Deception Decoder: Proposing a Human-Focused Framework for Identifying AI-Generated Content on Social Media

TL;DR

This paper introduces the Deception Decoder, a human-focused framework that helps social media users identify AI-generated misinformation across multiple media types, addressing limitations of automated detection methods.

Contribution

It presents a novel multimodal, systematic framework designed to assist users in detecting AI-generated content, shifting focus from automated to human-centered detection strategies.

Findings

Initial testing shows promising improvements in detection

Framework supports multiple media types including text, images, and videos

Further research needed for generalisability and long-term effectiveness

Abstract

Generative AI (GenAI) poses a substantial threat to the integrity of information within the contemporary public sphere, which increasingly relies on social media platforms as intermediaries for news consumption. At present, most research efforts are directed toward automated and machine learning-based detection methods, despite growing concerns regarding false positives, social and political biases, and susceptibility to circumvention. This dissertation instead adopts a human-centred approach. It proposes the Deception Decoder; a multimodal, systematic, and topological framework designed to support general users in identifying AI-generated misinformation and disinformation across text, image, and video. The framework was developed through a comparative synthesis of existing models, supplemented by a content analysis of GenAI-video, and refined through a small-scale focus group session. While initial testing indicates promising improvements, further research is required to confirm its generalisability across user groups, and sustained effectiveness over time.