AI Deepfakes and Information Asymmetry in Platform Economies A Multimodal Deep Learning Framework for Detection, Attribution, and Economic Impact Modelling

Abstract:

Generative intelligence has fuelled a meteoric increase of deepfakes in social networks, politics, and digital market, which poses formidable dangers for information validity, social belief, and market resilience. The current forensic system has primarily focused on unimodal detection metho ds and proved to be non resilient to the mixed impacts of multimodal attacks, attribution difficulty, and new diffusion models. For this reason, this paper presents MIDAS: Multimodal Integrated Detection and Attribution System with an EfficientNet based sp atial encoding module, a Vision Transformer for spatio temporal modelling , a DCT based frequency analysis module, a MFCC based audio
forensics module, and a novel cross modal attention fusion integrated on unique transformer architecture. The multimodal ev aluation on the standard FaceForensics++, DFDC, OpenForensics, and ASVspoof 2021 datasets was conducted by modelling binary classification, attribution detection, and economic asymmetry. The experimental results indicate that MIDAS attains classification a ccuracies of 97.3% on the FaceForensics++ dataset, 88.6% on the DFDC dataset, 96.1% on the OpenForensics dataset, and 95.4% on the ASVspoof 2021 dataset. Moreover, multimodal fusion greatly contributes to MIDAS for being robust to compressed images, unseen attacks, and inconsistency across modal types and the temporal transformer learning contributes the most according to the ablating studies. The results from simulated experiments with Information Asymmetry Index indicate a super linear erosion in trusts e xceeding the deepfake penetration threshold of = 0.20. Therefore, this research introduces a scalable deepfake verification and provenance system for forensic analysis, moderating, and responsible platform governing in the age of generated contents.

Keywords: Deepfake Detection, Multimodal Learning, Information Asymmetry, Digital Forensics, Transformer Networks, Platform Economies