Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

How do people watch AI-generated videos of physical scenes?

Danqing Shi, Lan Jiang, Katherine M. Collins, Shangzhe Wu, Ayush Tewari, Miri Zilka

TL;DR

The paper investigates how people watch AI-generated videos of physical scenes using eye-tracking to capture moment-by-moment gaze during video understanding and AI-detection tasks. It employs two stimulus sets (physics-based and professionally edited) comprising real and AI-generated videos, with 40 participants and 5-second clips, to compare gaze patterns and detection performance. The key finding is that gaze is driven more by perceived authenticity than by actual video authenticity, and simply being aware that AI generation is possible prompts an active anomaly-search strategy that alters media consumption. These insights contribute human-centered metrics for evaluating AI video realism and detection, suggesting that perception significantly shapes viewing behavior in an era of increasingly realistic AI media, and the authors provide a dataset to support future research.

Abstract

The growing prevalence of realistic AI-generated videos on media platforms increasingly blurs the line between fact and fiction, eroding public trust. Understanding how people watch AI-generated videos offers a human-centered perspective for improving AI detection and guiding advancements in video generation. However, existing studies have not investigated human gaze behavior in response to AI-generated videos of physical scenes. Here, we collect and analyze the eye movements from 40 participants during video understanding and AI detection tasks involving a mix of real-world and AI-generated videos. We find that given the high realism of AI-generated videos, gaze behavior is driven less by the video's actual authenticity and more by the viewer's perception of its authenticity. Our results demonstrate that the mere awareness of potential AI generation may alter media consumption from passive viewing into an active search for anomalies.

How do people watch AI-generated videos of physical scenes?

TL;DR

The paper investigates how people watch AI-generated videos of physical scenes using eye-tracking to capture moment-by-moment gaze during video understanding and AI-detection tasks. It employs two stimulus sets (physics-based and professionally edited) comprising real and AI-generated videos, with 40 participants and 5-second clips, to compare gaze patterns and detection performance. The key finding is that gaze is driven more by perceived authenticity than by actual video authenticity, and simply being aware that AI generation is possible prompts an active anomaly-search strategy that alters media consumption. These insights contribute human-centered metrics for evaluating AI video realism and detection, suggesting that perception significantly shapes viewing behavior in an era of increasingly realistic AI media, and the authors provide a dataset to support future research.

Abstract

The growing prevalence of realistic AI-generated videos on media platforms increasingly blurs the line between fact and fiction, eroding public trust. Understanding how people watch AI-generated videos offers a human-centered perspective for improving AI detection and guiding advancements in video generation. However, existing studies have not investigated human gaze behavior in response to AI-generated videos of physical scenes. Here, we collect and analyze the eye movements from 40 participants during video understanding and AI detection tasks involving a mix of real-world and AI-generated videos. We find that given the high realism of AI-generated videos, gaze behavior is driven less by the video's actual authenticity and more by the viewer's perception of its authenticity. Our results demonstrate that the mere awareness of potential AI generation may alter media consumption from passive viewing into an active search for anomalies.
Paper Structure (19 sections, 6 figures)

This paper contains 19 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Method
  3. Hypotheses
  4. Experimental Design
  5. Participants
  6. Materials
  7. Set 1: Physics videos
  8. Set 2: Professionally-edited videos
  9. Apparatus
  10. Procedure
  11. Data Collection, Preprocessing, and Storage
  12. Metrics
  13. Eye tracking metrics
  14. AI detection metrics
  15. Results
  16. ...and 4 more sections

Figures (6)

  • Figure 1: This study investigates human gaze behavior from 40 participants with diverse backgrounds when watching real-world and AI-generated videos of physical scenes. Participants are asked to watch videos normally for understanding or to detect whether the videos are AI-generated or not. Eye tracking data is collected during the experiments for analysis.
  • Figure 2: Study procedure
  • Figure 3: Human gaze behavior analysis when participants (a) completed video understanding (T1) and AI detection (T2) tasks; (b) watched AI-generated (AI=a) and real videos (AI=r); and (c) judged the video as AI-generated (J=a) or real (J=r).
  • Figure 4: In the qualitative analysis of the aggregated human visual attentions on sample AI-generated videos, people show task-dependent and judgment-dependent gaze behaviors.
  • Figure 5: Accuracy of AI detection: (a) User accuracy ranking of 40 participants; (b) Accuracy comparison between S1, physics videos, and S2, professional videos; (c) Confusion matrix. We see more errors due to judging real videos as AI.
  • ...and 1 more figures