In the digital age, information spreads fast but misinformation spreads like wildfire. The ease of communication has enabled the rapid transmission of fake news, and the phenomenon has become so prevalent that it’s taken on proportions of a pandemic – an information pandemic – or infodemic.

The problem escalates manifold when the medium of misinformation is video – something so compelling and seemingly true that even the smartest and most skeptical of minds could be taken for a ride. Known as deepfakes, these are AI-generated videos showing real people doing and saying fictional things.

Deepfakes is a combination of two words: deep learning and fakes. It is an emerging cyber threat. They are AI-generated or AI-manipulated videos that can make people say and do things that they never actually said and did. General Adversarial Networks (GANs) are commonly used to generate deepfakes. Deepfake creators have many techniques to choose from: Some may superimpose a target’s face onto existing footage, while others create completely artificial faces, while a newer variant, Deep Video Portraits (DVP) pushes the stakes higher by allowing video creators to create more lifelike, almost-undetectable footage by recreating the facial expressions of one person using the likeness of another person.

At present, deepfakes are growing at an alarming rate. There are 145k videos currently in circulation. Fake news spreads 6x faster than real news. There is 6820x annual growth in the volume of deepfakes since 2019. There are 27k porn deepfakes. Deepfakes are being weaponised by bad actors for various reasons such as revenge, public shaming, extortion, manipulating voter behaviour, and to trigger international conflict.

To combat this infodemic, undergraduate students from G.H. Raisoni College of Engineering, Nagpur, Atharva Peshkar, Atharva Khedkar, Rishita Mishra, and Yash Moharir has developed Detectd, an AI-based deepfake detection platform that will allow anyone to verify the authenticity of the media they consume. 

“Detecting deep fakes used to be a very challenging task for individuals because it required up expertise in artificial intelligence domain. But what we have done is that we've created this platform where the user can simply visit our website, upload a media file from his local system, whether a photograph or a video and the video is passed through our state of the art model," explained Atharva Peshkar. "The media is then analysed for AI forgery to find out whether the video is fake or not, and the results are returned to the user within minutes."

“As a result, we have reduced the defect detection time from about 15 to 20 minutes to just about three minutes at the max”, he added.

A major issue with most of the deepfake detecting solutions that are presently available in the market is their unreliability and the slow pace at which results are given. These solutions mostly use older technology while the technology to create deepfakes is advancing rapidly. Detectd currently has an accuracy of 96% and is the first of its kind in India and amongst very few available globally and at present it can provide results from as low as 3- minutes. Last year Detectd emerged as the World Finalists and India Champions in Microsoft Imagine Cup 2021.

As of today, the annual cost of disinformation is around $78b. It can be a big problem for governments and democracies and more advanced and accessible tools are needed to fight off this current wave of infodemic.

• Watch the full video interview of team Detectd with INDIAai