Lee, Gyeong-Geon

Educators and researchers have analyzed various image data acquired from teaching and learning, such as images of learning materials, classroom dynamics, students’ drawings, etc. However, this approach is labour-intensive and time-consuming, limiting its scalability and efficiency. The recent development in the Visual Question Answering (VQA) technique has streamlined this process by allowing users to posing questions about the images and receive accurate and automatic answers, both in natural language, thereby enhancing efficiency and reducing the time required for analysis. State-of-the-art Vision Language Models (VLMs) such as GPT-4V(ision) have extended the applications of VQA to a wide range of educational purposes. This report employs GPT-4V as an example to demonstrate the potential of VLM in enabling and advancing VQA for education. Specifically, we demonstrated that GPT-4V enables VQA for educational scholars without requiring technical expertise, thereby reducing accessibility barriers for general users. In addition, we contend that GPT-4V spotlights the transformative potential of VQA for educational research, representing a milestone accomplishment for visual data analysis in education.

Constructivist learning theories have emphasized learners’ interactions with the environment, which includes their peers. Therefore, student cooperation/collaboration has been considered crucial in science education. However, although there have been many science cooperative/collaborative learning (CCL) studies reported in Korea, there has been a lack of review studies done to delineate trends and elicit implications for future science education research. This study systematically reviewed empirical science and engineering CCL studies reported in Korea. The researchers collected literature via the Korea Citation Index repository and selected 121 papers to be reviewed. The analytical framework was adapted from the cultural-historical activity theory (CHAT). The results of a review by two science education experts showed patterns revealed in each component of CHAT, which led to a discussion aimed at comprehensively understanding Korean science and engineering CCL research. Implications for future research and teaching were elicited.

Large language models (LLMs) have demonstrated strong potential in performing automatic scoring for constructed response assessments. While constructed responses graded by humans are usually based on given grading rubrics, the methods by which LLMs assign scores remain largely unclear. It is also uncertain how closely AI’s scoring process mirrors that of humans or if it adheres to the same grading criteria. To address this gap, this paper uncovers the grading rubrics that LLMs used to score students’ written responses to science tasks and their alignment with human scores. We also examine whether enhancing the alignments can improve scoring accuracy. Specifically, we prompt LLMs to generate analytic rubrics that they use to assign scores and study the alignment gap with human grading rubrics. Based on a series of experiments with various configurations of LLM settings, we reveal a notable alignment gap between human and LLM graders. While LLMs can adapt quickly to scoring tasks, they often resort to shortcuts, bypassing deeper logical reasoning expected in human grading. We found that incorporating high-quality analytical rubrics designed to reflect human grading logic can mitigate this gap and enhance LLMs’ scoring accuracy. These results underscore the need for a nuanced approach when applying LLMs in science education and highlight the importance of aligning LLM outputs with human expectations to ensure efficient and accurate automatic scoring.

Today's science education faces the imperative task of developing students’ competency to navigate misinformation while broadening the scope of scientific literacy. Traditionally, the concept of epistemic dependence, which encourages public trust in professional scientists, has supported this goal. However, the current landscape of science challenges the notions of experts with unanimous opinions and ‘the public’ as passive recipients of scientific information. In response, this case study examines the LK-99 incident, which involved a claimed discovery in the historic room-temperature and ambient-pressure superconductor, employing the Hype Cycle as the analytical framework. Data were collected on internet search traffic, discourse within the scientific community, mass media articles, and social media posts from July to December 2023, utilizing various online data analytics platforms. The researchers (1) quantitatively identified patterns in search trends, document sentiments, and associated word tokens related to LK-99, (2) qualitatively analyzed the shifting standpoints of stakeholders, the scientific community, mass media, and social media, and (3) synthesized these findings within the Hype Cycle framework. The results illustrate how the misinformation about LK-99 rapidly spread online (phase 1), leading to disagreements among scientists and confusion among the public, alongside erratic behavior in the stock market (phase 2). Ironically, the stakeholders' positioning themselves as scientists facilitated the scientific community's falsification of the claim (phase 3). We discuss the methodological and theoretical implications of this case and propose a reconceptualization of epistemic dependence centered on the scientific community as a whole and its collectively committed process of resolving uncertainty and verifying knowledge claims.