Fair algorithms

Those who have data often also wield power. Miriam Fahimi is a doctoral student at the Digital Age Research Center (D^!ARC) and is working on the social effects that result from algorithms.

In total, some 15 doctoral students – spread across the EU – are working on the EU-H2020 project “NoBIAS – Artificial Intelligence without Bias”. The research is premised on the assumption that algorithms lead to discrimination. While developers of novel technologies often pay too little attention to social, legal and ethical issues, there is a need for interdisciplinary approaches to ensure greater algorithmic fairness.

But what is algorithmic fairness? “I am interested in algorithmic decision-making processes in different contexts. For instance, I want to know how fair decisions are made, assessed and validated. Who decides what is fair and what is not – and for whom is it fair?,” Miriam Fahimi tells us. She conducted her ethnographic research, amongst others, at an agency that assesses people’s creditworthiness. She explains the principle as follows: “When people apply for a credit, the agency calculates the default probability. Existing data, attributes and the respective weights of credit seekers determines the score, a point value that is used by banks for their credit decisions. This is a complex process and people applying for credit often do not understand how the score is calculated.” The company Miriam Fahimi shadowed strives to make the processes transparent and to build trust. Part of its strategy is to provide people with information on how to improve their score.

In many cases, the algorithms responsible for minor and major decisions in our digital daily lives are not transparent. “We know that well-known translation programmes have gender bias. For a long time, the English word ‘doctor’ was translated into the male word ‘Arzt’ in German. Or that hate speech detection works better in colonial languages, e.g. English or Spanish. All of this has implications for equity in our society“, Miriam Fahimi explains. The fact that the systems are nonetheless so successful has to do with the fact that they are practical: “A lot of problems arising through algorithms and platforms are known. But Amazon, Netflix, Youtube and Google are also functional and convenient. So we find ourselves in a process of continuous negotiation.“

The question of who achieves the highest visibility on the major search engines, for example, is often also a question of money. “Data are commodities. And those who have a lot of data also wield power. In most cases, we are the ones generating profit for these companies with consciously or unconsciously disclosing personal data. I think it is important to be aware of this and to question the social impact of these technologies,” Miriam Fahimi concludes.

Twelve of the doctoral students in the NoBIAS project have a background in Computer Science. Two are Legal scholars, one – Miriam Fahimi – works in the field of Science and Technology Studies, or STS for short.

Fahimi studied Development Studies (BA) and Economics (BSc) at the University of Vienna. She graduated with a Master’s degree in International Development, always interested in global justice issues. She found her way to technology and engineering while working on her Master‘s thesis: “I looked at the way in which digitalisation processes are transforming care work and undertook research on this topic in care settings.” While still in Cologne, Miriam Fahimi took up her work in the research group headed by Professor Katharina Kinder-Kurlanda. Along with Professor Kinder-Kurlanda, she then moved to the University of Klagenfurt and the newly founded “Digital Age Research Center” (D^!ARC). Miriam Fahimi feels at home in the field of research, although: “I need the intersection with concrete, social issues. If we work on fairness in algorithms, we can contribute to a fairer society as digitalisation increases.”