Predictive models based on big data are founded on the idea that knowing the past enables us to predict the future, but what kind of future are we constructing when the models are based on a deepfake version of the past? Correlation has overtaken causation, while homophily shadows differences across these models that are not only applied in social media platforms’ algorithms, but also in policing predictives or risk assessment systems. We discuss how these practices reproduce bias and generate self-fulfilling prophecies with Wendy Chun, professor of New Media, Director of the Digital Democracies Institute and author of numerous books. Is it possible to exit the default world of endless reproducibility and conceive of a different future from what has been left out in the past?