Improv and AI: Science grad connects comedy and science

Meet improv comedian and AI researcher Kory Mathewson, one of the newest members in the Faculty of Science alumni family

Katie Willis - 20 November 2019

What do improv and AI have in common? A lot more than you might think, according to new Faculty of science graduate Kory Mathewson. Mathewson, who just completed a PhD in the University of Alberta's Department of Computing Science, is well-known in both the improv and AI scenes in Edmonton for his work with comedy robot Blueberry.

Now. after completing a master's and PhD in the Faculty of Science, Mathewson is headed east to work as a research scientist with DeepMind in Montreal-and check out the city's famous comedy scene, of course.

Hear more from Mathewson on his time at UAlberta and his plans for the future.

Tell us about your experience as a graduate student in the Faculty of Science.

The University of Alberta's Faculty of Science graduate student experience was fantastic. I was completely supported from the first moment I expressed an interest in being a student in the Faculty. It felt as though the professors were excited to enable and accelerate world-class scientific inquiry.

The graduate student experience was further positively defined by the lab groups I participated in, including the Alberta Machine Intelligence Institute (Amii), Reinforcement Learning and Artificial Intelligence (RLAI) group, and the Bionic Limbs for Improved Natural Control (BLINC) lab. Each of these lab groups brought diverse perspectives together in safe, caring environments where collaboration and co-creation toward improving the lives of all people flourished.

UAlberta was the obvious choice for my PhD in computing science. I had done both my undergraduate and master's degree at UAlberta. I appreciated the calibre of the research happening in the Department of Computing Science. The department is able to quickly iterate on research ideas and progress toward deployments in large studies by working with strategic interdisciplinary partners. It was the strong leadership of the Faculty of Science as well as the inspiring work in reinforcement learning which solidified my decision.

Working with Richard Sutton and Patrick Pilarski and the rest of the amazing faculty and colleagues was incomparable. These are world leaders who continue to define entire fields of research for the global artificial intelligence research ecosystem. They instilled in me several important lessons, daily practices, and useful philosophies which helped to support and drive my research. They embodied thought-leadership, and I am humbled to have had a chance to work near to them in my development.

Tell us about your work with Blueberry.

Blueberry is a robot. A robot that performs improvised theatre. Working with a robot in live theatre is challenging, exciting, and always changing. My research focused on how humans and machines can co-create theatrical experiences. Specifically, I showed how both of these intelligent systems augment the capabilities and capacities of the other.

How did you come to think of improv as having the potential as illustrating the power of AI?

The idea came about, as many great ideas do, as I was talking to a mentor of mine over sandwiches and sodas in a cozy restaurant in downtown Edmonton. Adam Meggido (London) pushed me to deploy artificial intelligence on the improv stage as soon as possible, and iterate every year. He said, "It might not be great this year, but you can be sure next year it will be better." The idea was obvious, at least according to the improvisor in me, because great improvisors should be able to perform with anyone (or anything) as a scene partner. So, if I could work well with other improvisors, and even audience members, then why not a machine learning trained dialogue system and emotive robotic platform.

After I performed the show for the first time, I wrote a blogpost: Building an Artificial Improvisor. The post described the failures of the show and my learning from the process. Shortly after I posted it, I was contacted by Dr. Piotr Mirowski, in London, UK. He was working on something similar and was excited to collaborate. The rest, as they say, is history.

What do your studies and your improv work complement one another?

I believe that improvised theatre and artificial intelligence work so well together because they are both based on one simple right: trial-and-error. Both of the fields of thought focus on actual interactions and learning from the success and failure of those interactions. Also, the improvised stage is an ideal test-bed for machine learning systems. They are safe, dynamic, observable real-world domains with complex understanding required for both language and movement generation.

Great scientists are artistic in their thoughts, and great artists are scientific in their pursuits. The artist mind is fueled with creativity, innovation, novelty, and on the collision of ideas. The scientific mind builds with these tools when exploring the means by which hypotheses are tested and the data is collected. Creativity is required in science. It is integral. We have come up with all sorts of ideas of why and how things happen, but creativity allows us to understand variations in variables. Many breakthrough scientific milestones have been creative perspectives applied to existing problems. Most importantly, (at least) half of science is communicating details, and creativity allows scientists to effectively communicate their ideas, experiments, and results.

What's next for you after graduation?

I am now living in Montreal, Quebec working for DeepMind. I also continue developing, producing, and directing my artificial intelligence-based improvisational theatre show Improbotics around the world.

I am working as a research scientist with DeepMind with fantastic researchers. I am humbled and excited to get to work with world-class researchers on building safe AI systems which learn how to solve some of the most complex and interesting challenges.

There is a good number of solid theatre performers in Montreal who were also brought up in the Edmonton improv scene. You better believe that in my free-time I am cooking up some exciting home-grown, Prairie-style improvisation for the east coast.

What's your biggest piece of advice for those thinking about studying AI?

Focus on your why. Why do you want to study AI? Ask yourself that, write it down, and focus on the path to the goal. It is easy to get pulled in multiple directions, especially when a field has a large number of new researchers and exciting research directions. By focusing on your why, you will be able to stay focused, motivated, and inspire others with your vision.

Congratulations, Kory!

Mathewson's graduate supervisor Patrick Pilarski, associate professor in the Faculty of Medicine & Dentistry, along with computing science professors Michael Bowling and Richard Sutton, were the scientists behind DeepMind's first international research lab, opened in Edmonton in July, 2017. DeepMind Alberta was followed by DeepMind's Montreal lab, opened in October, 2017 and headed by Doina Precup from MILA.