Facebook, Microsoft, and other companies launch Deepfake Detection Challenge

2019-12-12

The proliferation of deepfake counterfeits, which utilize artificial intelligence algorithms to replace individuals' portraits in existing images, audio recordings, or videos with those of other individuals, is rapidly increasing. Deeptrace, a cybersecurity startup based in Amsterdam, reported in its latest statistics for June and July that there were 14,698 deepfake videos, surpassing the 7,964 recorded in December last year, representing an 84% growth in just seven months. This is undoubtedly a concerning development, as deepfakes could be used to manipulate public opinion during elections or falsely implicate individuals in crimes they did not commit. Furthermore, this technology has already been exploited to produce explicit materials and defraud companies of billions of dollars. To combat the spread of deepfakes, prominent internet companies such as Facebook, Amazon, Microsoft, and their artificial intelligence partners, along with renowned academic institutions like Cornell Tech, MIT, and Oxford University, will launch a global adversarial challenge at the NeurIPS 2019 conference in Vancouver this week. This is not Facebook's first involvement in similar initiatives. According to available data, the company has invested over $10 million to encourage participation in such competitions. As for AWS, it will contribute up to $1 million in service credits and provide participants with their choice of models. Google's Kaggle platform for data science and machine learning will host the challenge and leaderboard. Other companies will also provide support in their respective areas of expertise. In short, they are putting their money and resources where their strengths lie. In a publicly released blog post, Facebook's Chief Technology Officer, Mike Schroepfer, stated, "Deepfake technology can present AI-generated realistic-looking videos of people doing fictional things, which has significant implications for determining the authenticity of online content. However, the industry lacks a good dataset or benchmark to detect them. Our hope is to develop a technology that enables everyone to better discover and identify AI-altered videos." Starting today, registered participants can submit their code to a black-box validation environment for evaluation once they have completed their designs. While participants are not required to share their models to participate, they must agree to make their work open-source, as this is a fundamental prerequisite for joining the challenge. Of course, high-performing participants will be rewarded.

RECENT AI NEWS

RECENT AI TOOLS