Next-Generation Automated Health Behavior Coaches

Automated health behavior coaches (HBCs) potentially can provide a widely accessible, cost-effective means of promoting health behavior. Coaches are intelligent agents that “converse” with users, offering tailored feedback, advice, and empathy. Research subjects like coaches and comply with target behaviors, but interest and adherence wane over time. More research is needed on next-generation HBCs to improve coaching techniques, enhance user engagement, and extend adherence. However, the necessary technical tools and expertise reside in only a few research labs. In an effort to expand and accelerate research, we are developing an HBC Kit that will extend and specialize our more general ImpTM Kit. We propose 7 innovations for next-generation HBCs, demonstrate them in a lifestyle coach, and characterize authoring with the Imp Kit. We discuss planned extensions for the HBC Kit to enable a larger and more diverse community to create and evaluate a broader range of coaches. Automated Health Behavior Coaches Individual coaching is the gold standard for promoting health behaviors such as exercise, eating, and medication compliance [Bening 2000, Stewart 1995]. Coaches use client-centered techniques [Bandura 1997, Hardeman et al 2005] to inform clients, negotiate behavior change, and build a therapeutic alliance, leading to patient satisfaction, improved health behavior, and positive outcomes. Since compliance and associated benefits often end with treatment, longitudinal care is indicated [Evangelista and Shinnick 2008, Fox and Kilvert 2003, Glasgow et al 2001, Latner et al 2006, Nigg et al 2008, Ockene et al 2002]. But most people can’t afford this, especially for preventive care. Increasing resource constraints further motivate the search for effective, accessible, affordable interventions for long-term adherence to health behaviors [Cohen 2008]. Toward this end, researchers are investigating automated health behavior coaches (HBCs) [Bickmore and Giorgino 2006, Bickmore et al 2005, Bickmore and Picard 2005, deRosis et al 2006, Eyck et al 06, Friedman 1998, HayesRoth 2008, King and Castro 2007, King et al 2007, Liu and Ricard 2005, Miksch 96, Murray et al 2008, Pinto et al 2002, Tate et al 2006]. Prospective advantages include: scalability, lower cost, broad access, comprehensive data capture, reliable performance of theory-based methods, demographic customization, and individual tailoring. The most sophisticated coaches are intelligent agents that “converse” with users, offering tailored feedback, advice, and empathy. Research subjects like coaches and comply with target behaviors, but interest and adherence wane over time. Subjects say they lose motivation because they can’t speak freely and coach dialogue becomes repetitive and boring. More research is needed on next-generation HBCs to improve coaching techniques, enhance user engagement, and extend adherence. However, the necessary technical tools and expertise reside in only a few research labs. In an effort to expand and accelerate research, we are developing an HBC Kit that will extend and specialize our more general ImpTM Character Development Kit [Hayes-Roth et al 1999, Hayes-Roth 2008]. We propose 7 innovations for next-generation HBCs, demonstrate them in a lifestyle coach, and characterize authoring with the Imp Kit. We discuss planned extensions for the HBC Kit to enable a larger and more diverse community to create and evaluate a broader range of coaches. 7 Innovations for Next-Generation Coaches