Engineering AI for the public good

The University of Maryland (UMD) is spearheading efforts to make artificial intelligence (AI) safer, more sustainable, and economically beneficial. Maryland Engineering is developing creative, pragmatic solutions to address the strain AI places on natural resources and the need for a skilled workforce to maintain its infrastructure. One significant area of focus is data centers, which present unique challenges due to their immense size, heat generation, water usage, and noise production. Nii Attoh-Okine, chair of UMD's Department of Civil and Environmental Engineering, is developing a data center "report card" based on metrics like water use per square foot for companies to use for self-evaluation. He is also collaborating with industry to establish maintenance protocols adaptable to regional climate, water availability, soil structure, and seismic activity, aiming to train a workforce capable of handling this complex infrastructure. Additionally, UMD engineers are tackling the substantial energy consumption of large language models (LLMs). Assistant Professor Sanghamitra Dutta of Electrical and Computer Engineering is devising more efficient methods for designing and tuning LLMs, particularly those customized for specific tasks. Her work involves using strategic data points, such as "counterfactuals," in the knowledge distillation process, which reduces the total data points needed for training by half, thereby cutting energy use and improving performance. These innovations aim to alleviate the strain on electric supply and mitigate energy price spikes in communities hosting data centers.