Mar 23, 2022 21:00:00

AI-generated super-sophisticated fake data deepfake detection platform 'Reality Defender'

Reality is a platform that detects

deepfake, which is fake images, videos, and sounds created by artificial intelligence (AI), built by the world's top AI research team based on more than 100 million video, image, and voice assets. Defender '. Users can simply scan their assets through a web app or API provided by Reality Defender to detect if they are mixed with deepfake.

Reality Defender
https://www.realitydefender.ai/

Deepfake first became a hot topic at the end of 2017. A deepfake porn movie of a world-famous actress appeared on a porn site and became a hot topic as 'fake porn'. However, fake porn was immediately regulated, and even Reddit, where fake porn was born, was banned from posting just three months after its appearance.

A programmer who creates pornographic videos of famous actresses with AI has appeared, and it looks like a real perfection --GIGAZINE

Since then, the evolution of AI has been remarkable, and tools and services for creating fictitious monsters , fictitious look-alikes , anime beautiful girls , etc. using deepfake have appeared. It has also become clear that researchers at Microsoft , Facebook , and the Massachusetts Institute of Technology have created technologies to counter these deepfake.

Reality Defender is a platform to counter such deep fake. Reality Defender is a government agency such as the US Department of Homeland Security , the US Department of Defense , the US Army , and the United States Secret Service , news media such as ABC.com , The Washington Post , WIRED , and ProPublica , and companies such as Microsoft , LinkedIn , Facebook , and BOSE . It seems that these organizations will also help determine the authenticity of the data.

Reality Defender developers Ben, Gaurav, and Ali said, 'The latest advances in machine learning create images, videos, and sounds that people have never said or said. With the latest spread of this technology, anyone can create very realistic deep fake. Some deep fake is for experienced observers. It's possible to tell, but it's difficult to tell because many people haven't had that experience or aren't always looking at it, which means that if a malicious person uses deep fake, it distorts reality. , Financial transactions and personal and brand reputation, public opinion, and even national security can be compromised, 'says Hacker News .

Reality Defender developers Ben, Gaurav, and Ali are data scientists from Harvard University, New York University, and the University of California, Los Angeles, including Goldman Sachs, Google, CIA, Federal Deposit Insurance Corporation, and Defense. He has decades of experience in the fields of machine learning and cybersecurity, including at the Ministry of Science and Harvard University Applied Research. The three initially developed a way to create deepfake from the iPhone using deep learning , which was easier and more sophisticated than expected, so 'we started to create a model that distinguishes between real and deepfake.' It is written as.

However, by working on a model that detects deepfake, the developers write that 'it turns out that it's very difficult to develop a model.' He added, 'Companies like Facebook and Microsoft were trying to build a single model that could be called a silver bullet , a'single model that could solve everything.'In our view, deep. This approach doesn't work because the underlying technology of fake is constantly evolving, just as there is no single model to solve the malware problem. ' It emphasizes the complexity of the model for detection.

The three developers claim that a 'multi-model' approach that aggregates deepfake detection algorithms is needed to solve this problem. Ben, Gaurav, and Ali have combined the highest-scoring deepfake detection models with models created by partner organizations such as Microsoft and the University of California, Berkeley to build a scalable solution. , It seems that this has been made available as a web application. In addition, through this web app, 'disinformation from Russia to Ukraine and Europe and the United States', 'fake voice requesting wire transfer by bank executives', 'video of Malaysian government leader's scandalous behavior', 'fake pornography' It is said that it succeeded in detecting such things.

In addition, in order to respond to the evolution of deepfake technology, it states that it will incorporate third-party models and adjust them so that they can be integrated into the model stack.