Artificial Intelligence in Science and Chemistry Education: A Systematic Review

Synthesis: Erümit and Özdemir Sarıalioğlu (2025) systematically review empirical research (2014–2024) on AI applications in science and chemistry education, finding rapid growth from 2021–2024 with ChatGPT and conversational robots as the most-used tools. Studies predominantly target science education (n=11) over chemistry (n=5) and STEM (n=2), and their effects concentrate on learning-process outcomes — supporting online learning, facilitating understanding, providing Multimodal AI lab environments, and encouraging personalized learning. Researchers consistently flag ethical risks — gender/racial bias, hallucination, plagiarism, accuracy/reliability concerns, infrastructure and language gaps, and effects on writing and independent thinking. The review calls for increased research at secondary/middle-school levels, more pedagogical (vs. technical) studies in chemistry, and teacher training to foster conscious, ethical AI use.

Purpose and method

The review addresses a gap: most AI-in-education research examines generic effects, but subject-area reviews spanning science and chemistry are limited. It uses a systematic review design (PRISMA), searching Web of Science and Scopus for studies published 2014–2024 on AI applications in science/chemistry education.

• Screening: 255 records identified, 149 screened, 25 assessed for eligibility, 18 studies included.
• Distribution: 3 (2021), 1 (2022), 3 (2023), 11 (2024) — reflecting the post-ChatGPT surge.
• Methods used: experimental/quasi-experimental with control groups, case studies, laboratory studies, mixed methods, and self-study; several integrated instructional design models (ADDIE, rapid prototyping).
• Samples: mostly teacher candidates, middle/high-school students, and teachers.

AI tools and applications

• AI is used most in science education (n=11), followed by chemistry education (n=5) and STEM education (n=2).
• ChatGPT and conversational robots predominate; machine learning algorithms appear in two studies and the DALL-E image tool in one.
• Applications include AI-supported games and simulations, virtual laboratories, AI-based evaluation tools, and AI chatbots embedded as tutors or even as research participants.

Effects on learning outcomes

AI applications most often affect learning-process outcomes (n=9): supporting students' online learning, facilitating learning, aiding successful knowledge construction, providing Multimodal AI (auditory) lab environments, offering interdisciplinary learning experiences, and encouraging personalized and equitable science learning.

Other outcome clusters:

• Prompt engineering (n=6) — students develop prompt-writing skills, use AI as an intelligent personal assistant for writing, gain writing confidence, and get help with complex mechanical explanations.
• Professional skill development (n=5) — teacher candidates build evaluation expertise and teaching skills; AI supports classroom management, lesson-plan design, and professional development.
• Technology (n=3) — increased acceptance of AI among teachers/candidates/students, introduction of machine learning in chemistry, and quality-education opportunities.
• Active participation (n=2) — activates teacher–student and peer interaction and participation in interactive activities.

Risks, limitations, and ethics

Researchers emphasize recurring ethical challenges: gender and racial bias, hallucination, copyright infringement, plagiarism and biased information production, accuracy and reliability problems, technical infrastructure and language-support gaps, and impacts on human decision-making and writing skills. Notably, excessive reliance on AI-produced content can weaken students' independent thinking and decision-making. The review concludes that determining and communicating the ethical issues around AI use and raising student and teacher awareness of conscious use are essential.

Chemistry-education gaps

• Studies in chemistry education focus more on technical applications, revealing a need for pedagogical research addressing AI integration into chemistry teaching processes.
• More research is needed on AI for developing conceptual understanding, Problem Solving, and laboratory skills in chemistry and STEM.
• The teacher-candidate focus signals a strategic orientation toward future practitioners' AI literacy; the limited high-school and middle-school coverage is a significant gap in integrating AI into early education levels.
• Recommended: teacher-training programs, projects encouraging AI use in chemistry/STEM, and efforts to develop AI literacy for conscious tool use.

What this means for practice

• Teacher educators. Have candidates use ChatGPT and conversational robots as science tutors, then require them to verify answers against other sources — the reviewed studies show students copying AI output without checking its accuracy.
• Teacher educators. Build AI-supported games, virtual laboratories, and simulations into methods courses to create the multimodal, auditory lab environments that the reviewed studies link to successful knowledge construction.
• Teacher educators. Add AI literacy to the Technology component of TPACK, covering gender and racial bias in AI-generated images, hallucination, plagiarism, and copyright, so graduates can name and manage these risks.
• Researchers. Run experimental or mixed-method studies at middle- and high-school levels, where only 1 of 18 reviewed studies (5.5%) addressed middle school and 2 (11.1%) high school.
• Researchers. Study pedagogical rather than technical uses in chemistry education, measuring effects on conceptual understanding, problem solving, and laboratory skills.

Limitations

• The review covered only English-language journal articles in the social sciences indexed in Web of Science (n=132) and Scopus (n=123); non-English work, books, and conference papers were excluded.
• Only 18 studies met the inclusion criteria, and no methodology information could be found for 9 of them, limiting conclusions about the quality of the underlying evidence.
• The evidence base is skewed toward teacher candidates (44.4%, n=8); middle-school (5.5%, n=1) and high-school (11.1%, n=2) students are barely represented.
• Several risk/limitation categories — data dependency, effects on human decision-making, and skill development — rest on a single study each (n=1).

Citation

Erümit, A. K., & Özdemir Sarıalioğlu, R. (2025). Artificial intelligence in science and chemistry education: A systematic review. Discover Education, 4, 178.