Identifying AI generated content in the digital age: the role of watermarking

In brief

• Identification of AI generated content is critical to foster trust and transparency in the AI ecosystem.
• Policymakers across the world have recognized the issue and there have been multi-stakeholder consultations and open dialogues for measures to promote AI content detection mechanisms.
• Content detection mechanisms such as watermarking are gaining traction with voluntary commitments being announced by the technology companies alongside partnerships and standards being developed by the stakeholders to maintain the balance between innovation and governance. 

Importance of authenticating content

The identification of AI-generated content is crucial for several reasons. In fields such as education, law, medical diagnosis and financial services etc., the origin and reliability of information directly influences critical decisions. Additionally, the rise of disinformation and misinformation—false information spread intentionally or otherwise—poses significant risks to public trust, law and order, and national security. In the digital economy, preventing deception and financial fraud is vital, as is protecting intellectual property, promoting consumer trust and identifying synthetic content. 

AI content detection

Today, there are various mechanisms that have been developed for the detection of AI-generated content including metadata verification, retrieval-based detectors, watermarking and post-hoc detectors. AI watermarking is a AI content detection solution that is often discussed. Historically, the roots of watermarking can be traced back to the 1950s when identifiers were added to music, laying the foundation for the linkages between technology and intellectual property management. AI watermarking involves embedding low perturbation identifiers in the content which are only detectable using algorithms.

Highlight of global measures

Governments worldwide recognize the importance of addressing the challenges posed by AI-generated content and are taking measures to introduce safeguards. The European Union's AI Act, 2024 mandates the disclosure and labeling of AI-generated content in specific contexts, in an effort to promote greater transparency in the context of when people are interacting with AI systems.

In the United States, an executive order issued on October 30, 2023, outlines measures to protect Americans from AI-enabled fraud and deception. It calls for the development of standards and best practices for detecting AI-generated content and authenticating official communications. Further, the US Department of Commerce is tasked with the creation of guidelines for content authentication and watermarking, setting an example for both the private sector and other governments.

On the other hand, technology giants with large footprints in the AI ecosystem have called for multi-stakeholder and global consensus for standards, practices, and technologies integral to AI content detection. There are examples of industry led standards and partnerships being developed for AI such as the Partnership for AI and the C2PA standards pertaining to synthetic media. These efforts recognize the importance of collaboration for solving issues relating to interoperability and keeping frenzied technological advancements in check.