5. Humans are more likely to forgive each other than to
forgive machines.
If users are to trust AI systems, we must strive for
algorithmic transparency.
6. Trust is dynamic, Trust is dependent, Trust must be maintained and managed.
Likewise, mistrust is dynamic; it too can be maintained and it too should be
actively managed
A Taxonomy of Emergent Trusting in the Human–Machine Relationship | 2017 | Robert R. Hoffman
7. How to establish trust
• Users tend to trust other users’ experience
• Users trust their own experience
• Follow users’ actions
• Clearly identify data sources
• Show system certainty with which a recommendation is made
9. Humans seek explanations to satisfy certain purposes
or goals.
Clear explanations enrich users’ mental models which
in turn enhance performance as well as nurture trust
in AI.
10. Metrics for Explainable AI: Challenges and Prospects | 2018 | Hoffman et al.
Users questions User Goal
How does it work? Feeling of satisfaction at achieving and
understanding how the system made a specific
decision
What does it achieve? Understanding of the system’s functions and use
What will it do next? Feeling of trust based on the predictability of the
system
What would it have done if X
were different?
Resolution of curiosity at achieving an
understanding of the system
11. How to clarify AI
• Understand the persona
• Use “Local Explanations”
• Focus explanation
• Treat explanations as a process
13. Humans are constantly bombarded with big, bold,
noisy accentuating data.
Effective AI systems meet users’ needs while they
co-exist and correspond with an ecosystem of
products competing for attention.
14. The presentation format of recommendations affects how people perceive and
receive the recommendation.
The complexity of the information, the presentation format and appearance, the
recipients’ context at the time of presentation and their ability to attend to the
information all affect the impact of the information.
Designing Recommendations | 2019 | Elizabeth F. Churchill
15. How to simplify AI
• Favor text over visual explanations
• Use the right vocabulary
18. Humans are typically most comfortable when in
control.
“Black box” systems steer users away from their
comfort zone and into unplanned interactions,
confusing pathways and unpredictable outcomes.
19. We found that a large number of participants used the systems without their
intelligent features…This behavior was often used as a coping strategy for
problems with the system, but could also be a feature of intelligent systems to
stress user control or raise “algorithmic awareness”
When People and Algorithms Meet: User-reported Problems in Intelligent Everyday Applications | 2019 | Eiband et al.
20. How to help users control AI
• Provide insight into what to expect
• Support Undo and Redo
• Focus on failure—don’t assume success
• Ask for user feedback
• Allow editing
• Allow Users to Turn Intelligence on and off
22. Humans generally develop good rapport on the basis
common grounds.
To enhance user engagement, design interactions that
are as close to human behavior as possible.
23. Version 1:
Man: “Book it for June 31.”
Google Assistant: “There are only 30 days in June.”
Version 2:
Man: “Book it for June 31.”
Google Assistant: “Actually, June has only 30 days.”
Alexa, Should We Trust You? | 2018 | Judith Shulevitz
24. How to humanize AI
• Personify the system
• Make the user experience enjoyable and engaging (efficient is
no longer enough)
25. Strategically speaking, a brilliant data-driven algorithm
typically matters less than thoughtful UX design.
Thoughtful UX designs can better train machine
learning systems to become even smarter.
AI Won’t Change Companies Without Great UX, Harvard Business Review | 2017 | Michael Schrage
26. People dump AI advisors that give bad advice, while they forgive humans for doing the same | 2016 | Michael J. Coren
Learning to trust artificial intelligence systems Accountability, compliance and ethics in the age of smart machines| 2016 | Dr. Guruduth Banavar
Metrics for Explainable AI: Challenges and Prospects | 2018 | Hoffman et al.
Trust and Recommendations | Victor et al.
When People and Algorithms Meet: User-reported Problems in Intelligent Everyday Applications | 2019 | Eiband et al.
A Taxonomy of Emergent Trusting in the Human–Machine Relationship | 2017 | Robert R. Hoffman
Designing Recommendations | 2019 | Elizabeth F. Churchill
Why micro-interactions are important for UX | 2018 | Athina Ntosti
Microinteractions in User Experience | 2018 | Nielsen Norman Group, Alita Joyce, October 21, 2018
User Empowerment and the Fun Factor | |2002 | Jakob Nielsen
Micro-interactions: why, when and how to use them to improve the user experience | 2018 | Vamsi Batchu
Personification of the Amazon Alexa: BFF or a Mindless Companion? | Dr. Irene Lopatovska, Harri et Williams
Alexa, Should We Trust You? | 2018 | Judith Shulevitz
Microiteractions, Designing with details | Dan Saffer
Silence is gold
AI Won’t Change Companies Without Great UX, Harvard Business Review | 2017 | Michael Schrage
The Challenge of Crafting Intelligible Intelligence | 2019 | Daniel s. Weld and Gagan Bansal
References