Randi Williams, Hae Won Park, Cynthia Breazeal
This paper thoroughly described the activities they created as well as the assessments they used to measure the effectiveness of their activities. The way they assessed the children’s understanding of AI was especially thorough. One observation I had was that they used both multiple choice questions and semi-structured interviews. I assume they used interviews in tandem with multiple choice questions in order to gain a deeper understanding of the children’s learning that they can’t get out of just multiple choice questions. They also incorporated storytelling into their questions and made it more visual since their target age group was very young. Another observation is that the multiple choice assessments were not really long. I wonder if this was to account for how they were working with preschool children or if it was to keep the overall experience more fun. I wonder what length we should make our assessments when we work on our own projects since we will be working with middle school age children. In addition, they carefully structured each of the multiple choice questions to test a certain skill. Furthermore, they used a pretest that tested the children’s Theory of Mind skills to see if they had the relevant skills that would be necessary to understand the topics presented by the PopBots. Afterwards, they also tested their perception of robots to see how the PopBots activities affected their overall outlook on robots and AI. They also performed a detailed analysis on the Theory of Mind assessment, the PopBots assessments, and the perception of robots assessment to draw multiple conclusions.
Vaishali Mahipal, Srija Ghosh, Ismaila Temitayo Sanusi, Ruizhe Ma, Joseph E. Gonzales, Fred G. Martin
I liked that they focused on teaching the children about the inner working of CNNs to the extent they did in order to try to mitigate the black-box issue we read about last week. I think it would have been really interesting to see the kids playing with DoodleIt and see them gain a stronger understanding of how CNNs work. Also, I liked how they made it a website and ensured it would be easy to access, because this would make it more likely for a larger amount of people to be able to easily use this tool to learn about CNNs. When we work on our own projects, I will definitely need to consider how to make it widely accessible and easy to use. When I was talking to some graduate students doing cybersecurity research, they mentioned that research papers don’t usually mention specific programming languages or technologies used, so I found it surprising that this paper mentioned their use of Keras, Python, and some details about how they built their website. However, I did like that they talked about these things as it gives us a clearer idea of their project and we might find ourselves using similar technologies with our projects. I was initially wondering if they talked about the applications of CNNs that the children have probably interacted with, so I was glad to see how they made sure to include lessons about AI ethics, where they also discussed current applications of image recognition. It’s interesting that all of the papers we’ve read so far highlight the importance of teaching AI ethics, since AI is going to become such a huge thing and we want the children to be thinking about the ethical implications of AI. When we work on our own projects, we definitely need to consider how we can incorporate things that can also teach the middle school children about AI ethics.
Safinah Ali, Daniella DiPaola, Irene Lee, Jenna Hong, Cynthia Breazeal
This study aimed to define a learning trajectory that can be used to teach middle school children about GANs and the ethics of AI generated content. They focused on GANs since children are increasingly interacting with AI generated content, they want children to be better equipped to work with generative AI in the future, and they want children to be able to identify AI generated content in their daily lives. In this study, they created various interactive activities to teach children about GANs. The 1st activity teaches the students about generation, the different types of media that AI can generate, and a game where they have to guess whether something was generated by AI or not. In the 2nd activity, students learn that a GAN is made up of a generator and a discriminator, what each of them do, and then they play a game where some students play the role of generators and the others play the role of discriminators. In activity 3, the students learn about the applications of generative AI by interacting with various GAN tools. In the 4th activity, the students learned more about Deepfakes and ways to identify Deepfakes. In the 5th activity, the students created stories using various generative models. The researchers measured student learning through a pretest, data collected during the various activities, and a posttest. After the activities, they found that the students gained a good grasp of what the generator and discriminator do but did not grasp how models get trained through datasets. The students were able to learn a lot about the various applications of GANs after interacting with various tools during the activities, but they weren’t able to identify as many negative uses of generative AI as opposed to positive uses. After analyzing the results, the researchers think that including a lesson on how neural networks work will allow the students to better understand how the generator and discriminator work.
Each of the 3 potential project ideas has been situated in a Big Idea from Touretzky et al's Envisioning AI for K-12: What Should Every Child Know about AI?
A potential project could aim to teach children about Random Forest (classifier model), a powerful machine learning algorithm that uses multiple decision trees. We would have to ensure that we teach the students about how decision trees work, how the model constructs the decision trees based off of training data, and how the model uses the various decision trees to make predictions on the testing data. We would have to ensure that we teach the children about decision trees and how they are constructed at a high level that is understandable for them. We could go about teaching them about Random Forest by having them play with a simple and small dataset that they can construct decision trees around. We could have each student make a decision tree and then use them all together to show them how Random Forest uses majority voting to come to a prediction. If it is too complicated to create a tool about the Random Forest Algorithm, then we could teach them about Decision Trees instead.
This potential project would come under Big Idea 3 (Computers can learn from data) since it will cover how Random Forest creates decision trees that it uses based off of its training data.
Another potential project idea is to teach the middle school children about how AI is used to make bots to play against humans in certain games. We could focus on board games to teach them about how AI maintains a representation of the board. We could use games like checkers or connect 4. The kids could get to play the game against the bot while also seeing a visual representation of how the AI is “seeing” the board. This aspect of the project would come under Big Idea 2 (Agents maintain models/representations of the world and use them for reasoning).
Since the AI’s representation of something like Connect 4 will be quite simple (2x2 array), we could also incorporate in ways to teach the students about how the AI decides what move to make. We could have an avatar that represents the AI bot that tells the student a little about why it made the move it did, so that the students can learn more about how the AI decides what move to make.
A third potential project idea is to teach the middle school children about AI bias. We could go about doing this by showing the students a simple, sample dataset that an AI model was trained on, and then we could show an example of something that will cause the AI to make an incorrect prediction on because of limitations of the dataset. We could then discuss how this is an issue with various applications of AI and how it is affecting various things. This would make the children realize the limitations of AI and how we need to be careful with how it is used since there are limitations in its abilities. However, we would also need to highlight how the AI is not always incorrect and can also be very useful to prevent the children from developing a fear of AI.
This potential project would fall under Big Idea 5 (AI applications can impact society in both positive and negative ways) since it focusses on a limitation of AI that can impact society negatively.