=Paper=
{{Paper
|id=Vol-2448/SSS19_Paper_Upload_228
|storemode=property
|title=Interpretable AI for Well-Being Using Mobile Health
|pdfUrl=https://ceur-ws.org/Vol-2448/SSS19_Paper_Upload_228.pdf
|volume=Vol-2448
|authors=Peter Pirolli
|dblpUrl=https://dblp.org/rec/conf/aaaiss/Pirolli19
}}
==Interpretable AI for Well-Being Using Mobile Health==
https://ceur-ws.org/Vol-2448/SSS19_Paper_Upload_228.pdf
Interpretable AI for Well-Being Using Mobile Health
Peter Pirolli
Florida Institute for Human and Machine Cognition
ppirolli@ihmc.us
Abstract attrition and improve achievement of behavior change goals.
Mobile health (mHealth) and other highly pervasive, interac- Intelligent mHealth systems that utilize multiple ap-
tive and adaptive digital systems offer an enormous oppor- proaches, automate personalization, and increase social sup-
tunity for enabling theory-based health behavior-change in- port can be a scalable approach to current healthcare chal-
terventions that are replicable, scalable, and sustainable. Such
lenges.
advances could have great impact, and they offer great op-
portunities for advances in artificial intelligence (AI), psy- Novel computational theories of the psychological and
chology, and social science. Unfortunately, the technical social mechanisms of behavior change are needed in order
challenges of developing AI to support healthier behaviors to accelerate the development of mHealth systems so that
generally fall into the category of “not suitable for machine they can help people build healthier lifestyles (Riley et al.,
learning.” This presentation will summarize recent research
2011; Spruijt-Metz et al., 2015). In turn, we could use
with the Fittle+ mHealth systems and predictive models of
the daily effects of individual-level mHealth interventions. mHealth systems as experimental platforms for accelerated
The models are developed in a computational cognitive ar- development of such theories. These theoretical and empir-
chitecture and rely on Instance Based Learning Theory and ical foundations would then provide a basis for intelligent
ACT-R learning mechanisms. interactive agents that reason about causal models of the dy-
namics of individual human behavior--models that enable
the planning and delivery of interventions and assessments
Introduction
that optimize the acquisition and maintenance of healthy
Mobile health (mHealth) systems offer an opportunity for habits.
pervasive support of health behavior change in the actual
ecology of people’s everyday environments. Such advances
could have vast impact since individual and social behavior, Human Goal-Striving and Habit Formation
including poor diet, sedentary lifestyle, and social isolation
are central to the etiology and management of many health An unhealthy lifestyle can be viewed, in part, as a complex
outcomes, and yet are typically resistant to change (Klein et set of interrelated habits that need to be switched out for
al., 2014). It is estimated that 70% of health care costs are healthy ones, a few tiny habits at a time (Fogg & Hreha,
due to changeable behavior, and behavioral and environ- 2010). Commercial and health-care provider weight loss
mental factors account for more deaths than genetics (Riley, programs can often involve months to years of counseling,
Nilsen, Manolio, Masys, & Lauer, 2015). which suggests thousands to tens of thousands of elementary
However, the mHealth field is still relatively new and habits being acquired (Feltovich, Prietula, & Ericsson,
lacks integrated theories of long-term behavior change that 2006). The working assumption for our own mobile health
address multiple interventions and a multiplicity of individ- research is that to master the complex fabric of a new
ual, social, and environmental factors. As with digital health healthy lifestyle, one must master and weave together a new
interventions in general, mHealth systems suffer high attri- set of elementary habits
tion rates (Eysenbach, 2005). However, recent research sug- Unfortunately, modeling human behavior change and in-
gests that mHealth systems that adapt to the individual tervening in ways that shape healthier habits is enormously
(Konrad et al., 2015; Pirolli, 2016), use evidence-based in- challenging and not suitable for current machine learning
terventions (Pirolli et al., 2018), and include online social approaches (Brynjolfsson & Mitchell, 2017). This presenta-
support (Du, Venkatakrishnan, Youngblood, Ram, & tion gives a theoretical approach and several computational
Pirolli, 2016; Du, Youngblood, & Pirolli, 2014) reduce
models that provide an integrated account of multiple mech- Programs. Paper presented at the Proceedings of the Wireless
anisms associated with people striving to achieve healthier Health 2014 on National Institutes of Health, Bethesda, MD, USA.
http://dl.acm.org/citation.cfm?doid=2668883.2668887
behavior and their long-term habit formation. Interestingly,
Eysenbach, G. (2005). The law of attrition. Journal of medical
the mechanisms modeled in health-behavior change are also
Internet research, 7(1), e11. doi:10.2196/jmir.7.1.e11
implicated in well-known human cognitive biases (Lebiere
Feltovich, P. J., Prietula, M. J., & Ericsson, K. A. (2006). Studies
et al., 2013) and, in a sense, the success of the interventions of expertise from psychological perspectives. In K. A. Ericsson, N.
we have studied appears to be the result of taking advantage Charness, P. J. Feltovich, & R. R. Hoffman (Eds.), The Cambridge
of those biases. handbook of Expertise and Expert Performance (pp. 41-67).
I will present an overview of the Fittle+ mHealth systems Cambridge, MA: Cambridge University Press.
(Pirolli et al., 2018) that have been used to study several Fogg, B. J., & Hreha, J. (2010). Behavior Wizard: A Method for
evidence-based behavior change interventions. These sys- Matching Target Behaviors with Solutions. In T. Ploug, P. Hasle,
& H. Oinas-Kukkonen (Eds.), Persuasive Technology (Vol. 6137,
tems provide scaffolding interventions: Behavior-change
pp. 117-131): Springer Berlin Heidelberg.
techniques and associated mHealth interactions (e.g., SMS
Klein, W. M. P., Bloch, M., Hesse, B. W., McDonald, P. G.,
reminders; chatbot dialogs; user interface functionality; etc.) Nebeling, L., O’Connell, M. E., . . . Tesauro, G. (2014). Behavioral
that support the acquisition and maintenance of healthy hab- Research in Cancer Prevention and Control: A Look to the Future.
its. American journal of preventive medicine, 46(3), 303-311.
I will present models developed in the ACT-R computa- doi:10.1016/j.amepre.2013.10.004
tional cognitive architecture (Anderson, 2007) that address Konrad, A., Bellotti, V., Crenshaw, N., Tucker, S., Nelson, L., Du,
data collected about individual-level daily achievement of H., . . . Whittaker, S. (2015). Finding the Adaptive Sweet Spot:
Balancing Compliance and Achievement in Automated Stress
behavioral goals for improved eating and exercise using Fit-
Reduction. Paper presented at the SIGCHI Conference on Human
tle+ mHealth applications. The models refine the psycho- Factors in Computing Systems (CHI 2015), Seoul, Korea.
logical constructs of perceived goal difficulty, self-efficacy Kukla, A. (1972). Foundations of an attributional theory of
(Bandura, 1998) and intended effort (a kind of motivation) performance. Psychological Review, 79(6), 454-470.
(Kukla, 1972). The models provide a plausible account of doi:10.1037/h0033494
how intentions can lead to the initial effortful striving to Lebiere, C., Pirolli, P., Thomson, R., Paik, J., Rutledge-Taylor, M.,
carry out goals, how repeated execution of behaviors can be- Staszewski, J., & Anderson, J. R. (2013). A functional model of
come automated habits, and how specific intervention tech- sensemaking in a neurocognitive architecture. Comput Intell
Neurosci, 2013, 921695. doi:10.1155/2013/921695
niques support the development of habits. Together the
models map a trajectory from initial effortful pursuit of a Pirolli, P. (2016). A computational cognitive model of self-efficacy
and daily adherence in mHealth. Translational behavioral
behavior-change goal to new stable habits. medicine, 6(4), 1-13. doi:10.1007/s13142-016-0391-y
The motivation for developing these models is that they
Pirolli, P., Youngblood, G. M., Du, H., Konrad, A., Nelson, L., &
may serve as a foundation for intelligent mHealth interac- Springer, A. (2018). Scaffolding the Mastery of Healthy Behaviors
tion algorithms. By developing these in a cognitive architec- with Fittle+ Systems: Evidence-Based Interventions and Theory.
ture, the models are not only predictive, but provide insight Human–Computer Interaction, 1-34.
as to the underlying causal mechanisms, which is necessary doi:10.1080/07370024.2018.1512414
for reasoning about optimal personalized interventions that Riley, W. T., Nilsen, W. J., Manolio, T. A., Masys, D. R., & Lauer,
help people achieve healthy lifestyles. M. (2015). News from the NIH: potential contributions of the
behavioral and social sciences to the precision medicine initiative.
Translational behavioral medicine, 5(3), 243-246.
doi:10.1007/s13142-015-0320-5
References
Riley, W. T., Rivera, D. E., Atienza, A. A., Nilsen, W., Allison, S.
M., & Mermelstein, R. (2011). Health behavior models in the age
Anderson, J. R. (2007). How can the human mind occur in the of mobile interventions: are our theories up to the task?
physical universe? Oxford, UK: Oxford University Press. Translational behavioral medicine, 1(1), 53-71.
doi:10.1007/s13142-011-0021-7
Bandura, A. (1998). Self-efficacy: The exercise of control. New
York: W.H. Freeman. Spruijt-Metz, D., Hekler, E., Saranummi, N., Intille, S., Korhonen,
I., Nilsen, W., . . . Pavel, M. (2015). Building new computational
Brynjolfsson, E., & Mitchell, T. (2017). What can machine models to support health behavior change and maintenance: New
learning do? Workforce implications. Science, 358(6370), 1530- opportunities in behavioral research. Translational behavioral
1534. doi:10.1126/science.aap8062 medicine, 5(3), 335-346.
Du, H., Venkatakrishnan, A., Youngblood, G. M., Ram, A., &
Pirolli, P. (2016). A Group-Based Mobile Application to Increase
Adherence in Exercise and Nutrition Programs: A Factorial Design
Feasibility Study. JMIR Mhealth Uhealth, 4(1), e4.
doi:10.2196/mhealth.4900
Du, H., Youngblood, G. M., & Pirolli, P. (2014). Efficacy of a
Smartphone System to Support Groups in Behavior Change