Bias in the Context of Artificial Intelligence Systems: Analyzing the risks and contributors from a data perspective

Sammanfattning: As Artificial Intelligence (AI) is progressing to take over decision making in different industries, the threat that comes with the use of these systems is also increasing. One major threat is the risk of these systems acting biased, causing discrimination to parts of the population. To tackle the risk of AI systems acting biased it is important to understand how these biases originate in the first place. An analysis is made to understand where in the process the risk of bias takes place as well as highlighting the major contributor to bias in AI systems. Through a qualitative approach, practitioners currently working with AI and data were inter-viewed and presented with real-life examples of AI systems acting biased to help identify the reasons for the biased outcomes in these AI systems. The findings in this thesis indicate that data is a major contributor to bias in these systems, however, research has mostly been attributed to algorithms. Conclusively, this thesis found that there is a high risk of bias in the data collection, data preparation and model development stages in the AI systems.