Computer science junior Kathir Gounder spends much of his time in the Engineering East building (Bldg. 20), completing schoolwork and pondering the technicalities of artificial intelligence.
Gounder was enrolled in a graduate level course called Intelligent Agents (CSC 580). With his new skills, he creates art using machines called artificial neuron networks, which loosely resemble the computational powers of the human brain’s neurons.
Gounder gave an example of seeing a cat on the street — the brain’s neurons process the visual image of the cat and give an instinctual understanding of what people are seeing. But, people cannot explain how the neurons achieved that recognition on the most basic microscopic level.
This is where artificial intelligence comes into the picture. It possesses the capabilities to give art lovers more insight into how exactly brains abstract an image from visual stimuli. There is also the potential for artificial intelligence to produce original images of its own — some machines going so far as to mimic the styles of famous artists.
Using artificial neuron networks to carry out tasks that would otherwise be too dangerous for humans to perform is a central goal of the artificial intelligence community, Gounder said. There is also an opportunity to create new art through artificial intelligence analyzing other artists’ styles and then using it to create new original images of its own.
Gounder used artificial intelligence in Spring 2019 to generate original landscape pictures, a result of pitting two types of artificial neuron networks together — the generator and the classifier.
The generator took in millions of inputs from data collected over time and made outputs that shared the same characteristics of that data. In Gounder’s case, this data was hundreds of landscape pictures found on the internet.
Gounder’s generator made images meant to resemble landscapes, and the classifier inspected them to make sure they look like the real thing.
“[The generator] takes in like a random probability vector and outputs an image, and its job is to generate, say, faces of cats that are so good that it tricks the [classifier] into thinking those are real images,” Gounder said. “You basically put these two networks into like a fight.”
And just like human-made art, artificial intelligence art can take on several forms. Computer science graduate student Megan Washburn said she is interested in using artificial intelligence to generate music in video games, but even then, this can also be used to come up with new melodies for composers who are stuck on a song.
“A.I. in music can definitely boost creators’ work,” Washburn said. “For example, we can create an algorithm – like, say I wanted to stay in this key and in this time signature – we can create an algorithm to search that space and find something we might not have, as a composer, thought of previously.”
Washburn said she likes to think of artificial intelligence in music as a co-writer, and indeed the same can be said of artificial intelligence in other situations where humans are standing right beside it, assisting with its functionality.
Computer science professor Franz Kurfess said that while artificial intelligence machines are highly capable learners, their processes are still limited when compared to the learning capabilities of humans.
“The basic principle here is that you give those neural networks a set of examples where you have inputs and the expected outputs,” Kurfess said. “Based on these examples, they learn how to behave in situations that are covered by the range of inputs that you give it.”
Gounder initially became interested in creating art with artificial intelligence after reading several research papers on the subject. He saw his project as an opportunity to bridge the gap between STEM and liberal arts subjects by using code to create something artistic and, in doing so, step out of his comfort zone.
After completing the class in Spring 2019, he said it has increased his confidence in his abilities and expanded his understanding of the broad applications of artificial intelligence.
“It basically had a huge impact because now I feel a lot more confident in the sense that I can take on different subjects, and it obviously enhanced my STEM education,” he said.
His favorite part of the class was taking his artwork to Laguna Lake Park and selling it at the Shabang music festival.
His work, he said, inspired him to branch out into other areas such as languages. Now he works on teaching computers how to understand English.
“There is no reason an art student can’t walk up the street to the computer science building and contribute something, in the same way a computer science student can go to the art or biology department,” Gounder said.