Is There a Horse Named Intel PR?

November 25, 2022

I noted the information in “Intel Introduces Real-Time Deepfake Detector.” I like the real time angle. The subtitle caught my attention:

Intel’s deepfake detector analyzes ‘blood flow’ in video pixels to return results in milliseconds with 96% accuracy.

Milliseconds.

I am not saying that Intel’s FakeCatcher does not work on a small, curated video, maybe several.

But like smart cyber security technology, a system works because it recognizes what it knows about. What happens when a bad actor (maybe a disaffected computer science student at a forward leaning university cooks up a novel exploit? In my experience, the smart cyber security system looks dumb.

And what about the interesting four percent error rate? A four percent error rate. So if Intel is monitoring in real time the 500 hours of video uploaded to the Googley YouTube, the system incorrectly identifies only 20 hours of video per minute. What if those misidentified videos were discussing somewhat tricky subjects like missiles striking Poland or a statement about a world leader? Not even the whiz kids who fall in love with chatbots bandy about 96 percent accuracy. Well, maybe a whiz kid who thinks a chatbot is alive may go for the 100 percent thing. Researchers often have a different approach to data; namely, outputting results that are not reproducible or just copied and pasted from other documents. Efficiency is good. So is PR.

Let’s take a step back.

What about the cost of a system to handle, analyze, and identify a fake? I think most US television programming is in the business of institutionalized fakery. I can hear the rejoinder, “We are talking about a certain type of video?” That’s okay for the researchers, not okay for me.

The Intel PR item (which may itself be horse feathers or its close cousin content marketing) says:

Intel’s real-time platform uses FakeCatcher, a detector designed by Demir in collaboration with Umur Ciftci from the State University of New York at Binghamton. Using Intel hardware and software, it runs on a server and interfaces through a web-based platform. On the software side, an orchestra of specialist tools form the optimized FakeCatcher architecture.

Ah, ha. Academic computer razzle dazzle. I am not sure if the Intel news release is in the same league as the computer scientist in Louisville, Kentucky, who has published the ways to determine if I am living in a simulation. (See this IFL Science write up.) It is possible that the Intel claim is in some ways similar: Academics and big companies in search of buzz.

Intel’s announcement is really important. How do I know? I learned:

Deepfake videos are a growing threat.

This is news? I think it is a horse named “PR.”

Stephen E Arnold, November 25, 2022

Comments

Got something to say?





  • Archives

  • Recent Posts

  • Meta