Deepfake Detection by Human Crowds, Machines, and Machine-informed Crowds

TL;DR

This study compares human and machine performance in detecting deepfakes, revealing that combining human judgment with machine predictions improves accuracy, but model errors can mislead humans, highlighting complementary strengths and weaknesses.

Contribution

It provides a large-scale comparison of human and machine deepfake detection, and investigates how combined human-machine approaches can enhance accuracy.

Findings

Humans and models have similar accuracy but different error patterns.

Combining human judgments with model predictions improves detection accuracy.

Disrupting facial visual processing impairs human detection but not model performance.

Abstract

The recent emergence of machine-manipulated media raises an important societal question: how can we know if a video that we watch is real or fake? In two online studies with 15,016 participants, we present authentic videos and deepfakes and ask participants to identify which is which. We compare the performance of ordinary human observers against the leading computer vision deepfake detection model and find them similarly accurate while making different kinds of mistakes. Together, participants with access to the model's prediction are more accurate than either alone, but inaccurate model predictions often decrease participants' accuracy. To probe the relative strengths and weaknesses of humans and machines as detectors of deepfakes, we examine human and machine performance across video-level features, and we evaluate the impact of pre-registered randomized interventions on deepfake detection. We find that manipulations designed to disrupt visual processing of faces hinder human participants' performance while mostly not affecting the model's performance, suggesting a role for specialized cognitive capacities in explaining human deepfake detection performance.