Microsoft and Facebook disclosed the Deepfake Detection Challenge (DFDC) to create a technology which could be run to detect “deepfake” videos which are often AI-generated videos of events and people which has been modified to mislead the people.
When does the challenge start?
The challenge is expected to start in a Conference at Neural Information processing Systems (NeurIPS) in Vancouver, Canada. But, other related events will begin on October 2019.
Who created this challenge?
Microsoft, Facebook, the Partnership on AI, and academics from Cornell Tech, University of Oxford, MIT, University of Maryland, UC Berkeley, College Park, and a University located at Albany-SUNY.
What does the challenge consist?
Facebook is pouring around $10 million on this attempt. The challenge consists of:
1. Data sets for videos and deepfakes, which can be later used to rectify and enhance the user experience. Facebook also confirms that it is associating with the third-party vendors to generate a new data set for videos adopting “required and collected the consent of the paid actors”. Later Facebook will create a Subset of “tampered videos” by making use of AI. In this way, no facebook user data will be used in this data set.
2. Providing funding to various research collaborations and boost participation in the event by distributing prizes.
How will the data set as well as challenge parameters examined?
1. On October 2019 International Conference on Computer Vision(ICCV) will be held where targeted trial sessions will run.
2. On December 2019 full data set will be released at NeurIPS.
Who is supposed to run it?
Media Integrity and the association with AI’s new Steering Committee, a cross-sector combination of Microsoft, Facebook, WITNESS, etc.
What is the reason behind it?
Deepfake videos have become very popular as technology has become more and more sceptical. Though previously it seemed that it was only for the experts but, with the popularity of Chinese app Zao that uses only images to deliver a deepfake. Therefore it has been registered that it is no longer an expert’s toy anyone and everyone can use it. It is free and available. How the fake news and wrong information campaigns have spread all across the world, deepfake videos enrage the trouble of verifying a piece of online information.
Prominent deepfake scandals:
- In June 2019, Mark Zuckerberg’s deepfake video was posted on Instagram, where he was found delivering a malevolent speech describing the power of facebook. Instead, in reality, it was an altered video of him describing Russian interference on Facebook from 2017.
- A few days back, US House Speaker Nancy Pelosi was found in an edited video where she was seen drunk, but it wasn’t deepfake video.
- A Reddit user once deepfake a vulgar video of a popular actress by using an open-source machine learning tool called “TensorFlow”.