video

Computer vision systems now surpass the performance of human experts in fields like radiology and dermatology. Can they also help us discern real videos from deepfakes — videos manipulated by artificial intelligence? This Zoom presentation will provide an overview of the state-of-the-art machine-learning models for detecting deepfakes. It will also present evidence that reveals how most people are more accurate at spotting deepfakes than the best machine-learning models. We will examine results of a recent experiment suggesting that humanity's specialized ability for recognizing faces lies at the heart of our superior performance. (No face shown above belongs to a real person!)

Our speaker, MIT's Matthew Groh, is a PhD student and research assistant at the MIT Media Lab where he is a member of the Affective Computing Group. Before MIT, Matt cofounded Proprio Labs, worked as a data scientist at Qadium, RaiseMe, and DARPA, and was a research assistant at Innovations for Poverty Action and the World Bank.