Artificial intelligence assisted design of a novel cooperative learning technique for higher education

Synthesis: Artificial intelligence assisted design of a novel cooperative learning technique for higher education

Key Findings

• A mixed-methods explanatory sequential study with 67 sophomore prospective teachers at Hakkari University (Turkey) in the spring 2024–2025 semester: an experimental group of 31 Guidance and Psychological Counseling students (19 female, 12 male) versus a control group of 36 Primary School Mathematics Teaching students, in a theoretical "Curriculum Development in Education" course.
• The intervention was designed by prompting five AI chatbots (ChatGPT, Copilot, DeepSeek, Gemini, Qwen) with course documents and cooperative learning principles in a single, non-iterative prompt; the researcher selected Qwen's "Curriculum Concept Constellation Technique (CCCT)" using a four-criterion rubric (novelty, alignment with cooperative learning principles, course context fit, clarity and practicality) — the other four proposals were judged syntheses or adaptations of existing methods.
• CCCT has students collaboratively map curriculum concepts as "stars" into visual "constellations," working in heterogeneous teams of 4–6 with assigned roles (Star Finder, Constellation Designer, Metaphor Maker, Visual Artist, Navigator) across a two-week cycle ending in a gallery walk and group reflection.
• Academic achievement: the experimental group's post-test (M = 51.45, SD = 6.08) significantly outperformed the control (M = 43.89, SD = 10.76), p = 0.001, g = 0.839; only the experimental group improved significantly from pre-test (p = 0.005, g = 0.738) while the control showed a weak effect (g = 0.119).
• Co-AI Regulation in Education: experimental group rose from M = 61.32 to 65.45 (p = 0.043, g = 0.512); attitudes toward cooperativeness rose from M = 68.03 to 75.19 (p = 0.007, g = 0.751).
• Qualitative analysis yielded five themes — fostering student agency and accountability, social cohesion and interpersonal growth, cognitive and pedagogical engagement, collaborative learning dynamics, and structural/logistical challenges (unequal participation, time constraints).

Study Design & Method

The study used an explanatory sequential mixed-methods design in three phases. Phase I developed the technique: two documents (course content; cooperative learning foundations and existing techniques) were given to five AI chatbots asked to propose a new, original cooperative learning technique; proposals were evaluated by the researcher against a rubric (novelty/distinctiveness, alignment with the five essential elements of cooperative learning, fit with the theoretical course and two-session timeframe, clarity and practicality). Phase II implemented the selected CCCT over two weeks, with pre- and post-measurement: a 15-item multiple-choice achievement test (KR-20 = 0.70), a 20-item cooperativeness scale (α = 0.80), and a 19-item co-regulated learning questionnaire (α = 0.89). Pre-tests showed no significant baseline differences; the experimental group was arranged into six heterogeneous teams based on pre-test scores, gender, GPA, and prior cooperative learning experience. Phase III collected semi-structured interviews (four open-ended questions) with experimental-group students.

Key Results

• CCCT mechanics: groups identify 5–7 key concepts ("stars") such as John Dewey's philosophy, Behaviorism, societal needs, and developmental psychology, connect them into "constellations," and invent metaphors for the relationships (e.g., "Behaviorism is the gravitational force pulling all other ideas toward measurable outcomes"); role allocation, constellation creation, a gallery walk, and reflection structure the two weeks.
• Mechanism accounts: participants described CCCT making abstract theoretical content more accessible and memorable (e.g., "When the lesson was conducted with this technique, I actually got more efficiency and it was more fun," P13), attributing gains to interaction, visual elements, shared responsibility, and holistic understanding — consistent with the authors' framing in cognitive elaboration, dual coding, and Constructivism theories.
• Human–AI division of labor: the authors stress the process was "AI-generated" rather than "co-design" — chatbots were prompted once and the researcher retained all selection, adaptation, and implementation decisions, with human oversight judged necessary to contextualize AI output.

What this means for practice

• Teacher educators. Generate candidate instructional designs by prompting several AI chatbots once with course documents and cooperative-learning principles, then select with an explicit rubric — novelty, alignment with the five essential elements of cooperative learning, fit with the course and time frame, clarity and practicality — since that rubric pass rejected four of the five proposals as syntheses or adaptations of existing methods.
• Instructors. Teach abstract theoretical content through metaphorical, visual mapping: groups that identified 5–7 key concepts, linked them into "constellations," and named the relationships outperformed lecture-based peers on the achievement test (M = 51.45, SD = 6.08 vs. M = 43.89, SD = 10.76, p = 0.001, g = 0.839).
• Instructional designers. Make the division of labor explicit with named roles and a public feedback ritual — Star Finder, Constellation Designer, Metaphor Maker, Visual Artist, and Navigator in heterogeneous groups of 4–5, closing with a gallery walk — rather than leaving coordination to emerge on its own.
• Instructors. Track co-AI Regulation in Education and cooperativeness alongside achievement when adopting the technique: co-regulated learning rose from M = 61.32 to 65.45 (p = 0.043, g = 0.512) and attitudes toward cooperativeness from M = 68.03 to 75.19 (p = 0.007, g = 0.751), so the collaborative measures move with the content gains.
• Teacher educators. Keep selection, adaptation, and implementation decisions with the human instructor: the authors describe the process as AI-generated rather than co-design, because the chatbots were prompted once and the researcher retained every subsequent decision.

Limitations

• Small sample from two departments of a single university limits generalizability; unmeasured disciplinary differences between intact classes (counseling students' possible inclination toward metaphorical, reflective thinking vs. mathematics students' logical, sequential orientation) may partly explain effects.
• The control condition was traditional lecture-based instruction rather than an established cooperative learning technique, so the study does not show CCCT is superior to existing cooperative learning methods.
• The two-week intervention duration makes a novelty effect ("more fun," "more enjoyable") impossible to rule out; longitudinal studies are needed.
• No formal systematic review (e.g., PRISMA) or bibliometric analysis was used to establish the pedagogical gap motivating the new technique, and qualitative data relied on self-report, which may be subject to social desirability bias.

Citation

Tutal, Ö. (2026). Artificial intelligence assisted design of a novel cooperative learning technique for higher education.