ABSTRACT
Functional integration of human cognition and machine reasoning is an industry-wide problem where failure risks health or safety. Differences in human versus machine functioning obscure conventional integration. We introduce cognitive work problems (CWP) for rigorous, verifiable functional integration. CWP specify the cognitive problem that integrated designs must solve. They are technology-neutral, work objects, allowing people and computing to share and transform them in coordination. The end-to-end method is illustrated on a system that employs AI for remote patient monitoring (RPM) during COVID-19 home care. The CWP specified actionable risk awareness as the medical problem RPM must solve. Graphical modeling standards enabled user participation: CWP as finite state machines and system behavior in BPMN. For model checking, the CWPs logical content was translated to linear temporal logic (LTL) and the BPMN into Promela as inputs to the SPIN model checker. SPIN verified the Promela implements the LTL correctly. We conclude this CWP-derived RPM design solves the medical problem and enhances patient safety. The method appears general to many critical systems. © 2022 IEEE.