Adobe and UC Berkeley Develop Deepfake Detection Technology

After last week’s viral deepfake video of Facebook (NASDAQ:FB) CEO Mark Zuckerberg on the day of the House Intelligence Committee’s hearing on artificial intelligence (AI), questions are circling around the future of misinformation.

The video surfaced a month after Facebook refused to take down a deepfake video of Nancy Pelosi. US President Donald Trump later posted this deepfake of Pelosi on his Twitter (NYSE:TWTR) feed.

The video of Zuckerberg was posted on Facebook’s Instagram platform. “We will treat this content the same way we treat all misinformation on Instagram,” said a spokesperson for Instagram in an email statement to tech journalists.

Deepfake is a technique that applies AI technology to mirror human behavioral attributes. Deepfake technology, originally created in 2017, superimposes an image, for example of Pelosi, on another human subject to create a strikingly similar image.

On Friday (June 14), Adobe (NASDAQ:ADBE) published a blog post discussing the technology that it is developing to essentially counter deepfake technology. In partnership with UC Berkeley, Adobe is working on a program to detect when images have been altered. Adobe and UC Berkeley are feeding data into a convolutional neural network that gains a greater understanding of altered images.

A convolutional neural network is an algorithm that is used specifically for image recognition. The deep learning algorithm ascribes varying weights to different image aspects and learns to distinguish them by being trained by vast amounts of data.

“We started by showing image pairs (an original and an alteration) to people who knew that one of the faces was altered,” said Adobe researcher Oliver Wang in the post, “For this approach to be useful, it should be able to perform significantly better than the human eye at identifying edited faces.”

The partnership is working within Photoshop’s Face Aware Liquify feature. The project aims to equip individuals with technology that can detect misinformation and authenticity. Adobe is working towards democratizing image forensics.

Adobe and UC Berkeley have reported 99 percent accuracy in deepfake detection using their deep learning technology. In contrast, just 53 percent of deepfakes were identified by participants.

“Beyond technologies like this, the best defense will be a sophisticated public who know that content can be manipulated — often to delight them, but sometimes to mislead them,” said Gavin Miller, head of research at Adobe.

Don’t forget to follow us @INN_Technology for real-time news updates!

Securities Disclosure: I, Dorothy Neufeld, hold no direct investment interest in any company mentioned in this article.