Tucked away on the fourth floor of the University of Wyoming Engineering Building, students are working toward the future — artificial intelligence.
Jeff Clune came to UW in 2013 with a plan to create a new lab specific to advancing computer intelligence. The assistant professor of computer science quickly created the Evolving Artificial Intelligence Laboratory, which has produced various world-renowned research and ideas.
“One of things we’ve done a lot of work on in my lab, where we’re one of the labs on the forefront, is trying to understand how networks operate internally,” he said.
“Once a network has learned to do something, like classify deer and giraffes, you can try and understand how it’s doing that. We call it ‘AI neuroscience,’ because it’s actually very similar to what neuroscientists do.”
Arash Norouzzadeh began working on a program meant to identify wildlife in photos from camera traps in Africa.
“We have a data set that is pictures that are unlabeled, and we train (the program) to learn,” he said. “For testing time, we just leave these pictures and let it try and make predictions.”
The identities of animals are not integrated into the program — it learns to identify the animals itself. The program spent five days analyzing 50,000 wildlife photos in Mount Moran, UW’s supercomputer.
More than 90 percent of the time, the program can correctly identify the animal, anything from a wildebeest to giraffe.
Such a program can be tweaked and changed to instead identify and count elk and antelope herd migrations. A series of camera traps are used to track their movements, although it is very time-intensive.
“Someone has to sit down and just have the worst job of all time — look at 5,000 pictures of elk and count them or identify if they’re male or female,” he said. “We’re quite literally going to train the computer brain on the African images of animals and, just like a human, once you learn one skill, you’re better at applying it to another skill.”
Clune submitted a grant application to the National Science Foundation to fund the project but has yet to get approval.
“We’ve actually decided, because we’re so excited about the project, we’re just going to go ahead and work on it anyways and hope the money comes in later,” he said.
Clune showed off the program’s capabilities during Gov. Matt Mead’s visit earlier this month. Using a webcam, the computer successfully identified Mead’s phone, watch and belt buckle, although it said gun holster a couple times before getting it right.
In addition to Norouzzadeh and his work, five other full-time Ph.D. students work on their own individual projects, varying from the photo identification program to finding and modulating specific parts of a program — for example, keeping the command and information to make a robot take one step forward separate from the turn right command instead of shuffled together. Such research can make creating complex robotics programs significantly faster and easier.
While Ph.D. students normally occupy the lab, many more master’s students and undergraduates get involved during the school year.
A chess board topped with 3-D printed pieces sits in the corner of the lab. The black pieces are used by people — an orange actuating arm plays the white side. Currently, the robot can only moves the pieces, but that could change if some students decide to take the challenge, Ph.D. student Joost Huizinga said.
“We need some students to work actively with the robots,” he said. “I’m really hoping some of our AI students would like to take on the challenge. It’s always more difficult, because working with hardware takes more time, but it shouldn’t actually be too hard, I think.”
Clune is proud of the students and knows they will be successful after graduation, regardless if they go to academia or the private sector.
“One of the top three technology companies in the world (is) asking to hire our students,” he said. “There’s plenty of demand. Everybody knows this is the future and is trying to hire as much talent as possible. There’s been $50 billion in venture capital invested specifically in the type of artificial intelligence (we’re working on).”
This strong demand can actually be a boon to Wyoming, Clune explained.
“People can work remotely from their favorite town in Wyoming instead of moving to one of these large companies,” he said. “Wyoming is very forward-thinking in bringing very strong internet connectivity to the state. Especially because there’s so much demand, if a company wants you, you can say, ‘I want to work at home.’”