Neural fingerprinting makes media identifiable even when distorted by AI: SoundPatrol co-founders

By CNBC Television

Share:

Key Concepts

  • Neural Fingerprinting: An AI-driven technology that creates unique digital identifiers for intellectual property (IP) like music, video, and live events, allowing for identification even when distorted or manipulated.
  • Generative AI: Artificial intelligence models capable of creating new content, such as text, images, music, and video, which poses risks to copyrighted material.
  • Intellectual Property (IP) Rights: Legal rights granted to creators and owners of original works, including copyrights, patents, and trademarks.
  • Antipiracy: Measures and technologies designed to prevent the unauthorized copying, distribution, and use of copyrighted material.
  • Synthetic Media: AI-generated content that mimics or replaces real media, raising concerns about authenticity and copyright infringement.
  • Arms Race (in technology): A continuous cycle of development where one party's advancements necessitate counter-advancements from another, as seen in antipiracy technology versus AI manipulation.
  • Hype Cycle (Trough of Disillusionment, Plateau of Productivity): A model describing the stages of adoption for new technologies, from initial excitement to eventual widespread use and integration.

Sound.AI: Protecting IP in the Age of Generative AI

This segment discusses the emergence of new technologies designed to combat the risks posed by viral AI models to Hollywood's copyrighted material. The focus is on Sound.AI, an antipiracy startup co-founded by legendary Hollywood super agent Michael Oitz and AI researcher Walter De Brower.

The Genesis of Sound.AI

Michael Oitz, with his extensive background in Hollywood and tech investment, was inspired to co-found Sound.AI after a conversation with a Stanford professor about NFT technology. He realized the professor's neural fingerprinting technology could be adapted to protect IP such as songs, videos, and live sporting events. This realization led to the rapid establishment of the company.

  • Key Statement (Michael Oitz): "Can your neural fingerprinting uh work with uh IP like songs or video or live sporting events?" and Walter's answer was, "It's elementary."

The company quickly secured interest from major music labels, with Sir Lucian Grainge of Universal Music Group expressing immediate enthusiasm, stating, "I'm in." Similar positive reactions were received from Dennis Coker and Rob Stringer at Sony Music.

Walter De Brower on Neural Fingerprinting Technology

Walter De Brower, an AI researcher and co-founder, explains the technical underpinnings of Sound.AI's neural fingerprinting technology.

  • Core Technology: Neural fingerprinting is an AI model that requires vast amounts of high-quality data for training. It functions like a "super student" that learns rapidly and extracts its own rules.
  • Differentiation: Sound.AI's neural fingerprinting is specifically focused and optimized for music. This is achieved through:
    • Training Data: Utilizing millions of legally usable songs that form the backbone of popular music globally, provided by major labels. This data is crucial for both creating generative models and developing forensic models to control the use of generative AI.
    • Multi-Teacher Model: The AI model is trained by multiple "teachers," each specializing in different aspects of the IP. These include teachers for melody, harmony, rhythm, metadata parsing, lyrics, and voice analysis. The combination of these specialized teachers with extensive data creates a "super model."
  • Authentication Layers: Neural fingerprinting is only one of three authentication layers employed by Sound.AI.

Expansion Beyond Music

While Sound.AI is initially focusing on music labels, the technology has broader applications.

  • Phased Rollout: The company is adopting a "silo by silo" approach, starting with music due to its lower barrier to entry and readily available data.
  • Future Targets: Simultaneously, Sound.AI is engaging with movie studios, streamers, and sports leagues.
  • Sports League Engagement: Discussions have already begun with the Premier League in Europe, where it's estimated that each team faces approximately 500,000 to 1 million illegal downloads per game.
  • Broad Applicability: The technology is not limited to music and is expected to be highly effective for all types of IP.

Individual Content Creator Protection

The discussion touches upon the potential for individual content creators to utilize such services.

  • Accessibility: Michael Oitz suggests that the service will be available to anyone who wishes to use it.
  • Use Cases: While the "half-life" of some content might be short, scripted content, sporting events, and unique, longer-form content represent clear use cases for individual protection.

The AI Arms Race and Future Outlook

The conversation addresses the ongoing challenge of generative AI evolving and potentially circumventing antipiracy technologies.

  • Continuous Development: Walter De Brower acknowledges that as piracy and forgery methods improve, so too must antipiracy technologies. He describes this as an "arms race."
  • Projected Timeline for AI Integration:
    • 2025: Expected to be a "trough of disillusionment" for AI, where initial hype subsides and users grapple with the limitations and impact on jobs.
    • 2026: Anticipated to be the end of major IP lawsuits and the beginning of settlements.
    • 2027: Regulation is expected to set in, leading to the "plateau of productivity" and the emergence of a "brave new world" of AI integration.
  • Sensitivity of Sound.AI's Technology: Michael Oitz highlights the advanced capabilities of their technology, stating, "our technology can uh identify stolen IP as little as a 10-second sound bite." This demonstrates its effectiveness against even short, manipulated audio clips, such as synthetic songs.

Conclusion

Sound.AI is leveraging advanced neural fingerprinting technology to address the growing threat of IP infringement posed by generative AI. By focusing on robust data training, multi-layered authentication, and a strategic expansion plan, the company aims to provide effective solutions for major content industries and potentially individual creators, navigating the evolving landscape of AI and intellectual property protection.

Chat with this Video

AI-Powered

Load the transcript when you're ready to chat so the initial page stays lighter.

Ready to summarize another video?

Summarize YouTube Video