Empowering Displaced Children Through AI Education
By: Erica Hooshi
Overview
Taking place in Athens, Greece, ReadyAI held a camp at a refugee shelter to give these students an opportunity to learn about Artificial Intelligence (AI) and data science through our activity-based learning program. The camp took place for two hours every day for five days. The students ranged from the ages of 13 to 18 years old and came from different native countries and shelters.
Goals
The camp’s main goal was to allow the students to get a better understanding of what AI is and how it is being used through interactive activities, games, demonstrations, and lesson slides. I wanted the students to feel at home with us and each other in a comfortable learning environment. I hoped to not only bring awareness to the topic of AI for these students but also help them find their sense of belonging and confidence (in and out of the AI field).
Background
This was my first time seeing a refugee shelter as well as running a camp. I didn’t know what to expect, but knew I had to be flexible and adapt to whatever was to come. I previously worked with the ReadyAI team on the paper, “Exploring the Gendered Impact of Project-Based AI Learning on Students’ Self-Efficacy, Career Interest, and Content Knowledge” and presented this paper at the CITERS 2023 Symposium in Hong Kong. Yes, I had this background, but I had to think about how to apply it in a different context now. Rather than being on the research and reflection side of this project, I was now going to also be a part of the actual data-gathering experience. I was ecstatic to be able to have the opportunity to run one of these programs and observe what was going on behind the scenes of the data I was previously just reading.
The Camp
Overall Structure:
I had the opportunity to work with Amalia Toutzi to help run the camp. We decided it would be best to use Google Slides to help provide the structure to the lesson. The day started with going through the topic of conversation and learning objectives of the day to get the students curious. We proceeded with a KWL chart (K: what you already Know, W: what you Want to know, L: what you Learned) and filled out the K and W sections of the chart. The K was often filled with things learned the previous day, while the W was filled with questions that lingered from the previous lesson or questions they had about the topic of the new day. After the lesson, students engaged in online and offline demonstrations and activities for that day. Following each activity, there was a “What was going on?” and a “Big Idea” slide to allow the students to see the bigger picture. Throughout the camp, all five of the Big Ideas were incorporated into the lessons. At the very end of the lesson, we revisited the KWL chart to fill out the L section of the chart with what was learned that day. We usually ended with a reflection on the entire lesson as well as the main takeaways to help solidify what was learned.
Day 1: What AI does well and does not do well
Summary:
Since this was the first day, the main goal was to get the students comfortable with each other and us as well. We handed out the pre-survey and asked a few questions about some of the individuals to get a sense of their AI backgrounds and views. We then introduced ourselves and did icebreaker activities. Following this, we introduced them to the topic of AI and demonstrated examples of how AI can do certain tasks well and others not so well.
My Experience:
On the first day, the main objective was to get a feel of the classroom environment and set the tone for the camp. I wanted the students to feel excited and want to return with interest the next day. Not all of the students came from the same shelter, so it seemed like they weren’t open with each other yet. To get them to start sharing about themselves, I volunteered to go first. I explained to everyone that I was a fencer, and even demonstrated some actions with my arms, legs and imaginary sword. Many of the students thought it was cool and wanted to share about their sport or activity next to try and get a similar reaction from their peers.
At the start of the lesson, we easily had almost everyone’s attention, but it was challenging to keep everyone engaged and focused. To get an idea across, we had to repeat it multiple times throughout the lesson. The class was diverse, in terms of interests, education levels, ethnicities, ages, characters, and levels of respect. While some were eager, shouting out answers, others were quiet, with airpods in, or made negative remarks to get attention. Gathering everyone’s attention to me was something I had to learn to navigate and figure out for the next couple of days.
The way we demonstrated a task AI does well was through a Tic-Tac-Toe game Aaron Wong had programmed by using five rules for AI to follow. We allowed the students to volunteer to take turns trying to beat the AI program. They got a real kick out of this as they started huddling around the computer screen as they cheered for each other trying to win! Some of the students were so determined to win and not give up to prevent a detrimental rematch!
Day 2: Training data and machine learning
Summary:
On day 2 we asked the students to gather and organize their own dataset through a hands-on flashcard activity. This allowed them to be able to see the power of AI, as it is capable to organize datasets far greater than the one demonstrated in class. We also introduced bias and real-world examples, to get them to start seeing the bigger picture. Towards the end of the lesson, we gave a teaser for the next day’s lesson and had them play around with face filters and how it relates to AI.
My Experience:
On day 2, the students were more attentive as there wasn’t a survey that needed to be done first. I was surprised and elated to see that the students were volunteering a lot of bullet points for the K section of the KWL chart. We wouldn’t pressure them to answer; Instead, if they declined the first time we asked, they would happily answer the next time and agree to actively engage.
Through Google’s AI Experiment Quick Draw, we demonstrated how AI was trained with a lot of data to complete a task like recognizing an object that is being drawn by someone. Students took turns drawing and having Google guess what was being drawn. It was difficult for the students to understand that AI was able to guess the drawing by learning from previous datasets.
I had to improvise here since we hadn’t had any other demonstrations planned for this learning objective. I explained how Quick Draw would see what is being drawn first and its shape, and make a guess of what it is based on what it has previously seen in its dataset.
I walked up to the whiteboard and drew a three-way T-chart and asked for three volunteers to draw a flower. I then questioned the class about how they knew it was a flower that was being drawn. I questioned the similarities among the three flowers in terms of how they looked (a stem, center, and petals) and what was drawn first (the circle representing the center of the flower). I pointed out how AI would look out for those same similarities when identifying on its own. This improvised demonstration worked very well as I even heard some students say, “Ohhh!”
For the next activity, I shared my experience on face filters. The students were able to connect the dots from the previous demonstrations and conclude on their own that AI has been programmed to know where parts of the face are located on a human. This was really interactive as the students had a fun time trying on the different face filters and passing the device around for others to try. To get the students to understand some ethics and get them thinking, we asked about the possible positive and negative effects on society.
Day 3: Senses vs. Sensors
Summary:
On day 3 we introduced how sensors help AI perceive the world as well as how speech recognition and other sensors work. We first discussed the human senses and how they are important for our everyday lives in order to interact and adapt to a changing environment. We then made the connection that in order for AI to imitate a human, it also has to be able to sense what is around and change accordingly if needed.
My Experience:
Almost everyone was able to hurl out what the five human senses were when asked. They were so proud to shout it out in confidence and pride! We presented real-life scenarios where our senses are needed, then asked the students to their own examples. This worked well, as we gave them a chance to share their own stories or interests through this activity. Some gave scenarios of playing football or baking cookies!
We then discussed some real-world examples where we see sensors being used. Amalia and I purposely used examples that directly or personally related to the student’s interests, such as a robot that can solve calculus problems or a robot that can play football. This got everyone involved and excited.
We focused on discussing a few sensors and demonstrating how they were used. First, we started with the microphone. To create the bigger picture, we explained how in order for the AI to “hear”, you need to have proper enunciation, grammar, and context, otherwise, the AI will guess incorrectly. An example of one we did as a class is shown in the image below.
Next on our agenda, we performed two demonstrations in order for the students to understand that in order to interact with the environment, detailed instructions and data need to be listed. For the first demonstration, we brought out peanut butter, jelly and bread. For the second demonstration, we displayed a Jenga tower. The students were asked to list out rules of how to do either of these tasks, then another student or ourselves had to perform the task exactly as the rules we had just written. The students would see that the rules they initially had devised weren’t detailed enough for an AI to follow. For example, for PB+J rule number nine which said to “Close it” in reference to the two pieces of bread, I exaggerated and closed the sandwich in a variety of wrong ways on purpose. The students got a kick out of this and knew where they were wrong here. They needed to revise the rule to specify to put the peanut butter side of the bread to touch the side of the other bread with jelly on it.
Day 4: Generative AI and Art
Summary:
On day 4, machine learning and neural networks were introduced. We applied how an AI program distinguishes characteristics of various artists and artwork as well as AI having the capability to mimic and create its own art. Similar to previous discussions, we raised concerns about the ethics behind AI-generated art in terms of who is the artist/ who gets the credit. At the end of the lesson, each student created their own artwork and we used a digital tool powered by an AI neural network to see if it could detect the student’s art style.
My Experience:
We started off by looking at different art pieces and comparing the way various artists create their artwork in terms of colors, shapes, brush strokes, etc… This helped the students make the connection that AI learns from these specific attributes of artists and then is able to detect which artwork is theirs by going back to this data.
The next demonstration used Google’s Teachable Machine. We asked the students to pick their favorite foods. We first demonstrated how we could define classes for multiple foods or styles of food, upload several image samples for each class, train the machine learning model, and then test the model with new food images. This would test to see how accurately the model is able to recognize and classify the food. The model was able to get it right every time! (see the picture below)
We gave the students a shot at creativity and see a real-world example! This time we tested to see how accurately the model could recognize and classify the artwork by artist or style. We all tried to draw an image either similar to Picasso, Da Vinci, or a mix of both and see if the AI model would be able to pick up on our intended artist. The students had a fun time with this as some even made remarks about how this coloring/drawing activity was therapeutic.
The next few activities we proceeded with demonstrated how AI could be used to create. analyze, and enhance music and artwork. We looked at different AI art filters and saw how they were able to create art pieces based on what it was sensing. The students played around with this activity and even turned themselves into AI-generated anime art or cartoon characters. We then made the connection of AI to music. We played two different songs — one generated by AI to sound like the Beatles, while the other was an actual Beatles song. The students were all able to hear the difference but admitted how similar they were! This truly amazed them.
Screenshot of Google Teachable Machine example with favorite foods
Day 5: Hands-On final project
Summary:
The final day was more of a hands-on experience where the students got a taste of coding and creating their own projects! They were finally able to see the behind-the-scenes of how AI actually runs these tasks through a computer program. Their project would be made on Scratch (coding program) and presented on their Google Slides– where they discuss how all of the five Big Ideas were implemented. We also presented the final conclusions of the camp including post-surveys and reflections.
My Experience:
The class size was much shorter than in previous days, so we gave the students the option to either work in two groups or together. The unanimous decision was to work together! We opened Scratch and surprisingly many of them were familiar with it or had used it before. We decided to create a ChatBot where the program would be able to answer questions based on the story it was given. We played around with the code and made the character dance, speak, move, hear, and other endless abilities. Each of the students was given the chance to give their own mini story to set as the Text Variable in the code.
Reflection
Adjustments and Acknowledgements
Throughout the camp, many adjustments had to be made. The size of the class kept on changing since students were coming from different shelters and needed available volunteer teacher escorts. There was also a heat wave which impacted the student’s ability to attend. Although there were fewer students, this facilitated teaching and ensured that each individual received something valuable out of the lesson.
Some of the students complained about starting at 10 am as they were still sleepy and couldn’t focus. Many didn’t have breakfast and were fasting. Since it was a feat to maintain the attention of the students, we knew we had to change the structure of the lesson to be even more engaging than initially planned. We cut the lessons for the following days shorter and only included activities that were more hands-on and demonstration-based. We had the students volunteer and come up to the whiteboard often to stay involved.
As each day progressed, there was clearly a change in the way the students opened up, respected, and connected with us and each other. This stimulated the classroom environment. The trust and respect of the students were evidently gained through the way they greeted us! Unlike prior days, students started starting the day by saying, “Good Morning”, “Hello! How are you?”, and bid us adieu with a “Have a good day!”, “Thank you”, as well as rounds of applause at the end of the lesson. This was the highlight of the camp for me. It really demonstrated the bond we were able to create with the students and it was proof of how much they appreciated the impact of our presence.
Expectations
As a researcher, I expected to gain a different aspect and perspective on the data that was being collected.
I came into this camp with the expectation of learning how to interact with displaced refugee children as well as teaching and conducting a classroom. I expected to gain aspects of becoming a better leader as well as a peer to those around me.
Takeaways
I’ve learned how the path to effective leadership can be life-changing to not only the students but to me as well. I’ve become a better listener and colleague to those I tried leading. Through this camp leadership role, I’ve learned the best way to get everyone’s attention focused on you is to make it seem like you want to learn together with the students, rather than you teaching them as a superior. From the beginning to the end of the camp, I adjusted the way I taught the class from standing in front of the classroom, to actually sitting with the students and interacting with them closer. I could see how much of a difference this made in the way the students felt comfortable opening up more. Seeing the smiles on their faces was my goal rather than getting a learning objective across — whether through an inside joke or a fun demonstration. The smiles and laughs were enough to tell me they also understood the bigger picture.
I had just met Amalia for the first time in Greece on the first day of the camp. Being able to conduct a class or lesson in sync with someone you don’t know can be very daunting and difficult. However, we adapted well to each other and collaborated. We constantly bounced ideas with each other! The students witnessed our rapport and really appreciated it. We have become great friends and will continue our friendship on the east coast of the US as she starts MIT in the fall while I embark on my halfway point at Yale.
I have gained so much from this experience and cannot wait to take this along with me in my upcoming adventures and forever in my life.
This article was written by Erica Hooshi. Erica is AI Education Fellow at ReadyAI.org
Erica Hooshi is an upcoming undergraduate junior majoring in Chemistry at Yale University (BS Intensive program). At Yale, she is in a Computational Organic Chemistry lab as well as a Neuroscience Lab. She plans on attending Medical School and eventually tying in medicine with AI and technology.
To learn more about ReadyAI, visit www.readyai.org or email us at info@readyai.org.
