As experts warn that ai-generated images, audio and video could influence the fall election, OpenAI is launching a tool designed to detect content created by its popular image generator, DALL-E. But the prominent artificial intelligence startup recognizes that this tool is only a small part of what will be needed to fight so-called deepfakes in the months and years to come.
On Tuesday, OpenAI said it would share its new fake detector with a small group of disinformation researchers so they could test the tool in real-world situations and help identify ways it could be improved.
“This is to kick-start new research,” said Sandhini Agarwal, an OpenAI researcher who focuses on security and policy. “That's really necessary.”
OpenAI said its new detector could correctly identify 98.8 percent of images created by DALL-E 3, the latest version of its imager. But the company said the tool was not designed to detect images produced by other popular generators such as Midjourney and Stability.
Because this type of deepfake detector is based on probabilities, it can never be perfect. So like many other companies, nonprofits, and academic labs, OpenAI is working to combat the problem in other ways.
Like tech giants Google and Meta, the company will join the steering committee of the Coalition for Content Provenance and Authenticity, or C2PA, an effort to develop credentials for digital content. The C2PA standard is a kind of “nutrition label” for images, videos, audio clips and other files that shows when and how they were produced or altered, including with ai.
OpenAI also said it was developing ways to watermark ai-generated sounds so they could be easily identified in the moment. The company expects these watermarks to be difficult to remove.
Anchored by companies like OpenAI, Google and Meta, the ai industry is facing increasing pressure to account for the content its products generate. Experts are calling on the industry to prevent users from generating misleading and malicious material and to provide ways to trace its origin and distribution.
In a year packed with major elections around the world, calls for ways to track the lineage of ai content are growing increasingly desperate. In recent months, audio and images have already affected political campaigns and voting in places like Slovakia, Taiwan and India.
OpenAI's new deepfake detector may help fix the problem, but it won't solve it. As Ms. Agarwal put it: In the fight against deepfakes, “there is no silver bullet.”
