Rethinking homework in the AI era: A shift to in-class collaborative problem solving

Abstract

Collaboration and teamwork are essential skills for success in STEM disciplines, yet many classroom practices still rely heavily on individual work that has to be completed outside of class. Recent advances in generative AI tools challenge this traditional approach by enabling students to generate solutions with minimal conceptual engagement, thereby undermining homework's efficacy while highlighting the collaboration skills it often neglects. This study examines the impact of transitioning from traditional individual homework to collaborative, in-class problem-solving as a more authentic approach to improving student learning and performance. Over two consecutive years, the same undergraduate Thermodynamics course was taught using two different pedagogical models. In the first year, students completed homework individually outside of class. In the second year, individual homework assignments were replaced with structured, in-class group work. In this model, students worked in small groups to solve problems and subsequently presented their solutions to the class. The instructor facilitated discussions by guiding problem-solving strategies, checking conceptual understanding, and encouraging student engagement. While both models covered identical content with comparable problem difficulty, students in the collaborative format demonstrated stronger problem-solving skills and achieved approximately 5–10% higher exam scores, along with an overall course grade increase of about 8%. Student feedback consistently reported improved conceptual understanding through peer discussion and greater confidence in applying thermodynamics principles independently. These findings suggest that replacing homework with guided, in-class collaboration creates a more engaging learning environment and better prepares students for the collaborative and problem-solving demands of modern engineering practice, particularly in the AI era.

Keywords

Active learning collaborative problem-solving traditional homework student performance engineering education AI tools

Introduction

Homework has long been considered an indispensable tool of the science, technology, engineering, and mathematics (STEM) education that helps students practice newly learned concepts and master the course material.¹ Students’ mastery of this material is ultimately evaluated through examinations, which are intended to assess independent problem-solving ability.² However, research on homework effectiveness shows mixed findings. While some studies report that well-designed homework can improve learning outcomes,^3,4 others suggest that traditional homework often fails to promote deep conceptual understanding.^5–7

In recent years, the effectiveness of traditional homework has been further compromised by the widespread availability of online solution platforms and, more specifically, generative Artificial Intelligence (AI) tools. These resources allow students to generate step-by-step solutions rapidly, often without engaging in underlying reasoning.^8,9 As a result, many students submit correct solutions without developing conceptual understanding. This behavior creates a growing disconnect between homework performance and exam outcomes, where homework grades dramatically increase while exam scores remain significantly lower, even when homework and exam problems are closely aligned in structure and difficulty.

This disconnect suggests that traditional homework alone may no longer support authentic learning. In response, many educators have turned to active learning strategies that require students to engage directly with content during class, rather than passively listening to the teacher to receive information. Prior research indicates that active learning improves conceptual understanding of the material, critical thinking, motivation, and long-term knowledge retention compared to lecture-based instruction.^10–12

Among active learning strategies, collaborative learning has emerged as particularly promising in STEM education. In this method, students work together in small groups to solve problems through structured interaction and shared responsibility. The goal is to support both their own learning and that of their peers. Unlike traditional group work, collaborative learning is designed to promote individual accountability, meaningful interaction, and the development of specific skills. Research shows that this approach improves conceptual understanding, self-confidence, engagement, and communication skills.^13–15 Collaborative learning is especially effective in engineering education because it mirrors professional practice, where engineers solve problems collaboratively. These skills are central to applications such as energy system optimization^16,17 and modeling of complex physical and biological systems.^18,19

Peer interaction plays a critical role in the effectiveness of collaborative learning.²⁰ When students explain ideas to peers, they must clearly articulate their reasoning and identify gaps in understanding. This process strengthens critical thinking and supports transfer to new problem contexts. As a result, students demonstrate improved performance on conceptual and application-based exam problems that require applying knowledge to new or unfamiliar scenarios.²¹

Despite its benefits, collaborative learning also presents challenges. Research has identified issues related to group dynamics, including uneven participation, overreliance on stronger students, and social loafing, where some group members contribute less effort than they would individually.^22,23 Research also shows that collaborative learning is effective only when it is intentionally structured and carefully facilitated.²⁴ These findings highlight the critical role of the instructor in guiding discussions, asking probing questions, providing timely feedback, and monitoring group dynamics – elements that are central to the instructional approach examined in the current study.

To clarify the pedagogical framework used in the current study, it is useful to distinguish between problem-based learning (PBL) and project-based learning (PjBL). PBL typically focuses on solving specific, often ill-defined problems within a constrained time frame. In contrast, PjBL involves extended projects that culminate in a final product or presentation.²⁵ The instructional model examined here aligns more closely with PBL, as students engage in structured in-class problem-solving completed within a single class session.

This paper compares two instructional models implemented in an undergraduate Thermodynamics I course over two consecutive summer semesters. In the first year, students completed individual homework assignments outside of class. In the second year, traditional homework was replaced with structured, instructor-led, in-class collaborative problem solving. Students worked in small groups while the instructor guided discussion, monitored reasoning, and provided immediate feedback. Groups then presented their solutions to the class. Student performance and feedback were compared across the two cohorts to evaluate the impact of this instructional change.

Methodology

This study employed a comparative design conducted over two consecutive summer semesters in an undergraduate Thermodynamics I course. The course covered the first seven chapters of the standard Thermodynamics textbook,²⁶ including basic concepts, the first and second laws of thermodynamics, properties of pure substances, energy analysis of closed systems, mass and energy analysis of control volumes, and entropy. Each class session was 90 min long and recorded for student review. Each cohort consisted of 15 junior-level mechanical and aerospace engineering students.

The small enrollment was intentional. It allowed close instructor facilitation, direct observation of student reasoning, and careful refinement of the collaborative model during its initial implementation. The goal of this phase was not large-scale generalization, but focused evaluation of student engagement, learning behaviors, and instructional feasibility under controlled conditions.

To accommodate in-class collaborative problem-solving in Year 2, instructional time was reallocated rather than reduced. Core theoretical content and essential worked examples were still covered during class. Additional example problems, previously solved during class in Year 1, were recorded as short videos and made available through the course Learning Management System (LMS) for optional review. This adjustment preserved topic coverage while freeing a few class sessions for collaborative activities without increasing student workload or reducing instructional rigor.

Instructional models

The lecture format remained consistent in both years: core concepts were taught, and key examples were solved in class. The primary difference between the two years lay in the assignment and reinforcement stage.

Year 1: Traditional model – individual homework

Assignment: At the end of each chapter, five practice problems were assigned.

Procedure: Students completed all problems individually outside of class and submitted them by a set deadline. Homework was graded for correctness and completeness. There was no instructor supervision during problem-solving and no peer interaction.

Year 2: Collaborative model – structured in-class collaborative problem solving

Grouping and Assignment: Students were divided into groups of three. Each group was assigned one of the five practice problems.

Procedure: Groups were given 30 min of the 90-min class session to solve the problem collaboratively. Peer discussion was required. Given the small cohort size (five groups), the instructor circulated continuously among teams. The instructor used structured prompts such as: “What are the givens?”, “What are you asked to solve?”, “What principle applies here?”, “Can you explain your reasoning?”. These prompts helped prevent common issues like free-riding, superficial agreement, or getting stuck. The instructor verified that at least one member of each group could clearly explain the solution before moving to the next team, encouraging internal peer teaching. After the problem-solving period, each group selected a representative to take the podium and present its solution. Presentations included explaining the problem statement and the entire solution. Both the collaborative work and the quality of the class presentation were graded. This structure transformed the classroom into a mini engineering workshop, allowing students to learn from peers and correct misconceptions immediately. Because all sessions were recorded, students could later watch and review both instructor explanations and peer presentations.

To ensure a fair comparison between instructional models, the same problems were used in both years. Numerical values varied, but problem structure, difficulty, and required concepts remained consistent.

Example of an in-class collaborative problem

To illustrate the type of problem-solving activities used in the collaborative model in Year 2, this section presents an example drawn from the end-of-chapter 2 problems in the textbook²⁶: An oil pump, as shown in Figure 1, is drawing 44 kW of electric power while pumping oil with a density of 860 kg/m³ at a rate of 0.1 m³/s. The inlet and outlet diameters of the pipe are 8 cm and 12 cm, respectively. If the pressure rise of oil in the pump is measured to be 500 kPa, determine the useful mechanical power of the pump (Ẇ_pump).

Figure 1.

Example of an in-class collaborative problem.²⁶

Before assigning the problems, the instructor reviewed the main concepts from the chapter, including conservation of mass, the fluid mechanical energy equation, and the assumptions appropriate for steady, incompressible flow. Then, the students were instructed to begin with general governing equations and systematically simplify them using the given information and reasonable assumptions. This scaffolded approach encouraged students to derive the necessary relationships themselves, rather than looking up online solution sites or AI tools.

The assigned team approached the problem by modeling the pump as an open system. They identified the inlet (number 1) as the 8 cm diameter section with higher velocity and lower pressure, and the outlet (number 2) as the 12 cm diameter section with lower velocity and higher pressure. The useful power delivered by the pump corresponds to the rate of increase in the oil's mechanical energy:

{\dot{W}}_{p u m p} = Δ {\dot{E}}_{m e c h, f l u i d} = \dot{m} (\frac{P_{2} - P_{1}}{ρ} + \frac{V_{2}^{2} - V_{1}^{2}}{2} + g (Z_{2} - Z_{1}))

(1)

Since $\dot{m} = ρ \dot{V}$ , the oil is incompressible ( $ρ_{1} = ρ_{2}$ ), and the flow is steady ( ${\dot{m}}_{1} = {\dot{m}}_{2}$ ), the oil volumetric flow rate ( $\dot{V}$ ) remains constant. Therefore, the inlet and outlet velocities are:

V_{1} = \frac{{\dot{V}}_{1}}{A_{1}} = \frac{0.1 m^{3} / s}{\frac{π}{4} {(0.08 m)}^{2}} = 19.9 m / s

(2)

V_{2} = \frac{{\dot{V}}_{2}}{A_{2}} = \frac{0.1 m^{3} / s}{\frac{π}{4} {(0.12 m)}^{2}} = 8.84 m / s

(3)

Assuming a negligible change in potential energy, Equation (1) reduces to:

{\dot{W}}_{p u m p} = \dot{V} (Δ P + ρ \frac{V_{2}^{2} - V_{1}^{2}}{2}) = 36.6 k W

(4)

In the traditional model in Year 1, many students copy-pasted Equation (4) from online solution platforms or generated it using AI tools, treating it as a hydraulic power formula without understanding where it came from. Several students later reported that because they “did not know the underlying concepts” and they “could not remember the formula” during exams, they were unable to solve a similar exam problem independently.

In contrast, in the collaborative model in Year 2, students derived the equation themselves during the in-class activity. This approach helped them understand the reasoning behind the formula and eliminated the need to memorize it for exams. This example highlights how structured, in-class collaborative problem-solving can reduce reliance on external solutions while improving students’ ability to apply concepts independently.

Assessment

Both instructional models included three exams: Exam 1, Exam 2, and a cumulative Final Exam. The exams were designed with comparable difficulty and problem types to ensure a fair comparison between the two cohorts. All enrolled students completed all assessments. Student feedback was collected through conversations during and after class to better understand their perceptions of learning, engagement, and problem-solving confidence.

This study was conducted in accordance with institutional guidelines for educational research. All students provided consent allowing their anonymized grades and feedback to be used for research purposes.

Results and discussion

Quantitative results

Table 1 summarizes the student performance for both instructional models. Results are reported as mean ± standard deviation, where the mean represents the average score (out of 100 points) and the standard deviation reflects the score spread within each assessment. Larger standard deviations indicate greater variability in student performance.

Table 1.

Comparison of student performance between teaching models.

Assessment category	Year 1 (Traditional Model)	Year 2 (Collaborative Model)	Difference (Year 2 - Year 1)	p-value
Exam 1	85.6 ± 13.8	90.5 ± 9.7	+4.9%	0.33
Exam 2	73.0 ± 17.3	83.2 ± 17.0	+10.2%	0.15
Final exam	74.4 ± 23.5	83.1 ± 14.4	+8.7%	0.30
Overall course grade	77.4 ± 17.0	85.4 ± 11.7	+8.0%	0.20

Across all assessments, students in the collaborative model (Year 2) earned higher average scores than those in the traditional model (Year 1). The largest improvements were observed in Exam 2 (10.2%) and in the overall course grade (8.0%). These results indicate improved student understanding and retention of thermodynamics concepts under the collaborative model.

Two-tailed t-tests were used to compare performance between cohorts. Although all assessments showed higher mean scores in the collaborative model, none reached statistical significance at the conventional p < 0.05 level. This outcome is expected given the small sample size (15 students per year) and substantial score variability, as reflected by the standard deviations. Despite this, the consistent direction and magnitude of improvement across all assessments suggest a meaningful educational effect that would likely reach statistical significance with larger cohorts.

Figures 2 to 5 illustrate grade distributions for each assessment. Figure 2 compares Exam 1 results. A score of 70 (C-) is the minimum required for passing Thermodynamics I. In the collaborative model, only 13% of students scored below this threshold, compared with 25% in the traditional model. Additionally, 87% of students in the collaborative cohort earned A and B grades (80–100 points), compared with 66% in the traditional cohort. This early-semester improvement reflects the positive effect of collaborative in-class activity on students’ understanding of foundational thermodynamics material.

Figure 2.

Exam 1 grade distributions for the traditional and collaborative instructional models.

Figure 3.

Exam 2 grade distributions for the traditional and collaborative instructional models.

Figure 4.

Final exam grade distributions for the traditional and collaborative instructional models.

Figure 5.

Overall course grade distributions for the traditional and collaborative instructional models.

Figure 3 shows a similar trend for Exam 2. More than 70% of students in the collaborative model scored above the C- threshold, compared with 58% in the traditional model. The proportion of students scoring below 70 decreased substantially from 42% to 27%. This indicates stronger mid-semester comprehension under the collaborative model.

Final exam results, shown in Figure 4, also favored the collaborative model. In Year 2, 87% of students earned scores above 70, compared with 58% in Year 1. The traditional model exhibited a higher concentration of low scores, indicating weaker cumulative understanding when learning relied primarily on unsupervised homework.

Figure 5 presents the overall course grade distributions. Only 13% of students in the collaborative model earned a final grade below the C- threshold, compared with 25% in the traditional model – nearly double the proportion. This result is particularly important because a minimum grade of C- is required for enrollment in subsequent courses such as Thermodynamics II and Fundamentals of Heat Transfer. The collaborative model also produced a higher proportion of A and B final grades, indicating stronger overall preparation for follow-on coursework.

Taken together, these results show that replacing traditional homework with structured in-class collaboration improved both short-term performance and cumulative learning outcomes.

Homework performance analysis

In the traditional model, widespread access to online solution manuals and generative AI tools led to uniformly high homework scores that overstated students’ actual level of understanding, creating a false sense of preparedness. Consequently, homework completion did not reliably reflect students’ ability to apply thermodynamics concepts independently. In Year 1, the average homework score was 97.3 ± 3.1, substantially higher than all exam averages, demonstrating a weak alignment between homework performance and exam-based measures of conceptual mastery.

In the collaborative model, traditional homework was replaced with in-class collaborative problem-solving. These activities were graded based on solution correctness and completeness, as well as the presentation quality. The average collaborative activity score was 98.1 ± 2.0, comparable in magnitude and variability to homework grades observed in Year 1.

Despite differences in instructional format, both grading schemes produced similarly high activity scores, limiting their usefulness as discriminators of learning between models. In contrast, exam performance (completed individually and without external aids) revealed clear differences between cohorts. As a result, exam outcomes provide the most credible evidence that the collaborative model improved students’ conceptual understanding and ability to independently solve thermodynamics problems.

Qualitative observations

Qualitative observations supported the quantitative findings. One notable outcome was a stronger alignment between student preparation and exam performance in the collaborative model. In Year 1, many students appeared confident based on homework completion and high scores, but struggled to solve similar problems independently during exams. Instructor observations suggested that reliance on solution manuals and AI tools reduced opportunities for active reasoning and self-assessment.

In contrast, the collaborative model in Year 2 required students to solve problems during class, explain their reasoning to peers, and respond to questions from teammates and the instructor. This process made misunderstanding visible in real-time and encouraged deeper engagement with the material. Students who actively participated in group discussions consistently demonstrated stronger performance on related exam problems.

Instructor observations during grading further indicated that students showed better mastery of concepts directly connected to problems they had solved collaboratively in class. Students who regularly explained their reasoning performed well on corresponding exam questions.

Student feedback aligned with instructor observations. Many students reported that collaborative problem-solving helped them understand topics more efficiently, particularly when exposed to multiple solution approaches. One student stated that group work made them “understand the topic more efficiently,” especially when they could listen to different approaches from peers. Several students noted increased confidence when working independently on exams.

A limitation of the collaborative model was uneven participation within some groups. A small number of students participated minimally and relied on teammates. Because all group members received the same in-class problem-solving activity grade and no peer evaluation was implemented, this behavior was not penalized.

Conclusion

This study demonstrates that replacing traditional individual homework with structured, instructor-led in-class collaborative problem-solving improves student learning and academic progression in a rigorous engineering course. Students in the collaborative model achieved an 8.0%-point increase in overall course grade and notable gains on exams, including a 10.2%-point improvement on Exam 2. More students met the C- threshold required to advance in the mechanical engineering curriculum. These results are particularly important because Thermodynamics I is a prerequisite for Thermodynamics II and Fundamentals of Heat Transfer. Improved exam performance and fewer failing grades in the collaborative model indicate stronger conceptual understanding and better readiness for upper-level coursework. Through guided teamwork, explanation of reasoning, and immediate instructor feedback, students developed analytical, communication, and collaborative problem-solving skills essential for both academic success and professional engineering practice. In an educational environment increasingly influenced by AI-generated solutions, this approach promotes authentic learning by shifting problem-solving from unsupervised homework to supervised in-class reasoning. The findings suggest that transforming homework into a structured, collaborative classroom activity is an effective strategy for improving mastery, persistence, and preparation for future STEM coursework.

Future work

While the findings of this study are promising, they are intentionally based on a small summer Thermodynamics I course. The low enrollment allowed for close instructor facilitation, real-time feedback, and careful observation of group interactions during the initial implementation of the collaborative model. Future work will extend this same instructional approach to larger-enrollment sections, which typically enroll 70–80 students during fall and spring semesters. Scalability will be explored using trained teaching assistants, structured facilitation strategies, and technology-supported group management tools, informed by prior literature and insights from this study.

Another important direction is the incorporation of individual accountability mechanisms, such as peer evaluation. These measures may promote active participation and reduce uneven contribution within groups, further strengthening the effectiveness of the collaborative approach.

Future studies should also investigate how in-class collaborative problem-solving affects students’ use of online solution manuals and AI-based tools. Understanding this relationship is critical for designing learning environments that promote authentic problem-solving, academic integrity, and meaningful engagement in the AI era.

Footnotes

Nomenclature

ORCID iD

Mary Zadeh

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The datasets used and analyzed in the current study are available from the author upon reasonable request.

References

McDonald

. STEM education: a review of the contribution of the disciplines of science, technology, engineering and mathematics. Sci Educ Int 2016; 27: 530–569.

Winget

Persky

. A practical review of mastery learning. Am J Pharm Educ 2022; 86: 1114–1122.

Goldstein

. Does homework help? A review of research. Elem Sch J 1960; 60: 212–224.

Eren

Henderson

. The impact of homework on student achievement. Econom J 2008; 11: 326–348.

Fisher

Frey

. Better learning through structured teaching: a framework for the gradual release of responsibility. 3rd ed. Alexandria, VA, USA: ASCD, 2021.

Gill

Schlossman

. The lost cause of homework reform. Am J Educ 2000; 109: 27–62.

Marzano

Pickering

. Special topic / the case for and against homework. Educ Leadersh 2007; 64: 74–79.

Fajt

Schiller

. ChatGPT in academia: University Students’ attitudes towards the use of ChatGPT and plagiarism. J Acad Ethics 2025; 23: 1363–1382.

Krecar

Kolega

Jurcec

. Perception of ChatGPT usage for homework assignments: Students’ and Professors’ perspectives. J Educ Technol Educ 2024; 12: 33–60.

10.

Prince

. Does active learning work? A review of the research. J Eng Educ 2004; 93: 223–231.

11.

Freeman

Eddy

McDonough

, et al. Active learning increases student performance in science, engineering, and mathematics. Proc Natl Acad Sci U S A 2014; 111: 8410–8415.

12.

Zadeh

. Modernizing the thermal fluid sciences curriculum: implementation of a hands-on lab course in mechanical engineering. Int J Mech Eng Educ 2025. doi:10.1177/03064190251363071

13.

Mason

Singh

. Helping students learn effective problem solving strategies by reflecting with peers. Am J Phys 2010; 78: 748–754.

14.

Stockwell

Jiang

. Group problem solving in class improves undergraduate learning. ACS Cent Sci 2017; 3: 614–620.

15.

Lee

Hannafin

. A design framework for enhancing engagement in student-centered learning: Own it, learn it, and share it. Educ Technol Res Dev 2016; 64: 707–734.

16.

Pourhasanzadeh

Bigham

. Optimization of a micro gas turbine using genetic algorithm. In: ASME Turbo Expo: Power for Land, Sea, and Air, Vancouver, British Columbia, Canada, June 6–10, 2011, pp.929–937.

17.

Zadeh

. Thermo-economic-environmental optimization of a microturbine using genetic algorithm. Therm Sci 2015; 9: 475–487.

18.

Pourhasanzade

Sabzpoushan

. A cellular automata model of chemotherapy effects on tumour growth: targeting cancer and immune cells. Math Comput Model Dyn Syst 25(1): 63–89.

19.

Pourhasanzade

Sabzpoushan

Alizadeh

, et al. An agent-based model of avascular tumor growth: immune response tendency to prevent cancer development. Simulation 2017; 93: 641–657.

20.

Smith

Wood

Adams

, et al. Why peer discussion improves student performance on in-class concept questions. Science 2009; 323: 122–124.

21.

Felder

Brent

. Teaching and learning STEM: a practical guide. San Francisco, CA, USA: John Wiley & Sons, 2024.

22.

Karau

Williams

. Social loafing: a meta-analytic review and theoretical integration. J Pers Soc Psychol 1993; 65: 681–706.

23.

Karau

Wilhau

. Social loafing and motivation gains in groups: An integrative review. Individual Motivation within Groups: Social Loafing and Motivation Gains in Work, Academic, and Sports Teams, 2019, pp.3–51.

24.

Johnson

Smith

. The state of cooperative learning in postsecondary and professional settings. Educ Psychol Rev 2007; 19: 15–29.

25.

Doulougeri

Vermunt

Bombaerts

, et al. Challenge-based learning implementation in engineering education: a systematic literature review. J Eng Educ 2024; 113: 1076–1106.

26.

Cengel

Boles

. Thermodynamics: an engineering approach. 10th ed. New York, NY, USA: McGraw-Hill, 2023.