=Paper=
{{Paper
|id=Vol-3136/paper3
|storemode=property
|title=Integrating human-centered artificial intelligence in programming practices to reduce teachers' workload
|pdfUrl=https://ceur-ws.org/Vol-3136/paper-3.pdf
|volume=Vol-3136
|authors=Renate Andersen,Eli Gjølstad,Anders Mørch
|dblpUrl=https://dblp.org/rec/conf/avi/AndersenGM22
}}
==Integrating human-centered artificial intelligence in programming practices to reduce teachers' workload==
https://ceur-ws.org/Vol-3136/paper-3.pdf
Integrating Human-Centered Artificial Intelligence in
Programming Practices to Reduce Teachers’ Workload
Renate Andersen1, Eli Gjølstad1 and Anders I. Mørch2
1
Oslo Metropolitan University, Pilestredet 52 Oslo, 0167, Norway
2
University of Oslo, Gaustadalléen 21, Oslo, 0349, Norway
Abstract 1
Artificial intelligence (AI) has impacted every industry, including the education sector. In this
position paper, we explore how human-centered AI (HCAI) can be integrated in programming
activities in school rather than discussing the design of user interfaces of HCAI systems. Our
main proposal is to integrate HCAI in educational practices to reduce teachers’ workload by
providing meaningful scaffolds to the learners, connecting technology and domain knowledge.
Keywords
Human-centered AI, education, programming, school, scaffolding
1. Introduction
Human-centered artificial intelligence (HCAI) has received increasing attention in recent years,
including the national educational systems worldwide. The United Nations Educational, Scientific and
Cultural Organization (UNESCO)’s (2022) mandate is to focus on human-centered approaches to AI–
“AI for all,” where everyone can take advantage of the technological revolution underway and access
it. UNESCO states that the connection between AI and education involves three main areas: 1) learning
with AI (e.g., the use of AI-empowered tools in classrooms), learning about AI (AI technologies and
techniques), and preparing for AI (e.g., enabling citizens to better understand the potential impact of AI
on human lives [14]. Here, we focus on the first approach, “learning with AI.”
Although HCAI in education is an emerging field, there remains limited research. Nonetheless,
Yang et al. [17] state that the research trends have brought new applications of AI in education, e.g.,
adoption of machine learning and new deep learning algorithms. Furthermore, AI research can
potentially improve intelligent tutoring with more precise adaptation and personalization. When
focusing on HCAI, the emphasis is on “learning with AI,” e.g., Replika [7], a virtual friend or a chatbot
companion powered by AI; Thinkster [12], a virtual math tutor built with AI to create personalized
learning programs; and Cognii [2], a virtual learning assistant that engages students using
conversational AI. This position paper adds to the debate along this line of research. The research
questions addressed in this paper are as follows: 1) How is programming integrated in learning school
subjects? and 2) How can programming take advantage of HCAI to improve learning? This paper
focuses on exploring how HCAI can be integrated into programming activities in schools, not on the
design of user interfaces for AI systems. We ground our position and discussion of HCAI in research
as activities that stem from educational research on programming in schools.
Proceedings of CoPDA2022 - Sixth International Workshop on Cultures of Participation in the Digital Age: AI for Humans or Humans for
AI? June 7, 2022, Frascati (RM), Italy
EMAIL: renatea@oslomet.no (R. Andersen); Eli.Gjolstad@oslomet.no (E. Gjølstad); anders.morch@iped.uio.no (A. I. Mørch)
ORCID: 0000-0002-1206-2140 (R. Andersen); 0000-0002-1470-5234 (A. I. Mørch)
© 2022 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Wor
Pr
ks
hop
oceedi
ngs
ht
I
tp:
//
ceur
-
SSN1613-
ws
.or
0073
g
CEUR Workshop Proceedings (CEUR-WS.org)
30
2. Related work
2.1. Human-centered artificial intelligence
HCAI empowers developers to build and design AI systems that support human self-efficacy,
promote creativity, clarify, and distribute responsibility, and facilitate social participation [11], thus
putting humans at the center of design thinking. Shneiderman [11] underscores that the goal in HCAI
is to put human users at the center, emphasizing user experience design, measuring human performance,
and celebrating the new powers that people have. However, this contradicts the traditional AI view
where developers and researchers focus on building AI algorithms and systems for machine autonomy,
measuring algorithmic performance, and celebrating what AI can do on its own. Therefore,
Shneiderman [10] argues that HCAI presents three ideas that go beyond automatization: 1) the
possibility of high levels of human control and high levels of automation, 2) shift from emulating
humans to empowering people, and 3) governance structures for HCAI (reliable, safe, and trustworthy
systems).
It is important to highlight that at the heart of HCAI is to recognize that the way intelligent systems
solve problems, especially machine learning, is fundamentally alien to humans without computer
science knowledge [8]. This underscores the need for future generations of pupils to learn about
programming, computational thinking, and computer science to enable them to understand the
algorithms underlying advanced AI systems such as deep learning and be able to interpret them
effectively. Riedl [8] emphasizes that HCAI can be divided into two main aspects: 1) AI systems that
understand humans from a sociocultural perspective and 2) AI systems that help humans understand
themselves. In this paper, we focus on the first aspect, which is useful when discussing HCAI in the
context of an educational institution. Shneiderman [10]. presents an HCAI framework using three main
ideas: 1) design for high levels of human control and high levels of automation, 2) understand the
situations in which full human control or full computer control are necessary, and 3) avoid the dangers
of excessive human control or excessive computer control.
2.2. A sociocultural perspective on learning
In a sociocultural perspective on learning, learning is seen as context-bound, situated in
socialpractices, and mediated by symbolic and cultural artifacts. Hence, this approach emphasizes
participation in different social practices [9]. This paper takes on the view that learning can best be
understood as social interactions mediated by artifacts; more precisely, in our case, the learning
processes during programming are mediated by technological tools, resulting in social interactions
between the teachers, learners, and AI chatbot. A central concept within the sociocultural perspective
on learning is pedagogical scaffolding [16]. Hammond and Gibbons [3] define scaffolding as how
teachers and other seniors help and support peers by providing feedback in the learning process. Maybin
and colleagues [4] define scaffolding as different kinds of support the learners receive in their
interaction with parents, teachers, and other mentors as they move towards new skills or concepts. A
current direction in this area is automated text analysis, such as EssayCritic [6]. Another central concept
in a sociocultural perspective on learning is the zone of proximal development, which is a key concept
derived by Vygotsky [15] as ways in which individuals move between different stages of development
and their potential learning levels.
3. Methods
The empirical data presented in this article is derived from a design-based research project [5], in
which we were participant observers in a classroom that used block-based programming (MakeCode in
micro:bit) as an exploratory design space for solving physics tasks assigned by the teacher. The project
consists of four interventions over a 2-year period including a total of 130 pupils aged 12–16 years. The
pupils met 3 hours/week for 16 weeks over 2 semesters. Data were collected using video recordings of
classroom interventions. The data presented below is derived from the last intervention, fourth, where
31
we were in a physics lab and recorded a video of a group of pupils using programming for solving
physics tasks.
Thematic analysis was used to analyze the entire dataset and screen for emerging common topics.
Thematic analysis is a qualitative method for identifying and organizing patterns of meanings across a
dataset to enable the researcher to make sense of collective or shared meaning and experiences [1].
Examples of thematic codes that emerge when screening data are knowledge sharing, programming,
computational concepts, and collaboration.
4. Findings and design scenario to include human-centered AI
In this section, we will address the research questions: 1) How is programming integrated into
learning school subjects? and 2) How can programming take advantage of HCAI to improve learning?
We address the research questions by presenting two different data excerpts that show how pupils work
when using programming for solving physics tasks assigned by the teacher. Presenting the empirical
data extract address the first part of the research question exemplifying how programming is integrated
into school subjects. This is useful to provide a realistic context of how these data can be extended to
integrate HCAI as a scaffold to help the learner. However, in the table where the data is presented, we
have added a column named “AI chatbot,” which is our design scenario, being an example of how
HCAI can be integrated and connected to programming practices when using micro:bit in an educational
context.
When sketching out a design scenario for how a micro:bit can be extended to include HCAI, we
uncovered that there are at least two different yet relevant directions: 1) focusing on scaffolding the
domain-specific knowledge learning process (in our example, physics) or 2) scaffolding learning
programming and how to use the micro:bit (a third approach emphasizing collaboration is presented as
direction for further work). We will present one design scenario from each to cover both directions,
reflecting data extract 1 and data extract 2. Data extract 1 and 2 below derive from one physics class
where the students use micro:bit for solving physics related tasks. The AI chatbot presented in the fourth
column in the tables is not included per se in the study, as it is created as an add-on for suggesting a
future design scenario for reflecting on how an AI chatbot can be integrated in an already existing
technology used in school, the micro:bit. These data extracts derive from a research project where we
developed and implemented technology rich interventions in several K-12 classrooms, consisting of
pupils ranging 12-16 years old. We followed three classes (20 students in each class) over two years
and videorecorded our observations in the classroom of the students when they were working in groups
on using micro:bit for solving subject-specific tasks. The scenarios are based on the lessons we learned.
Table 1 below presents data extract 1 derived from a lesson in physics where four pupils are working
together on a task given by the teacher on how to program and use a micro:bit to measure conductivity.
The pupils are experiencing problems with the micro:bit and the code as the assembly is not working
properly in connection with measuring conductivity. Two of the four pupils are discussing the problem
that starts by one of them asking a question.
Table 1. Data extract 1. Pupils using micro:bit to measure conductivity.
Line Participant Verbal utterance (comments in parenthesis) AI chatbot
1 Student 1 “Shall we see on the micro:bit. It looks like What are you working on now?
that.” (Looks at the code on the screen)
2 Student 2 “Yes. Because the number is there.” (pointing
to the micro:bit)
“I think that’s correct.”
3 Student 1 “Okay.”
4 Student 2 “We can try it here again.” (Connects the wires Are you having trouble with how to connect
and measures conductivity) pin1 and pin2 to the micro:bit? Type yes, if
you need help.
5 Student 1 “Ok. Fine now! (Does it work?)”
6 Student 2 “But it shows that the air conducts electricity?” Are you still having a problem with
connecting pin1 and pin2 to the micro:bit?
32
I can see you have also connected a
resistor, is the resistance correct?
7 Student 1 “Okay, then we’ll see if we have connected
something wrong (with the micro:bit).”
8 Student 2 (Checks the wires) “Okay, the wires are Have you checked that the wires are
connected properly to the micro:bit, but what’s connected?
weird is to be seen here!” (Holds the pins up in
the air and micro:bit shows that it conducts
electricity)
9 Student 1 “Maybe the air does conduct electricity then.”
10 Student 2 “No!” (Frustrated)
“If the current had moved through then you
would have seen it. Then it would have jumped
lightning and you would have heard it.”
11 Student 1 (Looking at the code for the micro:bit) “Press I will help you solve the problem. I will give
B for conductivity and A for resistance ..., but you a list of possible solutions – type yes to
there is something wrong here … . Look here. the solution that you have tested.
I did not hold them close to each other and the
micro:bit shows 0.001! And it is physically
impossible.”
Table 2 presents data extract 2, where the pupils are discussing how to use micro:bit to test
conductivity on different objects.
Table 2: Data extract 2: Using micro:bit to test conductivity on different objects
Line Participant Verbal utterance (comments in parenthesis) AI chatbot
1 Student 3 “All the measures get the result of 0,001.” (Using Greetings, what are you using the micro:bit
the micro:bit to measure the conductivity on his for today?
own finger)
2 Student 4 “I’m not holding them (the pins on the micro:bit Right now, you are not measuring
that measures conductivity) to anything, and the Conductivity in the air, I can feel that I am
micro:bit shows 0.001, and it’s physically not connected to anything with pin 1 and
impossible. Okay, then we have to find out what pin2. Please plug the pins to the object you
happened.” (Using the micro:bit to measure the are going to measure conductivity through,
conductivity in the air) and I will give you the answer.
3 Student 3 “It probably has something to do with the fact that What are you measuring the conductivity of
there are salts (sodium compounds) and minerals now?
that conducts electricity relatively good.”
(Measures the conductivity of Farris (mineral
water)
4 Student 4 “We can look at the nutritional content and
ingredient list of Cola and Pepsi afterward to see
if there are any differences.” (Measures the
conductivity of Pepsi and Cola)
7 Student 3 “What do we think? Do we think the orange juice
or lemon has the highest conductivity?”
(Measures conductivity of a lemon)
8 Student 4 “Acids have pretty good conductivity I have
heard.”
5. Discussion and conclusions: Implications for future design
The main argument in this position paper is that integrating HCAI in education has a significant
potential to reduce teachers’ workload. As shown above, we present two data extracts regarding how
programming is practiced in schools in our country, and we extended our view of learning by adding a
column describing a design scenario concerning how an AI chatbot can scaffold these learning
processes. The AI chatbot integrated into the micro:bit can be interacted with in two ways: 1) requesting
33
the chatbot directly in the chat window (top-down invocation) or 2) it infers the need for scaffolding
based on the learner’s actions and what the learner says (bottom-up invocation). When the computer is
connected to the micro:bit, the chatbot uses the microphone on the computer, listens to the
conversations, and tries to suggest scaffolds connected to what the learners are talking about.
However, as we can read out from extract 1 (Table 1) and extract 2 (Table 2) the AI chatbot interacts
in a multi-user context, which impacts how the AI chatbot reacts. In extract 1 and 2 we have group
collaboration as a premise, meaning several students are working together to create the code, however,
there is only one student that interacts with the computer and creates the program. This implies that
there is only one student that directly interacts with the AI chatbot during this time. However, it is
important to reflect upon how one can design a chatbot that also takes the interaction among several
interacting pupils into consideration. Can a future design scenario be that the AI chatbot can take
questions from different pupils at the same time into consideration? It would be useful with an
interactive AI chatbot that also can handle social interaction in groups. Recent research on chatbots
shows that a role for a chatbot could be to encourage non-active pupils to be more active, engaging
them in the discussions with peers and with the chatbot, helping students to become better collaborators,
identified as an important 21st century skill [13].
Another interesting reflection around a future design issue with the AI chatbot is to examine how it
can take advantage of context-awareness. A context aware chatbot must seek to understand and support
the aims of the user. In our case, in extract 1 and 2, which is a programming context, it quite essential
that the AI chatbot can be aware of this context to be able to support the pupils in their specific learning
activity. This means that the AI chatbot should have some built in mechanism for adaptive learning that
enable to learn over time through interaction pupils of different need and background knowledge. The
IA chatbot we have profiled is a programming expert with the aim of scaffolding pupils in K-12/lower
and upper secondary school when learning programming.
Summing up, the research contribution with this position paper is discussing a scaffolding scenario
for using an integrated chatbot in the online micro:bit programming environment for pupils to learn
together how to use a new technology in an educational context from the challenging position of relating
the technology to domain knowledge (e.g., middle school physics as we have profiled here).
It is interesting to reflect on the impacts of HCAI on both learners and teachers. For instance, learners
get faster and instant feedback, with a greater chance of receiving more accurate answers. There are
several benefits of integrating HCAI into programming processes in schools, such as reducing teachers’
workload and enabling a more flexible and accessible teaching experience to students. However, one
of the most important challenges with creating an AI chatbot is providing it with accurate context to
enable effective interaction with the learners. As seen in previous research, scaffolding pupils is not
easy due to many complex factors that may impact the learning process. In conclusion, the main findings
in this paper are as follows:
• An AI chatbot can provide meaningful scaffolds to pupils when learning to program
• Exploring HCAI from a sociocultural perspective on learning leads to interesting aspects
of how HCAI can be designed as scaffolds in an educational context, e.g., classrooms.
• Integrating HCAI in the educational context can reduce teachers’ workload.
Future research regarding how an AI chatbot could be designed and implemented into the micro:bit
environment is warranted. Integrating HCAI in education, especially as an approach to “learning with
AI,” has great potential to reduce teachers’ workload in classroom settings.
6. References
[1] Braun, V., & Clarke, V. (2012). Thematic analysis. In H. Cooper, P. M. Camic, D. L. Long, A. T.
Panter, D. Rindskopf, & K. J. Sher (Eds.), APA handbook of research methods in psychology, Vol.
2. Research designs: Quantitative, qualitative, neuropsychological, and biological (pp. 57–71).
American Psychological Association. https://doi.org/10.1037/13620-004
[2] Cognii. (2022). Artificial intelligence for education. https://www.cognii.com/
34
[3] Hammond, J., & Gibbons, P. (2005). Putting scaffolding to work: The contribution of scaffolding
in articulating ESL education. Prospect, 20(1), 6–30.
[4] Maybin, J., Mercer, N., & Stierer, B. (1992). ‘Scaffolding’: Learning in the classroom. In K.
Norman (Ed.), Thinking voices: The work of the National Oracy Project (pp. 186–195). Hodder &
Stoughton.
[5] McKenney, S., & Reeves, T. C. (2018). Conducting educational design research. Routledge.
[6] Mørch, A. I., Engeness, I., Cheng, V. C., Cheung, W. K., & Wong, K. C. (2017). EssayCritic:
Writing to learn with a knowledge-based design critiquing system. Educational Technology &
Society, 20(2), 216–226.
[7] Replika. (2022). The AI companion who cares. https://replika.ai/
[8] Riedl, M. O. (2019). Human‐centered artificial intelligence and machine learning. Human
Behavior and Emerging Technologies, 1(1), e117. http://dx.doi.org/10.1002/hbe2.117
[9] Säljö, R. (2001). Learning in practice: A sociocultural perspective. Cappelen Damm Akademisk.
[10] Shneiderman, B. (2020a). Human-centered artificial intelligence: Three fresh ideas. AIS
Transactions on Human-Computer Interaction, 12(3), 109–124.
https://doi.org/10.17705/1thci.00131
[11] Shneiderman, B. (2020b). Human-centered artificial intelligence: Reliable, safe & trustworthy.
International Journal of Human–Computer Interaction, 36(6), 495–504.
https://doi.org/10.1080/10447318.2020.1741118
[12] Thinkster. (2022). Get hyper-personalized attention from an online math tutor. Daily.
https://hellothinkster.com/online-math-tutor.html
[13] Trent, S. & Campa, A. (August 2021). Schools Look for Help from AI Teacher’s Assistants.
Researchers are developing artificial intelligence aimed at keeping students engaged and saving
educators’ time. The Wall Street Journal. https://www.wsj.com/articles/schools-look-for-help-
from-ai-teachers-assistants-11628262062?mod=article_inline
[14] United Nations Educational, Scientific and Cultural Organization (UNESCO). (2022). Artificial
intelligence in education. https://en.unesco.org/artificial-intelligence/education
[15] Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes.
Harvard University Press.
[16] Wood, D., Bruner, J. S., & Ross, G. (1976). The role of tutoring in problem solving. Journal of
Child Psychology and Psychiatry, 17(2), 89–100. https://doi.org/10.1111/j.1469-
7610.1976.tb00381.x
[17] Yang, S. J., Ogata, H., Matsui, T., & Chen, N. S. (2021). Human-centered artificial intelligence in
education: Seeing the invisible through the visible. Computers and Education: Artificial
Intelligence, 2, 100008. https://doi.org/10.1016/j.caeai.2021.100008
35