Assessing the Influence of AI-Generated Responses on Academic Achievement: An Ethical Perspective and Self-Efficacy

Synthesis: Oladayo and Diri (2026) used a quasi-experimental design in Ughelli North, Delta State, Nigeria, to test whether AI-generated responses improve mathematics achievement and Self-Efficacy among secondary school students. From a population of 1,369 Senior Secondary School I students, 115 were sampled from two coeducational schools: an experimental group of 62 taught with ChatGPT-supported feedback on quadratic equations and a control group of 53 taught conventionally, over 6 weeks. The experimental group scored higher on the Quadratic Equation Achievement Test (mean 29.18 versus 24.06; t(113) = 7.38, p = .000), and students reported that AI responses boosted their confidence (grand mean 2.61 on a four-point scale). Male students outperformed female students on the posttest (30.95 versus 25.16). The authors argue that AI can support learning when teachers provide their own input and students engage ethically, while flagging Privacy, bias, and infrastructure as conditions that must be addressed in under-resourced contexts.

Key Findings

1. AI-supported group outperformed. The 62 students taught quadratic equations with ChatGPT-generated feedback scored a mean of 29.18 on the posttest against 24.06 for the 53 controls, a mean gain of 5.12.
2. The achievement difference was significant. An independent-samples t-test returned t(113) = 7.38 at p = .000, leading the authors to reject the null hypothesis of no difference between the two groups.
3. Self-efficacy rose. On a five-item questionnaire, students agreed that AI responses boosted their marks (2.67), that teachers encouraged their use (2.74), and that the tools raised self-esteem (2.73); the grand mean was 2.61.
4. Gender gap in achievement. Male students scored a mean of 30.95 versus 25.16 for female students on the posttest, a difference of 5.79 that was statistically significant (t(60) = 6.08, p = .00).
5. No gender gap in self-efficacy. The self-efficacy scores of male and female students did not differ significantly (t(60) = 0.08, p = .94), with means of 2.47 and 2.54.
6. Ethical conditions matter. The authors tie the benefit to teacher guidance, AI Literacy, privacy safeguards, and low-bandwidth access, warning that unmonitored AI can widen inequities for under-resourced Nigerian schools.

AI-supported feedback and mathematics achievement

The study compares two intact classes. The experimental group used ChatGPT on their phones to check steps and receive hints while solving quadratic equations, with teachers trained to guide that use; the control group received conventional chalkboard, teacher-centered instruction. After a pretest and three weeks of teaching, the achievement test — 25 objective questions worth 4 marks each, totaling 100 marks — separated the groups. The authors interpret the 5.12 mean gain as evidence that AI-generated responses can support Problem Solving and conceptual understanding when embedded in lessons, and they connect the timely, personalized explanations to Scaffolding within a learner's zone of proximal development. Yet they caution that AI must promote reasoning rather than answer copying, and that overdependence may erode motivation, critical thinking, and self-directed learning. This tension — AI as scaffold versus AI as shortcut — runs through the paper's framing.

Self-efficacy and confidence

Drawing on Bandura's construct of self-efficacy and on self-determination theory, the authors treat confidence as a key determinant of mathematics success. The five-item questionnaire probed enjoyment, perceived marks, teacher encouragement, discouragement, and self-esteem. Students agreed with four items and disagreed with the statement that AI responses discourage and mislead them (mean 2.36), yielding a grand mean of 2.61, which the authors read as evidence that AI responses positively influence self-efficacy in solving mathematics. The paper notes, however, that excessive reliance on algorithmic guidance may weaken autonomy and problem-solving capacity, particularly in demanding topics such as quadratic equations. It also stresses that AI lacks moral reasoning, empathy, and contextual sensitivity, so accountability for instructional decisions should remain with educators rather than being delegated to the system.

Gender differences and ethical cautions

Male students in the experimental group outperformed female students by 5.79 points on the posttest, a significant difference, although self-efficacy did not differ by gender. The authors read this as a warning that AI tools offering instant solutions may reinforce existing gender disparities, especially where female students face stereotype threat and lose the confidence-building opportunities that come from productive struggle. They place the study within broader equity concerns: disparities in electricity, internet connectivity, and digital devices in Delta State could let AI widen achievement gaps, and biased training data may produce unequal learning experiences. Their recommendations call on government to regulate AI in schools, on curriculum developers to reflect AI tools in syllabi, and on teachers to integrate AI while providing their own input — all under Global South resource constraints that the paper says demand local ethics committees and governance frameworks.

What this means for practice

• Instructors. Integrate AI-generated responses into mathematics lessons while providing your own input: the experimental group improved only when trained teachers guided how students used ChatGPT to check steps and get hints.
• Instructors. Ask AI to explain why an answer is correct rather than supply the solution, since the authors warn that instant answers can encourage copying over reasoning and may widen gender gaps.
• Curriculum designers. Review secondary mathematics syllabi to reflect AI-generated response tools as recommended teaching strategies, and build AI literacy and ethical guidelines into the curriculum.
• Policymakers. Establish regulations, local ethics committees, and governance frameworks before scaling AI adoption, and address the electricity, connectivity, and device gaps that could otherwise widen inequities.
• Researchers. Follow up with longitudinal and gender-disaggregated studies: this six-week study cannot show whether sustained AI use builds autonomy or breeds dependency.

Limitations

• The sample was small (115 students) and drawn from just two coeducational schools in Ughelli North Local Government Area, Delta State, so the findings are not generalizable.
• The authors report that financial and time constraints limited how many schools could be included in the study.
• The study lasted only 6 weeks and was restricted to one topic, quadratic equations, so it says nothing about other mathematics content.
• Achievement and self-efficacy were measured by a teacher-developed test and a five-item self-report questionnaire rather than independent or behavioral measures.
• Only coeducational schools were included; single-sex schools were excluded from the study.

Citation

Oladayo, C. E., & Diri, E. A. (2026). Assessing the Influence of AI-Generated Responses on Academic Achievement: An Ethical Perspective and Self-Efficacy of Delta State Secondary School Students, Nigeria. Journal of Research in Mathematics, Science, and Technology Education, 3(4), 276–288.