WORCESTER—Roee Shraga, an assistant professor of computer science and data science at Worcester Polytechnic Institute (WPI), has been awarded a $175,000 grant from the National Science Foundation (NSF) to explore how human involvement can improve the quality of data in artificial intelligence (AI) processes.
His research examines the “human-in-the-loop” concept, which emphasizes the importance of human participation in data discovery and preparation, an area often overlooked by fully automated systems.
The two-year project investigates how human input, such as labeling, prompting, and validating data, can help uncover biases that AI systems may miss. By drawing on cognitive psychology, Shraga aims to identify implicit biases from computer scientists and coders that can affect AI algorithms and result in unintended discrimination.
“We’ll be able to create a future framework that’s better for users,” said Shraga in an announcement. “It will lead to better data, better data sets, and a better user interface for people searching for data.”
Shraga’s study mainly focuses on “table union search,” a technique used by researchers to expand datasets by finding additional sources online. For instance, in healthcare, researchers may use table union searches to aggregate more patient data to enhance the reliability of their results. Shraga’s work examines how integrating human oversight with AI in this process can provide more beneficial and accurate data for users.
“Having good data is more crucial than ever, particularly in the age of artificial intelligence,” Steve Foskett, WPI public relations manager, told the Worcester Guardian. “As large language models and other forms of AI continue to devour data, it’s still quite important to have a human in the loop. Worcester Polytechnic Institute researcher Roee Shraga is using a National Science Foundation grant to find ways to make sure humans continue have the last word in data discovery and integration.”
Shraga’s research also addresses the role of large language models such as ChatGPT, which, despite their advanced capabilities, may actually require more human involvement to ensure accuracy. The “human-in-the-loop” approach will be critical in refining how these models interact with data and ensure they produce reliable results.
An interesting aspect of the study will also look at how AI models’ tendency to “hallucinate,” or generate plausible but incorrect data, could potentially benefit researchers. In cases where privacy concerns exist, Shraga suggests that generating synthetic yet realistic data tables could offer a solution, helping researchers while safeguarding sensitive information.
This study could have wide-ranging applications, from improving data integrity in scientific research to enhancing how AI systems function across various industries. Shraga’s research aims to lay the groundwork for AI systems that work more seamlessly with human input, ultimately producing better outcomes for both data scientists and end users.
News? Tips? Interesting story? Contact Editor Charlene Arsenault at carsenault@theworcesterguardian.org
