Drexel University

ASIST:ANT

The ASIST project is a consortium of top researchers funded by DARPA to advance the capacity of human-machine collaboration. Existing AI agents are designed to aid humans to achieve pre-planned goals by performing specific tasks. This project seeks to address this limitation and build AI agents that are more flexible to engage humans in real team-work. To do so, these agents must reason about and understand the human’s abilities, mental models, cognitive processes, cultural factors and environment. One of the innovations of the ASIST:ANT project is that we explore human-machine collaborations in a MineCraft-based environment. Our ultimate goal is to create AI agents that can build mental models of the human team members, and assist them with flexible teamwork.

Sponsors: DARPA

Partners: SIFT, University of Central Florida, Arizona State University

Balancing Individual and Group Needs in Personalized Exergames

This project investigates how to increase and sustain physical activity using personalized adaptive social exergames. Two-thirds of the adult population in the U.S. are affected by overweight and obesity, with sedentary behavior as a primary cause. In addition to addressing a public health issue, the technology developed in this project will advance theories in human behavioral science. The empirical data generated from the planned system can shed light on the dynamic nature of people’s social comparison processes and reactions. The approach is innovative in bootstrapping design theory, algorithmic innovation, and health behavior science in a synergistic way to make scientific advancement. Learn more…

Sponsors: National Science Foundation

Partners: WELL Center (Drexel University), Rowan University

Diagnostic Driving

Americans spend 70 billion hours behind the wheel every year. What if the way we drive can reveal our health?

In collaboration with CHOP and George Mason University, this project aims to detect whether a teen driver with ADHD is in a dangerous state (e.g., operating a vehicle while not on prescribed medication) based on their driving behavior. If we can notify the drivers in real-time, we can potentially prevent motor vehicle crashes, a lead cause of death for adolescents in the US. By using machine learning techniques developed for player behavior modeling in video games, we model driver behavior and non-invasively predict their ADHD diagnosis in real-time. The long term research goal of this project is to move from reactive health care to a preventive, proactive, and person-centered health care system. Our work can potentially lead to transformative changes in how we assist patient-drivers with other medical and post-surgical conditions

Sponsors: NSF, Children’s Hospital of Philadelphia, Diagnostic Driving Inc.

Partners: Children’s Hospital of Philadelphia, George Mason University, University of Central Florida

eXplainable AI For Social Bias Detection

AI algorithms are not immune to biases. Traditionally, non-experts have little control in uncovering potential social bias (e.g., gender bias) in the algorithms that may impact their lives. With the wide use of deep learning, the issues of interpretability, fairness and trust become more pressing.

In this project, we design interactive visualization tools to help non-experts understand key aspects of AI systems and reveal biases. Currently, we are focusing on developing an interactive visualization for non-experts to explore a semantic Neural Network’s (NN) decisions, in the context of profiling models for loan applications, to reveal potential bias. Our goal is to empower citizens to detect and reduce implicit social biases embedded in AI-infused tools. Learn more...

Sponsors: N/A

Partners: IT University of Copenhagen

Open Player and Community Modeling as a Learning Tool

Personalized learning is an active research area in intelligent tutoring systems and game-based learning. In most personalized systems, the learner does not know how the game categorizes them and why the game changes. We attempt to not only make this information accessible, but also use it to encourage reflection and better learning. Learn more …

Sponsors: National Science Foundation

Partners: University of California, Santa Cruz

iNNk: A Deep Learning Game

Turn project description into: “iNNK, a multiplayer drawing game where human players team up against a neural network (NN). The players need to successfully communicate a secret code word to each other through drawings, without being deciphered by the NN. With this game, we aim to foster a playful environment where players can, in a small but crucial way, go from passive consumers of NN applications to creative thinkers and critical challengers. Learn more...

Sponsors: N/A

Partners: IT University of Copenhagen

Past projects and more information can be found at PXL and AI & Games lab websites.