Statistical study argues AI deepfake science risks

statmodeling.stat.columbia.edu3 min read

The author argues that algorithmic deepfake science—whether generated by modern chatbots or conducted historically through human hoaxes and bad statistical methods—simulates the entire scientific process rather than merely falsifying a single data point. Both human and chatbot deepfakes rely on conventional forms of data and writing to construct convincing simulations that can bypass scrutiny within a social environment. Ultimately, the author suggests that studying past human-generated scientific deepfakes can help us understand and address the rising threat of AI-generated deepfake science. Open article →