in the culture of companies, but also in the systems themselves. As an example of misogyny, we look here at voice assistance systems such as Amazon's Alexa, Apple's Siri, and Google's Assistant, in addition to facial recognition software. A Unesco report from 2019 comes to the following conclusion regarding voice assistants: "The association of a female voice with traits such as patience, submissiveness and low-complexity responses may turn them into feminine traits in societal perception. It is also still completely unclear how voice assistants will affect children's understanding of roles and behavior in the long term."

Scientific investigations of AI systems and resulting indications of misinterpretations or problematic decisions by AI systems have already led to some improvements. The proposals of algorithmic decision-making systems are not always comprehensible to those affected. Increasingly, therefore, there are calls for transparency and fairness of the systems. Last but not least, the EU Commission's regulatory proposals also aim in this direction. AI systems themselves cannot distinguish between meaningful and meaningless outcomes, between fair and discriminatory outcomes. They have no consciousness and cannot "think" in a larger social, political or humanitarian context. Therefore, it is not enough to leave the solution of the problems to the tech companies alone. These issues do not only concern the tech industry, governments or NGOs. For AI systems to be used for the good of humanity, the critical voice of every individual, every person whose life AI tools and services affect, is needed.

Speaker: Prof.  Dr. Gudrun Schiedermeier, Landshut University of Applied Sciences