Microsoft-backed AI startup OpenAI said on Tuesday that it will launch a tool that can detect images created by its “text-to-image” generator, Dall-E 3. Currently, there is growing concern that AI-generated content will have a negative impact on this year's global elections.
OpenAI said that in internal testing, the tool was able to correctly recognize images created by Dall-E 3 in about 98% of the time, and can handle common modifications such as compression, cropping, and saturation changes.
OpenAI also said it plans to add tamper-proof watermarks to mark digital content such as photos or audio, and these marks are difficult to remove.
As part of this effort, OpenAI has also joined an industry group including Google, Microsoft, and Adobe, and plans to provide a standard to help track different media sources.
In April of this year, during India's ongoing general election, a false video of two Bollywood actors criticizing Indian Prime Minister Narendra Modi went viral online. Furthermore, in places other than India, including the US, Pakistan, and Indonesia, AI-generated content and deep falsification are also widely spread and increasingly used in elections.
Today, OpenAI and Microsoft also jointly announced the launch of a $2 million Societal Resilience Fund (Societal Resilience Fund) to promote AI education and literacy among voters and disadvantaged communities.
The two companies said that around 2 billion people around the world will vote in democratic elections this year, providing tools and information to help people navigate increasingly complex digital ecosystems and find authoritative resources is more important than ever.