What is 'Detect Fakes', a project to investigate how much humans can see through deepfake?

AI's deep learning brings unprecedented technology to humans, such

as instantly generating realistic facial photographs of people who do not exist in the world, creating ultra-high-precision sentences that are indistinguishable from humans. .. However, it is also true that fake photographs that abuse this technology have appeared. The project launched to detect such ' deep fake ' is ' Detect Fakes '.

DeepFakes, Can You Spot Them?

Detect Fakes is a research project launched by members of the Massachusetts Institute of Technology. This project focuses on how different humans and AI are when it comes to 'seeing through deepfake.'

If there is an AI that produces deepfake, there is also an AI that can detect deepfake. However, the accuracy of AI that detects deepfake may not be perfect, and depending on the type of deepfake generated, it may be more accurately discerned by the human eye.

Therefore, Detect Fakes has launched a website to investigate how deep fake humans can distinguish.

When you visit the website , you will see notes and procedures such as 'All data for research will be collected for research purposes'. Click the x mark to close the display.

Then, a man wearing glasses appeared. It is said that you can tell if this is deepfake, so first click the play mark in the middle to play it.

Then, noise ran in the video and the glasses suddenly disappeared ...

At the end, I got a different face.

This is obviously a deepfake, so move the slider to the right, say 'I think it's 100% made' and click 'Submit' to see the answer.

Then, 'This is Fake' was displayed, and it turned out to be a fake. I will distinguish deepfake like this. Click 'Next' to proceed.

The following video shows the scene of a speech by President Joe Biden saying that ' Obamacare (Affordable Care Act) is inadequate.' In particular, the video is not disturbed, and the voice is probably the person's own, so when I check it as 'genuine' ...

It was a fake.

Next, a video by former President Donald Trump saying, 'We have to spend more on school.' This time there was no audio, only video and subtitles. I thought this was also genuine, and when I checked it as 'genuine' ...

It was a fake again.

With this kind of feeling, we will distinguish whether the presented video or text is genuine or fake. In addition, the number of fake and real is half and half. If you are familiar with or familiar with the presented items, you can contribute to the subdivision of the data by checking 'I have seen / hear / read this before.' In the center and clicking 'Submit'. ..

When you proceed to the end, the message 'You have contributed to the misinformation research!' Is displayed.

in Science, Posted by log1p_kr