The MIT Media Lab gas something to show us.
Yikes. The video shows old pictures being brought "to life." Here's Einstein "talking" with a voice barely escaping cartoon German scientist territory, Van Gogh with a mystery accent, and Mona Lisa not actually saying, "It's a-me, Mario!"
While MIT acknowledges the potential for harm of deepfakery and AI animatronics, I'm not sure they really get it. The "technology can be used for positive purposes--to revive Albert Einstein to teach a physics class, talk through a career change with your older self, or anonymize people while preserving facial communication."
I don't really know how that second one could possibly work, and I'll give them the one about anonymizing people--basically AI-supported catfishing so you can build face-to-face trust while still hiding yourself--could have some limited useful applications (could totally reinvent the confessional).
But that first one? No. Because Albert Einstein is not going to teach a physics class. Some programmer, using an AInimated image of a face is going to teach a class. The level of non-understanding teaching here is pretty severe, the notion that it is just delivering some information through a conduit that doesn't even rise to the level of straight lecture. In this class, you won't get Einstein's actual voice, mannerisms, movement--but you'll be encouraged by the software to imagine that you are getting all those things. Can Fake Einstein respond to questions, comments, inquiries, and can he do it with the same incisive intelligence that Real Einstein would have?
Of course not--and that reveals some of the hubris here. These guys aren't imagining using AInimated faces to fill in for a mediocre lecturer of simple material in a low-level course--no, they want us to imagine that they can use software to reproduce one of the greatest minds in human history. Really, they're further ahead with Talking Mona Lisa, who's creating a voice and mannerisms out of whole cybercloth. We're talking about plain old Making Shit Up--but with a computer, so that makes it really cool and legit. Talking paintings that explain themselves might be cute, and at least not give the impression that we are seeing a real thing.
The "product" is marked with a traceable, human-readable watermark "to help prevent its malicious use." Won't do anything to help with well-meaning but not-good use.
I used to joke that I would never retire, but that when I died I'd have my body stuffed and mounted with added animatronics and a library of my old lessons, and I'd just stay in the classroom forever. Now it appears that my vision was too limited--just using my picture, programmers could replace me with AInimation that might not sound, think, or act like me, but would still have my face. There are probably some non-terrible applications of this, but it is still some creepy stuff.
Rather like deceased textbook authors who still manage to turn out new editions.
ReplyDelete