ADVANCED TECHNIQUES IN DETECTING AI-GENERATED SYNTHETIC MEDIA TO CURB FAKE NEWS PROPAGATION

Abstract

The trustworthiness of our information networks suffers due to quick AI-generating fakes that give rise to wrong information online. This work studies new ways to spot fake media through forensic methods and advanced AI plus blockchain technology. Our tests use the Face Forensics++ dataset that includes 1,000 original videos plus 1,000 deepfakes and proves the effectiveness of these techniques. Our results indicate that CNNs excel at finding irregular lighting patterns and artificial facial features with a 95.2% success rate. With an accuracy of 92.3% RNNs recognize video structures through time and detect voice irregularities. Despite the 88.6% accurate noise pattern analysis and error level analysis (ELA) methods with human interpretable insights they require better scalability and human assistance to succeed. Blockchain technology produces an untouched record of original material with 96.8% accuracy making it the best verification method. However, its acceptance and implementation need substantial infrastructure. Research finds that blockchain verification works best in trust yet needs broad industry support whereas other tests have efficient results yet limited interpretation or need extra resources and large sample sets. Sophisticated detection systems prove their ability to protect against synthetic media and false news reports. Despite existing problems, detection systems require further development including quickly spotting threats and protecting against AI and attack methods. Future study should focus on protecting detection systems against attackers and making them better at detecting information across different channels while remaining easy to understand. By reviewing synthetic media detection methods this study gives useful knowledge to both experts and industry leaders who fight misinformation