Developing Computational Thinking at School with Machine Learning: An exploration

Artificial Intelligence (AI) and Machine Learning (ML) have heavily irrupted in society, bringing new applications and possibilities while introducing some ethical problems. Governments and institutions around the world are working on the challenges posed by AI in all aspects, from economy to education. Therefore, introducing AI-related content at school and exploring how this kind of content can be taught becomes mandatory. In this paper we carry out a bibliographic revision of previous works done on ML, and then describe an educational resource developed by the institution of the first two authors (INTEF) aimed to teach ML in schools with Scratch and Machine Learning for Kids. The testimonials of three educators, who have implemented their own version of these resources, are depicted. More efforts should be made to introduce AI-related content in education.

You can get a copy of this paper at: https://www.researchgate.net/publication/337474585_Developing_Computational_Thinking_at_School_with_Machine_Learning_An_exploration

Transcript

1. Developing Computational Thinking at School with Machine Learning: An exploration Juan David Rodríguez-García (INTEF), Jesús Moreno-León (Programamos), Marcos Román-González (UNED), & Gregorio Robles (URJC) SIIE 2019, Tomar (Portugal), 21-23 November 2019
2. Artificial Intelligence (AI) and Machine Learning (ML) are everywhere… AI and ML have heavily irrupted in society, bringing new applications and possibilities while introducing some ethical problems
3. What about kids and AI/ML? Children are daily using, more or less consciously, software programs with AI/ML capabilities such as facial or speech recognition, recommending systems, spam detectors… Consequently?
4. AI/ML on education ≈ a priority concern for governments
5. Computational Thinking (CT) and Machine Learning (ML)?
6. Computational Thinking (CT) and Machine Learning (ML)? Algorithmic design and implementation by means of computer programming
7. Computational Thinking (CT) and Machine Learning (ML)? Algorithmic design and implementation by means of computer programming Model building and training/testing through ML techniques
8. What is our goal? To introduce AI/ML-related content at primary/secondary school and to explore how this content can be properly taught.
9. Our context… http://code.intef.es/inteligencia-artificial-en-el-aula-con-scratch-3-0/
10. Our context… http://code.intef.es/inteligencia-artificial-en-el-aula-con-scratch-3-0/ Scratch 3.0 ML4K
11. Our context… http://code.intef.es/inteligencia-artificial-en-el-aula-con-scratch-3-0/
12. A – Initial B/C - Conditionals D – Lists E – ML solution
13. A – Initial B/C - Conditionals D – Lists E – ML solution Step A – Initial setup https://scratch.mit.edu/projects/346448012/ The goal is to build a virtual assistant in Scratch that is able to recognize four orders in natural language intended to switch on/off a light and a fan
14. A – Initial B/C - Conditionals D – Lists E – ML solution Steps B & C – If/then rules https://scratch.mit.edu/projects/346441641/ Exact/literal comparison between the user input and the text strings that represent each order Unfeasible & unreadable!!!
15. A – Initial B/C - Conditionals D – Lists E – ML solution Step D – Extending rules through lists https://scratch.mit.edu/projects/346449755/ At this point, we propose to classify text expressions (user inputs) in four lists, one for each order Limit of rule-based (top-bottom) approach is reached!!!
16. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples So let’s move to a bottom-up approach!!! Now we will try to capture the rules through sample data analysis and to generate a model that encodes such rules https://machinelearningforkids.co.uk/
17. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/ We collect a set of well classified texts which constitutes the training dataset
18. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/ We collect a set of well classified texts which constitutes the training dataset
19. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/ The ML algorithm builds a model (≈ ‘learns’) by performing some analysis over the training dataset (…) when the model is trained, it can recognize and assign any new text to one of the defined classes with a certain level of confidence
20. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/ We export the ML trained model to a fork of Scratch platform, where some extra blocks are available in order to implement it https://machinelearningforkids.co.uk/scratch3/
21. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/ https://machinelearningforkids.co.uk/scratch3/
22. A – Initial B/C - Conditionals D – Lists E – ML solution Step E – Rules inferred through examples https://machinelearningforkids.co.uk/scratch3/ For the most advanced students… continuous learning!!! Instead of reject low confidence responses, the virtual assistant could ask the user what of the four allowed orders he meant to say. Afterwards, the user response should be added to the corresponding class and the model retrained to be improved as data training set grows with low confidence cases
23. Classroom implementations Three teachers, based on the activity and resources presented in our paper, have designed and implemented their own methodology to introduce ML in their classroom Primary School CEIP Carlos V (Seville, Spain) Secondary School IES Vicente Aleixandre (Seville, Spain)
24. Classroom implementations Three teachers, based on the activity and resources presented in our paper, have designed and implemented their own methodology to introduce ML in their classroom “It is important that at school we explain the reality in which we live. Artificial Intelligence is increasingly becoming a part of our lives. Hence, using tools as ML4K helps our students to be aware of it. After knowing ML4K from the AI video tutorials developed at INTEF, I thought for first time that this kind of content can be taught in school”
25. Conclusions & further research Thank you!!! It seems that AI/ML-related content can be taught at primary/secondary school through tools such as ML4K (…) much more formal- empirical evidence must be collected in this regard.