Researchers at the University of Washington developed an AI system that can create photorealistic videos of former President Barack Obama speaking by synthesizing mouth movements to match any audio input.
Researchers at the University of Washington created an artificial intelligence system capable of generating photorealistic videos of former President Barack Obama speaking words he never actually said. The system uses a recurrent neural network trained on approximately 14-17 hours of high-definition footage from Obama's weekly presidential addresses. The AI analyzes millions of video frames to learn how Obama's mouth, lips, teeth, and facial features move when he speaks different sounds. When given an audio clip of Obama's voice, the system synthesizes corresponding mouth shapes and grafts them onto existing video footage of Obama, creating realistic lip-synced videos. The technology was demonstrated by various researchers and used by filmmaker Jordan Peele and BuzzFeed to create a public service announcement about fake news, where a synthetic Obama appeared to call President Trump a 'dipshit.' The researchers chose Obama because of the abundance of publicly available high-quality video footage of him speaking directly to cameras. While the technology has potential legitimate applications like improving video conferencing and creating educational content with historical figures, it has raised significant concerns about the potential for creating convincing fake news videos and deepfake content for malicious purposes.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
AI system
Due to a decision or action made by an AI system
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed