Virtual human interactions provide an important avenue for training as emergent opportunities arise. In response to a new training need, we propose a framework to rapidly create experiential learning opportunities in the form of a question--answer chat interaction with virtual humans. This framework takes quickly generated case documents and breaks down the case into small tasks that can be crowdsourced by nonexperts. This framework can serve as a first step to rapidly bootstrapping new virtual humans. We have applied our framework to the task of preparing health care students and professionals to infrequent, but high-stakes, situations such as infectious diseases, cranial nerve disorders, and stroke.
Our framework was utilized by medical professionals interested in providing new training experiences to students and colleagues. Over the course of two months, these professionals created seven scenarios on a diverse range of topics that included Ebola, cancer, and neurological disorders. These scenarios were developed for multiple target audiences such as medical students, residents, and fellows. As a first step, each scenario utilized our framework and crowdsourced workers to create an initial corpus over the course of two days.
From these seven cases, we selected two to evaluate the quality of the resulting virtual-human corpuses. The two scenarios were compared to preexisting reference scenarios that have been in curricular use for several years. We found a reduction in author time commitment of at least 92% while creating a character that was at least 75% as accurate as its reference counterparts. The commitment reduction and accuracy achieved by our framework represents a first step towards rapid development of a virtual human. Our framework can then be combined with other creation processes for further virtual-human development in order to create a mature virtual human. As part of a virtual-human development process, our framework can help to rapidly develop new scenarios in response to emergent training opportunities.
Borish, Michael, and Benjamin Lok. "Rapid Low-Cost Virtual Human Bootstrapping via the Crowd." ACM Transactions on Intelligent Systems and Technology (TIST) 7.4 (2016): 47.