 "OpenAI")
By Shirin Ghaffary
OpenAI is rolling out a series of initiatives to prevent its products from being used for misinformation ahead of a major year for US and global elections.
On Monday, the artificial intelligence startup announced new tools that will attribute information about current events provided by its chatbot ChatGPT, and help users determine if an image was created by its AI software. The changes comes as concerns rise over the risks of so-called “deepfake” images and other AI-produced content that could misguide voters during campaigns.
“Protecting the integrity of elections requires collaboration from every corner of the democratic process, and we want to make sure our technology is not used in a way that could undermine this process,” the company wrote in a blog post on Monday.
The startup will start encoding images produced by its Dall-E 3 image-generator tool with provenance information — which refers to data about the origins of a piece of content, such as who produced the image and when it was made. That could help voters better understand whether images they see on the web are made with AI. The process will use a cryptographic standard established by the Coalition for Content Provenance and Authenticity, or C2PA. Adobe Inc., Microsoft Corp., Intel Corp and a handful of other companies co-founded the coalition in 2021.
OpenAI also said it will release an image-detection tool that people can use to look up whether an image was generated by Dall-E. The company will first roll out the tool to journalists, platforms and researchers for feedback. Historically, many detection tools that try to determine if specific images or writing were created by AI have struggled with accuracy. However, OpenAI Chief Technology Officer Mira Murati said in October that the image-detection tool the company has been working on had 99% accuracy in internal testing.
For ChatGPT, OpenAI will enable people to get real-time information about current events using the chatbot. Results will provide attribution and links to articles, features it currently doesn’t show by default in the app. The company said it’s “increasingly integrating with existing sources of information.”
More From This Section
OpenAI is in talks with dozens of media companies to make content licensing deals, including major media outlets like CNN, Fox Corp., and Time, Bloomberg previously reported. The startup has already reached agreements with Axel Springer SE and the Associated Press.
“As with imagery, transparency around the origin of information can help voters better assess information and decide for themselves what they can trust,” the company said in the post.
