Alona Fyshe: How to talk to a machine

Alona Fyshe still remembers the first time she discovered that machines can learn. She is well-versed in the subject now, as a Fellow and Canada CIFAR AI Chair at the Alberta Machine Intelligence Institute (Amii) and assistant professor at the University of Alberta.

But as a student interning with a research group at the U of A in the early 2000s, Fyshe was surprised to hear a professor say they had to test the accuracy of a computer model they had built.

"I had never been taught that a computer could do anything other than what you told it. I was like, 'don't you know what the program does? Didn't you write it?' 

‘The idea was mind-blowing to me," she says.

That was the beginning of Fyshe's fascination with machine learning.

After completing her undergrad and an MSc at the University of Alberta, Fyshe went south and joined Google's newly-opened office in Pittsburgh. She spent a few years working as a software engineer before crossing the street -- literally -- to finish a PhD at Carnegie Mellon University. 

Fyshe has long been fascinated with how machines interact with people, particularly when it comes to Natural Language Processing. NLP is a branch of AI research that focuses on how machines process and understand language; if you've ever used a voice command on your phone or used autocorrect, that technology is likely based on NLP.

Fyshe points out that while children do learn some grammar in school, most kids learn how to talk just by listening to others. There are countless unspoken rules, hidden meanings and other tricky particularities in language. Most people seem to learn them naturally. Computers, not so much. 

"Getting a computer to understand what you're saying, it's surprisingly complicated. And it is something humans do so simply," she says.

Fyshe says if we wish to get machines to understand language the way humans do, it makes sense to look to the human brain as inspiration. Much of her work focuses on taking insights from how people process language and meaning and using that data to improve artificial intelligence. But those insights don't just go in one direction -- her research also pushes forward our understanding of how people use and process language. 

After Pittsburgh, Fyshe spent several years working with the University of Victoria before returning to Edmonton to become part of one of Canada’s leading centres for AI research and application. 

"People are often surprised to hear there is so much AI in Edmonton. We have world-leading reinforcement learning people here. That's super cool, and I just wish more people knew about that," she says.

Fyshe will discuss her research and natural language processing in a keynote speech at Amii's inaugural AI Week in Edmonton, May 24 -27.