Human detection of machine-manipulated media

Matthew Groh, Ziv Epstein, Nick Obradovich, Manuel Cebrian, Iyad Rahwan

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


The speed at which misinformation can be produced is faster than it has ever been. By combining instance segmentation with image inpainting, researchers present an AI model that can automatically disappear objects such as people, cars, and dogs from images. Exposure to manipulated content can prepare people to detect future manipulations. After seeing examples of manipulated images produced by the target object removal architecture, people learn to more accurately discern between manipulated and original images. Participant performance improves more after being exposed to subtle manipulations than blatant ones. To publicly expose the realism of AI media manipulations, researchers have hosted a website called Deep Angel, where anyone in the world could examine their neural-network architecture and its resulting manipulations.

Original languageEnglish (US)
Pages (from-to)40-47
Number of pages8
JournalCommunications of the ACM
Issue number10
StatePublished - Oct 2021

ASJC Scopus subject areas

  • General Computer Science


Dive into the research topics of 'Human detection of machine-manipulated media'. Together they form a unique fingerprint.

Cite this