Week 1
Developing Middle School Students’ AI Literacy
(Lee, I., Ali, S., Zhang, H., DiPaola, D., & Breazeal, C.)
The novel approach to making a summer workshop presented by the DAILy program was particularly inspiring. The use of hands-on games to teach children about Artificial Intelligence was something that I am also considering for ideas in building my tool to teach children about AI. I had an idea to do something like the Neural Network game where students took part in training a neural network, but I would take a different approach to it and make a more in-depth tool. The decrease in engagement when students had to perform tasks that resembled their classwork like activities submitted with google classroom stood out, further reinforcing that the tool used will need to be stimulating and possibly gamified in order to keep students engaged in wanting to learn more about AI technology. Something that surprised me about the study was that participants were very interested in the possible ethical repercussions of bias when programming AI. Exposing students to Ethical issues early on is important to foster a generation that is not only interested in the technical side of AI, but also considers the moral implications of developing advanced AI technology. The idea of career training activities which “immersed students in exploring their future jobs” was very interesting because it is important to develop tools which can inspire children to want to study AI in order use it effectively. The DAILy program is something that I will keep in mind when working on my own tool to teach children about AI technology.
What is AI literacy? Competencies and design considerations
Long, D., & Magerko, B.
The paper on AI literacy serves as a guide for the most important components in teaching non-technical audiences about Artificial Intelligence. I will ensure to return to this paper often when considering ideas about building my own tool for teaching children about artificial intelligence. One of the areas that stood out to me the most was the section titled “How Do People Perceive AI?”, specifically the part titled “Children’s Perceptions of AI as it is most relevant to my own project. Accessibility for children without the prerequisite skills to be able to program AI was an issue that was highlighted. The paper stated that children were most engaged with conversational, personified AI systems, especially those with faces, however, there needs to be a balance of education containing a variety of types of AI. In correspondence with the other article the topic of ethical issues surrounding AI was discussed. In addition to the possibility of bias in AI model training discussed in the other paper, issues such as privacy, employment, misinformation, transparency, and accountability were key issues that were brought up. Employment was a new topic brought up in this paper that is very relevant with the current rate of advancement in AI technology. While progression in AI technology will lead to increased productivity, a key concern is that advances in automation will lower the need for human workers, putting many people. By teaching children about AI sooner, they will be at an advantage in the increasingly AI driven modern world.
ChatGPT Prompt: Suggest a tool I could develop for teaching children about AI make sure it inclues an interactive neural network. make the AI human like and social. possibly a pet
Creating a tool to teach children about AI with an interactive neural network is a fantastic idea! Here's a suggestion for a tool:
Tool Name: AI Pet Pal
Description:
AI Pet Pal is an educational tool designed to introduce children to the fascinating world of artificial intelligence in a fun and interactive way. The tool features a virtual pet powered by a simplified neural network, allowing children to learn about AI concepts in a playful environment.
Key Features:
Virtual Pet Companion:
Children can adopt their AI-powered virtual pet, which has human-like qualities and a social personality.
The pet responds to user interactions, such as talking, petting, or feeding, demonstrating basic AI functionalities.
Neural Network Interaction:
A simplified neural network is integrated into the pet's behavior, showcasing the fundamental concepts of AI decision-making.
Children can customize and train their pet by providing positive reinforcement for desired behaviors.
Teachable Moments:
The tool includes interactive lessons and challenges that guide children through the basics of AI, explaining concepts like training data, algorithms, and decision-making processes.
Storyline Adventures:
AI Pet Pal engages children in educational adventures where the pet faces challenges that require problem-solving and decision-making, reinforcing AI concepts.
AI Playground:
A virtual playground allows children to experiment with different AI parameters and observe how they influence their pet's behavior, fostering hands-on learning.
Progress Tracking:
The tool includes a progress tracker to monitor the child's understanding of AI concepts, encouraging continuous learning.
Parental Dashboard:
Parents can access a dashboard to review their child's progress, understand the AI concepts covered, and engage in discussions about the educational journey.
By combining the appeal of a virtual pet with interactive AI elements, AI Pet Pal aims to make learning about artificial intelligence enjoyable and accessible for children.
Week 2
Ruizhe Ma, Ismaila Temitayo Sanusi, Vaishali Mahipal, Joseph E. Gonzales, and Fred G. Martin. 2023. Developing Machine Learning Algorithm Literacy with Novel Plugged and Unplugged Approaches.
The paper details a study on novel ways to introduce middle school students to concepts in machine learning and data science. The highlight of the study is that it uses a plugged and unplugged approach in conjunction in order to teach children about AI technology. The plugged approach involved using visualization software for teaching machine learning subject. The unplugged approach students created decision trees to classify different types of pasta. One of the main topics covered in the study was decision trees which are used to make complex decisions by breaking them down into a sequence of simpler decisions. An advantage of teaching about decision trees are that they are easy illustrate and understand for new learners. Additionally, the k-nearest neighbors algorithm was introduced in the paper, which is where the model makes predictions based on the proximity of new data points to existing labeled data points. The addition of tangible activities for the students to participate in is a great idea for increasing engagement for younger audiences. The students were evaluated using surveys before and after their involvement in the program, along with active monitoring throughout it. The data showed meaningful improvement in students understanding of machine learning concepts after taking part in the program. The paper stated that the authors had two of the five AI4k12 big ideas, specifically Big Idea 2: Agents maintain models/representations of the world and use them for reasoning, and Big Idea 3: Computers can learn from data. The study stated that middle school students were chosen because they were developing greater critical thinking skills and students experience at those ages had a high chance to affect their “future educational and career paths”.
Idea 1: Interactive neural network application. A while back I saw a project on YouTube called Bibites: simulating digital life and it was a competition where participants made creatures that had a neural network which was either hand crafted or trained by evolving over many cycles. That inspired me to make a project where students could interact with a neural network to train ai software. Perhaps there a simplified implementation of a neural network, or visual elements could be present. This could be applied in many different ways, like the example given above. Depending on the implementation this could express multiple of the big ideas in AI especially, Big Idea #2: Agents maintain models/representations of the world and use them for reasoning and Big Idea #3: Computers can learn from data.
Idea 2: Personified AI teacher. A common theme I recognized was that students were more engaged when working with personified artificial intelligence systems, especially those that were social and had faces. I would want to work on software that leverages natural language processing techniques in order to achieve this goal. One of the most important features I would want to achieve would be allowing students to view what is happening under the hood, in an easy-to-understand way. This would also convey Big Idea #2: Agents maintain models/representations of the world and use them for reasoning and Big Idea #3: Computers can learn from data.
Idea 3: Fix a biased AI system. Another common Issue discussed in the papers we read are the ethical considerations and harmful scenarios that must be taken into account when educating children about artificial intelligence. The software would include a variety of scenarios where the AI is flawed or biased and students must mess with the logic in order for the artificial intelligence to function properly. This way students would be educated about what biases can arise when developing an AI system. This would mainly express Big Idea #5 - AI applications can impact society in both positive and negative ways.
Williams, R., Park, H. W., Oh, L., & Breazeal, C. (2019). PopBots: Designing an Artificial Intelligence Curriculum for Early Childhood Education.
The study including PopBots serves as an excellent foundation on how to properly handle assessing student learning outcomes. The paper sought to understand “how did developmental factors, like perspective taking skills, impact what children could learn about AI? and how did children’s perceptions of “thinking machines" change after they engaged in educational AI activities?”. The study obtained the answers to the outlined questions by recording students progress while completing the activities as they went through them and using probing questions while the students were engaged in the activities. Additionally, students completed various assessments including one on theory of mind, and questions about their perception of robots,
Vaishali Mahipal, Srija Ghosh, Ismaila Temitayo Sanusi, Ruizhe Ma, Joseph E. Gonzales, and Fred G. Martin. 2023. DoodleIt: A Novel Tool and Approach for Teaching How CNNs Perform Image Recognition.
What I found most interesting about the article on DoodleIt was the visual representaion of the layers involved in the process of CNNs, such as the “display of kernels, the resulting feature maps, and the percentage of the match at output neurons”. Seeing DoodleIt in action was very eye opening and gave me an idea on they type of project I would want to develop to teach children about artificial intelligence. I would want to take part in a project that could uncover the black box and demystify seemingly difficult to understand concepts in AI technology in order for middle school age students to be able to understand the underlying concepts in AI.
Week 3
Along with facial recognition software having trouble identifying individuals base on gender and skin tone, there are other areas where bias in AI technology can occur such as predictive policing, loan approval systems, and recruitment algorithms. In the past use of AI in predictive policing algorithms have led to discrimination, as reported by the MIT Technology Review. Two major problems that occur with predictive policing algorithms are biased data and feedback loops. Predictive policing algorithms utilize historical crime data to predict future criminal activity, however issues may arise due to past over-policing in certain areas or racial profiling. Additionally, using this data may cause a feedback loop where increased police presence in certain areas could lead to more arrests, therefore generating more data from those regions and reinforcing the original dataset. Besides the issues mentioned above, loan approval systems may use proxy variables that correlate with user’s race, ethnic background or socioeconomic status in order to determine whether or not to lend to them. Examples of these proxy variables could include zip codes, educational background, or employment history which could involuntarily lead to discrimination. A key issue with recruitment algorithms are they can exclude individuals base on their university rather than evaluating their qualifications.
Synthesis: Ruizhe Ma, Ismaila Temitayo Sanusi, Vaishali Mahipal, Joseph E. Gonzales, and Fred G. Martin. 2023. Developing Machine Learning Algorithm Literacy with Novel Plugged and Unplugged Approaches, Karalekas G, Vologiannidis S, Kalomiros J. Teaching Machine Learning in K–12 Using Robotics. Education Sciences, and Ng, D.T.K., Su, J. & Chu, S.K.W. Fostering Secondary School Students’ AI Literacy through Making AI-Driven Recycling Bins.
The papers that I decided to review were Teaching Machine Learning in K–12 Using Robotics by Georgios Karalekas, Stavros Vologiannidis, and John Kalomiros along with Fostering Secondary School Students’ AI Literacy through Making AI‑Driven Recycling Bins by Davy Tsz Kit Ng, Jiahong Su, and Samuel Kai Wah Chu. The paper that I originally chose was Developing Machine Learning Algorithm Literacy with Novel Plugged and Unplugged Approaches by Ruizhe Ma, Ismaila Temitayo Sanusi, Vaishali Mahipal, Joseph E. Gonzales, and Fred G. Martin. A major similarity I noticed between the papers was that all of them employed a hands-on approach. The original paper I read differed by having one of the methods be totally technology free. The AI driven recycling bin paper focused on a hands-on approach with maker education where participants worked on their own projects. In the last paper they designed robotic kits with machine learning capabilities to teach concepts of machine learning in k-12 classrooms. I was really impressed by the project-based method of teaching kids about AI in the recycling bin study. In my personal learning I found that the project based approach was the best to learn a concept deeply and be able to implement it myself. However, one of the downsides of that study is the time a project based approach would take to complete. The last paper gave examples of differences when teaching machine learning using traditional tools and educational robots.
Week 4
Shuchi Grover. Teaching AI to K-12 Learners: Lessons, Issues, and Guidance
The thing I found most interesting about this paper was the challenges that they faced in educating children about AI technology.
Saniya Vahedian Movahed, Fred Martin. Exploring children’s attitudes toward an age-tailored AI-powered chatbot
The study assessed student’s learning outcomes in a multitude of ways including analyzing interaction metrics and survey data, analysis of audio and screen recordings of student dialog, behavioral observations, and an responses to an open ended post interaction survey.
`
Week 8