Outreach spotlight: Artist versus The Machine

GSL/CIRES scientist Kirana Bergstrom led a Jupyter Notebook tutorial on Artificial Intelligence (AI) for 8th grade students called” Artist vs The Machine: Can You Beat an AI at Pictionary?” She led three in person sessions, each an hour long, for a total of more than 100 students.

Kirana is part of a small team at the forefront of AI science at GSL. Because of her passion for STEM, Kirana built her own image classification AI named DAISY (Doodle-classifying Artificial Intelligence for Science-Y stuff) and trained her to play a simplified game of Pictionary. She trained DAISY with drawings of weather (sun, rain, hurricane, lightning, etc.) from Google’s Quick, Draw! dataset.

Kirana first had the students draw a sun using sketchpad.io, then downloaded the drawing into DAISY to see how she classified it. It was an easy match. She then had them draw a hurricane. Most people don’t know how to draw a hurricane, so DAISY identified the drawing as a tornado instead. During the tutorial she talked through some basic AI terminology and ideas. DAISY is purposely trained to be only 80% effective to feed the conversation about interesting problems and concepts that come up when training such an AI. Kirana’s clever use of technology engaged students. At the end Kirana led discussions about how image classification AIs like DAISY can be used in meteorology for cloud type identification, and smoke vs cloud identification.

OAR Boulder is exploring ways to bring this activity to students virtually. It is currently available to run in a browser, by launching from Google Colab, using the link above. Feel free to play around with the tutorial and send your feedback to Kirana!