DeepMind, a London-based AI company owned by Google, is trialling a digital watermark tool that can detect and label artificially generated content.
Called SynthID, the tool embeds the label into the individual pixels in an image. Humans will be unable to see the mark but computers can detect it and determine that it is AI-generated.
SynthID, which for now is being trialled internally for Google’s own image-generating model, is an attempt by DeepMind to combat disinformation from AI content, a key concern with the rise of the technology.
However, DeepMind cautioned that its watermark system is not “foolproof against extreme image manipulation”.
While generative AI tools with the capacity to create realistic images have only been publicly available for a short time, there have already been hoaxes ranging from harmless jokes to political propaganda.
Earlier this year, an image of Pope Francis sporting a designer puffer jacket went viral before people realised it had been created by AI. Around the same time, pictures depicting a fictional arrest of Donald Trump circulated social media, fooling a number of people before it was reported as fake.
AI-powered disinformation has become all the more a significant concern as both the UK and the US approach major elections.
This month, the debate around fake images influencing politics in the UK was further fuelled when a Labour MP shared a photograph of Prime Minister Rishi Sunak pouring a pint of beer next to a disapproving woman. It was later found to have been altered to portray Sunak in a bad light.
Though not done with AI, the edited image sparked an aggressive response, with tech secretary Michelle Donelan taking to Twitter to share her disapproval.
“In the era of deepfakes and digitally distorted images, it’s even more important to be able to have reliable sources of information you can trust. No elected member of parliament should be misleading the public with fake images,” Donelan said. “This is pretty desperate stuff from Labour.”
The detection of fake images generated by AI will likely be discussed at the upcoming AI Safety Summit, which will bring together international leaders and experts to explore a global approach to regulating the technology.
Source: DeepMind tests AI content label amid disinformation fears (uktech.news)
IT Support Limerick PAQ IT
Your Limerick IT support specialists are equipped with the knowledge and experience necessary to handle a range of IT-related issues, from fixing hardware and software issues to providing guidance on network security and data management. Due to their commitment to delivering top-notch service and in-depth understanding of the local technological environment, PAQ IT is in a perfect position to support the region’s thriving tech industry. Contact us here if you have any enquiries.