It’s been over a year since fake videos started appearing on social media with well known actors reading speeches they never gave. With the upcoming U.S. election, there’s a genuine concern that AI could be used to influence the outcome. Once upon a time, metadata embedded in images or video was good enough. Now that data is often stripped when people upload content to social media. To address these sorts of issues, Digimarc and DataTrails are collaborating to prove the authenticity with a tamper-proof audit trail on a distributed ledger.
“In an era of GenAI, tools like robust digital watermarks and cryptographic proofs can transform our current digital ecosystem – removing uncertainty and suspicion to create greater authenticity and trust in digital content,” said Digimarc Chief Product Officer Ken Sickles.
The watermark which is invisible to the human eye, enables a user to retrieve the unaltered metadata of the media. It can also link to provenance data stored on DLT. Digimarc has provided numerous central banks with watermarking services to deter currency counterfeits for decades.
Watermark in the AI era
We wondered about watermarks in the AI era. Digimarc likely works just fine today. What if nefarious actors train their AI tools to learn how to recognize and mimic watermark patterns in the future? That’s likely hard. But perhaps not impossible in the not-too-distant future. However, technology will likely evolve in response.
Recent research from the Brookings Institute outlined how it is possible to corrupt or remove a watermark. It started from the position of an AI algorithm watermarking content to demonstrate it is not created by humans. However, in the Digimarc and DataTrails case, corrupting the watermark would defeat the object. The aim is to prove the content is genuine. So, the absence of a watermark means it’s fake.
Meanwhile, the non profit Partnership on AI has developed a framework for media transparency. Digimarc and DataTrails claim their collaboration adheres to the framework’s requirements.
