Multinational tech company Google is developing a virtual watermark tool that will help detect images generated by artificial intelligence. SynthID, the watermark built by Deepmind, a subsidiary of Google, will tell whether an image was created with AI or not. When embedded into images, it will be difficult for humans to see the watermark but an AI detection tool can easily spot it.
DeepMind CEO Demis Hassabis hinted that the tool could become a Chrome extension or it could be embedded into the browser. None of the above is concrete for now.
While AI buzz has increased over the past few years, risks of misinformation and other downsides have also surged recently. A few months ago, an image of Pope Francis in a puffer jacket emerged on social media. Despite being discovered to be fake, the revelation further raised concerns about AI image generators and the kind of problems they can cause if used by mischief makers.
Given that AI is rapidly advancing, Google and six other companies (Amazon, Anthropic, Inflection, Meta, Microsoft, and Open AI) struck a deal that would see them work to mitigate the risks of AI technology. Among the things the companies planned to do, introducing watermarks to help internet users detect AI images is the most significant.
With AI-generated images, telling whether they are real or not is an extreme sport. The fact that more and more tools for image generation have surfaced on the internet has not helped the situation one bit. Consider Midjourney, a popular option that currently has 14+ million users.
Although Google’s SynthID hopes to make that easy, DeepMind revealed that the tool may not be “foolproof against extreme image manipulation.” In an interview with the BBC, Pushmeet Kohli, the head of research at DeepMind said that its system makes changes to images in a manner that makes it difficult for the average human to discern the modifications.
He further stated that the watermark remains even after the image is cropped or edited. It is common practice on the internet to crop or edit images that contain watermarks.
“You can change the colour, you can change the contrast, you can even resize it… [and DeepMind] will still be able to see that it is AI-generated”, Kohli added.
Kohli also mentioned that this was an experimental launch, adding that the more people used it, DeepMind had a greater chance to tell how effective the tool was.
Read also: Pixel Fold, android 14 updates, AI announcements expected during the Google I/O 2023 event
Google’s Duet AI is here
Announced at this year’s I/O event, Duet AI, the generative AI tool for Google’s array of workspace tools is now available. Similar to CoPilot, a tool introduced by major rival, Microsoft, Duet AI will assist users with a variety of tasks.
They include creating copy, discovering new ideas, scraping email threads and calendar invites for vital information, and generating images for charts and slides. It works across Google’s offering of Workspace apps like Gmail, Meet, Docs, and Slides. The tool can also summarize discussions held on Google Meet.
Workspace clients can get Duet AI through a free trial office that lasts up to 14 days for 10 users. Google has not announced official pricing for this product, though that detail should be expected soon.
Technext Newsletter
Get the best of Africa’s daily tech to your inbox – first thing every morning.
Join the community now!
