The Effect of Generative AI as a Coding Assistant in Deep Learning Practicum on Code Quality and Conceptual Understanding

The quote only partly states this claim.

Students in the experimental group completed the final practicum project significantly faster than the control group (143.05 vs. 167.08 minutes, t = -4.12, p < 0.001).

“Students using AI completed tasks significantly faster.”

• Also stated in: § Balancing Productivity and Cognitive Engagement · plain

  “Although AI reduced task completion time significantly, the conceptual improvement indicates that reduced time did not compromise learning depth.”

• AssumptionRests on an assumption: The experimental and control groups did not differ in unmeasured factors that could affect code quality or conceptual understanding.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

AI interaction quality varied among students.

“AI interaction quality varied among students.”

May combine two claims.

The study demonstrates large effect sizes in both cognitive and performance domains.

“Demonstrates large effect sizes in both cognitive and performance domains.”

May combine two claims.

High-performing students tended to use AI for clarification and refinement, whereas lower-performing students relied more heavily on direct code generation.

“However, qualitative observation of AI logs revealed that high-performing students tended to use AI for clarification and refinement, whereas lower-performing students relied more heavily on direct code generation.”

• AssumptionRests on an assumption: The AI interaction logs are a valid representation of students' AI use patterns.

May combine two claims.

Generative AI can reduce extraneous cognitive load (e.g., syntax debugging) while preserving or even enhancing germane cognitive load associated with conceptual schema construction.

“It demonstrates that Generative AI can reduce extraneous cognitive load (e.g., syntax debugging) while preserving or even enhancing germane cognitive load associated with conceptual schema construction.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Intelligent tutoring systems and automated feedback tools enhance learning outcomes and reduce dropout rates for novice programmers.

“For novice programmers, intelligent tutoring systems and automated feedback tools have been shown to enhance learning outcomes and reduce dropout rates [11].”

• cites [11]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Developers using AI pair-programming tools complete tasks faster and report higher perceived productivity.

“Empirical findings suggest that developers using AI pairprogramming tools complete tasks faster and report higher perceived productivity [9].”

• cites [9]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Some studies report improvements in code readability and structural organization with AI-assisted programming.

“While some studies report improvements in code readability and structural organization [17]”

• cites [17]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Other studies highlight risks of shallow learning and dependency with AI-assisted programming.

“others highlight risks of shallow learning and dependency [18].”

• cites [18]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim. May combine two claims.

Participants were selected using cluster random sampling from two parallel classes with equivalent academic backgrounds.

“Participants were selected using cluster random sampling from two parallel classes with equivalent academic backgrounds.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Deep learning practicums require both implementation skills and theoretical comprehension.

“Deep learning practicums represent an ideal testing ground for such investigation because they require both implementation skills and theoretical comprehension.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Project-based and mastery-oriented approaches improve implementation skills and conceptual clarity in machine learning courses.

“Project-based and mastery-oriented approaches have been shown to improve implementation skills and conceptual clarity in machine learning courses [16].”

• cites [16]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Prior studies emphasize the importance of hands-on experimentation combined with conceptual reflection in deep learning education.

“In the context of deep learning education, prior studies emphasize the importance of hands-on experimentation combined with conceptual reflection [5], [15].”

• cites [5]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [15]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

Findings may differ in other computing domains.

“The study focused on deep learning; findings may differ in other computing domains.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Adaptive learning systems integrating AI have positive effects on student motivation and learning gains.

“Similarly, adaptive learning systems that integrate AI have demonstrated positive effects on student motivation and learning gains [12].”

• cites [12]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Both groups received identical instructional materials, learning objectives, assignments, duration of instruction, and assessment rubrics; the only difference was permitted use of Generative AI tools.

“Both groups received identical instructional materials, learning objectives, assignments, duration of instruction, and assessment rubrics.The only difference between groups was the permitted use of Generative AI tools.”

Classified as probably not a substantial claim.

Deep learning concepts are often perceived as abstract and cognitively demanding due to mathematical complexity and multilayered computational structures.

“However, deep learning concepts are often perceived as abstract and cognitively demanding due to their mathematical complexity and multilayered computational structures [4], [5].”

• cites [4]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [5]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Research on automation bias suggests individuals may uncritically accept AI-generated outputs, leading to reduced critical thinking and verification behaviors.

“Research on automation bias suggests that individuals may uncritically accept AIgenerated outputs, leading to reduced critical thinking and verification behaviors [14].”

• cites [14]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

AI coding assistants can improve development efficiency by reducing coding time and minimizing syntactic errors.

“Studies indicate that AI coding assistants can improve development efficiency by reducing coding time and minimizing syntactic errors [3], [8].”

• cites [3]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [8]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

AI assistance does not inherently undermine deep learning; its impact depends on how students interact with the tool.

“These findings suggest that AI assistance does not inherently undermine deep learning; instead, its impact depends on how students interact with the tool.”

• AssumptionRests on an assumption: The experimental and control groups did not differ in unmeasured factors that could affect code quality or conceptual understanding.

Classified as probably not a substantial claim. May combine two claims.

According to cognitive load theory, excessive external assistance may reduce productive cognitive struggle, potentially weakening schema construction and long-term retention.

“According to cognitive load theory, excessive external assistance may reduce productive cognitive struggle, potentially weakening schema construction and long-term retention [13].”

• cites [13]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Posttest conceptual understanding scores and code quality scores were normally distributed in both groups (Shapiro-Wilk p > 0.05).

“The Shapiro-Wilk test indicated that posttest conceptual understanding scores and code quality scores were normally distributed in both groups (p > 0.05).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

AI-based support systems have been associated with increased engagement and immediate feedback mechanisms.

“In educational settings, AI-based support systems have been associated with increased engagement and immediate feedback mechanisms [10].”

• cites [10]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Generative AI can enhance both technical code quality and conceptual mastery in deep learning education.

“This research provides controlled experimental evidence demonstrating that Generative AI can enhance both technical code quality and conceptual mastery in deep learning education”

• Also stated in: § Prospects for Development and Future Applications · boosted

  “this study demonstrates that Generative AI, when integrated thoughtfully into deep learning practicums, can enhance both technical code quality and conceptual understanding.”

• AssumptionRests on an assumption: The Generative AI tools used (ChatGPT/GitHub Copilot) are representative of Generative AI coding assistants in general.
• AssumptionRests on an assumption: The experimental and control groups did not differ in unmeasured factors that could affect code quality or conceptual understanding.