MAR24
What are some ways to spot deepfakes?
In the near term, you can still often trust your instincts about deepfakes. The mouth moves out of sync with the body, or reflections are at a different frame rate, etc.
In the medium term, we can use deepfake detection software, but it's an arms race, and the accuracy will likely decline over time as deepfake algorithms improve.
In the long term, deepfakes may eventually become indistinguishable from real imagery. When that day comes, we can no longer rely on detection as a strategy. So, what do we have left that AI cannot deepfake? Here are two things: physical reality itself and strong cryptography, which is about strongly and verifiably connecting data to a digital identity.
Cryptography is what we use to keep browsing histories private, passwords secret, and it lets you prove you're you. The modern internet could not exist without it. In the world of computation, AI is just an algorithm like all others and cryptography is designed to be hard for any algorithm to break.
We are still able to link a physical entity (a person) to a strong notion of digital identity. This suggests that 'is this a deepfake?' may not be the right question we should be asking.
If 'is this a deepfake' is the wrong question, what is the right one?
The right questions to ask are: Where is this image coming from? Who is the source? How can I tell?
The sophistication of deepfakes may eventually evolve to the point where we can no longer distinguish between a real photo and an algorithmically generated fantasy.
In this world, the focus of the conversation should be less on the content of the image but on where it came from, i.e., the source, the communication channel, the medium. In that sense, Marshall McLuhan's old wisdom that "the medium is the message" still applies—perhaps now more than ever.
https://techxplore.com/news/2024-04-deepfake-wrong.html
