Don’t worry if you got them all wrong! Now let’s look at why it’s so hard to tell the difference between real and fake when it comes to AI generated images.
The “fake images” were actually all deepfakes. Now according to the general internet’s definition, deepfakes are synthetic media that have been digitally manipulated to replace one person’s likeness convincingly with that of another. In other words, deepfakes are where AI is trained to human faces and learns how to replicate the likeness of a face through intense generative methods and Deep Learning.
The deep part of deepfakes comes from Deep Learning. If you’re unfamiliar with Deep Learning, it’s where AI is taught to process data in a similar way to the human brain. Deep Learning models are known for being able to find intricate patterns in text, pictures, audio and more as well as produce accurate conclusions and insights on that data.
So how are these strange deepfakes created? Well, they are reliant on 2 separate types of Artificial Intelligence: Auto-encoders and General Adversarial Networks (GANs).
Auto-encoders are a type of neural network capable of swapping faces, which is crucial for creating deepfakes. The two main parts of auto-encoders are: encoders and decoders. For swapping faces, two autoencoders are used.
A GAN (or General Adversarial Network) on the other hand is simply a type of machine learning model and a popular framework for creating generative artificial intelligence.
According to Alan Zucconi, it’s important to remember that the technology used for deepfaking is not constrained on faces. It can be used, for instance, to turn apples into kiwis.
But deepfakes can be used as a technologically interesting tool (as Zucconi is saying)— or as a malicious weapon. Unfortunately, the latter seems to be the more popular option, with the proportion of fraudulent deepfakes more than doubling from 2022 to (Q1) 2023, according to Sumsub.
If you’re interested in a far more in-depth and expanded look into solely the technology behind deepfakes, I recommend Zucconi’s blog posts.
So now you have a topic to start a conversation on when you have nothing to say — and you’re also enriched with knowledge about AI! Follow me for more stimulating but brief blog posts from which you can learn a lot. Leave a clap and a comment if you found this article helpful. Thanks everyone!
See you in the next one!