Fake videos of real people -- and how to spot them | Supasorn Suwajanakorn

Summary

In this thought-provoking talk, Supasorn Suwajanakorn discusses the potential of AI-generated holograms and 3D models, using photos and videos to create lifelike digital replicas of individuals. He explores the possibilities of preserving the legacies of influential figures, such as Holocaust survivors or historical icons like Richard Feynman, and even bringing loved ones back through interactive experiences. While showcasing impressive technology that can synthesize facial expressions and speech, Suwajanakorn also addresses the ethical concerns and potential misuse of such technology, stressing the importance of responsible usage and countermeasures to prevent the spread of fake media.

Takeaways

• 😀 The technology can create realistic 3D models of individuals using only existing photos and videos, without the need for 3D scanning.
• 😀 This approach allows for high-quality reconstruction of facial expressions and details, even from static images or videos.
• 😀 AI can learn to mimic an individual’s unique speech patterns and mannerisms using video footage and audio recordings.
• 😀 This technology has potential applications such as preserving the legacies of historical figures, allowing them to 'speak' again, and making their teachings more accessible.
• 😀 The ability to create digital replicas of people opens the door for interactive experiences, such as conversations with holograms of Holocaust survivors for educational purposes.
• 😀 Famous individuals, like Richard Feynman or even loved ones, could be brought back to continue their impact through AI-generated content.
• 😀 By analyzing large collections of photos and videos, AI can refine and create models with an accurate representation of a person's likeness.
• 😀 The generated models are capable of realistic speech synthesis and detailed facial movements, with mouth movements synchronized to audio inputs.
• 😀 The technology raises ethical concerns, particularly regarding the potential misuse of AI to create deepfakes or manipulate public figures.
• 😀 Researchers are developing countermeasures, like Reality Defender, a browser plugin that flags potentially fake content, to protect against the harmful use of synthetic media.

Q & A

What is the ultimate goal of the technology presented in the script?

The ultimate goal is to create accurate models of individuals that replicate their mannerisms, expressions, and speech patterns, enabling realistic interactions with digital versions of people.

How does the technology reconstruct 3D models from photos and videos?

The technology analyzes large collections of photos and videos to build a refined 3D model. It uses an iterative process to capture fine details, like facial wrinkles, and to ensure that the model accurately reflects expressions.

Can the technology be applied to any person?

Yes, the technology can be applied to anyone as long as there is a sufficient collection of photos and videos available, which can be used to create a model of the person.

How does the system handle different expressions and facial details?

The system can adjust to different facial expressions by analyzing a sequence of photos and refining the model to capture variations such as creases and wrinkles, ensuring it accurately represents the person's facial features in different scenarios.

What is the significance of synthesizing audio from video footage?

Synthesizing audio allows the system to generate a realistic mouth movement that matches the person’s speech patterns. This makes the digital representation more lifelike and enables the creation of a speaking model based on just audio inputs.

What are the potential uses of this technology?

Potential uses include bringing back historical figures to teach, allowing authors to read their books in any language, providing interactive advice from loved ones who have passed away, and generating realistic digital versions of people for various applications.

What are the ethical concerns related to this technology?

The main ethical concern is the potential for misuse, such as creating fake videos or spreading misinformation. These digital models can be easily manipulated to mislead others, which raises questions about the authenticity and trustworthiness of media.

What steps are being taken to prevent misuse of the technology?

To address the concerns, the team is developing countermeasure tools like Reality Defender, a browser plug-in that flags potentially fake content, alongside AI-based methods and human moderators to detect fake images and videos.

What did Supasorn Suwajanakorn mean by 'scaling to anyone'?

By 'scaling to anyone,' Suwajanakorn refers to the ability to apply this technology to any individual, using publicly available images and videos, making it possible to create a digital version of anyone, not just celebrities or public figures.

What are the potential educational benefits of this technology?

The technology could revolutionize education by allowing digital recreations of historical figures, scientists, and teachers to engage with students in a highly interactive and personalized way, potentially reaching millions of people globally.