=Paper=
{{Paper
|id=Vol-2448/SSS19_Paper_Upload_227
|storemode=property
|title=Open Affect-Responsive Systems: Toward Personalized AI to Beat Back the Waves of Technostress
|pdfUrl=https://ceur-ws.org/Vol-2448/SSS19_Paper_Upload_227.pdf
|volume=Vol-2448
|authors=David Agogo,Leona Chandra Kruse
|dblpUrl=https://dblp.org/rec/conf/aaaiss/AgogoK19
}}
==Open Affect-Responsive Systems: Toward Personalized AI to Beat Back the Waves of Technostress ==
https://ceur-ws.org/Vol-2448/SSS19_Paper_Upload_227.pdf
Open Affect-Responsive Systems:
Toward Personalized AI to Beat Back the Waves of Technostress
David Agogo,1,* Leona Chandra Kruse 2
1 Information Systems and Business Analytics Department
Florida International University
Miami FL 33199, USA
2 Institute of Information Systems
University of Liechtenstein
9490 Vaduz, Liechtenstein
Abstract individual. We call this class of tools the Open Affect-Re-
We review existing system-based solutions to the growing sponsive Systems (OARS).
problem of technostress. Based on an analysis of 102 digital Used interchangeably, the terms technostress and digital
applications for stress management, we find several signifi- stress broadly refer to both immediate and drawn out stress
cant limitations in the approaches of these tools and sparse responses attributable to potential or actual technology use
evidence of their effectiveness in dealing with technostress. (Agogo & Hess 2008). In practice, there are an abundance
Thereafter, we propose a blueprint for an autonomous soft-
ware agent that not only addresses the root of technostress by
of digital-based solutions addressing this problem, but we
building user resilience towards technostress, but also gener- do not have any systematic account on their mechanisms.
ates contextually rich information that system creators and
organizations can act upon to be more responsive to the ex- Research Findings
periences of individual users. The operation of the OARS We analyzed 102 digital applications that are available on
(Open Affect Responsive Systems) is described with a user popular application stores or are referred to in articles about
story.
dealing with technostress. We found seven common mech-
anisms among them (Figure 1) that generally follow one of
Background three approaches: (1) modification of IT features and its use
Practically everyone who uses technology is becoming more routines; (2) modification of individual reactions to IT
vulnerable to technostress as technology continues to embed stressors; and (3) temporary disengagement from IT such as
itself into our everyday lives. This has prompted the creation online/offline venting (cf. Pirkkalainen, et al. 2017). Note
of digital support tools to help people address this problem. that each tool can apply more than one mechanism. Given
Most of the available tools target general stress management their technological nature, can these tools in fact inject more
and promote wellbeing by simulating offline relaxation- stress into the issue of dealing with stress? To answer this
based interventions. Of the few that specifically target tech- question, we peaked behind the veil at the theoretical mech-
nology as a cause of stress, the majority focus on monitoring anisms that justified how these different classes of tools
and controlling user exposure to their devices. However, this were designed.
mechanism itself is likely counterproductive. Constant mon- Some tools (35%) were created based on widely acknowl-
itoring and abundance of data constitutes a form of surveil- edged intervention approaches (e.g., cognitive-behavioral
lance that increases feelings of pressure and triggers more therapy (CBT) and mindfulness), while others (36%) didn’t
technostress. There is need to switch focus from addressing explicitly refer to neither theory nor intervention approach
acute symptoms of technology-induced stress to figuring out that would evoke confidence in their effectiveness. The ma-
ways to address the root of the problem. jority (70%) were static systems, with pre-programmed re-
We propose a blueprint for a digital support tool that not sponses while others were adaptive (30%), with most of
only addresses the root of technostress by building user re- those applying artificial intelligence (AI) at their core
silience towards technostress, but also generates contextu- (24%). Of that subset, apps applied AI for different purposes
ally rich information that system creators and organizations - from identifying patterns in users’ emotional state based
can act upon to be more responsive to experiences of
* Corresponding author: dagogo@fiu.edu
1: Monitoring (e.g., Life Charge App and Welltory)
2: Simulating offline relaxation intervention (e.g., Pocket Yoga, Col-
orfil, and Fidget Spinner)
3: Information and guidance (e.g., Head to Health)
4: Virtual support group (e.g., Beyond Blue and 7 Cups)
5: Gamification (e.g., Forest: Stay Focused)
6: Controlling exposure (e.g., Digital Detox)
7: AI as counsellor (e.g., Wysa and Tess)
Figure 1: Mechanisms to Mitigate Technology-related Stress in Commercial Digital Applications
on their interaction with their mobile device to acting as a here focus on hormesis to instantiate OARS and demon-
virtual counsellor and conversational agent. strate its use.
Unfortunately, using AI as a constant monitor and inter- Hormesis is the principle underlying Stress Inoculation
preter of behavior can lead to increased contact with tech- Therapy (SIT). It describes a biological phenomenon where
nology that may in turn trigger negative affective responses. exposure to low doses of a toxic substance can actually have
At the same time, scholars (e.g., Weizenbaum 1976) have a beneficial effect, although exposure to those same toxins
warned that users may build strong attachment and depend- in larger amounts might prove lethal (Meichenbaum 2007).
ency to their AI counsellor. This is despite how far off AI Such approach has been recommended for a broad range of
tools still are from being truly conversational and assistive issues and found to be "at least moderately effective" (Flax-
for health purposes (Strickland, 2018). We believe the po- man & Bond 2010). SIT itself involves exposing individuals
tential for the use of AI in helping users to deal with tech- to milder forms of stress to bolster coping mechanisms and
nostress is still nascent. Before this can be achieved, there confidence in future coping behavior.
is need to think systematically about the architecture of a For the system to leverage hormesis approach to help im-
system in which AI plays a theoretically supportable role in prove users' ability to deal with technostress, the system
warding off the waves of technostress. must be capable of delivering periodic low doses of typical
IT stressors to users. When users asynchronously indicate
Architecture of an Open Affect Responsive System they are experiencing an issue with the system (e.g. using a
OARS are a class of autonomous software agents are de- hotkey), OARS can restore system to its normal functioning
signed to drive improvements on the individual user level, and provide users with guidance to reframe such situations
system level as well as the organizational level. OARS have in the future. If implemented according to this and other de-
a four-stage system architecture (identify, formulate, evalu- sign guidelines we propose, such operation of an OARS
ate and learn) that is iterative and employs AI to learn adap- should increase the preparedness and confidence of users in
tively. These four stages occur across five subsystems which the face of future unanticipated IT breakdowns. In the fol-
are independent modules that can be developed separately lowing section we offer a descriptive vignette of a user’s ex-
and in parallel to deliver a fully functional OARS (see Fig- perience with the proposed OARS, along with a screenshot
ure 1). Where possible, OARS integrate user feedback (col- of the system prototype in action (Figure 2).
lected as asynchronous pull data, instead of the synchronous
push of constant monitoring – although that form of input OARS in Action (Hormesis Approach)
may be possible as well). Such nudge-based user feedback
can be used as labelled training data for constant learning Jane logged onto her computer to complete the months
and improvement of the OARS, as well as the development accounts. She had recently installed a new accounting
of user phenotypes which can make a personal AI possible. software and was hoping the experience went
The architecture of OARS supports the application of mul- smoothly. During the installation, she had enabled the
tiple theoretically supported resilience-building mecha- OARS add-on that came with the software. Her under-
nisms to make users less vulnerable to technostress. Based standing was that she could press the ctrl-f12 hotkey if
on contemporary stress management literature, we discern the system was not running as desired and her per-
three promising mechanisms for delivery via OARS: active sonal AI would drop in to help out. Within a few mo-
stress management (CBT), mindful monitoring (Acceptance ments, she noticed the system felt
and Commitment Therapy (ACT)), and hormesis. Let us
Figure 2: Screenshot of Prototype OARS in action
a bit unresponsive. Her attention started to drift from desirable and stress-free work environment. This, in turn,
the task at hand and even though she didn’t realize it, would pay back in better performance.
her heart began racing slightly as the nerves emerged.
At that instant, the faded outline of a small notification
window began to gently fade into view at the bottom
right of the screen. It caught her eyes and she absent References
mindedly hit the hotkey while continuing to scroll Agogo, D., & Hess, T. J. (2018). “How does tech make you feel?”
through the application. A few seconds later, the full a review and examination of negative affective responses to tech-
notification faded into view with the message “Trying nology use. European Journal of Information Systems, 27(5), 1–
to identify what the issue is…”. She ignored it and con- 30.
tinued working. A few short moments later, the notifi- Flaxman, P. E., & Bond, F. W. (2010). A randomised worksite
cation message changed to “Did that fix the issue for comparison of acceptance and commitment therapy and stress in-
you?”. She paused for a micro-second as if to remind oculation training. Behaviour research and therapy, 48(8), 816-
herself of the issue she had previously experienced, 820.
Meichenbaum, D. (2007). Stress inoculation training: A preventa-
then she leaned back into her seat and continued work-
tive and treatment approach. Principles and Practice of Stress Man-
ing, the system seemed a little snappier. What Jane agement, 3, 497–518.
didn’t realize at that time was that the OARS had cre- Pirkkalainen, H., Salo, M., Makkonen, M., & Tarafdar, M. (2017).
ated a temporary processor bottleneck to simulate the Coping with Technostress : When Emotional Responses Fail.
slowing down on the processor that happened occa- Strickland, E. (2018, June 25). Layoffs at Watson Health Reveal
sionally during the final computation phase of running IBM’s Problem With AI. IEEE Spectrum, (June 2018). Retrieved
the accounts. She would realize this in a few moments from https://spectrum.ieee.org/the-human-os/robotics/ artificiall-
when she closed the application and received a final intelligence/layoffs-at-watson-health-reveal-ibms-problem-with-
status message from the OARS “These experiences ai
Weizenbaum, J. (1976). Computer power and human reason: From
make you unique!”.
judgment to calculation. San Francisco, SF: W. H. Freeman.
Conclusion
In summary, we can view OARS as personalized AI tailored
to individual and organizational needs. The system learns
from prior experience and improve its capability and user
training. We also expect organizational learning to occur
that results in a better understanding of the states and needs
of its individual members, therefore creating a more