Key points are not available for this paper at this time.
Abstract Artificial Intelligence (AI) and its subfield, Machine Learning (ML), have become ubiquitous in our lives by playing a critical role in applications that range from automatic identification to predicting future events. Early exposure to AI and ML is essential for K-12 students as it cultivates academic interest and equips them with the foundational understanding necessary to adeptly incorporate these technologies into their future careers. This study aimed to design, implement, and evaluate ML4Math, a program that seamlessly integrates ML into the high school mathematics curriculum. ML4Math promotes active student participation through culturally engaging content and hands-on experience in programming ML models. Through this program, students learned about image classification, from the macro level (e.g., everyday uses of ML) to the micro level (e.g., detailed mathematical concepts underlying ML). Moreover, they built image classifiers by collaborating with generative AI tools designed to assist programming novices. To construct their image classifiers, students employed generative AI to fill in the five sections of incomplete Python code provided to them. This pilot study was conducted with fifteen high school students in Florida over the course of two days. To evaluate the effectiveness of the ML4Math program, students were required to take a knowledge test and complete a survey both before and after participating in the program; however, parts of the survey were conducted only after the program. This test consists of seven questions, both multiple-choice and short-answer questions, assessing their basic knowledge of ML and understanding of mathematical questions related to ML. Additionally, students completed a 5-point Likert scale survey gauging their self-efficacy and engagement toward AI, mathematics, and the ML4Math program. A paired samples t-test and Hedges' g were used to analyze the results of the pre- and post-knowledge tests as well as the surveys on self-efficacy and engagement. Initially, in the pre-test, the average number of correct answers was 3.14 (SD = 1.167), which improved to 4 (SD = 1.881) in the post-test. The paired t-test indicated a significant improvement (t(14) = 3.122, p < .01, g = 1.092). Additionally, the ML4Math program successfully enhanced students' self-efficacy toward integrating ML with mathematics and employing generative AI in their learning. A significant improvement in self-efficacy regarding their use of code generation AIs was observed(t(14)= 2.797, p < .05, g = 1.117). Furthermore, the results of the post-only survey provided that the ML4Math program might introduce ML into math education engagingly, as 64% of participants agreed enjoying learning math and AI together, with none disagreeing. Also, 64% felt motivated to learn more about AI, while only 7% disagreed. The use of generative AI enabled students to program ML models interactively. Students employed generative AI not just to fill incomplete code but for broader educational tasks: asking general questions, copying and pasting the whole/part of the script, and debugging. These results suggest the potential benefits of integrating ML with algebra in high school education, making the learning process engaging and enjoyable. Incorporating ML into mathematics curricula allows students to engage with ML from fundamental concepts to practical applications, which can inspire further interest in learning about AI in the future. Generative AI has made programming more accessible to beginners by removing technical barriers. Yet, it's crucial to recognize that generative AI is prone to inconsistencies, like varying responses to identical prompts, and students must grasp basic programming tenets, including indentation. The limitations due to the small number of participants suggest that further study with larger participants is necessary to validate and generalize these findings.
Oh et al. (Sun,) studied this question.