December 9, 2024 |

Hunter College DAIR Lab: Advancing Responsible AI in Healthcare and Infrastructure

By Emily Blackburne

From predicting life-threatening pregnancy complications to protecting New York City���s aging infrastructure, the Distributed Artificial Intelligence Research (DAIR) Lab at Hunter College lab is addressing some of society���s most pressing challenges with innovative artificial intelligence (AI) technology and research. Founded in 2003, the lab is led by , a computer science professor at Hunter College and the CUNYGraduate Center, and leverages the power of AI while championing an approach that prioritizes responsibility, fairness, and transparency.

Made up of undergraduate and graduate Hunter College students, the primary mission is to develop intelligent single and multi-agent decision support AI systems that can predict outcomes earlier and with more accuracy for better complex decision-making.

���Our work focuses on early prediction and preventive actions, ensuring decisions are not only robust but also actionable within the constraints of real-world conditions��� Raja explains��.

By integrating AI alongside expert knowledge, the lab addresses critical areas like healthcare and civil infrastructure.

One of the DAIR Lab���s main projects is funded by the National Institute of Health (NIH) to improve maternal health outcomes using AI. In collaboration with Columbia University Medical Center, the lab developed algorithms that can , a life-threatening pregnancy complication, as early as 11 weeks, instead of the typical 20 weeks. By analyzing biomarkers and other clinical data, the system enables physicians to take preventive measures well before the condition manifests to potentially

Raja stresses the importance of fairness in these systems. Early versions of the preeclampsia model revealed biases due to underrepresentation of certain demographics in the training data.

���It gives me goosebumps to think about unnecessary treatments or missed diagnoses during pregnancy. This underscores why it���s critical to double-check models for equity����� she stresses.

To address this, her team employs data augmentation techniques, including generating synthetic data to balance the absence of certain groups.

Integral to this effort at DAIR are CUNYGraduate Center computer science Ph.D. students, and Loodaricheh works on data augmentation, or in transforming huge datasets into pixelated images to program AI computer models. Jia centers her research on addressing missing data using ensemble learning techniques that tailor imputations to specific data types. This is especially helpful for synthetically filling in data gaps for underrepresented groups who may not have enough data collected from traditional medical surveys or studies, or when patient data isn���t fully provided, such as survey questions left blank.

Beyond healthcare, the DAIR Lab also addresses urban challenges such as the aging infrastructure and flooding. In collaboration with geographers at Hunter and Cornell University, Raja���s team uses distributed AI approaches to predict which components of New York City���s water distribution and sewer systems are most likely to fail and pose potential health and economic risks. New York���s infrastructure consists of some pipes dating as far back as the 1890s and poses especially unique urban challenges. ���It is a struggle when we have an older city like New York��� Raja says, ���there are different building materials used over time.��� The lab���s work relies heavily on distributed AI, a methodology where decision-making is determined by multiple agents���both human and environmental. By incorporating diverse data points like pipe age, material, location, and environmental stressors, her team���s programmed algorithms can predict water pipe failures and prioritize repair schedules. Repairing pipes is critical to prevent potential urban flooding, and the DAIR team is working on algorithms to simulate potential flooding scenarios using a multi-agent AI computer model, which informs emergency response and long-term urban planning. Using publicly available data and through a potential collaboration with 勛圖弝け/NYU���s initiative, the team is now working on developing a dashboard and smartphone application that gives real-time updates during floods, benefiting both residents and emergency responders.

The DAIR lab also applies distributed AI methods for traffic congestion management, especially to mitigate ���selfish routing,��� where drivers unintentionally clog new roads. Similar to the Waze navigation app, Raja notes that distributed AI allows for real-time coordination, optimizing traffic flow for everyone on the road.

Central to the DAIR lab���s philosophy is the principle of responsible AI.

���It must be fair, transparent, and as bias-free as possible,��� ��Raja emphasizes.

Her projects prioritize interpretability, ensuring that AI systems can explain their recommendations simply���whether that���s in advising a physician on prescribing an expensive medical test or in identifying which failing infrastructure needs immediate attention. The DAIR Lab���s in-house app ���Bayesian Network Inspector���,developed as part of a master���s thesis by Hunter M.A. student , helps with this interpretability, by using original Bayesian networks to model relationships between data pieces. This allows users to recognize key features that affect the outcome, even if data is scarce. This could allow for synthetic information to be constructed for long-term planning, where the physical data is lacking or inaccessible.

Whether predicting adverse health outcomes or safeguarding urban infrastructure, the DAIR lab���s work exemplifies how AI can be used for society���s benefit���when guided by a commitment to equity, transparency, and collaboration. ���The goal is not to replace the experts, but to help them with AI��� Raja says. The DAIR Lab is not just working on researching today���s problems but also in developing practical AI-driven applications for a better prepared and resilient future.