Trust-based Decision-Making for Energy-Aware

Trust-based Decision-Making for
Energy-Aware Device Management
Stephan Hammer, Michael Wißner, and Elisabeth André
Human Centered Multimedia
Augsburg University, Germany
Motivation
Smart environment that is able to support users in saving energy by
proactively performing energy-aware adaptations.
Sensors:
- to recognize situations
such as „user leaves
room and light is on“
- Examples:
- Smart Plugs
- Brightness
- Ultrasound
HomeMatic CCU to control
electronic appliances
Displays
Motivation
Problem:
If the system performs an adaptation which:
– the users do not understand,
– the users consider inconvenient,
– makes the users feel they are no longer in control,
–…
then the users’ trust in the system might be impaired, resulting in
lesser acceptance of and, in the worst case, disuse of the system.
Our Goals
• Develop a user model, which:
– chooses adequate actions to reduce energy consumption
– models user trust in adaptive environments
– chooses that action that will result in the highest user trust
 User Trust Model (UTM)
• Initialize the UTM with data gathered in an online survey
• Evaluate users’ experience, acceptance, and trust
towards a system that uses the UTM in a real setting
Building the UTM What is „Trust“?
• Trust is a very subjective concept
• Trust is a non-deterministic concept
• Trust is a multi-dimensional concept:
–
–
–
–
–
–
–
Comfort of use
Controllability
Transparency
Reliability
Security
Credibility
Seriousness
Building the UTM –
Example: Device = Light
Applicationspecific layer
Generic part
(applicable for different
kinds of self-adaptive
systems) [1]
Initializing the UTM –
Gathering Empirical Data
• Online survey (38 Participants)
– Descriptions of concrete system reactions in concrete situations
– Example: “You leave your desk for a short time (for example to
get something from a shelf) and your display is switched off
automatically.”
a)
b)
c)
Switch off Display
Ask To Switch off Display
Via Mobile Phone
Do Nothing
a)
b)
c)
d)
Switch off Light
Ask Via Mobile Phone
Ask Via Display
Do Nothing
Initializing the UTM –
Gathering Empirical Data
• Online study (38 Participants)
– Descriptions of concrete system reactions in concrete situations
– Example: “You leave your desk for a short time (e.g. to get
something from a shelf) and your display is switched off
automatically.”
– Ratings for the following statements (5-point Likert scale):
• Q1: I understood why the system was reacting in this way.
• Q2: I had control over the system.
• Q3: I found the system comfortable to use.
Initializing the UTM
Questions 1-3 =>
Evaluating the UTM –
User Study
24 Participants (18 male, 6 female, Age: 23-33)
Setting:
•
•
•
•
“Typical” day in an office
Different tasks
Changing context
After each system
reaction:
– Transparency, User Control,
Comfort of Use, Trust
– Preferred system action
 User Experience and User
Trust
Evaluating the UTM –
Results
Ratings on a 5-point Likert Scale
• System actions (Light):
– Consistently high ratings concerning Transparency,
Controllability, Comfort of Use and Trust
– Lowest average rating (M: 3.92, SD: .86):
• Criterion: Trust
• Situation: User is leaving the room
• System action: Ask to switch the light off via the
user’s mobile phone
• Reason: No Feedback on Phone
– System actions and users’ preferences differed
• Reason: Repeated confirmations of system actions
via the mobile phone are uncomfortable and
obtrusive.
Evaluating the UTM –
Results
Ratings on a 5-point Likert Scale
• System actions (Display):
– System reactions matched the users’ preferences
in all situations
– Users wanted the system to decide autonomously
– Only moderate ratings concerning Controllability
(M: 2.5 – 3.46)
– Lower ratings concerning Trust (M: 3.63 – 3.88)
• Reasons: No Feedback when leaving, No authentication
mechanism when arriving
– Still high ratings concerning Transparency (M:
3.79 – 5.0) and Comfort of Use (M: 4.0 – 4.58)
Evaluating the UTM –
Further Results
• Participants were satisfied (M: 3.96; SD: .68)
• Participants did not feel:
– distracted (M: 2.00; SD: 1.00)
– restricted (M: 1.83; SD: 1.07)
– observed (M: 2.33; SD: 1.18)
• Participants acknowledged that the system:
–
–
–
–
supported them in saving energy (M: 4.71; SD: .54)
behaved adequately (M: 4.38; SD: .70)
was unobtrusive (M: 3.71; SD: 1.10)
was transparent (M: 4.96; SD: .20)
Conclusion
• User Trust Model (UTM):
– Generic approach for trust-based decision-making for the
adaptation of smart environments
– Based on an empirically grounded Bayesian Network which
aims at maintaining user trust
• Construction, initialization with empirical data, integration in
an office setting
• User Study:
– UTM succeeded in maintaining users’ trust in a smart
office environment
Future Steps
• Further analysis of the collected data:
– Influence of user-specific attitudes (e.g. trust disposition) on
preferences concerning system actions and trust dimensions
(e.g. Distrust towards technical systems -> Higher level of
control by the user)
• Decision-making for more than one user
UMAP 2014
Thank you!
Any Question?
For more detailed information about the generic part of the UTM:
[1] Kurdyukova, E., Andre, E., Leichtenstern, K.: Trust management of ubiquitous multi-display
environments. In Krueger, A., Kuik, T., eds.: Ubiquitous Display Environments. Cognitive Technologies.
Springer (2012)
http://www.informatik.uniaugsburg.de/en/chairs/swt/se/projects/oc-trust/
http://www.it4se.net/