Growing up in Romania with two computer scientist parents, Doina Precup hated hearing about computers. "I was actually frustrated because my mom and dad would talk about computer science and I wouldn't understand a thing in the house," she said.
But once she started learning computer science in ninth grade, everything started to click into place. AI always held a particular fascination for her because of her love of science fiction.
"Once I started learning how to design algorithms, I thought it was very cool and creative and I was good at it, so I pretty much decided straight away that's what I was gonna do," she told Business Insider.
Once Precup started pursuing computer science at university she really found her calling: a kind of AI called reinforcement learning.
Read more: Meet the coding prodigy who has Prince Harry and Meghan Markle on her side in the fight to boost women in tech
"It's actually a lot like training animals how to do tricks by offering rewards," is how Precup describes reinforcement learning. "We basically train computers in the same way by allowing them to interact with their environment and providing incentives for succeeding at the task that we want them to do. But basically, the computer is allowed to experiment and learn by trial and error."
Precup has now been a professor at Canada's McGill University for almost 20 years, but a few years ago she got herself a new gig heading up DeepMind's research team in Montreal.
DeepMind was acquired by Google in 2014. It has gained huge acclaim for various projects, including the famous AlphaGo algorithm, which is capable of beating human masters at the ancient Chinese board game Go.
Precup was personally scouted by founder and CEO Demis Hassabis. The two chatted about the job over a Google hangouts call, and Precup took up the role in October 2017.
"I quite enjoyed the academic life, but when the possibility came up to have a DeepMind group in Montreal I thought it was really too good to pass up because a lot of the cutting edge research right now in machine learning is actually happening at DeepMind, especially in reinforcement learning," she said. "So now I do both."
As well as managing a team of dozens in DeepMind's Montreal office, Precup spends her time mathematically designing algorithms to think more like we do.
"I'm interested in making reinforcement learning more efficient, and part of that is having algorithms that learn a little bit more like people do," she said. She does this by placing the AI in simulated environments, such as mazes.
Getting more women into AI at an early age
Precup said that in the field of AI research, intake of female bachelors students is already low, and it drops off even more at post-graduate level. "For research specifically the pipeline just is not big enough," she said.
She believes the answer to getting more women in at undergrad level could lie in alternative courses. McGill University offers joint courses, and she's seen a lot of women come through to computer science that way. "They happen to take the intro to computer science class and if they really like it they will switch into one of these joint programs," she said.
Precup herself runs an AI for good course during the summer, and has seen women come through that program and even spin-out companies from the projects they work on there.
"I really believe there need to be some alternative ways to reach out to people who can feed the pipeline. Not necessarily going through the regular schooling system," said Precup.