The CSIRO says it has developed an algorithm to prevent deepfakes by shielding online images from artificial intelligence (AI) learning models.
It says it has developed an algorithm that will ‘tweak’ images uploaded online so they are shielded from artificial intelligence (AI).
The research is the result of a three-way partnership between the CSIRO, Cyber Security Co-operative Research Centre (CSCRC) and the University of Chicago.
The CSIRO says their tests show it is possible to subtly change image content (like photos and artwork) to make it unreadable to AI while remaining visually unchanged to the human eye.
CSIRO scientist Dr Derui Wang says the research offers a new level of certainty for anyone uploading content online.
“Existing methods rely on trial and error or assumptions about how AI models behave,” he says.
“Our approach is different; we can mathematically guarantee that unauthorised machine learning models can’t learn from the content beyond a certain threshold.”
In a podcast interview with ABC Radio, he says it works by changing the pixels of the images before uploading so it will disturb AI learning.
He also told the broadcaster that the algorithm can tell data owners exactly how learnable their image data is to AI algorithms.
ABC INTERVIEW: Click through to hear the four-minute interview in full
“That’s a powerful safeguard for social media users, content creators, and organisations,” he says.
Dr Wang says it can be applied automatically at scale.
“A social media platform or website could embed this protective layer into every image uploaded,” he explains.
“This could curb the rise of deepfakes, reduce intellectual property theft, and help users retain control over their content.”
The CSIRO says artists, organisations and social media users can protect their work and personal data from being used to train AI systems or create deepfakes.
CSIRO ON CONTENT PROTECTION
The use of AI to create deepfake images and videos, pornographic and otherwise, has been raising red flags worldwide.
Social media platforms like Facebook are experiencing a rapid growth in deepfake videos including ones that ‘age’ the images of long-dead celebrities to depict their appearance today.
A social media user will be able to use this new algorithm to apply a protective layer to their photos before posting, preventing AI systems from learning facial features for deepfake creation, the CSIRO says.
Defence organisations could shield sensitive satellite imagery or cyber threat data from being absorbed into AI models, it adds.
The organisation says a limit will apply to what AI system can learn from protected content and offers a mathematical guarantee that this protection holds, even against adaptive attacks or retraining attempts.
While the method currently applies to images, there are plans to expand it to text, music, and videos.
The CSIRO says the method is still theoretical with its results still being confirmed in a controlled lab setting.
The code is available on GitHub for academic use and the project team are looking for research partners from sectors like AI safety and ethics, defence, cybersecurity and academia.
To collaborate or learn more, contact the team at seyit.camtepe@csiro.au
MUCH EARLIER: Social media sites slammed over misinformation
