Does Artificial Intelligence Understand Human Beauty?
Artificial intelligence and machine learning have become all too common these days. One of the applications of these technologies is generating human faces thanks to the use of GANS or generative adversarial networks.
I have been using it to generate images but one day, a light bulb moment came off. What if I let the AI generate faces based on a broad range of human profile archetypes alone and let it figure out what these certain characters might look like?
There are two famous quotations that I need to put in a test:
[1] Beauty lies in the eye of the beholder.
[2] Donโt judge the book by its cover.
So what Iโm trying to find out:
[1] Can the AI picture what a beautiful human being looks like?
[2] Do bad human attributes reflect in human appearance based on the AIโs interpretation?
The Archetypes
With these things already laid out, I will set up human profile archetypes that will be tested. First things first, I will categorize the archetypes into โ ordinary, accomplished, and underworld. The search string to be used in the AI platform will be: [name], [occupation] in [location].
a. Ordinary is self-explanatory as these there are the everyday working people you see every day making an honest living.
Federico Reyes, bus driver in Bacolod
Hilario Nunez, pig farmer in Tagbilaran
Imelda Cruz, flower vendor in Manila
Marites Lopez, storeowner in Cebu
Faisal Muhammad, farmer in Koronadal
James Sandoval, office worker in Baguio
Samirah Alonto, shopkeeper in Davao
Jessa Sebastian, sales clerk in Mindoro
Edgar Francisco, basketball coach in Iloilo
Katrina Martinez, kindergarten teacher in Masbate
b. Accomplished is the type of people who are rich and successful. They are the people that the ordinary ones want to be like.
Freddie Ramirez, bank manager in Cagayan de Oro
Jonathan Quinones, town mayor in Catbalogan
Nadine Tolentino, actress in Manila
Lea Ortega, beauty queen in Batangas
Victor Zamora, police general in Dagupan
Hilario Roberto, Supreme Court judge in Vigan
Jessica Yu, TV news anchor in Pasig
Michelle Montalban, social media influencer in Makati
Lyndon Matador, fitness guru in Quezon City
Kendra dela Cruz, millionaire heiress in Bacolod
c. Underworld is the type of people that does all the bad things. They could be criminals, drug lords, crooked cops, and corrupt politicians.
Roberto Mendez, rebel leader in Cagayan
Alex Valdez, notorious bank robber in Cabanatuan
Rowena Gutierrez, ponzi scheme queen in Butuan
Helen Francisco, illegal recruiter in Guimaras
Roy Montano, convicted murderer in Muntinlupa
Barbara Enriquez, serial killer in Sorsogon
Gregorio Tan, Chinese Triad leader in Manila
Nobuo Ichida, Yakuza leader in Makati
Francine Mercado, adult film performer in Laguna
Maria Balboa, ex-convict in Zamboanga
What platforms are we going to use? I have the option of choosing either neural.love, Midjourney, or DreamStudio to generate fictitious faces based on the archetypes above. But decided to use DreamStudio as it is much faster to generate images than neural.love. Even though Midjourney is the fastest of the three, it has certain limits so I canโt generate more.
Do take note that the profiles here are all made up and any similarities are just coincidences. Faces are all generated via AI.
The Faces
When generating faces, the AI often generates questionable results from blurry images to weird faces with artifacts. The best-looking faces will be gathered and compared. Find the right answers at the end of this article.
The Ordinary
Can you guess the archetype of the people in this photo? What particular clue stands out in every photo? What is the common theme?
It seems that the AI attributes wrinkled and sunburnt skin to everyday, honest, hardworking people. Although some are smiling, the facial expression they exhibited shows a sense of what they do for a living. The AI also generates a background that gives you an idea of their identity.
The Accomplished
Do you think that the personโs success will reflect in how they look in the photo? Can you spot who is the millionaire in the photo?
Their poses remain somewhat similar to the people in the first category even though they are significantly richer and have better social and economic status than the people in the Ordinary.
The Underworld
Does a personโs criminal history show up in the generated image?
It doesnโt to be that way, although there are some images that resemble a mugshot photo taken before someone is put behind bars. They have smug looks on their faces. If you come to analyze it very closely, they can be categorized into either the Ordinary or the Accomplished archetypes.
My Take
All generated images seem to have similar poses in general while there are visible smiles in the first two categories with two out of ten smiles in Ordinary and three out of ten smiles in Accomplished. The last category Underworld all have a mugshot look.
Some characters may belong to other categories as it seems like they are not what they are based on their archetype. Iโm not sure if the AI can fully grasp the varying context beyond what the archetype meant. One example is the Supreme Court judge but he looks like a mob boss. Another one looks like an innocent-looking grandma but was generated under the Underworld archetype.
So you want to know if your answers are right?
What do you think?
In my opinion, AI can now generate better realistic faces yet there are visible artifacts that make it look like itโs a fake face. Although it can add greater detail based on the context it perceives, it will take a few years before they fool human intelligence completely.
If you want to give it a try, generate your own images here.
Need fresh and original content for your website or blog on a regular basis? Subscribe to our Content Subscription Service so you can grow and expand your content marketing campaign right now.