While generative AI shows promise in supporting early phases of knowledge construction through educational discourse, its ability to function as a true collaborator remains limited, requiring careful attention to tool design, pedagogical integration, and learner characteristics to facilitate deeper cognitive engagement beyond basic information exchange.

Objective

This systematic literature review aimed to explore how Generative AI (GenAI) facilitates knowledge construction through discourse in educational settings. The study examined four primary research questions: (1) the research contexts in which GenAI is studied, (2) the learning contexts where GenAI is implemented, (3) the emergent processes of knowledge construction supported by GenAI, and (4) the impacts of GenAI on knowledge construction, including its various roles and influencing factors. The overarching goal was to determine whether and how GenAI can serve as a meaningful collaborator in educational discourse that promotes deep and sustained knowledge construction.

Methods

The researchers conducted a systematic literature review following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) protocol. They searched four major bibliographic databases (Scopus, ACM Digital Library, IEEE Xplore, and Web of Science) for peer-reviewed journal articles and conference proceedings published between November 2022 and June 2024. The search string combined terms related to GenAI ("generative AI" OR "ChatGPT" OR "Chatbots") with knowledge construction terms ("knowledge construction" OR "knowledge building").

The screening process involved eliminating duplicates, excluding studies based on titles and abstracts, and conducting full-text reviews. From an initial 137 potential articles, the researchers ultimately selected 28 studies for analysis. Data analysis employed a structured coding framework with nine multi-categorical coding elements, guided by Ertmer's framework for technology integration barriers and Onrubia and Engel's four-phase model of knowledge construction (Initiation, Exploration, Negotiation, and Construction). Two researchers independently coded the studies with high interrater reliability (Krippendorff's alpha > 0.77), resolving disagreements through consensus. Thematic synthesis following Braun and Clarke's approach was used to identify patterns across studies.

Key Findings

Research and Learning Contexts:

• Developmental research was the predominant methodology (39.3%), reflecting the field's focus on creating and refining GenAI tools for specific educational objectives
• Studies concentrated on higher education (39.3%) and K-12 settings (32.1%), with emphasis on STEM fields (42.9%) and general disciplines (35.7%)
• GPT-based models, particularly ChatGPT, were most frequently employed (50.0%), followed by task-specific GenAI (28.6%)
• Studies examined GenAI in both individual and collaborative learning contexts relatively equally
• The primary purpose was examining learning experiences (71.4%), with fewer studies on research applications (25.0%) and teaching (3.6%)

Knowledge Construction Processes: GenAI demonstrated varying levels of support across the four knowledge construction phases:

• Phase 1 (Initiation): All 28 studies (100%) showed GenAI supporting initial exploration of concepts and ideas, with four studies (14.3%) showing only this phase
• Phase 2 (Exploration): 24 studies (85.7%) demonstrated GenAI helping learners identify inconsistencies and explore alternative perspectives
• Phase 3 (Negotiation): 21 studies (75.0%) showed GenAI collaborating with learners to reconcile differences and synthesize information, though nine studies (32.1%) progressed only to this phase
• Phase 4 (Construction): Only 12 studies (42.9%) demonstrated GenAI supporting learners in developing shared understanding and applying newly constructed knowledge to real-world tasks

This distribution reveals that while GenAI effectively supports initial and intermediate phases, it has limited representation in enabling the most complex cognitive processes required for deep knowledge construction.

Roles of AI in Knowledge Construction: The analysis identified three distinct roles GenAI can play:

• Collaborator (58.9%): Actively participating as a peer in knowledge construction, engaging in meaningful dialogue and co-construction
• Facilitator (30.4%): Serving as a scaffolding tool, guiding learners through complex tasks and providing procedural support
• Assistant (10.7%): Providing efficiency and convenience, primarily supporting rather than actively participating in knowledge construction

Factors Influencing Effectiveness: Three categories of factors emerged as critical:

Tool-Related Factors:

• Prompt design (71.4% of studies) – Well-crafted, contextually relevant prompts were critical for steering learners toward higher-order thinking
• Learner control/agency (57.1%) – Allowing learners to control interactions enabled deeper engagement
• Social-emotional support/tone (42.9%) – Empathetic and supportive responses encouraged sustained discourse
• Interface design and usability (32.1%) – Well-designed interfaces reduced cognitive load
• Humanized features (28.6%) – Natural conversational patterns fostered collaboration
• Adaptability (17.9%), reuse/reset mechanisms (17.9%), and hinting mechanisms (14.3%)

Pedagogical Factors:

• Guided instruction (60.7%) – Educator-provided guidance was crucial for scaffolding learner engagement
• Structured integration into learning (35.7%) – Systematic embedding within course activities promoted intentional use
• Timing/duration of intervention (35.7%) – Properly timed and sustained interactions supported progression through cognitive phases
• Alignment with pedagogical principles (21.4%), explicit instructions to GenAI (21.4%), and collaborative features (25.0%)
• Ethical considerations (10.7%) – Data privacy and overreliance concerns

Learner-Related Factors:

• Trust/doubt in AI responses (28.6%) – Levels of trust significantly affected depth of engagement
• AI literacy (21.4%) – Learners with relevant information and skills leveraged GenAI more effectively
• Prior knowledge (14.3%) – Strong foundational knowledge enabled deeper engagement
• Motivation (10.7%) and self-regulated learning skills (10.7%)

Implications

This research makes several significant contributions to understanding GenAI's role in educational discourse:

1. Age-appropriate integration needed: The concentration of studies in higher education settings suggests current GenAI limitations make it more suitable for learners with developed critical thinking skills. Educational institutions need comprehensive, age-appropriate guidelines considering cognitive development stages.
2. Multifaceted interaction recognition: GenAI's effectiveness depends on the complex interplay of tool design, pedagogical integration, and learner characteristics rather than the technology alone. This emphasizes the need for holistic approaches integrating social, emotional, and cognitive dimensions.
3. Limited collaboration in advanced phases: While GenAI shows unique potential compared to earlier conversational technologies in supporting higher knowledge construction phases (43% reaching Construction phase vs. isolated cases in prior studies), it still struggles with enabling the deepest levels of cognitive engagement required for true collaboration.
4. Critical role of human mediation: Guided instruction and systematic integration emerged as essential, highlighting that instructor guidance remains crucial in AI-supported learning environments. Professional development programs should equip educators with strategies to integrate GenAI aligned with pedagogical best practices.
5. Expanding beyond learning contexts: There's a notable gap in research on GenAI's role in teaching and research activities, despite widespread adoption by educators and researchers. Future research should examine how GenAI transforms pedagogical approaches and research methodologies.
6. Theoretical advancement: The study expands understanding of technology integration in education by demonstrating how GenAI supports "conversational scaffolding," guiding learners to reconcile viewpoints and construct coherent understanding through iterative dialogue.

Limitations

The researchers acknowledge several important limitations:

• Narrow demographic focus: Most studies explored individual learning contexts, overlooking GenAI's potential in teacher-mediated and collaborative learning scenarios. Future research should investigate how teachers can leverage AI tools and how AI supports group interactions.
• Exploratory stage research: The review was restricted to a 19-month timeframe (November 2022-June 2024), capturing only preliminary studies focused on initial implementation and short-term effects rather than sustained long-term impacts.
• Process-focused analysis: Studies primarily focused on immediate discourse between learners and GenAI, often overlooking final learning artifacts (assignments, projects, essays) that indicate how learners internalize and apply constructed knowledge.
• Cognitive depth not explicitly assessed: While the study examined knowledge construction processes, it did not explicitly assess cognitive complexity levels. Integration with frameworks like Bloom's taxonomy could provide deeper insights.
• Learner-driven burden: Results often showed knowledge construction was contingent on learner-driven engagement, indicating GenAI may not yet function as a fully autonomous collaborator without placing undue burden on learners.
• Limited scope: The 28 studies analyzed, while systematically selected, represent a relatively small sample that may not capture the full breadth of GenAI applications or evolving capabilities.

Future Directions

The researchers recommend several avenues for continued investigation:

1. Longitudinal designs: Employ long-term studies examining sustained interaction with GenAI and its influence on deeper learning processes, tracking learners' engagement across different knowledge construction phases over extended periods.
2. Holistic outcome analysis: Adopt approaches analyzing both interaction processes with GenAI and end products of learning (assignments, projects) to understand the extent GenAI contributes to meaningful learning and knowledge transfer.
3. Cognitive depth assessment: Integrate frameworks like Bloom's taxonomy with knowledge construction models to examine how each phase corresponds to specific cognitive engagement levels, capturing both collaborative construction processes and cognitive quality.
4. Teacher-mediated and collaborative contexts: Investigate how GenAI can be integrated into teacher-facilitated instruction and collaborative learning scenarios, exploring how teachers can leverage AI tools for differentiated instruction and how AI supports group interactions.
5. Intelligent adaptive systems: Focus on designing more intelligent, adaptive GenAI systems that can proactively support learners across the entire learning cycle without placing undue burden on learners to direct the process.
6. Comparative studies: Investigate how the quality and complexity of final assignments differ when learners use GenAI as a collaborative tool compared to traditional methods or human-peer collaboration.
7. Broader application contexts: Expand research beyond learning to examine GenAI's role in teaching practices and research methodologies, potentially revealing new paradigms in education and academic inquiry.
8. Ethical integration: Develop frameworks addressing data privacy, algorithmic bias, and potential overreliance on AI-generated content to create safe, effective learning environments.

Title and Authors: "Questioning the Role of AI as Collaborator: A Systematic Literature Review of Generative AI-Supported Knowledge Construction" by Yeonji Jung (Texas A&M University) and Sung-Hee Jin (Hanbat National University, South Korea)

Published On: September 29, 2025

Published By: Interactive Learning Environments (Taylor & Francis Group)