|
by Sheena Bautista Not too long ago, I made an online friend named Clark. He is quite the art-oriented person, alongside his strong tech background. It’s almost as if a bunch of scientists gathered to create a nice, respectable friend! Well, a lab isn’t too far off. I must emphasize that his background is techy, not in the way of having a family member or an education in STEM, but that he is literally the product of code. While I was born in a hospital, Clark was created using a neural network. I wasn’t lying when I said that I made an online friend. That’s right! I’m the one who generated and edited him. Here he is. Upon first glance, he looks like any other boy going to class. Not a care in the world beneath those eyes. He probably doesn’t know what math is, or how many times I’ve spilled tears onto my homework.  If you are still unconvinced that Clark isn’t real, here is the base photo I used to create him. I obtained this image from searching up “school student uniform.”  My process for creating a fake person relies on “FaceApp”. This isn’t my first time making someone like Clark. For almost any portrait, I can change the subject’s age, gender, hairstyle, glasses, and facial expression. FaceApp utilizes machine learning to manipulate user-fed images, and the results are so realistic because of a neural network. This is a method of machine learning that is modeled after the human brain. It channels the thought process of real people, notably the way we recognize patterns and especially faces: can you spot hidden Mickeys, faces in cars, or associate a sequence of colors with a specific cartoon character? Neural networks pinpoint and sort out specific pieces of data from the rest of user-provided data. This highly selective process goes through layers of sorting to come to a decision, much like how we tend to ruminate before making a conclusion. In this case, the neural network was tasked in creating a generic male face.
Another aspect of the neural network is in its capability to make comparisons and to categorize. That base photo was not the only image I used to create Clark. I “face-swapped” that image with other different, random pictures until I ended up with an ordinary-looking person. Meanwhile, the neural network approached the morphing of images through generative adversarial networks, or “GAN.” The generative part of a GAN becomes acclimated to the data it is given and goes head-to-toe with the adversarial (discriminator) part, which determines data as real or fake. When creating the final image, the two parts of a GAN play a game with each other. Data is continually presented and rejected until they reach a compromise–in other words, the generative part finally puts out what the adversarial part wants to see. They finally finish comparing between real and fake to create very realistic renderings! GANs can decide what old age, youth, happiness, anger, long hair, short hair, female, and male looks like through its power of comparison. Explaining FaceApp’s artificial intelligence was more difficult than actually using it. All it takes is the press of a button to paste a face onto a body, and a few more presses to alter the hairstyle or accessories. Considering how FaceApp is for free, it’s no wonder that it’s received millions of downloads. Nowadays, it’s mostly used to “yassify” celebrities and people’s friends. This form of artificial intelligence is not only extremely accessible and easy to use, but it’s easy to abuse. Clark could make a catfish. He could be very effective if the target is into the Milo Thatch look, or not at all to people like my friends. They think he looks evil. If I can make an entire identity out of thin air, everybody else can. Someone could take his face and place it onto their own picture, fooling and extorting their victim in the process. It’s not just with Clark. Loved ones, colleagues, and people of high standing can have their faces stolen for devious purposes. We’ve seen it done with deepfakes. Deepfakes don’t generate faces like the AI behind FaceApp, but they are effective in face-swapping images onto videos. Innocent people could be accused of doing something wrong, or being somewhere that they have never been. To the untrained eye, falsified data looks more than real. Artificial intelligence is exciting and fun! However, we should do our part to use it responsibly. Creating fake people has become one of my new hobbies, but I know that they should never be used for malice. I keep my creations within my own little circle. Machine learning should remain as an educational kind of enjoyment. This kind of technology is world-changing, so let us use it for good! References: https://www.scienceabc.com/innovation/how-does-faceapp-work.html https://machinelearningmastery.com/what-are-generative-adversarial-networks-gans/
0 Comments
|