Last month, 24 girls aged 11–17 and their parents attended the fourth annual—and second virtual—“Introduce Your Daughter to AI” event, which is the sixth installment of an educational series hosted by the Women in Computing (WiC) networking group at the US Department of Energy’s (DOE’s) Oak Ridge National Laboratory (ORNL).
On June 24, ORNL research scientists Pravallika Devineni and Olivera Kotevska began the event by explaining how machine learning algorithms, a form of artificial intelligence (AI), “learn” from data to make predictions or categorize information. The girls learned that this process is prevalent in many areas of everyday life, from travel apps that optimize driving directions based on traffic conditions to online retailers that use data from previous purchases to provide personalized shopping suggestions.
Participants, parents, and organizers gathered online for the second year in a row. Credit: Amy Coen/ORNL, U.S. Dept. of Energy
“I would have loved if there was a similar event available when I was growing up,” Kotevska said. “It’s important to introduce kids and teenagers to topics they may not be aware of when their curiosity is high, and they are less likely to be intimidated.”
To illustrate an artificial learning curve, the instructors described a hypothetical algorithm designed to determine whether images include pigs. When asked what characteristics differentiate pigs from other animals or inanimate objects, the girls named defining features, such as curled tails and pointed ears. But because the output will only be as accurate as the training data, participants learned that requiring overly specific criteria, such as the color pink, increased the algorithm’s error rate and caused it to misidentify pigs with other color variations.
Additionally, the instructors explained that algorithms must be trained on robust and diverse datasets to avoid exhibiting bias. Real-world consequences of bias in AI include discrepancies in facial recognition software’s ability to identify individuals regardless of age, race, or gender and variations in voice recognition software’s understanding of different accents.
“We wanted to stress that providing the right kind of data is what makes a model unbiased,” Devineni said. “It is our responsibility as researchers to not blindly trust a model and to use our intuition to recognize when bias might be present.”
Next, the organizers led multiple machine learning activities. Using the teachable machine tool, the girls took pictures of physical objects using their computer cameras and trained an algorithm to automatically classify new images.
Participants also experimented with the sketch-rnn drawing completion tool, a fan favorite from previous years. Sketch-rnn enables users to begin drawing one of more than 100 preprogrammed subjects, which range from cats to cruise ships, and then the built-in machine learning algorithm quickly identifies the selected model and finishes the drawing. The girls also drew numbers one through nine to test the accuracy of a digit classification tool.
Finally, the girls attended a live virtual tour of the nation’s fastest supercomputer, Summit, which is located at ORNL’s Oak Ridge Leadership Computing Facility, a DOE Office of Science user facility.
Katie Bethea—who leads the Computing and Computational Sciences Directorate’s User Access, Outreach, and Communication group—held discussions and answered questions concerning astronomy and climate models, the amount of power needed to operate Summit, and where supercomputers go when they retire. Many participants were especially fascinated by Summit’s closed-loop system in which water heats up while cooling the machine and is then cooled down again before being reused.
“They were so interested in infrastructure concepts like using water to cool a machine because we’re usually taught at a young age not to mix water and electronics,” said computer scientist and WiC cochair Swaroop Pophale. “It was really great to see their curiosity emerge.”
By providing a safe space for girls to learn about seemingly complex topics they might have avoided otherwise, the organizers hope to instill confidence in the next generation of women, cultivate their interest, and potentially encourage them to pursue careers in science, technology, engineering, and math, or STEM, fields.
“An important outcome is that they see AI or software in general as a realistic career choice, and that’s a big deal,” said computer programmer Amy Coen. “By investing in them, we hope to increase the number of women in the professional pipeline.”
WiC has hosted either “Introduce Your Daughter to Code” or “Introduce Your Daughter to AI” since 2016 to teach daughters of ORNL staff members about the world of computer science. WiC will host the AI event again next year, and organizers hope to eventually expand the series to cover additional topics, such as parallel programming.
“It’s important to me to show these girls what a female scientist looks like, especially because many kids have never met a scientist at all,” Devineni said.
Along with Coen, Devineni, Kotevska, and Pophale, the following people also served as event organizers: Andy Berres, Rachel McDowell, and Katie Schuman. Bethea, along with Lauren Davidson and Megan Fielden, also contributed.
UT-Battelle manages Oak Ridge National Laboratory for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science.
