Search for: All records

Artificial Intelligence (AI)-driven Digital Health (DH) systems are poised to play a critical role in the future of healthcare. In 2021, $57.2 billion was invested in DH systems around the world, recognizing the promise this concept holds for aiding in delivery and care management. DH systems traditionally include a blend of various technologies, AI, and physiological biomarkers and have shown a potential to provide support for individuals with various health conditions. Digital therapeutics (DTx) is a more specific set of technology-enabled interventions within the broader DH sphere intended to produce a measurable therapeutic effect. DTx tools can empower both patients and healthcare providers, informing the course of treatment through data-driven interventions while collecting data in real-time and potentially reducing the number of patient office visits needed. In particular, socially assistive robots (SARs), as a DTx tool, can be a beneficial asset to DH systems since data gathered from sensors onboard the robot can help identify in-home behaviors, activity patterns, and health status of patients remotely. Furthermore, linking the robotic sensor data to other DH system components, and enabling SAR to function as part of an Internet of Things (IoT) ecosystem, can create a broader picture of patient health outcomes. The main challenge with DTx, and DH systems in general, is that the sheer volume and limited oversight of different DH systems and DTxs is hindering validation efforts (from technical, clinical, system, and privacy standpoints) and consequently slowing widespread adoption of these treatment tools. 

This paper presents an intensive case study of 10 participants in the US and South Korea interacting with a robotic companion pet in their own homes over the course of several weeks. Participants were tracked every second of every day during that period of time. The fundamental goal was to determine whether there were significant differences in the types of interactions that occurred across those cultural settings, and how those differences affected modeling of the human-robot interactions. We collected a mix of quantitative and qualitative data through sensors onboard the robot, ecological momentary assessment (EMA), and participant interviews. Results showed that there were significant differences in how participants in Korea interacted with the robotic pet relative to participants in the US, which impacted machine learning and deep learning models of the interactions. Moreover, those differences were connected to differences in participant perceptions of the robot based on the qualitative interviews. The work here suggests that it may be necessary to develop culturally-specific models and/or sensor suites for human-robot interaction (HRI) in the future, and that simply adapting the same robot's behavior through cultural homophily may be insufficient. 

As data science is an evolving field, existing definitions reflect this uncertainty with overloaded terms and inconsistency. As a result of the field’s fluidity, there is often a mismatch between what data-related programs teach, what employers expect, and the actual tasks data scientists are performing. In addition, the tools available to data scientists are not necessarily the tools being taught; textbooks do not seem to meet curricular needs; and empirical evidence does not seem to support existing program design. Currently, the field appears to be bifurcating into data science (DS) and data engineering (DE), with specific but overlapping roles in the combined data science and engineering (DSE) lifecycle. However, curriculum design has not yet caught up to this evolution. This working group report shows an empirical and data-driven view of the data-related education landscape, and includes several recommendations for both academia and industry that are based on this analysis.