meta_pixel
Tapesearch Logo
Log in
Science Quickly

Are You Better Than a Machine at Spotting a Deepfake?

Science Quickly

Scientific American

Science

4.31.4K Ratings

🗓️ 15 March 2022

⏱️ 12 minutes

🧾️ Download transcript

Summary

New research shows that detecting digital fakes generated by machine learning might be a job best done with humans still in the loop. 

Transcript

Click on a timestamp to play from that location

0:00.0

This is Scientific Americans' 60-Second Science. I'm Sarah Vitac.

0:14.0

Early last year, a TikTok of Tom Cruise doing a magic trick went viral.

0:19.0

I'm going to show you some magic. It's the real thing.

0:24.0

I mean, it's all the real thing.

0:34.0

Only it wasn't the real thing. It wasn't really Tom Cruise at all. It was a deep fake.

0:41.0

A deep fake is a video where an individual's face has been altered by a neural network to make an individual do or say something that that individual has not done or said.

0:53.0

That is Matt Groh, a PhD student and researcher at the MIT Media Lab.

0:59.0

Just a bit of full disclosure here. I worked at the Media Lab for a couple years and I know Matt and one of the other authors on the research.

1:07.0

There's a lot anxiety and a lot of worry about deep fakes and our inability to, you know, know the difference between real or fake.

1:16.0

But he points out that the videos posted on the Deep Tom Cruise account aren't your standard deep fakes.

1:22.0

The creator, Kris Ume, went back and edited individual frames by hand to remove any mistakes or flaws left behind by the algorithm.

1:31.0

It takes them about 24 hours of work to edit each TikTok. This makes the videos eerily realistic.

1:37.0

But without that human touch, a lot of flaws show up in algorithmically generated deep fake videos.

1:43.0

Being able to discern between deep fakes and real videos is something that social media platforms in particular are really concerned about as they need to figure out how to moderate and filter this content.

1:54.0

You might think, okay, well, if the videos are generated by an AI, can't we just have an AI that detects them as well?

2:02.0

And the answer is kind of yes, but kind of no. It's actually a fairly complex task.

2:08.0

And so AI is getting really good at a lot of specific tasks that have lots of constraints to them.

2:14.0

And so AI is fantastic at chess, AI is fantastic at go, AI is really good at a lot of different medical diagnoses.

2:22.0

Not all, but some specific medical diagnoses, AI is really good at.

2:26.0

But video has a lot of different dimensions to it.

2:30.0

But a human face isn't as simple as a board game or a clump of abnormally growing cells.

2:35.0

It's three-dimensional varied. It's features create morphing patterns of shadows and brightness and it's rarely at rest.

...

Please login to see the full transcript.

Disclaimer: The podcast and artwork embedded on this page are from Scientific American, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of Scientific American and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.

Copyright © Tapesearch 2025.