Early prediction and timely decision-making of acute diseases are critical to enabling early intervention and improving clinical outcomes (for example, a sepsis patient may benefit from a 4% higher chance of survival if diagnosed 1 hour earlier). Developing machine learning (ML) models for clinical decision-making on Electronic Health Records (EHRs) presents several significant challenges: 1) existing models are trained mostly on EHR data from intensive care units (ICUs), which are not generalizable for sepsis onsets in emergency rooms and hospital wards; 2) most existing tools simply output prediction result as a risk score, without sufficient explanation or confidence interval for it, which is not trustworthy for physicians; 3) existing systems often ignore the human workflow by neither providing actionable insights to physicians nor enabling interactive explorations from physicians, which limits their clinical usages. To address these challenges, we propose a Human-Centered Artificial Intelligence (HCAI) system to collaborate with human domain experts in the high-stake and high-uncertainty decision-making process. Specifically, we 1) create a deidentified database with complete visits and long-term EHR history for patients with sepsis risk; 2) develop early sepsis risk prediction models with uncertainty quantification and active sensing; 3) design and implement a physician-centered AI prediction module and user interface for early sepsis human-AI decision making; and 4) design and conduct controlled usability evaluations to quantitatively and qualitatively measure the clinical outcome and user satisfaction. This project integrates human-AI collaboration design, novel ML algorithms, and data visualization tools for improving early prediction and decision-making for sepsis, which hold great promise for leading new insights into human-AI systems for clinical decision support. RELEVANCE (See instructions): Sepsis, which can be caused by bacteria, fungi, or in the case of COVID-19, a virus, is a life-threatening condition with high mortality rates and expensive treatment costs. This project will develop a physician- centered deep-learning algorithm to predict sepsis onset and a user interface for effective human-AI collaboration. As a result, this work relates to the mission of the NIAID and will make a relevant public health impact by delivering early, life-saving care to the bedside of sepsis patients, and will lead to a useful clinical decision support tool for physicians.