This document summarizes research on human-centered AI in AI education (AIED) systems. It discusses the need for transparency, interactivity, and collaboration between humans and AI in AIED. Some key points: 1) Early "expert systems" lacked transparency and trust, motivating research on explainable, transparent, and human-centered AI. 2) Modern research aims to make learner and content models visible, allow user control of AI parameters and recommendations, and provide explanations for AI decisions. 3) Several AIED systems are discussed that collaborate with users, visualize models, and give users control over navigation, rankings, and social comparisons to improve learning outcomes.

1. Human-Centered AI in AIED
Peter Brusilovsky
with Sergey Sosnovsky, Julio Guerra,
Jordan Barria-Pineda, Kamil
Akhuseyinoglu,
PAWS Lab
School of Computing and Information
University of Pittsburgh

2. • A problem with a long history in AI
– The experience of “expert systems”
• Modern issues
– Possible biases of AI-based decisions with
no ability to inspect
– Lack of trust to recommendations coming
from AI
– Limited ability to control or impact AI 2
AI vs Human?

3. AI or Humans + AI?
3

4. • Transparent AI
• Explainable AI
• Human-in-the Loop AI/ML
• Human-Centered design of AI systems
• “Natural” communication with AI systems
• User interfaces for recommender systems
– Controlling and explaining recommendations
5
Growing stream of research

5. What’s about AI in Education?
6
A classic view on an AI-Ed system architecture

6. Transparency and Interactivity
Two Sides of the Same Coin
Visualize Explain
Control
Collaborate
7
Transparency
Interactivity
No full transparency
without interactivity
Interaction is challenging
without transparency

7. What are Possible Solutions?
• Collaborate
– Work together with AI to make best decisions
• Visualize
– What is the current state of the learner model that drives AIED
decisions?
– What is the content model of the current activity?
• Control
– Edit or negotiate your learner model
– Express your goals and preferences in the learning process
• Explain
– Why a specific learning activity item is good at the current point?
8

8. Collaborate!
Adaptive Hypermedia: Human-AI Collaboration
9

9. Human-AI Collaboration
10

10. Navigation vs. Adaptive Sequencing
11
Human makes navigation decisions AI makes navigation decisions

11. Adaptive Navigation Support
AI provides information,
human makes an informed decision
AI and human collaborate

12. ELM-ART: Adaptive Annotation (1996)
Weber,
G.
and
Brusilovsky,
P.
(2001)
ELM-ART:
An
adaptive
versatile
system
for
Web-based
instruction.
International
Journal
of
Artificial
Intelligence
in
Education
12
(4),
351-384.

13. ELM-ART: Evaluation
• No formal classroom study
• Users provided their experience
• Drop-out evaluation technology
• 33 subjects
– visited more than 5 pages
– have no experience with Lisp
– did not finish lesson 3
– 14/19 with/without programming experience

14. ELM-ART: Value of Annotation
Mean number of pages which the users with experience in at least one
programming language completed with ELM-ART
Users with better starting knowledge work twice as longer
when they can collaborate with AI

15. ELM-ART: Value of Sequencing
Mean number of pages which the users with no experience in
programming languages completed with ELM-ART
Users with no experience work twice as longer with direct AI
guidance

16. NavEx: Adaptive Annotation
Brusilovsky,
P.
and
Yudelson,
M.
(2008)
From
WebEx
to
NavEx:
Interactive
Access
to
Annotated
Program
Examples.
Proceedings
of
the
IEEE
96
(6),
990-999.

17. Adaptive Annotation Can:
• Reduce navigation efforts
• Reduce repetitive visits to learning content
pages
• Encourage non-sequential navigation
• Increase learning outcome
• For those who is ready to follow and advice
• Make system more attractive for students
• Students stay much longer without any reward

18. VISUALIZE!
Making learner and content models visible
19

19. Making Hidden Visible
20
A classic view on an AI-Ed system architecture

20. Open Learner Models
21
Bull, S., Brusilovsky, P., and Guerra, J. (2018) Which Learning Visualisations to Offer Students? In: V. Pammer-Schindler, M. Pérez-Sanagustín, H.
Drachsler, R. Elferink and M. Scheffel (eds.) Proceedings of 13th European Conference on Technology Enhanced Learning, EC-TEL 2018, Leeds, UK,
September 3–5, 2018, Springer, pp. 524–530.

21. InterBook: Open Content Model
22
Brusilovsky,
P.
and
Anderson,
J.
(1998)
ACT-R
electronic
bookshelf:
An
adaptive
system
for
learning
cognitive
psychology
on
the
Web.
Proceedings
of
WebNet'98,
World
Conference
of
the
WWW,
Internet,
and
Intranet,
Orlando,
FL,
November
7-12,
1998,
AACE,
pp.
92-97.

22. List of annotated
links to all quizzes
available for a
student in the
current course
Refresh
and help
icons
QuizGuide: OLM+ANS
Sosnovsky,
S.
and
Brusilovsky,
P.
(2015)
Evaluation
of
Topic-based
Adaptation
and
Student
Modeling
in
QuizGuide.
User
Modeling
and
User-Adapted
Interaction
25
(4),
371-424.

23. QuizGuide: Adaptive Annotations
• Target-arrow: OLM
– Number of arrows – level of
knowledge for the specific
topic (from 0 to 3).
Individual, event-based
adaptation.
• Color Intensity: ANS
– Learning goal (current,
prerequisite for current,
not-relevant, not-ready).
 Topic–quiz organization:

24. Mastery Grids: Personalized
Practice System with OSLM
Loboda, T. D., Guerra, J., Hosseini, R., & Brusilovsky, P. (2014, September). Mastery grids: An open source social educational progress
visualization. In European conference on technology enhanced learning (pp. 235-248). Springer, Cham.
Learning content
OSLM Features

25. Topic-Level vs. Concept-level OLM
26
Guerra Hollstein, J., Barria Pineda, J., Schunn, C., Bull, S., and Brusilovsky, P. (2017) Fine-
Grained Open Learner Models: Complexity Versus Support. In: Proceedings of Proceedings of the 25th
Conference on User Modeling, Adaptation and Personalization, Bratislava, Slovakia, ACM, pp. 41-49.

26. Transparency and Interactivity
Two Sides of the Same Coin
Visualize Explain
Control
Collaborate
28
Transparency
Interactivity
No full transparency
without interactivity
Interaction is challenging
without transparency

27. CONTROL!
Allow the user to control various parameters of the AI process to
better adapt personalization / presentation for the current context.
Avoid guessing! Let the user contribute!
29

28. What Can Be Controlled?
30
Data Process Presentation
User Model Ranking
Source Fusion
What/how
to present

29. Indirect Control: Editable OLM
31
Weber, G. and Brusilovsky, P. (2001) ELM-ART: An adaptive versatile system for Web-based instruction. International Journal of
Artificial Intelligence in Education 12 (4), 351-384.

30. RelevanceTuner: Ranking Control in a
Hybrid Social Recommender
Tsai, Chun-Hua and Peter Brusilovsky (2018) Beyond the Ranked List: User-Driven Exploration and Diversification of Social
Recommendation. In 23rd International Conference on Intelligent User Interfaces, 239--50. Tokyo, Japan: ACM.

31. Mastery Grids: Open Social LM
33
Brusilovsky,
P.,
Somyurek,
S.,
Guerra,
J.,
Hosseini,
R.,
Zadorozhny,
V.,
and
Durlach,
P.
(2016)
Open
Social
Student
Modeling
for
Personalized
Learning.
IEEE
Transactions
on
Emerging
Topics
in
Computing
4
(3),
450-461.

32. Student-Controllable Presentation
of Social Comparison
Default Class Average
Change to Lower progress
Change to Higher progress
Akhuseyinoglu, K., Barria-Pineda, J., Sosnovsky, S., Lamprecht, A.-L., Guerra, J., and Brusilovsky, P. (2020)
Exploring Student-Controlled Social Comparison. In: Proceedings of European Conference on Technology
Enhanced Learning EC-TEL 2020, Cham, 14-18 September, 2020, Springer International Publishing, pp. 244-258.

33. Student-controllable Social
Comparison: Average
35

34. Student-controllable Social
Comparison: Lower
36

35. Study Details
Classroom study at Utrecht University in early 2020
– Introduction to computational thinking course
– Python Programming
Single condition
Python Grids offered as optional practice
To increase motivation
– 2% bonus offered towards homework submissions
– One Question and One Problem in each topic
Analyzed logs of 44 students
– Practiced with 1+ learning content
37

36. Total 706 group changes (n=40, M=17.65, SD=19.09)
– 91% (n=40) 1+ changes
– 77% (n=32) 5+ changes
– 34% (n=15) 10+changes
Total 139 ranked list view (n=40, M=3.48, SD=7.74)
38
Did Students Use Control Features?

37. Results
40
Control features used considerably
– Preferred more average and higher
– Performed frequent changes
Positive association with more practice
– Practiced more with all activity types
Positive upward social comparison
Better navigational support
Topics to catch
the group
Topic that student
worked more than
group

38. EXPLAIN!
Make it more clear for students why specific recommended
learning activities are recommended and how they relate to their
knowledge and learning goals
42

39. Transparent Educational Recommender?
43
Lack of
Transparency !

40. ATEC Worksho
2 0 1 9
Los Angeles
44
Explanations with OLM: Success
Barria-Pineda,
Jordan,
and
Peter
Brusilovsky.
2019.
"Explaining
Educational
Recommendations
Through
a
Concept-level
Knowledge
Visualization."
In
Proceedings
of
the
24th
International
Conference
on
Intelligent
User
Interfaces:
Companion,
103--04.
New
York,
NY,
USA:
ACM.

41. 1
45
Explanations with OLM: Balance
Higher adoption of
recommended
learning content
Barria-Pineda, J., Akhuseyinoglu, K., Želem-Ćelap, S., Brusilovsky, P., Klasnja Milicevic, A., and
Ivanovic, M. (2021) Explainable Recommendations in a Personalized Programming Practice System. In: 22nd
International Conference on Artificial Intelligence in Education, AIED 2021, Springer, pp. 64-76.

42. 46
Work with non-recommended
activities
Lower adoption of
non-recommended
learning content
Explanations with OLM: Visual

43. The presence of recommendations and explanations could
increase student engagement with knowledge-relevant
learning content leading to a higher success rate.
48
Explanatory features of educational recommendations were
used extensively for exploring the learning materials.
Students with lower entry-levels of
knowledge !
Explanations provide students an opportunity
for reflecting on the appropriateness of
the content for supporting each step of their
learning
Especially students with higher
entry-levels of knowledge !
Explanations with OLM: Summary

44. Explanations with OLM: Remedial
Knowledge estimates as bar chart
Related concepts highlighted
49
Recent success rate as bar color
Warning sign on “struggled”
concepts

45. Explanations with OLM: Remedial
Textual explanations
# of “struggled” concepts
# of “proficient concepts”
(Knowledge Est. > .66)
50
Barria-Pineda, Jordan, Kamil Akhuseyinoglu, and Peter Brusilovsky. 2019. "Explaining Need-based Educational
Recommendations Using Interactive Open Learner Models." In International Workshop on Transparent Personalization
Methods based on Heterogeneous Personal Data, ExHUM at the 27th ACM Conference On User Modelling, Adaptation
And Personalization, UMAP '19. Larnaca, Cyprus.

46. Transparency and Interactivity
Two Sides of the Same Coin
Visualize Explain
Control
Collaborate
51
Transparency
Interactivity
No full transparency
without interactivity
Interaction is challenging
without transparency

47. Questions?
52

48. Readings
• Brusilovsky, P. and Anderson, J. (1998) ACT-R electronic bookshelf: An adaptive system for learning cognitive psychology on the
Web. Proceedings of WebNet'98, World Conference of the WWW, Internet, and Intranet, Orlando, FL,, pp. 92-97.
• Weber, G. and Brusilovsky, P. (2001) ELM-ART: An adaptive versatile system for Web-based instruction. International Journal
of Artificial Intelligence in Education 12 (4), 351-384.
• Brusilovsky, P. and Yudelson, M. (2008) From WebEx to NavEx: Interactive Access to Annotated Program Examples.
Proceedings of the IEEE 96 (6), 990-999.
• Loboda, T. D., Guerra, J., Hosseini, R., & Brusilovsky, P. (2014). Mastery grids: An open source social educational progress
visualization. In European conference on technology enhanced learning (pp. 235-248). Springer, Cham.
• Sosnovsky, S. and Brusilovsky, P. (2015) Evaluation of Topic-based Adaptation and Student Modeling in QuizGuide. User
Modeling and User-Adapted Interaction 25 (4), 371-424.
• Brusilovsky, P., Somyurek, S., Guerra, J., Hosseini, R., Zadorozhny, V., and Durlach, P. (2016) Open Social Student
Modeling for Personalized Learning. IEEE Transactions on Emerging Topics in Computing 4 (3), 450-461.
• Guerra Hollstein, J., Barria Pineda, J., Schunn, C., Bull, S., and Brusilovsky, P. (2017) Fine-Grained Open Learner
Models: Complexity Versus Support. In: Proceedings of Proceedings of the 25th Conference on User Modeling, Adaptation and
Personalization, Bratislava, Slovakia, ACM, pp. 41-49.
• Tsai, Chun-Hua and Peter Brusilovsky (2018) Beyond the Ranked List: User-Driven Exploration and Diversification of Social
Recommendation. In 23rd International Conference on Intelligent User Interfaces, 239--50. Tokyo, Japan: ACM.
• Barria-Pineda, J., Akhuseyinoglu, K., and Brusilovsky, P. (2019) Explaining Need-based Educational Recommendations Using Interactive
Open Learner Models. In: Proceedings of International Workshop on Transparent Personalization Methods based on Heterogeneous Personal Data,
ExHUM at the 27th ACM Conference On User Modelling, Adaptation And Personalization, UMAP '19, Larnaca, Cyprus, June 09, 2019, pp. 273–277.
• Akhuseyinoglu, K., Barria-Pineda, J., Sosnovsky, S., Lamprecht, A.-L., Guerra, J., and Brusilovsky, P. (2020)
Exploring Student-Controlled Social Comparison. In: Proceedings of European Conference on Technology Enhanced Learning EC-TEL
2020, Cham, 14-18 September, 2020, Springer International Publishing, pp. 244-258.
• Barria-Pineda, J., Akhuseyinoglu, K., Želem-Ćelap, S., Brusilovsky, P., Klasnja Milicevic, A., and Ivanovic, M. (2021)
Explainable Recommendations in a Personalized Programming Practice System. In: I. Roll, D. McNamara, S. Sosnovsky, R. Luckin and
V. Dimitrova (eds.) Proceedings of 22nd International Conference on Artificial Intelligence in Education, AIED 2021, Utrecht, The
Netherlands, June 14-18, 2021, Springer, pp. 64-76. 53