Author: Georgia Lascaris
Project: Me the A.I #2.0
School: 2nd P.S of Nea Erythraia, Athens
Students: 6th grade
Subject: Artificial Intelligence & Machine Learning
Implementation: 12 – 26 February 2020
This STEM activity is one of the various activities of the European eTwinning project “ME THE A.I 2.0” held in cooperation between Greece, Croatia, Turkey, Lithuania, Spain, Latvia, Cyprus, and Italy. The project started in September 2019. It aims to introduce students with the basics of Artificial Intelligence through various activities and, more specifically, make them:
1. Realize how AI works, the range of possible AI uses, and its limitations.
2. Reflect on the ethical issues raised by AI (stereotypes, biases, personal data).
3. Explore how it can help humans with solving the world’s biggest challenges.
Machine Learning Model Activity
In the past month, the students were in the activities about the safe use of the Internet. Therefore, we decided to use AI for helping us detect harmful pictures that appear on the Internet. For this purpose, we used the free web tool “Machine Learning for Kids”. The idea was to create and train a Machine Learning Model (M.L) to recognize ‘harmful‘ and ‘harmless’ pictures. And then, code a game in Scratch which would use this M.L to categorize pictures.
Firstly, the students had to create a collection of images regarding harmful and harmless Internet content. The students searched for the right pictures on the Internet. The web search was restricted allowing to search only for emoticon pictures and PEGI icons so the students wouldn’t be exposed to violent images. All partners schools co-operated to create a collaborative picture collection.
On the website “Machine Learning 4 Kids”, the teacher set up a classroom-based project to allow students to work together on the same project. Next, the students trained their Machine Learning Model by labelling all the collected pictures as being ‘harmful’ or ‘harmless’. After the training, they were able to test their model with pictures that were different from the training data.
After, students created a game with scratch and played it. It was their favourite activity in the whole project. The game used the trained M.L model to categorize new pictures as ‘harmful’ or ‘harmless’.
Lastly, based on their prior activities on AI, the students tried to explain some errors that their Machine Learning Model made. Indeed, a few pictures failed to be recognized accurately. The model was able to distinguish pictures that had different shapes or colours. But it failed to recognize pictures having similar shape and colour but different small details (yellow happy face, yellow sad face).