Science-AI Convergence: Block Coding for Physics Education

The integration of Artificial Intelligence (AI) into education is rapidly increasing, driven by the need to equip students with skills to solve real-world problems using interdisciplinary approaches. This article explores the development and application of a Science-AI convergence class that utilizes block coding to enhance students' understanding of physics concepts, specifically the motion of a damped pendulum. By using the KNIME platform, students can build AI models to predict the pendulum's position, fostering a deeper understanding of both physics and AI principles. This approach aims to make AI more accessible to high school students, enabling them to engage with complex scientific concepts in an innovative and engaging way.

The 2022 revised science curriculum emphasizes the importance of AI-integrated inquiry activities to cultivate students' ability to solve scientific problems in everyday life and society based on convergent thinking. Integrating AI into science education allows for the application of modern scientific practices within the curriculum. Previous research has explored data-driven convergence classes using programming languages like Python to create neural network models. However, these approaches often require a deep understanding of coding, which can be a barrier for many students. This article addresses this challenge by using KNIME, a block coding platform that simplifies the process of building and analyzing AI models, making it more accessible to students with limited coding experience.

The Science-AI convergence program was developed around the concept of a damped pendulum, a fundamental topic in physics. The program involves several key steps: (1) Selecting the damped pendulum as the inquiry topic; (2) Analyzing pendulum motion exploration activities within textbooks; (3) Constructing a dataset by collecting position and velocity data of the pendulum using Tracker software; (4) Building an AI model using KNIME to predict the pendulum's position; (5) Evaluating the model's prediction results. This structured approach allows students to understand the underlying physics principles while engaging with AI technology.

The Science-AI convergence class was implemented with high school students, and their experiences were analyzed through in-depth interviews. The results highlighted several key themes, including students' motivation to participate, their experiences and changes in understanding, and the challenges and limitations they faced. Students reported increased engagement and a deeper understanding of both physics and AI concepts. However, some students found the initial learning curve of KNIME challenging, requiring additional support and guidance. Overall, the program was successful in fostering a positive learning experience and promoting interdisciplinary thinking.

The findings of this study have significant implications for the development and implementation of Science-AI convergence programs. The use of block coding platforms like KNIME can lower the barrier to entry for students with limited coding experience, making AI more accessible and engaging. The structured approach to data collection, model building, and evaluation provides a clear framework for students to follow. Furthermore, the integration of real-world data and hands-on activities enhances students' understanding of both physics and AI principles. The study also highlights the importance of providing adequate support and guidance to students as they navigate the challenges of learning new technologies.

This research demonstrates the potential of Science-AI convergence programs to enhance students' understanding of physics concepts and promote interdisciplinary thinking. By using block coding platforms like KNIME, students can build AI models to predict the motion of a damped pendulum, fostering a deeper understanding of both physics and AI principles. Based on the findings of this study, several recommendations can be made for future research and practice: (1) Continue to develop and refine Science-AI convergence programs that integrate real-world data and hands-on activities; (2) Provide adequate support and guidance to students as they navigate the challenges of learning new technologies; (3) Explore the use of other block coding platforms and AI tools to enhance the learning experience; (4) Conduct further research to evaluate the long-term impact of Science-AI convergence programs on students' academic achievement and career aspirations.

KNIME (Konstanz Information Miner) is an open-source software widely used for data integration, processing, and analysis, enabling machine learning without extensive coding knowledge. Its graphical user interface (GUI) allows users to connect various nodes to build data analysis and AI models. KNIME offers thousands of nodes and shared workflows, facilitating collaboration and model comparison. Its offline capability and compatibility with languages like Python and R provide flexibility and autonomy in learning. KNIME's visual workflow simplifies the coding process, making it easier to approach machine learning concepts.

The Multi-Layer Perceptron (MLP) is a type of artificial neural network used in this study. It consists of an input layer, an output layer, and multiple hidden layers. The MLP model learns by adjusting weights and biases through a process called backpropagation, minimizing the error between predicted and actual values. The number of neurons in the hidden layers is typically determined using a specific formula to avoid overfitting. The performance of the model is evaluated using metrics such as Root Mean Square Error (RMSE). MLP models can be used for various tasks, including predicting the motion of objects and developing predictive models in clinical medicine.

To collect data for the damped pendulum, a spring pendulum was constructed, and its trajectory was quantified using Tracker software. The pendulum was submerged in a graduated cylinder filled with water to induce damping. The position and velocity data were collected over time, resulting in a dataset of 581 data points. This data was then used to train and test the AI model. The process of collecting and analyzing the data helps students understand the relationship between time, position, and velocity in damped harmonic motion.

The MLP model was built using KNIME's workflow, with nodes representing the input layer, hidden layers, and output layer. The data was pre-processed using normalization to ensure that the position and velocity values were on the same scale. The dataset was divided into training and testing sets. The RProp MLP Learner node was used to train the model, and the MultiLayer Perceptron Predictor node was used to generate predictions. The model's performance was evaluated using R-squared and RMSE. The results showed that the MLP model was able to accurately predict the position of the damped pendulum, with an R-squared value of 0.992 and an RMSE of 0.01.