Trust, Usefulness, and Dependency on AI in Programming: A Hierarchical Clustering Approach

AI summary

Overview

Research area: AI in programming education, with a focus on adoption in underrepresented and developing regions; categorized under AI Safety & Ethics.

Technical level: Beginner-Friendly. The abstract names hierarchical clustering as the analytic method but does not describe the algorithm, features, or validation in technical detail, so it is accessible to a general audience.

Scope: A survey-based study of 508 first-year programming students in Pampanga, Philippines, that uses hierarchical clustering to group students by their trust in, perceived usefulness of, and dependency on AI programming tools.

What This Paper Is About

AI tools are reshaping how programming is taught, but most evidence about how students relate to those tools comes from well-resourced settings — adoption in underrepresented countries is not well understood. This paper asks how first-year programming students in one Philippine province perceive AI tools in terms of trust, usefulness, and dependency. The goal is to identify distinct student profiles so that educators and institutions can integrate AI into programming education more equitably and effectively.

Key Contributions

1. Provides empirical survey evidence on AI perceptions from an underrepresented region, addressing a gap the authors identify in existing research.
2. Applies hierarchical clustering to student responses rather than treating the student population as a single average, yielding four distinct profiles defined by differing trust and usage intensity.
3. Reports that usage frequency and positive perception do not move together — high-frequency users were not necessarily more trusting or more convinced of AI's usefulness — pointing to a more complex relationship between how much students use AI and how they feel about it.
4. Translates the findings into concrete institutional recommendations: infrastructure development, training programs, and curriculum integration.

Main Findings

• Four student profiles emerged: Hierarchical clustering grouped the 508 respondents into four unique profiles that differ in their levels of trust and intensity of AI tool usage.
• Benefits acknowledged, dependency low: Students recognized the benefits of AI tools, but their dependency on them remained low.
• Context constrains adoption: The authors attribute low dependency to limited infrastructure and insufficient exposure rather than to student attitudes alone.
• Usage does not equal trust: High-frequency users did not necessarily report greater trust or greater perceived usefulness, which the authors read as evidence of a complex relationship between usage patterns and perception.
• Intervention is needed for impact: The study concludes that maximizing AI's educational value requires deliberate support — infrastructure, training, and curriculum integration — rather than assuming adoption will follow from tool availability.

Methodology in Plain English

The researchers surveyed 508 first-year programming students in Pampanga, Philippines, collecting their perceptions of AI tools along three dimensions named in the abstract: trust, perceived usefulness, and dependency. They then analyzed the responses with hierarchical clustering, a technique that groups participants so that students with similar response patterns land in the same group and dissimilar ones are separated. Rather than reporting only an overall average, this approach surfaces subgroups — here, four profiles — that may need different kinds of support. The abstract does not specify the survey instrument, the exact variables fed into the clustering, the distance measure, or how the number of clusters was chosen.

Why This Matters

Impact on research: The study extends the AI-in-education literature beyond well-studied, well-resourced contexts, supplying empirical data from a developing region. Its finding that usage intensity and trust diverge challenges the intuitive assumption that more exposure automatically produces more favorable perceptions, and it gives future work a set of testable student profiles to confirm, refine, or refute.

Real-world applications:

• Curriculum design: Institutions can tailor AI integration to different student profiles instead of applying one policy to everyone.
• Infrastructure planning: The low dependency attributed to limited infrastructure signals where investment in labs, connectivity, and tool access would matter most.
• Training programs: Targeted training can address insufficient exposure, which the authors identify as a barrier alongside infrastructure.
• Equity-focused policy: Governments and universities in developing regions can use these findings to plan AI adoption that does not widen gaps between resourced and under-resourced institutions.

Industry relevance: Employers and AI tool developers benefit from understanding where future programmers' familiarity and trust actually stand. The disconnect between heavy use and low trust suggests that tool makers cannot assume frequent usage means satisfied or confident users, and that localized onboarding, documentation, and support may be needed in markets where exposure is limited.

Future Directions

• Test whether improving infrastructure and exposure actually raises dependency and trust, since the study identifies these as the limiting factors but does not evaluate interventions.
• Track students over time to see whether the four profiles are stable or whether students shift between them as they gain programming experience.
• Investigate the mechanisms behind the usage–trust disconnect: why do frequent users fail to report greater trust or usefulness? Qualitative interviews or follow-up surveys could probe this.
• Replicate the clustering approach in other underrepresented regions and institutions to determine whether the same four profiles generalize or are specific to this sample.
• Examine how the identified profiles relate to learning outcomes, which the abstract does not address.

Target Audience

Programming educators and curriculum designers, particularly those working in developing regions; education researchers studying technology adoption and AI literacy; university administrators and policymakers responsible for infrastructure and curriculum decisions; and AI ethics and safety researchers interested in equitable deployment of AI tools. The paper is written at an accessible level and does not require a background in clustering or statistics to follow its main claims.

Authors’ abstract

While AI tools are transforming programming education, their adoption in underrepresented countries remains insufficiently studied. Understanding students' trust, perceived usefulness, and dependency on AI tools is essential to improving their integration into education. For these purposes, this study surveyed 508 first-year programming students in Pampanga, Philippines and analyzed their perceptions using hierarchical clustering. Results showed four unique student profiles with varying in trust and usage intensity. While students acknowledged AI tools' benefits, dependency remained low due to limited infrastructure and insufficient exposure. High-frequency users did not necessarily report greater trust or usefulness which may indicates a complex relationship between usage patterns and perception. This study recommends that to maximize AI's educational impact, targeted interventions such as infrastructure development, training programs, and curriculum integration are necessary. This study provides empirical insights to support equitable and effective AI adoption in programming education within developing regions.