See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/373924038

Machine Learning Tool for Kids: A Contribution to Teaching Computational

Thinking in Schools

Article · September 2023

CITATIONS READS

0 255

4 authors:

Jorge Pacheco João Paulo Ferreira

Governo Regional dos Açores Governo Regional dos Açores - Direção Regional da Educação
1 PUBLICATION 0 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

Helder Tavares Maura Miranda

University of the Azores University of the Azores
1 PUBLICATION 0 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jorge Pacheco on 20 November 2023.

The user has requested enhancement of the downloaded file.

 Machine Learning Tool for Kids :
A Contribution to Teaching Computational
Thinking in Schools
Jorge Pacheco João Ferreira Hélder Tavares Maura Miranda
University of Azores University of Azores University of Azores University of Azores
Ponta Delgada, Portugal Ponta Delgada, Portugal Ponta Delgada, Portugal Ponta Delgada, Portugal
[email protected] [email protected] [email protected] [email protected]

Abstract - Several authors have mentioned that we are CT allows people to face complex challenges, assist in their
in the age of Thinking with Artificial Intelligence (AI) and decision-making and help find solutions to solve problems in
that it is necessary to introduce AI, at different levels of a faster, more efficient and systematic way.
education, in order to prepare young people for the digital The Machine Learning for Kids platform was used in the
society of the XXI century. The school plays a key role study, which aims to encourage students to explore machine
in achieving this objective, otherwise we are encouraging learning in a playful and educational way, when students
info-excluded citizens, unable to follow the professions of the perform tasks in the field of AI. According to (Garcia et
future. Several world organizations note a mismatch between al., 2019) the ML4K platform is considered a pioneering
the curriculum and educational practices. Fortunately, tool in teaching this subject to children and young people.
Portugal has been slowly moving towards the integration Rock Paper Scissors is a hand game in which two players
and renewal of new technologies in our education system choose between three different options: each option can win
and we are in a process of digital transition. This study or lose to one of the other options, and the objective of the
aims to understand the potential of AI and the correlation game is to choose the option that beats the option chosen by
with Computational Thinking (CT) in an educational the opponent. It’s a simple and fun game that can be played
context, with a 9th grade class, from the 3rd Cycle of anywhere with just two people. Therefore, the objective of
Basic Education, in a public school on the island of S. this study is to evaluate the development of the CT, using the
Miguel and through a machine learning activity, based on ML4K platform and tool, with the application of an machine
the well-known game Rock Paper Scissors, using the AI learning activity and with ML4K’s Scratch3, based on the
platform and tool, Machine Learning for Kids (ML4K). Rock Paper Scissor game, with the use of a script available
The methodology used in this case study is qualitative on the aforementioned platform (Machine Learning for Kids,
and contributed to recognize AI and machine learning as 2023), adapted and translated on the Web & IA platform
a resource for development of CT. The results achieved (Web & IA). The choice of ML4K is justified because it has
with the study show that the use of AI topics promotes a friendly and intuitive interface for children, and because it
new ideas and new knowledge in students, which can help is used by several schools around the world. Regarding the
in understanding the basic concepts of CT. These results structure of this document, and after the introductory chapter,
are analyzed, discussed and related to another study in a literature review is presented that supports the concepts
the area and confirmed according to the proposed objective. covered. Subsequently, the methodology adopted to carry out
this study is described and the results obtained in the activity
Keywords: Artificial Intelligence, Computational Thinking, are analyzed. Finally, conclusions are presented.
Machine Learning, Learning Model
II. S EARCH E XPRESSION

I. I NTRODUCTION The search expression created and tested on Google Scholar

was “Machine Learning” AND “learning” AND “computa-
Currently, the technological advances that are being felt tional thinking”. The previous search expression uses the
throughout the world are transforming the society in which Boolean operator AND to combine the key terms “Machine
we live, creating forms of interaction and collaboration be- Learning for kids”, “learning” and “computational thinking”.
tween humans and technology. According to Wing (2006), the The expression searches for articles that address the use of the
ability to think computationally is not limited to just the use Machine Learning for kids tool as a contribution to teaching
of technology. The acquisition of skills associated with the computational thinking.
 III. L ITERATURE R EVIEW abilities, because it is accessible and easy to use. The plat-
form helps children stimulate their creativity, critical thinking
and programming in a visual programming language, which
According to Garcia et al. (2019) the proliferation of
contributes to the creation of machine learning solutions to
new technologies is changing the way we live, work and
problems and interactive experiences. ML4K has been used in
learn, shaping a world where humans and technology
schools around the world, enabling children to create machine
work together to discover and innovate. The CT emerges
learning models, for example, such as image recognition and
as an important response to the impact of technology on
as a project in Scratch, MIT APP Inventor and Python. It
society, helping people understand and use technology more
also supports different languages, enabling the integration of
effectively and responsibly. Yadav et al. (2016) emphasizes
children from different countries with different cultures.
that it is essential to develop CT in students at different
levels of education, as a way of preparing them for the
current and future world. From the perspective of Wong
et al. (2020) AI is a branch of computer science that is
increasingly omnipresent behind the scenes of our daily
lives and for Denning (2017) learning is a skill that occurs
when people get involved with it and practice it. According
to Caruso and Cavalheiro (2021), it is important to address
AI concepts in CT activities, as a motivating element “ (. . . )
and highlighting that the algorithms themselves, on which
AI is based, intrinsically depend on the CT. Therefore,
to understand how the first works, even superficially, it is
necessary to have notions of the second. In a complementary
way, AI is a highly motivating topic to be used in CT
teaching practices in the classroom, especially for children
and young people”. When students have contact with an
Fig. 1. ML4K Home Page with Login option
experience associated with AI, they understand that the
technology is not limited to its use; in a way, it helps
them better understand the underlying AI processes in our
society (Touretzky & Gardner-McCune, 2022). The AI4K12
initiative created a nationwide curriculum in the US and
also globally for teaching AI, and defined it in five big
ideas – Perception, Representation and Reasoning, Learning,
Natural Interaction and Social Impact (Touretzky et al.,
2019). Learning has three categories – Supervised Learning,
Unsupervised Learning and Reinforcement Learning.
Supervised learning is a machine learning technique in
which a model is trained on a set of classified data, where
each sample is associated with a pre-determined label or
class. The objective of supervised learning is to learn the
relationship between input variables and output variables, so
that the model can predict the output for new samples that Fig. 2. ML4K management page, where teachers can manage access and
supervise learning models.
were not seen during training (Saravanan & Sujatha, 2018).
As can be seen in figure 1, the platform’s home page is
IV. M ACHINE L EARNING FOR K IDS shown and figure 2 shows the management options for the
teacher, after logging in.

Among the various platforms on the Web, in addition to

Google’s Teachable Machine, there is ML4K. The ML4K
platform was developed by Dane Lane (former AI program- V. ROCK PAPER S CISSORS G AME
mer at IBM) and is supported by an IBM Watson API. It
is free, which allows children and teachers to adapt its use This game consists of playing with three shapes made with
to their specific needs and was created to help children learn your hand: Rock – Closed Hand; Paper – Open hand; Scissors
machine learning in a fun and educational way, using a simple – index and middle fingers visible. The objective of the game
and intuitive interface. Using the ML4K platform in teaching is to play by trying to guess in advance the shape of the
can help promote the inclusion of children with different hand, which our opponent will do in the next move, with
rock beating scissors, paper beating rock and scissors beating – Decomposition
paper. – Pattern Recognition
– Algorithm
– Logical Reasoning
– Critical Thinking
– Troubleshooting
– Algorithmic Thinking

The initial and final interviews were audio recorded, last

approximately 10 minutes each and have their own structure,
related to the topic under study and which contributes to
the assessment of students and the impact of the activity
(Amado, 2013). The sample was 10 students from a 9th
grade class, based on the knowledge of block programming,
which these students generally already have and by including
emerging technologies in the essential learning of Informa-
Fig. 3. Hand Shapes of Rock Paper and Scissors tion and Communication Technologies of the school year,
among others the AI. This class is made up of eighteen
students, 10 of whom participated in the study, six females
VI. M ETHODOLOGY and four males, aged between 14 and 16 years old. This
investigation took place at a public basic and secondary
The methodology used is the qualitative case study, and school, firstly including an initial interview and at the end
is used as a data collection technique, the interview of of the activity, the final interview using a previous script and
the structured type (which was previously validated with 2 a brief introduction to AI and machine learning. Next, the
students), with audio recording and registration in a digital activity that created a project as a machine learning model
form. The list of questions is identical to the interviews and in ML4K was carried out. It was carried out in a computer
the respective topics. The list of interview topics is presented. room, which has the necessary and appropriate technological
resources for carrying out the activity. The activity lasted
approximately 90 minutes, and students were divided into
Interview – Content Exploration Topics 5 groups of 2 students. Each group created their machine
Machine Learning Examples (Artificial Intelligence) learning image recognition model in ML4K, capturing the
– Master Algorithm three hand shapes through a webcam and thus introducing
– Decision Tree training data, performing learning with testing and importing
– Neural Networks the model into ML4K’s Scratch3 . Then, in order to simplify
– Big Data the construction of code blocks, they selected the template
– Deep Learning (among others) for the Rock Paper Scissors game, existing
– Supervised Learning on the platform. The students modified the code blocks,
– Unsupervised Learning strictly necessary to run the game, with the previously created
– Natural Language Processing (NLP) learning model and tested it. Throughout the activity, machine
– Recognition Numbers, Text, Image, Sounds learning concepts were related to the pillars of the CT –
Other examples of Artificial Intelligence Problem decomposition, Pattern recognition (prominent in the
learning model), Abstraction, Algorithm and also concepts
– Google Assistant, Siri, Alexa (Voice Assistants)
associated with the CT.
– ChatGPT (Virtual Assistants)
– Smart Robots
– Computer Vision
– Autonomous Cars
– Route Planning
– Genetic Algorithms
– Expert Systems
– Decision Making
– Machine Learning
Fig. 4. Learning project phases
– Knowledge Representation
– Personalized Online Recommendations Figure 4 shows the Train, Learn & Test and Make (Model
Computational Thinking Examples import) phases in ML4K.
– Abstraction
 Fig. 8. Students in the testing phase of the ML4K learning model.

Fig. 5. Learning Model Categories

Figure 8 shows students testing the learning model in

Figure 5 shows the Rock Paper Scissors categories with ML4K.
their associated images. The template includes standards for
each category.
VII. DATA ANALYSIS

The data analysis presented in this article aims to answer

specific questions related to the study.

Fig. 6. Project imported into Scratch3 and with blocks added from the
Learning Model

Figure 6 shows the ML4k Scratch3 project, with the Fig. 9. Results obtained in the Initial / Final interview
specific blocks of the learning model, as well as other blocks
necessary for image capture. The execution of the Rock We found, through the initial interview, that 10% of the
Paper Scissors game project is based on events between the students knew machine learning or machine learning. At the
different authors end of the activity, 100% of the students acquired the concepts
mentioned above.

Fig. 7. Execution of the project in Scratch3 Fig. 10. Results obtained in the Initial / Final interview

Figure 7 shows an execution of the project, with the rules of The concept of Artificial Intelligence was mentioned by
the game, in which there is movement of the player, in this 100% of students as having knowledge.
case of scissors and the computer - rock, and the respective
result - you lose, is presented.
 Fig. 11. Results obtained in the Initial / Final interview

In the initial interview, 10% of the students had knowledge

about the CT and in the final interview, everyone indicated
that they already knew. Fig. 12. Artificial Intelligence

Can you provide examples of machine learning ?

Examples of Google Assistant, Siri, Alexa, followed by
Description Number of Students intelligent robots were mentioned by students in the initial
Master Algorithm 1 and final interviews and in the final interview 3 students
Decision Tree 0 identified machine learning.
Neural networks 0
Big Data 0
Deep Learning 0
Supervised Learning 0
Unsupervised Learning 0
Natural language processing 0
Number and text recognition 1
Tabela I
R ESULTS OBTAINED IN THE I NITIAL INTERVIEW

In the initial interview, only one student indicated examples

of machine learning.

Description Number of Students

Master Algorithm 6
Decision Tree 0 Fig. 13. Pillars and concepts of Computational Thinking
Neural networks 0
Big Data 1 In the initial interview, only 1 student identified the
Deep Learning 1
Supervised learning 0
4 pillars of the CT. In the final interview, all students
Unsupervised learning 0 identified the pillars Abstraction, Decomposition and
Natural language processing 0 Algorithm and 8 students gave Pattern Recognition as
Number and text recognition 9 an example. Four students indicated Problem Solving as
Tabela II
R ESULTS OBTAINED IN THE F INAL INTERVIEW a pillar of CT and 1 student indicated Algorithmic Thinking.

.
VIII. C ONCLUSION
In the final interview, most students indicated number and text
recognition as examples, followed by the Master Algorithm, Based on the results obtained and relating to (Caruso &
Big Data and Deep Learning. Cavalheiro, 2021) in which “programming projects applying
AI are a powerful tool in CT development. Based on
the statistical analysis of the responses to questionnaires
administered before and after students participated in
the workshops with LearningML, there was a significant
improvement in learning concepts regarding AI.” In this
sense, the results obtained indicate that this study was
an added value and could be a promising and innovative
approach to the teaching and learning of AI and the
development of the CT. In the analysis carried out between
the use of the platform and the ML4K tool with the CT, it is
concluded that ML4K is a valuable contribution to teaching
the CT. As future work, we plan to use the ML4K platform
in the 3rd cycle and Secondary, with activities related to the
AI context, promoting the development of the CT.

R EFERENCES

Caruso, A. L. M., & Cavalheiro, S. A. da C.

(2021). Integração entre Pensamento Computacional
e Inteligência Artificial: uma Revisão Sistemática de
Literatura. Anais do XXXII Simpósio Brasileiro de
Informática na Educação (SBIE 2021), 1051–1062.
https://doi.org/10.5753/sbie.2021.218125
Denning, P. J. (2017). Remaining trouble spots with com-
putational thinking. Communications of the ACM, 60(6),
33–39. https://doi.org/10.1145/2998438
Garcia, J. D. R., Leon, J. M., Gonzalez, M. R., &
Robles, G. (2019). Developing Computational Thinking at
School with Machine Learning: An exploration. 2019 Inter-
national Symposium on Computers in Education (SIIE), 1–6.
https://doi.org/10.1109/SIIE48397.2019.8970124
Amado, J. (2013). Manual de investigação Qualitativa em
Educação (Imprensa da Universidade de Coimbra, Ed.).
Machine Learning for Kids. (2023, Janeiro 20). Rock,
Paper, Scissors. https://machinelearningforkids.co.uk/.
Saravanan, R., & Sujatha, P. (2018). A State of Art
Techniques on MachineLearning Algorithms: A Perspec-
tive of Supervised Learning Approaches in Data Classi-
fication. 2018 Second International Conference on Intelli-
gent Computing and Control Systems (ICICCS), 945–949.
https://doi.org/10.1109/ICCONS.2018.8663155
Touretzky, D., & Gardner-McCune, C. (2022). Artificial
Intelligence Thinking in K-12 in Computational Thinking Ed-
ucation in K-12: Artificial Intelligence Literacy and Physical
Computing.
Touretzky, D., Gardner-McCune, C., Martin, F., & Seehorn,
D. (2019). Envisioning AI for K-12: What Should Every Child
Know about AI? In Proceedings of the AAAI Conference on
Artificial Intelligence, volume 33,9795-9799.
Web & IA. Guião Rock, Paper, Scissors. (7 de março de
2023). https://webia.uac.pt
Wing, J. (2006). Computational Thinking. Communications
of the ACM.49(3): 33-35.
Wong, G. K. W., Ma, X., Dillenbourg, P., & Huan, J.
(2020). Broadening artificial intelligence education in K-12.
ACM Inroads, 11(1), 20–29. https://doi.org/10.1145/3381884

View publication stats