This Person Does Not Exist
Synapse #26: AI-generated Fake News
Here at Synapse we are interested in learning about human nature not only to scratch a curiosity but also to live an examined life. Part of that process, I would contend, is thinking carefully about what is real in our world and how our senses may be limited.
It’s fascinating to me how incredibly intelligent our brains are compared to the rest of the animal world—and yet at the same time how easily we are fooled. For example, humans are notoriously blind to change and have a surprisingly narrow sense of attention. For example, Daniel Simons famously studied inattention blindness when he asked participants to watch a video count the number of players wearing white who passed the ball. In this video, a gorilla walked through the middle of the scene and, surprisingly, went unnoticed by the vast majority of the participants.
Tricking our eyes has been used for decades to great effect in media industries with special effects such as CGI. However, at least up until a couple of years ago, one object that special effects couldn’t fool into believing was the human face. Perhaps this is why even as animation has become photo-realistic, studios such as Pixar don’t even attempt to accurately portray the human face.
Our brains are impressively adapted to recognizing and reading faces as it is such a critical part of our social life as a species. A couple of years ago, even the most expensive Hollywood studios using cutting-edge technologies and the world’s experts have been unable to fool anyone with a synthetically altered face. Take Netflix’s The Irishman, for example, which famously spent millions of dollars and a lot of man hours to ‘de-age’ its characters:
Even if you can’t exactly describe it, I think most people would be able to immediately tell that the face on the right has been altered in some way. Traditionally, the human face has been extremely difficult to fake without someone noticing.
That is… until now.
Synthetic Media and Deep Fakes
Recently… very recently… there have been advancements in Artificial Intelligence technologies that have gotten good enough to fool our senses. If you navigate over the thispersondoesnotexist.com and refresh, each time you will be presented with what looks like a portrait photograph of a random person.
Except that every picture is not actually of a real person and was instead synthetically generated by AI with perfect precision.
I really challenge each of you to navigate to that link and find a photo that looks not quite right to you.
Clearly, fakery can be accomplished easily with still photographs, but what about video?
Well, remember how I wrote above that The Irishman painstakingly de-aged its characters using a top Hollywood studio and a huge budget? Well, just recently a single Youtuber took just 7 days and some free software to do a better job than Netflix at fooling us—all thanks to the power of modern AI. Check out the video below:
The shocking thing about this technology is that it has already been democratized for free and will continue to be widely available now and in the future. According to Nina Schick on the Making Sense podcast, AI-generated media will soon be good enough that anyone in their basement can—for free—make content as realistic as the most prominent Hollywood studios.
Soon, ‘fake news’ will mean more than biased or flawed journalism—it could mean the propagation of false video and audio that sounds indistinguishable from the real thing.
If you haven’t been convinced yet, here is an AI generated video of former president Barrack Obama:
It’s easy to see how this could be problematic for society—especially one with social media and algorithmically propagated content. Social media has already made it difficult to know what is the truth. Just a couple weeks ago we had a coup attempt in our nation’s capital. I shudder to think of the content of these peoples’ information diets and the misinformation that they have been exposed to that compelled them to storm the capitol building. Just imagine how many more people would have been hurt had a fake video of Trump calling for violence been released online. Our information ecosystem is already horrendously corrupt by bad actors such as foreign governments and I fear the impact of this technology on our society.
This recent phenomenon is what Nina Schuck calls the “Infocalypse” and I encourage all of you to read her very accessible and incredibly vital book on the coming impact of these ‘Deep Fakes’ on our society.
🙏 A Small Favor
Did you enjoy this week’s edition of Synapse? Forward it along to even just one friend and encourage them to subscribe—it would mean so much to me!
🔗 Links
💡 Yes, there is such a thing as an ‘objective truth’
Here, H R Venkatesh writes a thoughtful essay that I think is closely related to this week’s topic. I also encourage each of you to check out his newsletter Media Buddhi which is about “how to stay safe and sane in a world of information overload–one idea at a time.”
↘️ Misinformation dropped dramatically the week after Twitter banned Trump and some allies
Speaking of misinformation, here is some evidence that our information ecosystem has become a little less corrupt following the removal of one of its largest and most powerful sources.
🎮 We wouldn’t be able to control super intelligent machines
Speaking of losing control to AI, here computer scientists show it would be fundamentally impossible for humans to control a super-intelligent AI.
🐦 Tweet of the week
⚡️P.S. If you're new here and want to read more of the Synapse Newsletter each Sunday, subscribe below!⚡️
ThisPersonDoesNotExist.com are sometimes good, but usually still easy to detect as fakes. If you were just to glance at the image, maybe 50% wouldn't be obvious fakes. The others scream out. For one, some 'faces' appear with lots of amusing deformities. But even the relatively good ones have magic melting eyeglasses, impossible jewelry, or clothing that merges into hair. Their algorithms generally don't get foreheads or neck hair right yet. But is it impressive? Yes.