Hey so what’s going on with deepfakes these days? DARPA was doing some really interesting research on them last year, but I haven’t heard much about it since then...

Launched in Sept. 2019, the two-year DARPA project — called the Semantic Forensics (SemaFor) program — will “develop & train an algorithm that should ultimately be able to identify 5,000 fake items hidden among 250,000 news articles and 250,000 social media posts.”

Also, DARPA just bought a $1 million AI tool meant to “decode and predict the emotions of allies and enemies,” with the ultimate goal of “advising generals on major military decisions.”
https://twitter.com/forbes/status/1285990252499304449?s=21 https://twitter.com/forbes/status/1285990252499304449

They’re “operating largely under stealth” but also testing their emotion manipulation/detection software with “various major publications from CNN and Time Inc. to the BBC.”

Oh. Ok then...

What could possibly go wrong with using major media networks to stealthily study the causes of “abject terror”?

Emotion detection for riot control...?

At least one major developer is open to “a more controversial application“ of the AI involving the use of emotion detection to “look out for patterns of agitation in a group of protesters to predict if a riot was about to erupt.”

But don’t worry, it won’t be “determinative” — it will only be used as a “guide” to inform the human operator, and certainly there’s no room for error there...

In all seriousness, this is actually really cool technology from a behavioral science perspective, but it scares me to think about it outside of a lab. And in the wrong hands, it could be absolutely catastrophic — hence the concerns about hacking & intellectual property theft.