Give Randomness a Chance: The Benefits of Randomness in Algorithmic Decision Making

Humans have an ambivalent relationship towards randomness and chance. We embrace the concept of chance and luck as positively surprising events that might change our life for the better without effort and control. However, as part of our evolutionary conditioning, we often perceive randomness as a threat to our ability to "fight or flight" as it introduces uncertainty and unpredictability into our environment, making it difficult for us to anticipate and prepare for potential dangers. Hence, modern society makes use of digital technologies to reduce randomness through predictive analytics, personalisation, control systems, and quality control algorithms.  However, the absence of chance in the digital space comes at a price. Algorithmic predictions in digital applications can have several downsides. Algorithmic recommendations reduce diversity, while filter bubbles and the risk of over-personalisation emerges and privacy and security concerns arise, as algorithms used to make predictions require large amounts of personalised data. More importantly, excessive algorithmic decision making in public domains can perpetuate and amplify societal biases, create "feedback loops" that lock people into certain outcomes, and lead to a lack of accountability and transparency. In this work, I suggest “corrective  randomness” as a policy tool. Corrective randomness describes a fine-tuned amount of randomness that is added to algorithmic decision making and predictive analytics in order to downscale adverse effects such as over-personalisation, filter bubbles, “vicious” feedback loops or loss of serendipity with reasonable impact on the functionality and economic performance of the respective digital technologies. I showcase the concept and application of “corrective randomness” as a tool to balance between accuracy and fairness, accountability, and privacy for a range of real-world decision making systems, including e-commerce, fraud detection, and criminal justice. I suggest that emerging legal frameworks in the European Union (AI Act and Data Act) should incorporate “corrective randomness” as a policy tool for regulating the adverse effects of algorithmic decision making and predictive analytics in industry and public domains.