Group Interaction and Group Dynamics. What is a Group? Individuals Group Team.
INTERACTION WITH AI MODULE 2...Midterm report –group assignment Content –5-7 pages • A...
Transcript of INTERACTION WITH AI MODULE 2...Midterm report –group assignment Content –5-7 pages • A...
-
INTERACTION WITH AI –MODULE 2Session 2
Asbjørn Følstad, SINTEF
-
2
Interaction withAI – module 2
Interaction design
Four sessions
Design of interactionwith AI
Asbjørn Følstad
Understandinginteraction with AI
Morten Goodwin
September 22
October 20
October 13
October 6
-
Midterm report - individual assignment
Three topics:
• Characteristics of AI-infused systems.
• Human-AI interaction design.
• Chatbots / conversational user interfaces.
Language: English or Norwegian.
Max. pages: 6
Min. articles referenced 4.
-
Midterm report – group assignmentContent – 5-7 pages• A description of the group, who you are - names.
• A description of what area of “interaction with AI” you are interested in working with.
• (new) Background section: Position your work relative to existing knowledge and practice
• Minimum 1 maximum 2 questions that you want to address. Please write some sentences about the questions. These questions can change and evolve later in the midterm report and in the final report - as you go about investigating your questions.
• (updated) Method section – overall approach, design process(optional, but encouraged), data collection methods
• (new) Sketches and/or prototypes (optional, but encouraged)
• (new) Findings (progress, initial outcomes)
• (updated) Minimum five references to literature.
Appendices – approx. 1 page each• Appendix 1: Chatbot design task – briefly describe the process and
outcome. Detail reflections and lessons learnt.
• Appendix 2: Machine learning task – briefly describe the process and outcome. Detail reflections and lessons learnt.
Brief status on the group task– each group say a few words
-
5
Agenda
Previous
Today
Interacting with AI – an overview
Chatbots – interacting with AI in naturallanguage
User-centred design of AI
User-centred design of chatbots
1
2
3
4
-
6
User-centred design of AI
-
Individual assignment – task 1:
Characteristics of AI-infused systems
• AI-infused systems are ' systems that have features harnessing AI capabilities that are directly exposed to the end user' (Amershi et al., 2019).
• Drawing on the first lecture of Module 2 and the four mandatory articles (Amershi et al. (2019), Kocielnik et al. (2019), Liao et al. (2020), Yang et al., (2020)). Identify and describe key characteristics of AI-infused systems.
• Identify one AI-infused system which you know well, that exemplifies some of the above key characteristics. Discuss the implications of these characteristics for the example system, in particular how users are affected by these characteristics.
What are the characteristicsof AI-infused systems?
Identify one system and usethis to exemplify
-
8
Challenges in human-centreddesign of AI
-
9
Challenges in human-centreddesign of AI
Capability uncertainty – due to data capture(How to sketch?)
Output complexity – due to adaptive character(How to prototype?)
-
Yang, Q., Steinfeld, A., Rosé, C., & Zimmerman, J. (2020). Re-examining Whether, Why, and How Human-AI Interaction Is Uniquely
Difficult to Design. In Proceedings of the 2020 CHI conference on human factors in computing systems (Paper no. 164).
-
11
Design for user-centred explainableAI (XAI)?
Varying user needs for explanation
Explanations needs to be in terms relevant to users
Developers struggle with gap between algorithmic output and explanation
-
Liao, Q. V., Gruen, D., & Miller, S. (2020). Questioning the AI: Informing Design Practices for Explainable AI User Experiences. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (paper no. 463). ACM.
Question bank to support user-centred design of XAI
-
13
VS.
Learning | Improving | Black box | Fuelled by large data sets
Dynamic Mistakes inevitable Data gathering through interactionOpaque
-
14
Guidelines for Human-AI Interaction
-
https://aidemos.microsoft.com/guidelines-for-human-ai-interaction/demo | Find by Google search: demos human ai interaction
https://aidemos.microsoft.com/guidelines-for-human-ai-interaction/demo
-
Learning | Improving | Black box | Fuelled by large data sets
16
Google Maps - Timeline
-
17
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
18
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
19
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
20
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
21
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
22
Learning system - design for change
• M1: make clear whatthe system can do
• M2: make clear howwell the system cando what it can do
• Explain dynamiccharacter (?)
Learning | Improving | Black box | Fuelled by large data sets
-
23
Learning | Improving | Black box | Fuelled by large data sets
-
24
Mistakes inevitable -design for uncertainty
• M9: Support efficient correction
• M10: Scope services when in doubt
Learning | Improving | Black box | Fuelled by large data sets
-
25
Mistakes inevitable -design for uncertainty
• M9: Support efficient correction
• M10: Scope services when in doubt
Learning | Improving | Black box | Fuelled by large data sets
-
26
Mistakes inevitable -design for uncertainty
• M9: Support efficient correction
• M10: Scope services when in doubt
Learning | Improving | Black box | Fuelled by large data sets
-
27
Mistakes inevitable -design for uncertainty
• M9: Support efficient correction
• M10: Scope services when in doubt
Learning | Improving | Black box | Fuelled by large data sets
-
28
Mistakes inevitable -design for uncertainty
• M9: Support efficient correction
• M10: Scope services when in doubt
Learning | Improving | Black box | Fuelled by large data sets
-
29
Learning | Improving | Black box | Fuelled by large data sets
-
30
Difficult to understand and validate output –design for explainability
• M11: Make clearwhy the system didwhat it did
Learning | Improving | Black box | Fuelled by large data sets
-
31
Difficult to understand and validate output –design for explainability
• M11: Make clearwhy the system didwhat it did
Learning | Improving | Black box | Fuelled by large data sets
-
32
Difficult to understand and validate output –design for explainability
• M11: Make clearwhy the system didwhat it did
Learning | Improving | Black box | Fuelled by large data sets
-
33
Learning | Improving | Black box | Fuelled by large data sets
-
34
Data wanted –design for data capture
• Accommodategathering of data from users
• … but with concernfor the risk of beinggamed
• Make users benefitfrom data
• Privacy by design
Learning | Improving | Black box | Fuelled by large data sets
-
35
Data wanted –design for data capture
• Accommodategathering of data from users
• … but with concernfor the risk of beinggamed
• Make users benefitfrom data
• Privacy by design
Learning | Improving | Black box | Fuelled by large data sets
-
36
Data wanted –design for data capture
• Accommodategathering of data from users
• … but with concernfor the risk of beinggamed
• Make users benefitfrom data
• Privacy by designhttps://www.technologyreview.com/s/610634/microsofts-neo-
nazi-sexbot-was-a-great-lesson-for-makers-of-ai-assistants/
Learning | Improving | Black box | Fuelled by large data sets
-
37
Data wanted –design for data capture
• Accommodategathering of data from users
• … but with concernfor the risk of beinggamed
• Make users benefitfrom data
• Privacy by design
Learning | Improving | Black box | Fuelled by large data sets
-
38
Data wanted –design for data capture
• Accommodategathering of data from users
• … but with concernfor the risk of beinggamed
• Make users benefitfrom data
• Privacy by design
Learning | Improving | Black box | Fuelled by large data sets
-
39
User-centred design of AI –automagic or explicit?
-
Individual assignment – task 2:
Human-AI interaction design
• Amershi et al. (2019) and Kocielniket al. (2019) discuss interaction design for AI-infused systems. Summarize main take-aways from the two papers.
• Select two of the design guidelines in Amershi et al. (2019). Discuss how the AI-infused system you used as example in the previous task adheres to, or deviates from these two design guidelines. Briefly discuss whether/how these two design guidelines could inspire improvements in the example system.
https://aidemos.microsoft.com/guidelines-for-human-ai-interaction/demo
Find by Google search: demos human ai interaction
https://aidemos.microsoft.com/guidelines-for-human-ai-interaction/demo
-
Individual assignment – task 2:
Human-AI interaction design
• Amershi et al. (2019) and Kocielniket al. (2019) discuss interaction design for AI-infused systems. Summarize main take-aways from the two papers.
• Select two of the design guidelines in Amershi et al. (2019). Discuss how the AI-infused system you used as example in the previous task adheres to, or deviates from these two design guidelines. Briefly discuss whether/how these two design guidelines could inspire improvements in the example system.
Select two of the design guidelines.
Identify an AI-infused system and use the two guidelines to
discuss the design of thissystem
-
42
Erica Virtue, product designer, FB: Designing with AI.
At Facebook, AI is everywhere.
Behind the scenes …
- Translate text- Recognize what is in images- Filter out spam- Understand intent behind
posts -> improve FB- (decide on content in feed?)
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
-
43
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
-
44
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
Explore concepts
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
Add tag to request?
-
45
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
Explore concepts
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
Tutorial?
-
46
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
Explore concepts
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
Automagic!
-
47
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
Explore concepts
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
Automagic + opportunities for adaptation and feedback
-
48
Erica Virtue, product designer, FB: Designing with AI.
Facebook recommendations
How to design for includingrecommendations in dialogue?
Lessons learnt
https://medium.com/facebook-design-business-tools/designing-with-ai-3f7652619f4
Look for existing behaviour
If you don’t notice the AI, you’re doing it right
Don’t depend on perfection
-
49
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
-
50
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
-
51
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
Expectation confirmationmodel
Expec-tation
Perfor-mance
perception
Confir-mation
Satis-faction
Bhattacherjee, A. (2001). Understanding informationsystems continuance: an expectation-confirmationmodel. MIS quarterly, 351-370.
-
52
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
Explore concepts
AI accuracy indicator
-
53
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
Explore concepts
AI explanations
-
54
Kocielnik et al. (2019). Designs for expectation setting with AI
Scheduling assistant
Design of system for meetingrequest detections in email
Explore concepts
AI control
-
55
Two fundamentallydifferent approachesto the design of AI-infused systems
Automagic (FB recommendations)
Show, explain, adjust(email meetingrequests)
-
56
Two fundamentallydifferent approachesto the design of AI-infused systems
Automagic (FB recommendations)
Show, explain, control(email meetingrequests)
-
57
Chatbots –conversational interaction design
-
Individual assignment – task 3:
Chatbots / conversational user interfaces
• Chatbots are one type of AI-infused systems. Based on the lectures, and the mandatory articles, discuss key challenges in the design of chatbots / conversational user interfaces.
• Revisit Guidelines G1 and G2 in Amershi et al. (2019). Discuss how adherence to these could possibly resolve some of the challenges in current chatbots / conversational user interfaces.
• Optionally, you may read Følstad & Brandtzaeg (2017), Luger & Sellen(2016), and Hall (2018) from the optional literature to complement your basis for answering.
Key challenges in the design of chatbots
-
59
Chatbot interactiondesign with importantimplications and challenges
-
60
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
61
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
62
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
63Høiland, C. (2019) “Hi, can I help?” An exploratory study of designing a chatbot to complement school nurses in supporting youths’ mental health. Master Thesis. UiO.
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
64Social Health bots (www.sintef.no/socialhealthbots)
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
65
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
66https://www.wired.com/2016/03/fault-
microsofts-teen-ai-turned-jerk/
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
67
Implications
Conversation as design object
Necessary to move from UI design to service design
Necessary to design for networksof humans and bots
-
68
Interaction withAI – module 2
Interaction design
Four sessions
Design of interactionwith AI
Asbjørn Følstad
Understandinginteraction with AI
Morten Goodwin
September 22
October 20
October 13
October 6
-
69
End session 2