Georgia Tech is accelerating the future of electric vehicles using Azure OpenAI Service

When it comes to innovation, the age-old question persists: Is it a tool that drives change, or is it the ingenuity of individuals that defines the tool’s purpose? Maybe it’s the journey that shapes the destination, rather than the other way around.

In the global landscape of electric vehicle (EV) adoption, one question endures: What constitutes the elusive “if you build it, they will come” moment? Is it the vehicles themselves, or is it necessary to first build public trust in electric vehicle charging infrastructure? This is the challenge Dr. Omar Asensio, an Associate Professor and Director of the Data Science and Policy Lab at the Georgia Tech who currently serves as a Visiting Fellow with the Institute for the Study of Business in Global Society (BiGS) at the Harvard Business School, has been working to solve.

Utilizing Microsoft’s Azure OpenAI Service, Asensio was able to automatically discover EV charging behavior in 72 languages globally. By partnering with Microsoft, Georgia Tech’s Institute for Data Engineering and Science (IDEaS) created the Cloud Hub to advance the use of Azure for research and education. That partnership allowed Asensio and his team to unlock valuable insights using Azure OpenAI Service to help drive policy development and EV charging infrastructure enhancements.

Asensio’s work spans 10 years, addressing climate change by devising solutions to curtail greenhouse gas emissions with technology and behavioral strategies. With a primary focus on accelerating electrification in the transportation sector, his research has coincided with advancements in EV battery technology and increased vehicle affordability. However, a major roadblock for many prospective EV adopters is the lack of charging stations.

“The question is: ‘How do we understand, at a large scale, ways to make it easier for consumers to have confidence in public infrastructure?’” Asensio said. “That is a major issue holding back electrification for many consumers.”

A decade ago, electric vehicle sales barely accounted for 1% of the total new vehicles sold. Fast forward to 2023, and EVs make up a remarkable 14% of all new car sales worldwide. With EV infrastructure poised to attract a staggering $100 billion in investments by 2040, electrification is witnessing an unprecedented surge in both public and private funding. This electrification strategy, embraced by major automakers, has prompted the need for insights into the consumer charging experience.

In his work to better understand this experience, Dr. Asensio’s team found an active community of EV drivers who are connected through a variety of apps and platforms, creating a treasure trove of unstructured data. This community is so vocal it has its own lingo. The volume of information was substantial, but it was estimated that human experts would require 99 weeks to extract the salient data points, which wasn’t realistic. Azure OpenAI Service was pivotal in advancing the research.

Asensio started by seeing if AI could understand what factors contributed to negative consumer experiences at public chargers and compared the classifications to responses from an online panel of 1,000, 18-and-older adults in the US. The first experiment produced results that were just above 50%, no better than random chance that the AI could understand negative reviews with domain specific lingo used by EV drivers.

The team then trained the AI by providing context prompts and specific examples (using reinforcement learning from human feedback) to create “superhuman classification” where the expert-trained AI eventually exceeded the performance of human experts. Now, instead of costly government surveys or simulations based on small amounts of data (how most research studies are conducted), “we can just fine-tune expert-trained AI models to detect all these important phenomena happening at a very large scale.”

The collaboration with Azure OpenAI Service has been transformative, saving federal, state, and local governments significant amounts of time and money that would have been needed to get a proper sampling. The ability to conduct extensive research through AI, parsing large quantities of data in minutes rather than months, is a significant advancement for the research community. Using AI offers scalable solutions that ensure quicker results and cost reductions.

In the most recent publication at the AAAI Fall Symposium on Climate Change and AI, Asensio and his team show that generative AI models can effectively extract pricing mechanisms automatically from messy, operator descriptions with high accuracy, and at substantially lower cost of three to four orders of magnitude lower than human curation at USD 0.006 pennies per observation.

“What we’re doing with Azure is a lot more scalable for research,” Asensio said. “I mean, we literally hit a button, and within 5-10 minutes, we had classified all the US data. Then I had my students look at the European data, and then we had a snapshot of performance in Europe with urban and non-urban areas. Most recently, we aggregated evidence of stations across East and Southeast Asia, and we used machine learning to translate the data in 72 detected languages.”

One critical issue with charging stations is reliability. Asensio discovered that networked charging stations exhibit higher reliability compared to non-networked stations. These findings help to identify where government intervention may be necessary to improve reliability, standardize pricing, and enhance interoperability. The goal is to ensure that public investment in communities creates reliable experiences, ensuring government incentives effectively address these challenges.

Asensio’s team is also working to use the machine learning models to predict problems before they happen to ensure a more rapid, real-time response from charge-point operators.

The next step in Asensio’s work, and what he’s addressing as a Visiting Fellow at Harvard Business School, is how federal policies can impact marginalized communities especially those susceptible to climate indicators such as poor air quality and a lack of strategies for workforce development. Research indicates that rural communities and small urban clusters, which typically have limited EV infrastructure, are often deemed “charging deserts” due to insufficient infrastructure investment and charging-station maintenance.

“There’s been talk about creating Community Benefit Agreements (CBAs) that would have provisions like, in exchange for an incentive to install a charger, you agree to share anonymized data about the performance of that charger with some kind of centralized data aggregator, that can help provide more transparency about charger reliability and pricing,” Asensio said.

This might create local requirements and potential incentives, like those faced by gas stations, where charge point operators must provide a reliable service.

“As more and more of the economy electrifies, solutions to extract insights from distributed data sources will become increasingly important to model relationships between energy and transportation systems from an emissions perspective,” Asensio said.

Access and equity are critical next steps in building consumer trust when it comes to EV adoption, he added. While current incentives for new EV purchases are helpful, many potential buyers remain skeptical due to concerns about charging infrastructure.

In the ever-evolving landscape of electric vehicle adoption, the question of what constitutes the "if you build it, they will come" moment remains pivotal. Dr. Omar Asensio's pioneering research, with the assistance of Microsoft's Azure OpenAI Service, demonstrates the power of combining innovative tools with human creativity. By tapping into the global community of EV drivers and utilizing advanced AI, Asensio's research has shed light on the intricate dynamics of EV charging behavior. With machine learning models predicting problems, improving reliability, and addressing equity concerns, this research marks a significant milestone in accelerating electrification and environmental sustainability. As EVs become increasingly prevalent, it is this fusion of technological tools and creative insights that paves the road to a cleaner, more accessible, and equitable transportation future, while also setting a precedent for future innovations using Azure OpenAI Service.