When a gamer fires up a console, the response is instantaneous. This feature is a hallmark of modern technology, but for Ojasvi Khanna (B.A. ‘22), a senior data scientist at Microsoft XBox, that seamless user experience comes at a cost. As the tech industry navigates the massive energy requirements of the artificial intelligence era, Khanna is working at the intersection of predictive modeling and environmental ethics to help mitigate the impact.
“The ‘always on’ feature is now a unique selling point for all tech devices,” said Khanna. “That has a huge environmental footprint.”
Many companies rely on carbon offsets to meet their commitments to sustainability while waiting for a technical solution. That is where Khanna comes in.
One of her primary contributions at Microsoft was the development of a Long Short-Term Memory model, a type of machine learning designed to predict user behavior. The model analyzes patterns to determine when a specific player is likely to return to their console. By doing so, the device can remain in a deep power-saving mode during inactive hours and switch to "always on" just before the user typically starts a session. Khanna explains this optimization can reduce a device’s carbon footprint by approximately 20% without affecting the player’s experience.
Khanna developed a focus on the intersection of technical rigor and environmental stewardship while a student at UC Berkeley. The data science major’s focus on interdisciplinary problem-solving provided the framework that positioned Khanna to view climate challenges through a computational lens.
Today, her focus is shifting upstream, toward the growing demand on technological infrastructure. Data centers, vast buildings packed with servers, are one of the fastest-growing sources of energy demand in the world. As AI systems, cloud services and digital entertainment expand, so does the load on these facilities.
One of the most pressing environmental challenges posed by rapid data center expansion is water usage. Data centers generate immense heat and require constant cooling, often diverting water from local communities. The water that is available for residents can cause health risks.
"People in these remote towns cannot use their tap water anymore. They need to buy water crates to drink, to cook, to even wash their hands," Khanna said. "If they do, it ends up causing lots of health issues."
While at Berkeley, Khanna competed in a water-cleaning challenge with an environmental engineering team through the student org, Cal Enviro. Her group placed third internationally and she developed a lasting fascination with the science behind water potability and consumption.
Now, Khanna envisions using machine learning to ensure water is used to its maximum potential in data centers before being cycled out, reducing impact on local environments. Still, Khanna is clear-eyed about the limits of optimization.
“At some point, you have to address the core issue: what energy sources are powering these systems in the first place.”
That’s why she’s closely watching the industry’s growing interest in renewables and nuclear power, and thinking about where her skills might be most useful next. She encourages others in the field, and students planning to join it, to consider how their own expertise may be applied to this emerging challenge.
“We’re all part of the problem,” she says, “but we can also be part of the solution.”
For Data Science Undergraduate Studies inquiries:
brandilyn@berkeley.edu
