Natasha Jaques BA’12, BSc’12 has come a long way since she first became fascinated by computers as a youngster. Today, after completing a PhD from MIT, she lives in Silicon Valley and works for Google Brain, one of the behemoth’s research teams focused on artificial intelligence
It’s not the most elegant piece of programming code Natasha Jaques has written but, to be fair, it was her first program and she was only seven years old at the time.
“We had this old computer – a Franklin 87 – with a black screen and a flickering green cursor,” she reminisces. Jaques’ father, a lawyer, was always interested in computers, and he passed that fascination on to his children. That first program she wrote allowed the computer to ask questions and react to the responses. If it inquired “How are you?”, and you typed “I’m sad”, it would respond, “What’s wrong?” If it asked your name and you answered with Sam, Jaques’ brother’s moniker, it would tell you, “You stink.”
Jaques and her brother were so obsessed with computers that their parents imposed a timer system – each one got a 30-minute turn on the machine. “The only way we went outside was during the other person’s turn,” she laughs.
Despite her love of computing, Jaques was undecided about her major when she started undergraduate studies at the University of Regina. “I had no idea what I wanted to do. I was taking classes in everything – philosophy, sociology, psychology and, of course, computer science.”
Then a professor suggested that she didn’t have to choose just one discipline or even one faculty. An extra year of university would earn her two degrees: a bachelor of science in computer science and a bachelor of arts in psychology. She was thrilled to follow this path, and it’s one that has garnered her great success in her career. “It’s a bit of an odd combination, but it’s like I’m a special butterfly. Businesses will say, ‘We don’t have a computer scientist and a psychologist.’”
After convocating from the University of Regina with her two degrees in 2012, Jaques earned an MSc from the University of British Columbia in 2014 and a PhD from Massachusetts Institute of Technology (MIT) in 2019.
“MIT was great in terms of connecting you with a ton of opportunities. There were so many smart people to learn from. Through MIT, I was able to meet the top researchers at Google, creating career opportunities,” she says. Jaques interned at DeepMind and Google Brain, with the latter hiring her on as a research scientist in 2019. She’s also currently a postdoctoral fellow at the University of California, Berkeley, where she’s combining her unique background of psychology and computer science to do research in reinforcement learning and affective computing.
Reinforcement learning involves building artificial intelligence (AI) agents that have intelligent capabilities. Affective computing uses computer science, machine learning and artificial intelligence techniques to detect and model human emotions and signals, for example, detecting a person’s level of stress. “I’m interested in artificial intelligence agents that can interact effectively with humans and do complex tasks, making a series of decisions to accomplish a task,” Jaques says.
The research agenda she’s working on at Berkeley is about how to use social learning to improve artificial intelligence. “What sets humans apart from other animals in terms of intelligence is social intelligence. There was an experiment with one-year-old children, where an adult would go up to a door carrying heavy books and would be unable to open the door themselves. The child would automatically open the door for them. They perceive what you are trying to do and want to help you. Apes don’t have those abilities. Learning and co-operating makes us unique,” she says.
“In certain niche areas, such as detecting tumours, AI can be better than a human. They can analyze MRI scans and medical images, and detect tumours with more accuracy than a team of oncologists. But there’s no system that would be able to both detect tumours and play golf or move. In terms of walking and talking like a human, it’s not even close,” she says.
A real-world example of an application for reinforcement learning is autonomous driving cars. An autonomous car can learn to model the behaviour of other cars on the street. “If an ambulance comes, and others are pulling to the side of the road, the autonomous car should do that, too. Artificial intelligence should take cues from other agents and learn from them. That’s an example of why I think my research agenda is important,” Jaques notes.
Another example of reinforcement learning is controlling the switches in a power grid to improve energy efficiency. “Building electrical grid energy efficiency is really cool,” says Jaques, who has co-authored a paper on tackling climate change with machine learning. “Most grids have a combination of carbon-intensive sources and renewable sources. Solar and wind are unpredictable and there aren’t good batteries or storage for them. If we had better modelling of when solar and wind were available and of consumer demand – when everyone is going to turn on their dishwasher, for example – we could have better planning on how to turn on and off resources. By using artificial intelligence, we could save energy and reduce emissions.”
Jaques is also excited about recent developments in artificial intelligence that involve modelling language. “You can input text and ask it to predict what text comes next. If you do that with enough data, the model starts being able to answer questions. I expect that they’ll improve things like machine translation and web searches, and eventually improve things like Siri and Alexa. Far into the future, I hope reinforcement learning will bring improvements in robotics,” she says.
Jaques has received several accolades for her research, including an honourable mention for best paper in 2019 at the International Conference on Machine Learning – one of the top conferences in the field – and a best paper award at the NeurIPS ML for Healthcare workshop. She was also part of the team that received best demo at NeurIPS 2016. In computer science, it’s more prestigious to present at conferences than to publish a paper in a journal of machine learning research. “Computer science moves very fast; things are changing so rapidly, and the conference cycle allows you to publish things faster,” she says. Her work has also been featured in Quartz, the MIT Technology Review, Boston Magazine and on CBC Radio.
She recalls her time at the University of Regina fondly. “It was a wonderful place to learn. I had a good community there. The U of R allowed me to flourish without too much pressure.” It was where she discovered her love of research, being granted her first research opportunity by Howard Hamilton, computer science professor and director of the Laboratory for Computational Discovery. “That experience is the reason I went to grad school. He put me on to machine learning research. In the first year of my undergrad, I was doing genetic algorithms and reinforcement learning. Now I do that full-time.”
Hamilton is a strong supporter of undergraduate researchers. “It’s an important opportunity for them, because they see through at least one small research project and get an idea of what it means. Many students in their first year think that research is something you do in the library. It’s a revelation to them when we’re doing a creative computer program to do something we don’t know how to do yet. It makes a difference in the potential they see in research to learn this early on,” he says.
Hamilton notes that Jaques was “the most forward-looking undergrad student that I’ve encountered. In her era, she was the one who caught my attention for having this long-term plan for her career. She was able to make use of her time and resources because she had a goal. She was always crossing back and forth between computer science and psychology. She had a good plan.” Her plan was very specific: she told Hamilton that she was going to do two degrees in computer science and psychology and then go to MIT to get her PhD. “She’s the only one, halfway through her first year, to have her whole career mapped out.”
In her time at the University of Regina, she also participated in a pilot project called the Supplemental Instruction program, headed by chemistry and biochemistry lecturer Stephen Cheng. Jaques was one of the first undergraduate students he hired. “She was exactly what we wanted – she had the personality, intelligence and people skills we needed,” he says. Students were assigned to a first-year class and attended the lectures. After each class, they would do three sessions per week outside the classroom. “In those sessions, Natasha would teach the lecture, go over examples and get the students to solve the problems on their own,” Cheng says. He was so impressed by her work that he later wrote recommendation letters for Jaques to do her MSc and PhD.
The day she convocated with her undergraduate degrees, the Department of Computer Science hired Jaques to teach a first-year class. Hamilton says, “It’s highly unusual for us to hire someone like that. But she was already providing special tutoring for all the students who took CS110. She went to the lectures for the class each time it was taught, and she was available in a friendly way for the students to talk to. I said, ‘Given you’ve attended the class all the way through four times, I’m sure you can teach it.’ When she taught CS110, she got sparkling reviews from the students.”
That experience will be useful as Jaques applies for faculty positions in the coming year. She plans to continue in the industry as a researcher as well, given that many universities allow professors to work 20 per cent of their time outside the institution.
Teaching will give Jaques an opportunity to make a difference in an area that she’s very passionate about: inspiring more women to go into computer science. “The field is super male-dominated. In my undergrad, my psychology classes were 90 per cent women; in computer science, I was the only woman in a class of 30 to 40 men. When I graduated from the U of R, only 10 per cent of computer science degrees were granted to women. Ten per cent of papers accepted to conferences are by women. I think that’s a tragedy, because computer scientists have a lot of job security. It’s a hot area,” she says, noting that while she was doing her PhD, companies were desperate to hire anyone with expertise in machine learning. “In 2012, there was a revolution in artificial intelligence with deep learning and neural network techniques. Canadian professors were instrumental in this. The industry hired so many profs and salaries skyrocketed. There are so many opportunities in this field; so few women participating makes me very sad.”
Jaques does her part by participating in mentorship programs such as Girls Who Code. “It would be nice if more women would see it as a valid career choice that’s a lot of fun.” She would also like to see more ethnic diversity in the field.
Currently living in California, she enjoys the different flowers blooming each month and the fact that she can go running outside year-round. On the downside, there are “too many people, too much traffic and forest fires. You can go for a run outside in January, but not in August, because the air quality is too bad.” She loves hiking, backpacking and all things outdoors. “When work gets too stressful, I go somewhere with no cell service and bring my Kindle.” With her work at Google and Berkeley being remote because of the COVID-19 pandemic, she can work from anywhere in the world. At the time of the interview for this article, she was working from Mexico on a surf trip.
While she doesn’t know where in the world her career will take her next, there’s always a special place in her heart for Regina. “I love how green it is, how uncrowded it is. When I was home last, I went for a bike ride. There were miles of super-safe, super-green, open parks. If you go biking in Boston, you can’t find a park where you don’t hear cars, and you’re likely to get hit by a car when you’re out biking. In New York, it’s hard to see a blade of grass. Regina is beautiful and I miss it.”