A theoretical and empirical analysis of tensions between learning objects and constructivism

Abstract

Learning Objects (LOs) have long aimed to make digital education scalable and reusable, yet their alignment with constructivist learning remains contested. This study offers a structured comparison of traditional LO design principles and constructivist learning metaphors—acquisition, participation, and knowledge creation—to examine how emerging research directions position themselves within this educational technology landscape. We analyse how emerging research directions—symbolic AI, generative AI, hybrid AI (Retrieval-Augmented Generation), and constructivist-oriented LO research—align with or challenge these learning metaphors. We then explore how these directions influence the relationship between LOs and constructivist pedagogy. Our findings show that while some AI-based approaches reinforce structured, predefined learning, others—and especially constructivist-oriented LO models—support more adaptive, collaborative, and student-centred designs. Empirical findings from teacher interviews reveal that teachers’ conceptions of learning vary by context—often defaulting to transmissive models under technological constraints, but aligning more closely with participation and knowledge creation metaphors when reflecting on pedagogical theory. These combined and somewhat surprising findings underscore the need for LO frameworks that are pedagogically flexible—that is, able to support both structured and open-ended designs, adapt to varying teaching contexts, and empower learners through meaningful engagement.

1 Introduction

Learning Objects (LOs) were originally introduced to enable modular, reusable, and cost-effective learning content, inspired by principles from software engineering and the vision of a learning object economy, based on large-scale sharing and repurposing of digital content across platforms and institutions (Campbell, 2003). The Learning Management System (LMS) market in particular embraced this model, prioritising standardisation, metadata tagging, and content interoperability (Brady & O’Reilly, 2020). While technically efficient, this lineage has often been criticised for foregrounding reuse and control at the expense of pedagogical depth and learner agency (Brady & O’Reilly, 2020).

Constructivist learning theories offer an alternative perspective, positioning learning as an active, contextualised, and socially mediated process in which learners co-construct knowledge through exploration, collaboration, and reflection (Sawyer, 2005). As elaborated by Sfard (1998) and Paavola, Lipponen, and Hakkarainen (2002), this view is captured in the metaphors of acquisition, participation, and knowledge creation, each emphasising learner agency, meaning-making, and social interaction in distinct ways. From this vantage point, the rigid structure and transmission-oriented logic of traditional LOs appear poorly aligned with core constructivist principles.

This disconnect is well documented. A systematic review by Papastergiou and Mastrogiannis (2021) found that while constructivist principles are occasionally present in LO-based interventions, most implementations continue to follow behaviourist or cognitive paradigms. Notably, the constructivist-aligned examples they identified tended to arise from research-driven, educationally grounded design efforts, rather than from mainstream edtech innovation.

In sum, the field remains divided. On one side stands an established LO infrastructure shaped by standardisation, market logics, and technical constraints. On the other, a body of constructivist theory and design practices offers compelling alternatives—but often with limited uptake in scalable technologies.

However, in recent years, the development of AI-driven learning technologies as well as more constructivist-oriented LO research have introduced new opportunities for revisiting these tensions. Research on symbolic AI (Brusilovsky & Peylo, 2003; Ritter, Tehranchi, & Oury, 2019), generative AI (Banh & Strobel, 2023; Thüs, Malone, & Brünken, 2024), hybrid AI (Gao, Xiong, Gao, Jia, Pan, Bi, & Wang, 2023; Lewis, Perez, Piktus, Petroni, Karpukhin, Goyal, & Rocktäschel, 2020), and constructivist-oriented LO design (Ilomäki, Lakkala, & Paavola, 2006; Vanoostveen, Desjardins, & Bullock, 2019) has opened up new ways of understanding how LOs can evolve beyond their original conception.

At the same time, existing uses of LO-based technologies in schools continue to reflect entrenched metaphors and assumptions about teaching and learning, often shaped more by technical affordances than pedagogical insight. As Lakoff and Johnson (2003) emphasise, such metaphors are not merely linguistic devices but deeply embedded conceptual structures that guide how people think, act, and make design decisions—including in education.

This article seeks to bridge these perspectives by examining both theoretical developments and empirical data. The study is situated in two interrelated domains: (1) how different research directions influence the relationship between LOs and constructivism; and (2) how teachers’ conceptions of teaching and learning reflect and shape their use of LO-inspired systems.

Our overall research interest is:

• - How can the alignment between traditional LOs and constructivism be improved?

The article supports two principal approaches to addressing this question. The first approach seeks to identify discrepancies between LOs and constructivism so that these can be addressed through technological and pedagogical innovation. The second approach explores how LOs and constructivist principles shape teachers’ thinking. Rather than resolving these discrepancies through design, its aim is to raise teachers’ awareness of how different learning designs influence their conceptions of teaching and learning.

In response to the tensions outlined above—between scalable LO infrastructures and constructivist pedagogical ideals—this study investigates how six key positions in the field relate to one another: traditional LO design principles, constructivist learning theories, symbolic AI, generative AI, hybrid AI (RAG), and constructivist-oriented LO research. The study’s contribution lies in a novel framework that combines learning metaphors, LO design principles, and emerging AI paradigms to analyse how pedagogical ideals and technological models align, conflict, or co-evolve. Our comparative framework examines these positions from two complementary angles: (1) how symbolic, generative, hybrid, and constructivist-oriented LO research align with or challenge constructivist learning metaphors, and (2) how they support, modify, or challenge core LO design principles. Rather than treating either LO design or constructivism as a static reference point, the analysis traces patterns of mutual adaptation—highlighting points of convergence, tension, and reinterpretation across theoretical and practical domains.

Our research questions are as follows:

• - RQ1: What are commonly discussed perspectives and principles that underlie LO and constructivist learning theories?

• - RQ2: How do traditional LO principles and emerging LO research directions align or conflict with constructivist learning theories?

• - RQ3: How do the affordances of LO-based learning technologies affect teachers’ perspectives on teaching and learning?

  • RQ3a: How do teachers’ conceptions of teaching and learning correspond with mature understandings of learning according to constructivist learning theories?

  • RQ3b: How do teachers’ reflections on LO-based teaching compare with their perspectives when encouraged to draw on their broader pedagogical knowledge?

RQ1 is addressed through the literature review in Sects. 2 and 3. RQ2 builds on the same theoretical and conceptual foundation and is further explored through the comparative analyses in Sect. 6.1 to 6.3. RQ3a and RQ3b are explored through a focus group interview, analysed in Sect. 5 and further discussed in Sect. 6.4.

In this paper, Sects. 2 and 3 present the theoretical background. Section 2 outlines core perspectives in constructivist learning theory. Section 3 examines the origins and principles of Learning Objects (LOs), and introduces emerging research directions—spanning new technologies, practices, and theoretical perspectives—that may help bridge tensions between traditional LO design and constructivist learning. Section 4 describes the research method, and Sect. 5 presents the empirical findings. Section 6 offers an extended discussion: Sect. 6.1–6.3 analyse theoretical and technological developments in relation to constructivist learning principles, while Sect. 6.4 shifts to the empirical level—exploring how teachers’ conceptions of knowledge, teaching, and learning vary depending on whether their reflections are constrained by technology or informed by pedagogical theory. Section 7 concludes the paper with key implications and suggestions for future research.

2 Constructivism and learning metaphors

From the broader frameworks of cognitive and social constructivism, we derive several principles, which we detail in Sect. 2.1. Subsequently, in Sect. 2.2, we explore learning metaphors that aim to bridge the conceptual divide between Learning Objects (LOs) and constructivism.

2.1 Constructivist principles

It is beyond the scope of this article to provide an exhaustive overview of constructivist learning and teaching theories. Instead, our aim is to highlight central aspects that illustrate the complexity of constructivism and illuminate important differences between LOs and constructivism, as reflected in the informants’ descriptions of teaching and learning in these contexts. This overview identifies certain ideal-typical traits to underscore constructivism’s unique features in contrast to LOs. It is crucial to note that these principles are not isolated; they often overlap, coexist, and are deeply integrated, as in the different types of scaffolding environments. The selected principles straddle both constructivism as a learning theory and the didactic principles derived from it. Due to space limitations, we do not discuss the didactic implications of the principles. This approach aligns with the article’s ambition to elucidate the principal differences between LOs and constructivism, rather than discussing the implementation of these perspectives. Our selection is guided by relevant literature and the need to provide a comprehensive yet focused foundation for comparison.

Connecting new and old knowledge

All learning is based on experience (Cunningham & Duffy, 1996). Teachers, therefore, should help learners activate and build on their prior knowledge (Limón, 2001) through the following interdependent principles.

Active learning

In contrast to traditional lectures during which students passively receive information from the teacher (Prince, 2004), learning should be perceived as an active process (Karagiorgi & Symeou, 2005; Mayer, 2002) both cognitively and in the promotion of student engagement (Papastergiou & Mastrogiannis, 2021; Prince, 2004).

Authentic learning

Authentic learning emphasises real-world tasks and problems, embedding learning in realistic and relevant contexts (Honebein, 1996; Karagiorgi & Symeou, 2005; Papastergiou & Mastrogiannis, 2021; Savery & Duffy, 1995), as elaborated in the next three principles.

Context dependency

Learning and context are perceived to be inseparable (Cunningham & Duffy, 1996; Greeno, 1997; Savery & Duffy, 1995; Sfard, 1998), as stressed by the participation metaphor.

Learning vs. complexity

The task or learning design should reflect the complexity of the environment students will face after education (Savery & Duffy, 1995). Constructivism thus often addresses complex, messy, ill-structured, open-ended problems (Barrows & Kelson, 1993; Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995).

Multiple perspectives

There are numerous ways to think about and solve problems and construct knowledge. Authentic learning preserves and facilitates this openness and encourages students to appreciate, use and assess different perspectives (Cunningham & Duffy, 1996; Honebein, 1996; Karagiorgi & Symeou, 2005; Savery & Duffy, 1995).

Ownership and students’ voice

Supporting students’ voice and ownership of the overall problem or task and the problem-solving process (Savery & Duffy, 1995) can increase their sense of responsibility (Hmelo-Silver & Barrows, 2006; Savery, 2006) and stimulate self-regulated learning (Schraw, Crippen, & Hartley, 2006; Zimmerman, 1990).

Learner agency and epistemic agency

Learner agency refers to students’ capacity to make intentional choices and take responsibility for their learning within supportive sociocultural contexts (Greeno, 1997). Epistemic agency further highlights learners’ control over knowledge-building processes—such as setting goals, choosing strategies, and evaluating progress—often in collaboration with others (Scardamalia & Bereiter, 2006). Constructivist approaches value such agency for fostering ownership, metacognition, and co-construction of meaning.

Externalise and articulate

Learning improves when learners externalise and articulate their developing knowledge (Bransford, Brown, & Cocking, 2000; Sawyer, 2005). Major concerns are the forms of articulation that are most beneficial to learning and how one can support students’ on-going articulation processes (Sawyer, 2005).

Reflection or metacognition

Articulation should lead to reflection or metacognition, which promotes deeper understanding and awareness of one’s knowledge, thinking and learning strategies (Savery & Duffy, 1995; Sawyer, 2005).

Bridging concrete experience and abstract general knowledge

Teaching should support the dialectic process of connecting and bridging students’ everyday experience and schools’ scientific knowledge (Sawyer, 2005; Vygotsky, 1986; cf. Figure 1).

Fig. 1

Bridging concrete experience and abstract general knowledge

Collaborative learning

Social constructivism views learning as a social, collaborative process (Cunningham & Duffy, 1996; Honebein, 1996; Karagiorgi & Symeou, 2005; Vygotsky, 1978). Collaborative learning is important as it encourages learners to articulate, reflect on and justify their thoughts, to compare and understand multiple perspectives on issues, to resolve differences through negotiation and constructive dialogue, and to build shared knowledge and meanings (Karagiorgi & Symeou, 2005; Luckin & Holmes, 2016).

Support and challenge learners’ thinking

The purpose of the justification and negotiation processes promoted by principles such as multiple perspectives and collaborative learning is to support and challenge learners’ thinking (Cunningham & Duffy, 1996; Savery & Duffy, 1995). Support of learners’ thinking is based on the ideas of scaffolding (Van de Pol, Volman, & Beishuizen, 2010) and the zone of proximal development (Vygotsky, 1978). To challenge students’ thinking one can impose cognitive conflicts through the presentation of anomalous data and contradictory information and opinions (Limón, 2001).

2.2 Learning metaphors

This discussion focuses on four general learning metaphors that are intended to capture the essence of learning and reflect key changes in learning theory over the past hundred years: (1) behaviourism (particularly in 1910–1960 s); (2) the cognitive revolution (began in the 1950 s); (3) social learning theories that emphasise adaption to social practices (began in Western context in 1970 s, but originated in Vygotsky’s work from the 1930 s); and (4) social learning theories that emphasise disruption, transformation and knowledge advancement (began in the mid 1980 s). The connections between the learning theories and these four metaphors are shown in Fig. 2.

Fig. 2

Connections between learning theories and learning metaphors

Whereas the conduit metaphor does not acknowledge that learning involves an active process of knowledge construction, the three other metaphors are all based on constructivism as a fundamental premise for how people learn (Sfard, 1998). However, in an important difference, the acquisition metaphor builds on cognitive constructivism (Piaget & Inhelder, 1969), whereas the participation and the knowledge creation metaphors are based on social constructivism (Vygotsky, 1978). Cognitive constructivism thus perceives knowledge as individually formed and residing in the mind, whereas the participation and the knowledge creation metaphors view knowledge as more socially derived and residing in the environment (Alexander, 2007). The metaphors and their strengths and weaknesses are briefly described.

2.2.1 Conduit metaphor

Linguist Reddy (1979) conceived of the conduit metaphor and brought attention to the extensive use of metaphorical language in everyday speech. Reddy (1979) initiated a broader, systematic metaphorical mapping of the human conceptual system (Sfard, 1998). A widely held belief deeply rooted in everyday language is that knowledge is (like) an object that can be managed (e.g. encoded, stored, retrieved, transmitted and distributed) in a thing-like manner (Reddy, 1979). This metaphor is based on three analogies: ideas, meanings and knowledge are like objects, linguistic expressions are like containers, and communication and learning are like sending an object from A to B through a container (Lakoff & Johnson, 2003).

The conduit metaphor, also referred to as the ontological metaphor (Lakoff & Johnson, 2003) and the metaphor of objects (Sfard, 2012), has strong appeal because portraying intangibles (e.g. learning) as objects expands opportunities to talk and reason about them:

Understanding our experiences in terms of objects and substances allows us to pick out parts of our experience and treat them as discrete entities or substances of a uniform kind. Once we can identify our experiences as entities or substances, we can refer to them, categorise them, group them, and quantify them—and, by this means, reason about them. (Lakoff & Johnson, 2003, p. 26)

However, the conduit metaphor also has severe drawbacks. Among others, it trivialises the efforts required to achieve learning, particularly by the learner (whom Reddy (1979) refers to as the listener). To the extent that the metaphor attends to this question, it acknowledges only the efforts of the teacher (whom Reddy (1979) calls the speaker).

2.2.2 Acquisition metaphor

Research drawing on this metaphor challenges the focus of behaviourism by highlighting the roles of language, memory and the mind and by using the computer (rather than animal studies) as an analogy to explore thinking and learning. The acquisition metaphor was inspired by information processing theories and computational theories of the mind, comparing the mind to computers (Carello, Turvey, Kugler, & Shaw, 1984).

The main concern of this metaphor is how to get something into and out of the head (Sfard, 1998). How is external information transformed and processed for the learner to acquire knowledge, meaning, understanding and learning? How are recall and activation of knowledge affected by different information processing strategies? The acquisition metaphor addresses how information is encoded, stored and retrieved. It has been used to consider issues such as the functions of different memory structures, bottlenecks and barriers within and between memory structures, and strategies to improve the functioning of memory structures (Anderson, 2005).

The acquisition metaphor originally paralleled the conduit metaphor by viewing learning as an individual property amassed in the mind, perceived as a container, but has since evolved. It now accentuates active engagement and meaning-making, transitioning towards a constructivist approach that underscores the learner’s active role in knowledge construction (Sfard, 1998).

2.2.3 Participation metaphor

We can distinguish between weak and strong versions of the participation metaphor (Cole, 1996). The weak version views learning dependent on the context, which is believed to be stable and independent of the learner. The context and its impacts on learning, therefore, can be adequately accounted for by a set of variables. The strong version instead claims that the context and the learner or learning practice are interdependent, inseparable and co-constitutive (Cole, 1996). The context cannot be reduced to variables as it is not a stable entity but an infinite set of potential resources that may or may not be activated during the learning process. The strong version seeks to understand when, how and why certain contextual resources (fail to) become activated and what the impacts on learning are.

Rather than a property (knowledge, know-what and knowledge about) in the head (the acquisition metaphor), the participation metaphor perceives learning as an aspect of doing (knowing, know-how and knowledge of) and participation in social practices (Scardamalia & Bereiter, 2006; Sfard, 1998). Knowing and know-how are inscribed in cultural discourses, artefacts and social practices and are distributed among all the members of communities of practice.

Learning is seen as the process of gradually improving one’s ability to participate as a fully competent member in different social practices (Lave & Wenger, 1991). Learning and participation are mediated by both material (physical and technical tools) and intellectual tools (language, signs and symbols; (Säljö, 2002; Wertsch, 1994). The unit of analysis thus is the isolated individual but the ‘individual(s)-acting-with-mediational-means’ (Wertsch, 1991, p. 12).

2.2.4 Knowledge creation metaphor

Neither the acquisition metaphor nor the participation metaphor sufficiently addresses the creation of new knowledge. The acquisition metaphor focuses on existing knowledge (the knowledge creation metaphor focuses on the generation of new knowledge), and the participation metaphor focuses on adapting to existing practices (the knowledge creation metaphor is concerned with learning processes that transcend existing practices). The culture and practices addressed by the participation metaphor appear to be relatively stable: no substantial and deliberate changes or cultural transformations are involved in the learning process, and the focus is typically the intergenerational transmission of knowledge or, more generally, the transmission of knowledge between old-timers or experts and newcomers. This context is quite different from modern knowledge societies characterised by swift changes and an absence of clear-cut distinctions between old-timers and newcomers; both groups are newcomers as they are lifelong learners who must continuously surpass their earlier knowledge and achievements.

The knowledge creation metaphor has also been called the knowledge advancement metaphor (Paavola et al., 2002) and the artefact creation metaphor of learning (Paavola & Hakkarainen, 2005). Knowledge creation indicates the central role of collaborative knowledge creation. Knowledge advancement emphasises that 21 st -century learning often concerns the advancement of new knowledge rather than the reproduction of pre-existing knowledge. Artefact creation stresses that the artefacts created by students mediate knowledge advancement. An artefact becomes a “partner” in the learning discourse that “acts back” and stimulates collaborative reflections that parallel the progressive steps of their development, as illustrated by Paavola and Hakkarainen (2005) neologism trialogical learning (see Fig. 3).

Fig. 3

Three metaphors of learning (Paavola & Hakkarainen, 2005, p. 539)

The discussion in Sect. 6.1 further compares the four metaphors presented above in Sect. 2.2 and summarised in Fig. 8.

The subsequent section of the theory and literature review examines Learning Objects (LO).

3 Learning objects

In this section, we investigate the metaphors, contexts, processes, and principles associated with traditional LOs. This lays the foundation for the later discussion on the discrepancies between traditional LOs and learning metaphors based on constructivism, and the potential to bridge the gap between them. The early 2000 s were the formative years for traditional LO research – the foundational perspectives and concepts underpinning much of the subsequent debate stem from this period.

3.1 Learning object metaphors

Metaphors play important roles in conceptualising and communicating the core ideas of LOs. The term LOs itself is a metaphor based on an analogy to objects in object-oriented programming (Friesen, 2004; Parrish, 2004). The main disadvantage of this metaphor is that it does not communicate well outside the developer community. The following metaphors have broader appeal.

The LEGO metaphor has been especially influential but has also faced criticism (Hodgins, 2002; Parrish, 2004; Wiley, Recker, & Gibbons, 2002; Wiley, 2003), resulting in a search for more powerful analogies, as discussed later. The LEGO metaphor assumes that anyone can combine LOs to meet their specific needs and that LOs are just the right size and have standard ‘connectors’ (Parrish, 2004, p. 61). Simplicity is the strength of the LEGO metaphor. Its weakness is its many false assumptions, such as that any LO is combinable with any other LO, LOs can be assembled in any manner one chooses, and LOs are so fun and simple that even children can assemble them (Wiley, 2003).

Construction work in the building industry

Hodgins (2002) compared LOs to the modern building industry’s extensive use of premanufactured artefacts and highly component-based approaches: ‘On average, 85 to 95% of the total amount of materials in every building built in the past ten years … are pre-built components. … This means that almost all of the material in any building is premanufactured and sitting in a warehouse awaiting delivery BEFORE the building is conceptualized, designed, or built’ (p. 3). Hodgins (2002) envisioned a similar revolutionary development in the LOs industry.

Atomic and molecular structure

Similar to LEGOs, atoms are small things that can be combined and recombined to form larger things. However, unlike LEGOs, ‘not every atom is combinable with every other atom. Atoms can only be assembled in certain structures prescribed by their own internal structure. Some training is required in order to assemble atoms’ (Wiley, 2003, p. 17).

Film montage

Parrish (2004) argued that we need a metaphor drawn from human communication and proposed film montages (a sequence of images in motion pictures). Parrish (2004) asserted that like instructions, the components of films can be combined in any number of ways that have unique effects on viewers, and the film montage metaphor acknowledges ‘cultural norms for how to tell a story’ (p. 62). This metaphor has the advantage of taking into account other human dimensions related to education training (e.g. the economy, effort, expertise and creativity; Parrish, 2004).

3.2 Learning object principles

Common and partly overlapping design principles associated with LOs include the following.

Reusability

Perhaps most distinctive feature of LOs is the aspiration to make them reusable learning resources. All other LOs principles are related to reusability.

Learning object economy

A goal of LOs is to promote a LOs economy (Campbell, 2003; Collis & Strijker, 2001) based on possibilities to sell content objects and reduce development costs through reuse and sharing rather developing products from scratch (Longmire, 2000).

Scalability/accessibility

Closely associated with the LOs economy is the principle of scalability or accessibility, which concerns the possibility of adapting LOs for larger audiences without a proportional cost increase (Longmire, 2000; Parrish, 2004).

Metadata

Accessibility to LOs is often considered in the context of learning object repositories based on LOs metadata standards (Neven & Duval, 2002). Metadata facilitate database storage and referencing (Polsani, 2006) and rapid updating, searching and content management by filtering and selecting only content deemed to be relevant (Longmire, 2000).

Adaptivity

Adaptivity for LOs can refer to either the individual learner, multiple contexts, or to both (Longmire, 2000; Parrish, 2004). When referring to individual learners, adaptivity targets individual competency gaps determined by the match between a core competency model and the learner’s competency model (Longmire, 2000). When referring to multiple contexts, adaptivity stresses the need to design flexible resources for use across contexts rather than only one context (Longmire, 2000).

Interoperability/portability

LOs need to be independent of the delivery media and knowledge management systems (Polsani, 2006) to operate across hardware and software systems (Rehak & Mason, 2003), including content providers and learning management systems.

Durable

Hardware and software upgrades should not require modification of LOs (Campbell, 2003; Rehak & Mason, 2003).

Single purpose/objective

Ideally, an LO should do one and one thing only; that is, it should satisfy a single learning objective (Boyle, 2003; Hodgins, 2002; Longmire, 2000). Granularity has the same premise but also considers how a single learning objective affects the size of a LO.

Granularity

Refers to the degree to which learning content is broken down into small, self-contained units. In the context of learning objects, granularity concerns the size and scope of an LO, which directly affects its reusability: the finer or smaller the granularity, the more easily a LO can be adapted across different contexts (Hodgins, 2002). Ideally, a LO should address only one (main) idea or concept (Polsani, 2006).

Minimised coupling/nonsequentiality

Reusability is affected not only by the size of LOs (granularity) but also by the degree of dependencies between them. Ideally, LOs should be self-contained, free-standing and not reliant on other LOs to be understood and applied (Longmire, 2000; Polsani, 2006).

Generativity

Whereas granularity addresses the decomposition of complex instruction into smaller, self-contained modules (i.e., LOs), generativity refers to the recomposition of such modules into new instructional sequences (e.g., lessons or courses) to meet pedagogical goals (Gibbons, Nelson, & Richards, 2000). This traditional view of generativity aligns with a linear instructional design model. However, as Hodgins (2002) and later constructivist and AI-informed research suggest, the concept has gradually expanded to include more dynamic, learner-driven forms of creation, remixing, and reorganisation—leading to more adaptive and context-sensitive learning experiences. These developments reflect two interrelated shifts: from purely technical flexibility toward learner agency, and from standardised sequencing of instructional content toward the co-construction of situated learning experiences. Later sections (particularly 6.3) revisit how different research traditions—symbolic, generative, hybrid, and constructivist-oriented—support, modify, or challenge the LO principles.

Modularity/customisability

A final principle implied in several preceding principles is that learning resources should be easy to customise. LOs’ modularity helps realise this goal (Longmire, 2000).

3.3 Learning objects: contexts and theoretical limitations

Learning Object (LO) research typically differentiates between three key contexts (Weitl, Kammerl, & Göstl, 2004):

• 1. The original author, who creates and structures the LO.

• 2. The teacher, who reuses and adapts the LO in instructional settings.

• 3. The student, who interacts with the LO as part of their learning process.

Most LO research has predominantly focused on contexts 1 and 2, addressing issues of content creation, modularity, and standardisation. In contrast, context 3 (the student perspective) has received limited attention, despite its central role in constructivist learning theories.

3.3.1 LO metaphors and their bias toward authors and teachers

As discussed in Sect. 3.1, several metaphors have shaped the conceptualisation of LOs, emphasizing efficiency, reusability, and scalability. These metaphors—including Object-Oriented Programming (OOP), LEGO blocks, and construction analogies—strongly align with the needs of authors (context 1) and teachers (context 2). However, they fail to account for the situated, experiential, and interactive aspects of learning central to students (context 3).

The dominant focus on modularity and standardisation assumes that learning can be broken into discrete, combinable components, but this perspective does not fully reflect how learners actively construct knowledge in specific contexts. This misalignment between traditional LO principles and constructivist learning theories is a key limitation in current LO frameworks.

3.3.2 Theoretical limitations and emerging directions in LO research

Early LO research has been criticised for weak integration with learning theory. Frameworks often referenced pedagogical ideas only implicitly, leading to a dominance of standardisation and reuse over learner-centred approaches. For instance, while Collis and Strijker’s (2004) model includes multiple contexts, it largely overlooks student learning—the core of constructivism. Sampson and Papanikou’s (2009) extension offers little improvement in this regard.

LO research has struggled to integrate modern constructivist paradigms, reinforcing structured instructional models over learner-driven exploration. This challenge is particularly relevant in light of technological advances, which have reshaped both LO research and its theoretical underpinnings. Given these limitations, two major strands of research have sought to address the tensions between LO principles and learning theories:

• 1. AI-driven approaches (symbolic, generative, and hybrid AI) – Advances in artificial intelligence have introduced new adaptive mechanisms, automation, and personalisation strategies into LO frameworks, influencing their pedagogical and theoretical foundations (Sect. 3.4).

• 2. Constructivist-oriented LO research – Later constructivist LO research has explored ways to align LO with learner-centred, situated learning, emphasising flexibility, learner control, and contextual adaptation (Sect. 3.5).

The following sections examine these responses to the theoretical tensions in LO research.

3.4 LO principles in light of AI and ICT developments

AI-driven approaches—including symbolic AI, generative AI, and hybrid AI—have introduced new adaptive mechanisms, automation, and personalization strategies into LO frameworks. These developments have the potential to redefine LO principles, impacting their design, use, and alignment with learning theories.

3.4.1 Symbolic AI in adaptive learning systems

Adaptive learning systems (ALS) and intelligent tutoring systems (ITS) personalise instruction by analysing learner behaviour and adjusting content delivery in real time. These systems typically rely on symbolic AI, which uses rule-based models, domain ontologies, and expert systems to track learner progress and adapt instructional strategies accordingly (Graesser, Chipman, Haynes, & Olney, 2005; Ritter et al., 2019). Adaptive hypermedia systems (AHS) further extend this approach by enabling modular sequencing and branching logic within LO frameworks (Apoki, Al-Chalabi, & Crisan, 2019).

Symbolic AI aligns well with traditional LO principles, emphasising structured feedback, modularity, and measurable learning outcomes. Systems like AutoTutor (Graesser et al., 2005) and ACT-R-based tutors (Ritter et al., 2019) exemplify this alignment in precision-oriented domains such as medicine, aviation, and compliance training. However, the reliance on predefined learning pathways (Denvir & Brown, 1986) can limit adaptability in open-ended, creative, or interdisciplinary contexts, where constructivist learning paradigms are more appropriate.

Magnisalis, Demetriadis, and Karakostas (2011) provide a broad review of adaptive educational technologies built on symbolic AI. Their analysis highlights the effectiveness of rule-based systems in supporting learner modelling, scaffolding, and instructional efficiency—further illustrating symbolic AI’s pedagogical and practical potential in technology-enhanced education.

A recent review by Sharma, Nguyen, and Hong (2024) synthesises empirical studies on symbolic AI–based learning environments and their role in supporting both self-regulated and socially shared regulation of learning. These systems typically rely on explicit feedback, performance tracking, and structured scaffolding to facilitate collaborative knowledge construction, particularly in contexts where learners jointly plan, monitor, and evaluate their progress.

While symbolic AI does not generate new content, it enables reliable and context-sensitive adaptation through structured metadata and semantic search. Interoperable LO repositories play a key role by enabling personalised selection of existing, quality-assured resources, based on learner profiles, prior knowledge, and instructional goals (Gudoniene, Staneviciene, & Motiejunas, 2022). These mechanisms contrast with generative AI by prioritising stability, traceability, and instructional consistency.

Taken together, these affordances reinforce symbolic AI’s continued relevance for LO models that prioritise individual progression, formative assessment, and pedagogical transparency—particularly in domains where structure and control are valued over open-ended exploration.

3.4.2 Generative AI in educational contexts

The rise of generative AI introduces new possibilities for dynamic and flexible learning experiences. Unlike symbolic AI, which applies rule-based logic, generative AI creates new content based on probabilistic models trained on vast corpora of text, code, and other modalities—commonly referred to as large language models (LLMs) (Banh & Strobel, 2023). These models can support problem-solving, discussion-based learning, and automated feedback generation (Thüs et al., 2024).

Recent research also highlights how generative AI can support dialogic and collaborative learning. Lee, Tan, and Teo (2023) demonstrate how GPT can scaffold idea development and stimulate diverse contributions within inquiry-based group tasks. Sharples (2023) proposes that generative AI be understood not just as a tool, but as a social actor in educational settings—capable of participating in collaborative learning through roles such as “Socratic opponent” or “collaboration coach.” Building on this, Cress and Kimmerle (2023) argue that generative AI can contribute to co-constructive knowledge practices, provided users engage in reflective prompting and iterative dialogue (i.e., inputting strategically phrased queries to guide generative AI outputs). While these systems lack semantic understanding, they may still support collaborative knowledge construction in well-orchestrated settings.

Generative AI also enhances personalisation by enabling the creation of tailored instructional materials. Brehmer and Buonassisi (2024) describe how AI-assisted LO design allows teachers to generate adaptable content more efficiently. Mohamedhen, Alfazi, Arfaoui, Ejbali, and Nanne (2024) show how multi-agent systems combine deep learning and learner profiling to personalise LO selection—illustrating another application of generative AI for adaptive learning.

Despite these affordances, generative AI also presents notable risks. Its tendency to hallucinate or fabricate information makes it unreliable in disciplines requiring strict factual accuracy (Modran, Bogdan, Ursuțiu, Samoila, & Modran, 2024). This suggests that generative AI may be more suitable for exploratory learning and creative disciplines, while symbolic AI remains essential in structured, high-stakes contexts (Halkiopoulos & Gkintoni, 2024).

These developments indicate that generative AI may reshape how LO principles are interpreted and applied in learning environments that value open-ended inquiry, collaboration, and adaptive learner engagement.

3.4.3 Hybrid AI and the role of retrieval-augmented generation (RAG)

To mitigate the weaknesses of generative AI while retaining its strengths, hybrid AI architectures have emerged. One prominent model is retrieval-augmented generation (RAG), which combines symbolic AI’s rule-based reasoning with generative AI’s language flexibility (Gao et al., 2023; Lewis et al., 2020).

In educational contexts, RAG-powered systems retrieve curated, authoritative content before generating responses, reducing hallucination risks while preserving contextual adaptability (Modran et al., 2024; Thüs et al., 2024). This allows AI systems to deliver more accurate, traceable, and pedagogically appropriate outputs—especially in high-stakes or knowledge-intensive learning environments.

Some, Yang, Bain, and Kang (2025) provide a comprehensive review of techniques for integrating large language models with structured knowledge systems, such as knowledge graphs and RAG. They show how hybrid AI approaches enhance factual grounding, context awareness, and pedagogical robustness—making them well suited for learning scenarios that demand both precision and flexibility.

These developments suggest that hybrid AI may enable new configurations of LO design, where structured and open-ended learning coexist within a unified system. This opens promising pathways for applying LO principles across diverse pedagogical models.

3.5 Constructivist-oriented LO research

While Sect. 3.4 examined AI-driven LO research, this section focuses on constructivist-oriented LO research and its role in supporting student-centred, inquiry-based learning. Traditional LO research emphasised modularity, reusability, and interoperability, whereas later constructivist-oriented LO studies have explored situated learning, student agency, and collaboration.

In this paper, we use the term constructivist-oriented LO research to refer to studies that integrate learning objects into student-centred, inquiry-driven, or situated learning environments. These approaches often emphasise learner agency, contextual adaptation, and interactive knowledge construction. While many such models reinterpret or adapt traditional LO principles to better support constructivist pedagogy, others incorporate them without explicitly challenging their underlying assumptions. This distinction underpins the analytical structure of the sections that follow.

3.5.1 Constructivist-oriented LO models within structured frameworks

Some LO models incorporate constructivist elements while retaining structured sequencing, metadata tagging, and predefined learning paths. These approaches support learner-driven adaptation while maintaining instructional structure.

Papastergiou and Mastrogiannis (2021) show how interactive LOs enhance engagement in physical education, demonstrating that modularity and flexibility can coexist. Similarly, Gaviria-Chavarro, Rojas-Padilla, and Vergara-López (2023) explore virtual LOs for teaching statistical methods, integrating adaptive features that encourage exploration.

While incorporating constructivist elements, these models still rely on predefined configurations for coherence. Unlike AI-driven LO models such as those by Mohamedhen et al. (2024) and Modran et al. (2024), which personalise LO selection through machine learning, these studies focus on pedagogical integration rather than automation.

3.5.2 Open-ended constructivist LO models

Other studies prioritise contextual adaptation and student-driven learning over structured LO frameworks. Ilomäki et al. (2006) and Vanoostveen et al. (2019) explore LOs that support inquiry-based and problem-oriented learning, shifting the focus from standardisation to student agency. Topali and Mikropoulos (2023) examine Scratch-based LOs that promote experimentation and iterative problem-solving in programming. Domínguez Romero and Bobkina (2021) investigate multimodal LOs in flipped classrooms, enhancing interactive, student-led learning.

These approaches emphasise flexibility and learner autonomy, adapting LOs to specific contexts rather than adhering to predefined structures. While some AI-driven models, such as those by Mohamedhen et al. (2024), Apoki et al. (2019), Gudoniene et al. (2022) and Modran et al. (2024), incorporate personalisation, they do not fully align with constructivist LO models that prioritise learner agency over algorithmic sequencing.

Taken together, recent research on AI-based and constructivist-oriented LO models suggests a broad spectrum of alignment with LO principles informed by constructivist theory. While symbolic AI often supports structured progression and feedback, generative AI emphasizes exploration and dialogue, and hybrid AI combines these elements. In contrast, constructivist-oriented LO models place learner agency and contextual adaptation at the centre, often outside algorithmic control. Across both strands, varying degrees of alignment, modification, or tension with key learning metaphors and LO principles are evident—underscoring the importance of pedagogically informed integration.

3.6 Theoretical tensions and transition to analysis

Sections 2 and 3 examined how LO research engages with learning theory—particularly constructivism—and identified key tensions between standardisation and learner-centred pedagogy. Building on this, Sect. 6 will analyse how the four recent research directions introduced in Sect. 3.4 and 3.5—symbolic AI, generative AI, hybrid AI, and constructivist-oriented LO approaches—may reinforce, mitigate, or reframe this longstanding tension.

4 Method

This section presents our research design and analytical framework.

4.1 Research design

Our research design adopts an abductive approach (Reichertz, 2014), oscillating between theoretical exploration and empirical investigation to iteratively refine insights (see Fig. 4). This approach integrates both top-down (theoretical) and bottom-up (empirical) methodologies, combining a literature review with qualitative analysis to address the broad and interdisciplinary nature of our research topic.

Fig. 4

Research design

The study operates at the intersection of educational technology and constructivist pedagogy, with a specific focus on Learning Objects (LOs). While many studies examine the opportunities and challenges of educational technology in general, this research also considers the technical requirements and constraints specific to LOs. By engaging with both pedagogical and technical dimensions, the study offers a more holistic understanding of how digital learning tools align with and challenge teaching practices and educational theory.

The abductive process illustrated in Fig. 4 highlights iterative movement across the study’s research questions. RQ1 provides a foundational understanding of the perspectives and principles underpinning both LO design and constructivist theories. These inform the conceptual mapping of tensions in RQ2. Insights from RQ1 also contribute to RQ3a, which explores teachers’ conceptions of teaching and learning. In turn, RQ3a feeds into RQ3b, which examines how these conceptions differ across contexts. Through this iterative process, each research question builds on and informs the others, facilitating a comprehensive and nuanced understanding of the phenomenon under study.

This abductive framework allows us to move reflexively between theoretical constructs and empirical findings, ensuring that the research remains both conceptually grounded and sensitive to the complexities of real-world practice. The two core components of the design are presented in the following sections: the literature review and theoretical analysis (Sect. 4.2), and the empirical analysis (Sect. 4.3).

4.2 Literature review and theoretical analysis

We developed a conceptual framework by combining key principles and metaphors from traditional LO design (identified in Sect. 3.1 and 3.2) with constructivist learning metaphors (analysed in Sect. 2.1 and 2.2). This framework, summarised in Fig. 8, enables a structured comparison of how different research directions position themselves in relation to these two pedagogical dimensions. Our framework encompasses both the simple metaphors by which people live (Lakoff & Johnson, 2003; Reddy, 1979) and the more mature, theory-based metaphors that guide researchers and educational practitioners in facilitating learning (Paavola & Hakkarainen, 2005; Paavola et al., 2002; Sfard, 1998). The breadth of the framework permits nuanced analyses of gaps not only between different theoretical orientations to learning but also between intuitive everyday conceptions and formalised educational theory.

This study provides a structured review of key developments in LO research and its intersections with learning theories and AI-based approaches. The review is comprehensive for early LO research and for core concepts in constructivist learning theory, as these form the basis for our theoretical framework and are used to expose tensions between pedagogical principles and LO design assumptions. However, the discussion of AI-driven LO research (symbolic, generative, and hybrid AI) and more recent constructivist-oriented LO research is selective, focusing on key theoretical developments and representative studies rather than aiming for exhaustive coverage. This ensures a balance between historical depth and emerging trends, providing a structured yet dynamic overview of tensions between LO principles and evolving pedagogical paradigms.

4.3 Empirical analysis

The empirical part of the study investigates how teachers conceptualise knowledge, teaching, and learning in relation to Learning Objects (LOs) and constructivist pedagogy. The goal is to explore whether and how teachers’ conceptions resonate with or challenge the tensions between traditional LO principles and constructivist ideals as identified in our theoretical framework.

4.3.1 Research setting and participants

Data were collected through a semi-structured focus group interview with eleven teachers at a large, urban upper secondary school in the capital of Norway, known for its diverse student population. The school’s head of professional development assisted with recruitment to ensure diversity in gender, age, teaching experience, and disciplinary backgrounds. The final group consisted of three women and eight men, aged from their late twenties to early sixties, with teaching experience ranging from 1.5 to 37.5 years. Participants represented a wide range of subjects, including Norwegian, English, History, Social Studies, Psychology, Religion, Mathematics, Physics, Natural Sciences, and Physical Education.

4.3.2 Rationale for method and data collection

The focus group method was chosen because it supports dynamic interaction and co-construction of meaning among participants—well aligned with our interest in uncovering implicit pedagogical assumptions. Group dynamics allow for spontaneous elaboration, clarification, and contestation of perspectives, which may not emerge in individual interviews. The session lasted 90 min, was video recorded, and later transcribed for analysis.

To minimise priming effects and allow participants to express their existing perspectives freely, the interview was structured to move from practice-oriented to more abstract and pedagogically charged themes. This gradual shift was supported by five thematic prompt groups (i.e., sets of open-ended, discussion-oriented questions designed to frame each topic), each introduced using brief screen-shared materials. In the final prompt group (Prompt E), selected screenshots were used to illustrate relevant features of the platform. While these prompts were not presented to participants as distinct categories, they later served as an analytical scaffold for interpreting the conversation. The five thematic areas were:

• Prompt A – initial questions on resource reuse.

• Prompt B – reuse constraints and trade-offs.

• Prompt C – constructivism and 21 st-century skills.

• Prompt D – contextual reuse and student perspectives.

• Prompt E – learning platform affordances for reuse (with references to itslearning).

This design allowed us to trace how participants’ conceptions of teaching and learning developed during the session—not as a result of directive questioning, but as a response to gradually shifting themes that encouraged deeper reflection. The full interview guide is included as Supplementary Information (SI), Appendix B.

4.3.3 Analytical approach

Data were analysed using a combination of interaction analysis (Jordan & Henderson, 1995) and thematic analysis (Braun & Clarke, 2006; Corbin & Strauss, 2008). Interaction analysis focused on turn-taking, argumentation patterns, and collaborative meaning-making. Thematic analysis was used to identify patterns in teachers’ utterances about knowledge, learning, and teaching.

The coding followed Corbin and Strauss (2008) multi-layered structure—open, axial, and selective coding—but was adapted to an abductive logic. In the first phase, open coding was used to categorise utterances according to descriptive themes. In vivo coding was employed to retain participants’ phrasing, particularly where metaphors were used. In the second phase, utterances were grouped into three interpretive levels—simple, intermediate, and advanced conceptions—based on their degree of pedagogical reflexivity and awareness. Finally, selective coding aligned these conceptions with the theoretical dimensions from our framework, integrating empirical patterns with conceptual categories.

This iterative, abductive process allowed us to explore how teachers’ reflections moved between surface-level practical concerns and more abstract pedagogical ideals—sometimes shifting during the session itself as theoretical constructs were gradually introduced by the interviewer.

To support transparency, we provide representative examples of how statements were categorised and interpreted. Coding was carried out iteratively by both authors, initially using descriptive and in vivo codes to capture patterns in how teachers talked about reuse, adaptation, and pedagogy. These codes were then clustered abductively into the three interpretive categories—simple, intermediate, and more mature understandings—based on their degree of pedagogical reflexivity and alignment with theoretical principles outlined in Sect. 2. The resulting categories are illustrated through representative excerpts in Figs. 5, 6 and 7, which show how empirical patterns informed the analytical structure presented in Sect. 5.

While grounded in participants’ language and experiences, the thematic analysis also incorporated theory-informed interpretation, allowing for consistency between the abductive coding process and the conceptual discussion in Sect. 6. In that section, the categories are re-engaged and further elaborated through theoretical lenses—including constructivist learning principles and metaphor theory—thus extending the empirical distinctions introduced in Sect. 5 into a broader conceptual framework.

4.3.4 Limitations

As with all qualitative case studies, this study has limitations in scope and generalisability. The focus group consisted of a small and context-specific sample from a single school, and the findings should therefore be interpreted with caution. While efforts were made to ensure diversity among participants, the cultural and institutional specificity of the Norwegian school context may limit transferability. Furthermore, the study does not aim to offer a comprehensive design framework for learner-centred LOs, but rather to expose conceptual tensions and identify conditions for pedagogical reflection and change. The study followed recognised ethical guidelines for research involving human participants. All participants gave informed consent, were made aware of their right to withdraw, and were assured anonymity in reporting.

5 Results – empirical analysis of teachers’ perspectives on teaching and learning

The analysis presented in this section identifies three levels of understanding of resource reuse in teaching practice: simple, intermediate, and mature. These levels emerged gradually over the course of the interview and can be directly traced back to the thematic framing provided by the interview prompts. Specifically, statements associated with simple understandings were primarily elicited in response to Prompt A; intermediate understandings emerged partly in Prompt A (lines 1 and 2) and more fully in Prompt B (lines 3 and 4); while mature understandings arose in response to Prompt C (lines 1–10) and Prompt D (lines 11–21). See Sect. 4.3 and Appendix B for a full overview of the prompt structure.

How do teachers’ conceptions of teaching and learning correspond with mature understandings of learning according to constructivist learning theories? (RQ3a)

Teachers described a range of perspectives on learning and reuse, ranging from simplistic views (Sect. 5.1) to more nuanced, theory-informed conceptions (Sect. 5.3). In between, we also find an intermediate level (Sect. 5.2).

5.1 Simple understandings

The first group of responses, categorised as “simple understandings,” emerged during the early part of the session, in response to Prompt A, which included general questions about whether and how participants reused teaching resources. These statements typically described reuse in pragmatic and procedural terms—such as copying, borrowing, or saving time—without reference to underlying pedagogical goals or student learning outcomes.

We refer to this category as simple because it reflects a surface-level understanding of reuse, often conceptualising knowledge, teaching, and learning through object metaphors (e.g. knowledge as content, learning as transfer, and teaching materials as containers or tools). This framing is evident in the use of language that treats resources as static objects to be reused with minimal modification (see the boldface phrases in Fig. 5). Compared to intermediate and mature understandings (see Figs. 6 and 7), simple understandings have limited capacity to stimulate critical or innovative pedagogical reflection.

The excerpts presented in Fig. 5 are representative of this category.

Fig. 5

Examples of simple understandings

These excerpts and the following examples are further discussed in Sect. 6.4.

5.2 Intermediate understandings

Intermediate understandings began to surface during the latter part of Prompt A (lines 1 and 2) and became more explicit in responses to Prompt B (lines 3 and 4), where questions addressed constraints, trade-offs, and the need for adaptation when reusing materials.

Positioned between the simple and mature categories, intermediate understandings reflect a partial shift toward pedagogical reflection. While they acknowledge that reuse is not always straightforward, and may require modification or contextual consideration, they offer only limited alignment with core principles of constructivist learning (see Fig. 6).

Fig. 6

Examples of intermediate understandings

Notably, these understandings tend to remain focused on the teacher’s work—such as design effort, lesson planning, and classroom management—rather than on student learning or agency. This tension is further discussed in Sect. 6.4.2.

The strength of the intermediate understandings is that they acknowledge the need to adapt learning resources to their particular contexts, topics and student groups, to which simple understandings pay little or no attention. However, in applying constructivist thinking, teachers with intermediate understandings tend to concentrate on their own efforts rather than students’.

5.3 More mature understandings

In the second half of the interview, participants began to articulate more pedagogically reflective and conceptually complex understandings of reuse. This shift occurred gradually, beginning in response to the questions in Prompt C, which invited reflections on constructivism and 21 st-century skills, and continuing into Prompt D, which addressed contextual and student-oriented perspectives on learning and reuse. Specifically, lines 1–10 in Fig. 7 were articulated during the segment prompted by the questions in Prompt C, while lines 11–21 arose during the segment associated with Prompt D.

These responses were marked by a clear change in tone and content. Participants used more evaluative and personally invested language, frequently contrasting the pedagogical implications of reuse with more traditional, transmission-oriented models of teaching—sometimes explicitly, but often implicitly, through qualifiers such as more independent, more meaningful, or at a higher level. Rather than simply reacting to prompts, teachers began to express their own convictions and teaching philosophies, often referencing their actual classroom practices to illustrate a more learner-centred view of teaching and knowledge construction.

The mature understandings presented above reflect this shift and are organised chronologically. Due to space limitations, the examples are summarised, with an emphasis on preserving the teachers’ original phrasing (see Fig. 7).

Fig. 7

Mature understandings

6 Discussion

RQ2 is discussed in Sect. 6.1; RQ3a in Sect. 6.4.1 to 6.4.3; and RQ3b in Sect. 6.4.4.

6.1 How do traditional LO principles and emerging LO research directions align or conflict with constructivist learning theories? (RQ2)

As shown in Sect. 3.3, LO process frameworks tend to focus on the logistical tasks of teachers—such as finding, modifying, aggregating, and managing resources—while giving limited attention to learners and learning processes. Although student interaction is occasionally acknowledged (e.g., Weitl et al., 2004), the teacher remains the primary actor. Our empirical findings (Sect. 5.1 and 5.2) confirm this emphasis, and the conceptualisation of principles (Sect. 3.2) reflects the same tendency. Pedagogical strategies for activating students’ own learning processes remain largely underdeveloped.

Considering the contexts, processes, and design principles addressed in traditional LO research, the common LOs metaphors provide limited pedagogical guidance. We argue that robust didactics for LOs can be better served by incorporating the learning metaphors presented in Sect. 2.2 and by building on principles associated with constructivist learning approaches (Sect. 2.1). Additionally, it is crucial to integrate traditional LO design principles (Sect. 3.2) that align with constructivist ideals, ensuring a comprehensive framework that leverages the strengths of both perspectives.

Our literature review shows that learning metaphors play key roles in identifying and analysing aspects of learning and can guide educational interventions. It, therefore, is surprising that the traditional LOs research literature has so few conceptual frameworks based on learning metaphors. Two exceptions are Griffiths and García (2003) and Collis and Strijker (2004), but they conducted their work more than 20 years ago. We agree with Griffiths and García (2003) that approaches to LOs generally have been based on the conduit metaphor. We also agree with Collis and Strijker (2004) that we need to bring LOs closer to the more mature learning understandings found in the acquisition and the participation metaphors. Collis and Strijker’s (2004) discussion is based on Sfard’s (1998) seminal distinction between acquisition and the participation metaphors. Sfard’s (1998) acquisition metaphor seems to encompass both the conduit and the acquisition metaphors, and her participation metaphor seems to encompass both the participation and the knowledge creation metaphors. However, the conduit and the knowledge creation metaphors are poorly described in Sfard’s (1998) analysis (the conduit metaphor is however addressed in (Sfard, 1994, 2012). We extend the analyses of Griffiths and García (2003) and Collis and Strijker (2004) by combining all three metaphors and adding the knowledge creation metaphor (see Fig. 8).

Fig. 8

Comparison of the four learning metaphors. [1] (Paavola & Hakkarainen, 2005) [2] (Sfard, 1998, p. 7) [3] (Reddy, 1979) [4] (Lakoff & Johnson, 2003, p. 10) [5] (Collis & Strijker, 2004, p. 7) [6] (Paavola et al., 2002, p. 1) [7] (Sfard, 1998) [8] (Engeström, 1987) [9] (Scardamalia & Bereiter, 2006) [10] (Sfard, 1998, p. 5) [11] (Collis & Strijker, 2004, p. 7) [12] (Paavola et al., 2002) [13] (Anderson, 2005) [14] (Piaget & Cook, 1952) [15] (Stahl, Koschmann, & Suthers, 2006) [16] (Lave & Wenger, 1991) [17] (Vygotsky, 1978) [18] (Damşa, Kirschner, Andriessen, Erkens, & Sins, 2010) [19] (Ludvigsen & Mørch, 2003)

Our framework addresses shortcomings of previous research. Collis and Strijker (2004) did not recognise that the participation metaphor provides insufficient guidance for the development of expansive learning. We address this by including the knowledge creation metaphor (Ilomäki et al., 2006; Paavola & Hakkarainen, 2005; Paavola et al., 2002). Neither Ilomäki et al. (2006) or Collis and Strijker (2004) included the conduit metaphor, while Griffiths and García (2003) included it but only focused on its weaknesses and metaphors’ explanatory and normative roles vis-à-vis learning theory. However, the conduit metaphor also plays descriptive, predictive and explanatory roles vis-à-vis everyday thinking (Lakoff & Johnson, 2003). It helps illustrate why and how everyday (mis)conceptions of knowledge, teaching, and learning diverge from learning theory and yet continue to profoundly influence our thinking, discourse, and actions (see Sect. 6.4).

In sum, we perceive the traditional view of LOs as closely associated with the conduit metaphor, whereas constructivism builds on three alternative learning metaphors (acquisition, participation, and knowledge creation). Recent research directions—including symbolic AI, generative AI, hybrid AI, and constructivist-oriented LO research—offer new ways of engaging with constructivist learning principles. The next two sections present complementary analyses: Sect. 6.2 examines how each research direction aligns with key constructivist learning metaphors—acquisition, participation, and knowledge creation—while Sect. 6.3 explores how the same directions relate to the traditional LO principles.

6.2 Emerging LO research directions and their alignment with learning metaphors

This section explores how symbolic AI, generative AI, hybrid AI, and constructivist-oriented LO research relate to constructivist learning metaphors (Paavola et al., 2002; Reddy, 1979; Sfard, 1998) and the framework in Fig. 8 (in Sect. 6.1). While the conduit metaphor traditionally shaped LO principles (cf. Section 3.2), the acquisition, participation, and knowledge creation metaphors offer alternative perspectives on how LO-based learning evolves within a constructivist paradigm.

Each research direction aligns differently with these metaphors, emphasizing certain aspects of constructivist learning while downplaying others. To reflect these variations, the analysis applies four alignment levels:

• ✓ Strong Alignment – Fully supports and operates within the metaphor.

• ↺ Moderate Alignment – Supports some aspects but diverges in others.

• ⚠ Weak Alignment – Has some relevant features but largely deviates.

• ✗ Minimal Alignment – Has virtually no connection to the metaphor.

Note on alignment levels

While the categories of Strong, Moderate, Weak, and Minimal Alignment describe the current positioning of each research direction, some emerging trends suggest potential shifts toward stronger alignment. For instance, recent developments in generative AI indicate increasing applicability to collaborative, knowledge-building contexts, while hybrid AI continues to refine its balance between structured and adaptive learning. Similarly, certain strands of constructivist LO research are progressively incorporating more explicit knowledge creation frameworks. Cases showing a potential shift toward stronger alignment are marked with (✓) in the analysis below.

6.2.1 Symbolic AI and learning metaphors

Symbolic AI relies on rule-based, structured learning models (Denvir & Brown, 1986), making it highly compatible with acquisition-based learning but less adaptable to participatory or knowledge-building approaches.

✓ Strong alignment with the acquisition metaphor

Symbolic AI reinforces structured knowledge acquisition by providing stepwise cognitive scaffolding (Brusilovsky & Peylo, 2003). ITS systems like AutoTutor (Graesser et al., 2005) and ACT-R cognitive models (Ritter et al., 2019) align with Piagetian assimilation and accommodation, helping learners build on prior knowledge systematically. The metadata-driven retrieval mechanisms of Apoki et al. (2019) and Gudoniene et al. (2022) further support structured content access, which mirrors schema development in acquisition-based learning.

⚠ Weak alignment with the participation metaphor

Although adaptive learning systems (ALS) personalize instruction, they still rely on predefined expert-driven learning pathways, limiting learner autonomy (Ritter et al., 2019). Minimal focus is placed on peer collaboration or co-construction of knowledge, as many adaptive hypermedia systems remain instructor-led and standardised (Brusilovsky & Peylo, 2003; Sharma et al., 2024). Even when user-driven modifications are permitted, the underlying learning structure is not inherently designed to facilitate collaborative participation, making it misaligned with the participation metaphor (Magnisalis et al., 2011).

✗ Minimal alignment with the knowledge creation metaphor

Symbolic AI primarily transmits predefined knowledge rather than fostering knowledge co-creation. Traditional rule-based AI systems structure learning around fixed pathways, limiting opportunities for students to actively generate new knowledge structures (Magnisalis et al., 2011). The focus remains on individualized content delivery rather than collaborative knowledge-building, reducing learner agency in the construction of new concepts (Sharma et al., 2024). While these characteristics have been persistent (Brusilovsky & Peylo, 2003), contemporary research confirms that Symbolic AI systems still struggle to support socially shared regulation of learning and co-constructed knowledge development in digital learning environments (Sharma et al., 2024).

6.2.2 Generative AI and learning metaphors

Generative AI enables dynamic, AI-generated content (Banh & Strobel, 2023), but its alignment with constructivist learning metaphors varies based on implementation.

⚠ Weak alignment with the acquisition metaphor

Unlike symbolic AI, generative AI does not follow rigid cognitive scaffolding principles, as it relies on probabilistic content generation rather than predefined instructional pathways (Banh & Strobel, 2023; Thüs et al., 2024). Learners are not guided through structured knowledge sequences, making generative AI less compatible with traditional acquisition models that rely on progressive cognitive development (Halkiopoulos & Gkintoni, 2024; Modran et al., 2024).

↺ (✓) Moderate alignment with the participation metaphor

Traditionally, generative AI has focused on individualised learning and personalisation, rather than collaborative knowledge-building. However, emerging research highlights its potential for AI-mediated collaborative learning environments (Sharples, 2023), where AI can facilitate peer interaction and structured student discussions. Generative AI can support student-led discussions and peer-assisted learning, resembling social constructivist pedagogies (Brehmer & Buonassisi, 2024). AI-enhanced student discourse (Lee et al., 2023) suggests that participatory models of generative AI are emerging, potentially expanding its alignment with the participation metaphor (✓).

↺ (✓) Moderate alignment with the knowledge creation metaphor

Generative AI enables adaptive content generation, allowing learners to build upon prior knowledge dynamically (Banh & Strobel, 2023). It can assist in knowledge-building practices, supporting iterative improvement and creative problem-solving (Thüs et al., 2024). However, generative AI remains AI-driven, often constraining student epistemic agency in the knowledge creation process. While generative AI introduces dynamic content creation and personalised learning paths, its primary applications have traditionally been individualised. However, recent studies suggest a growing potential for integrating generative AI into collaborative, knowledge-building communities (Cress & Kimmerle, 2023; Lee et al., 2023; Sharples, 2023) (✓).

6.2.3 Hybrid AI (RAG) and learning metaphors

Hybrid AI, particularly Retrieval-Augmented Generation (RAG), integrates symbolic and generative AI (Gao et al., 2023), balancing structured content retrieval with AI-driven adaptation.

↺ moderate alignment with the acquisition metaphor

Hybrid AI enhances structured knowledge retrieval while maintaining some adaptive capabilities (Lewis et al., 2020). Retrieval-Augmented Generation (RAG) combines rule-based content retrieval with generative flexibility, making it well-suited for adaptive acquisition models (Gao et al., 2023). RAG-powered systems retrieve curated, authoritative learning materials, ensuring accuracy and structured knowledge development while allowing for some adaptability (Modran et al., 2024).

↺ Moderate alignment with the participation metaphor

RAG-powered AI tutors can support collaborative inquiry by retrieving relevant knowledge resources for group discussions (Thüs et al., 2024). They enhance peer-driven learning by providing contextually relevant knowledge scaffolding, helping learners engage in guided exploration rather than passive content consumption. However, most implementations remain individually focused, as retrieval mechanisms primarily facilitate interaction with curated resources rather than direct peer collaboration. This means that participatory learning remains AI-mediated rather than socially constructed (Modran et al., 2024).

⚠ (✓) weak alignment with the knowledge creation metaphor

RAG supports iterative learning cycles by retrieving curated knowledge that informs ongoing student inquiry (Gao et al., 2023; Thüs et al., 2024). However, its role is more about knowledge refinement than true knowledge creation, as it retrieves rather than generates new epistemic artifacts (Lewis et al., 2020). The integration of LLMs with knowledge bases improves data contextualization and model accuracy but does not inherently transform these systems into platforms for original knowledge creation (Some et al., 2025). Since RAG relies on external databases, its ability to foster original knowledge innovation is limited, making it more aligned with structured knowledge retrieval than open-ended, social knowledge-building approaches (Modran et al., 2024). However, evolving AI retrieval systems suggest that RAG may increasingly support more advanced forms of knowledge-building (Some et al., 2025) (✓).

6.2.4 Constructivist-Oriented LO research and learning metaphors

Constructivist LO research spans all three metaphors, with a strong emphasis on participation and knowledge creation (Ilomäki et al., 2006).

↺ Moderate alignment with the acquisition metaphor

Constructivist LOs retain cognitive scaffolding mechanisms, aligning partially with Piagetian constructivism (Papastergiou & Mastrogiannis, 2021). However, they prioritise adaptive, contextualised learning over rigid knowledge structures (Domínguez Romero & Bobkina, 2021). Ilomäki et al. (2006) highlight that teachers frequently adjust LOs to align with local learning needs, reducing reliance on standardised instructional sequences. Gaviria-Chavarro et al. (2023) demonstrate that statistical LOs were most effective when adapted to students’ prior knowledge, favouring flexibility over rigid structure.

✓ Strong alignment with the participation metaphor

Studies emphasise collaborative learning and situated knowledge construction (Ilomäki et al., 2006). Socially constructed learning aligns closely with the participation metaphor, fostering peer interaction and shared knowledge-building (Vanoostveen et al., 2019). Constructivist LO models increasingly integrate networked, collaborative learning environments, enabling student-driven interaction (Papastergiou & Mastrogiannis, 2021).

↺ (✓) Moderate alignment with the knowledge creation metaphor

Some research supports trialogical knowledge practices, where LOs facilitate iterative idea improvement (Topali & Mikropoulos, 2023). However, not all constructivist LO research explicitly adopts knowledge creation frameworks. While many studies integrate situated and collaborative learning, some prioritise adaptability and engagement rather than new knowledge production (Papastergiou & Mastrogiannis, 2021). For example, Domínguez Romero and Bobkina (2021) emphasise structured learning objectives and modularity, supporting learner-centred adaptation but not necessarily fostering epistemic agency. Recent studies suggest a growing incorporation of explicit knowledge-building models, particularly through collaborative LO design (Vanoostveen et al., 2019) and adaptive LO refinement processes (Gaviria-Chavarro et al., 2023), indicating a possible trajectory toward stronger alignment (✓).

Final thoughts

This analysis clarifies that no single research direction fully embraces all constructivist metaphors. While constructivist-oriented LO research spans all three metaphors, AI-driven approaches have traditionally operated within structured acquisition models but are now increasingly exploring participatory and knowledge-building possibilities.

While the previous section focused on how emerging LO research directions align with constructivist learning metaphors, this alignment alone does not fully determine their implications for LO design. The four directions also engage with core LO principles in distinct ways, shaping how LOs are conceptualised and used in practice.

6.3 Comparing the research directions’ relationship to the LO principles

This section provides a structured synthesis of how symbolic AI, generative AI, hybrid AI (RAG-based), and constructivist-oriented LO research interact with the twelve traditional LO principles outlined in Sect. 3.2. Rather than restating the full analytical assessments provided in the supplementary analysis (available as Supplementary Information (SI), Appendix A), the focus here is on synthesising how each direction supports (✓), modifies (↺), or challenges (✗) the LO principles across different educational contexts and design paradigms.

Figure 9 provides a visual overview, categorising each direction’s relationship to the LO principles as follows:

• ✓ Supports LO principle – The research direction aligns with and reinforces the principle.

• ↺ Modifies LO principle – The research direction adapts or alters the principle while maintaining some foundational aspects.

• ✗ Challenges LO principle – The research direction fundamentally opposes or replaces the principle.

Fig. 9

Comparing research directions’ relationship to traditional LO principles

The following analysis expands on the patterns in Fig. 9, highlighting key differences across the four research directions in their relationship to LO principles.

6.3.1 Reusability

✓ Symbolic AI

Emphasises structured, rule-based content that enables consistent reuse across varied learning contexts (Brusilovsky & Peylo, 2003; Graesser et al., 2005).

✗ Generative AI

Shifts from reusable LOs to dynamic generation, reducing reliance on predefined content (Banh & Strobel, 2023).

↺ Hybrid AI

Retrieves and adapts modular LOs, preserving reuse while enabling contextual flexibility (Lewis et al., 2020; Thüs et al., 2024).

✗ Constructivist LO research

Emphasises contextual adaptation, often at the expense of general reusability (Ilomäki et al., 2006; Vanoostveen et al., 2019).

Comparative perspective

Symbolic AI aligns closely with traditional reuse ideals. Generative and constructivist approaches favour dynamic creation or contextual tailoring. Hybrid AI balances both by enabling reuse with adaptive flexibility.

6.3.2 Learning object economy

✓ Symbolic AI

Reduces per-learner cost by scaling structured LOs through standardised systems (Gudoniene et al., 2022; Ritter et al., 2019).

✗ Generative AI

Automates LO creation, lowering manual labour but increasing computational costs (Banh & Strobel, 2023).

↺ Hybrid AI

Enhances cost-efficiency by reusing existing resources via AI-assisted retrieval and adaptation (Gao et al., 2023; Modran et al., 2024).

✗ Constructivist LO research

Prioritises pedagogical fit and local adaptation over production efficiency (Vanoostveen et al., 2019).

Comparative perspective

Symbolic and hybrid AI support cost-efficient scaling. Generative AI reduces labour through automation but introduces computational costs. Constructivist approaches prioritise pedagogical relevance over efficiency.

6.3.3 Scalability & accessibility

✓ Symbolic AI

Ensures broad access through standardised metadata and instructional structures (Apoki et al., 2019; Graesser et al., 2005).

↺ Generative AI

Expands access by personalising content on demand, though risks consistency loss at scale (Halkiopoulos & Gkintoni, 2024).

✓ Hybrid AI

Supports scalable delivery while tailoring outputs via context-aware retrieval (Thüs et al., 2024).

✗ Constructivist LO research

Emphasises local adaptation over wide-scale deployment (Papastergiou & Mastrogiannis, 2021).

Comparative perspective

Symbolic and hybrid AI best support scalability and access. Generative AI increases personalisation but risks inconsistency. Constructivist approaches prioritise local relevance over broad deployment.

6.3.4 Metadata

✓ Symbolic AI

Relies on structured metadata to enable standardised retrieval and integration across platforms (Apoki et al., 2019; Gudoniene et al., 2022).

↺ Generative AI

Uses metadata selectively to guide content generation and improve relevance, but often bypasses rigid tagging (Brehmer & Buonassisi, 2024).

✓ Hybrid AI

Combines traditional metadata with AI-driven categorisation to enhance retrieval precision (Gao et al., 2023; Thüs et al., 2024).

✗ Constructivist LO research

De-emphasises metadata in favour of educator judgement and contextual tagging (Ilomäki et al., 2006; Domínguez Romero & Bobkina, 2021).

Comparative perspective

Symbolic and hybrid AI maintain strong metadata reliance, while generative AI blends metadata with contextual inference. Constructivist approaches resist rigid classification, favouring human-centred interpretation.

6.3.5 Adaptivity

✗ Symbolic AI

Uses rule-based adaptivity that follows predefined logic but lacks flexibility in handling learner diversity (Graesser et al., 2005; Ritter et al., 2019).

✓ Generative AI

Enables dynamic, real-time adaptation based on learner input and context (Halkiopoulos & Gkintoni, 2024; Mohamedhen et al., 2024).

↺ Hybrid AI

Integrates structured retrieval with AI-enhanced personalisation to support layered adaptivity (Modran et al., 2024; Thüs et al., 2024).

✓ Constructivist LO research

Fosters human-led, context-specific adaptivity driven by teacher and learner agency (Ilomäki et al., 2006; Gaviria-Chavarro et al., 2023).

Comparative perspective

Symbolic AI offers controlled but rigid adaptivity. Generative and hybrid AI enhance flexibility through automation, while constructivist approaches favour human adaptation in social contexts.

6.3.6 Interoperability & portability

✓ Symbolic AI

Prioritises platform-agnostic content through metadata and LO standardisation (Brusilovsky & Peylo, 2003; Gudoniene et al., 2022).

↺ Generative AI

Introduces flexible content creation, but lacks consistent standards for seamless portability (Modran et al., 2024).

✓ Hybrid AI

Maintains interoperability by combining structured LO retrieval with adaptive generation (Lewis et al., 2020; Gao et al., 2023).

✗ Constructivist LO research

Prefers deep integration into local environments, limiting cross-platform reuse (Ilomäki et al., 2006; Vanoostveen et al., 2019).

Comparative perspective

Symbolic and hybrid AI align with traditional interoperability ideals, while generative AI sacrifices standardisation for flexibility. Constructivist approaches diverge most, favouring pedagogical depth over system compatibility.

6.3.7 Durability

✓ Symbolic AI

Promotes stable, rule-based structures that maintain long-term content relevance (Denvir & Brown, 1986; Brusilovsky & Peylo, 2003).

✗ Generative AI

Continuously produces dynamic content, requiring ongoing revision and quality control to maintain relevance (Banh & Strobel, 2023; Brehmer & Buonassisi, 2024).

↺ Hybrid AI

Maintains structural durability through stable LO retrieval, while allowing selective updates via AI augmentation (Thüs et al., 2024).

✗ Constructivist LO research

Emphasises immediate contextual relevance over long-term standardisation (Ilomäki et al., 2006; Gaviria-Chavarro et al., 2023).

Comparative perspective

Symbolic AI is most closely aligned with durability, offering long-lasting structures. Generative and constructivist approaches challenge this by favouring continual evolution. Hybrid AI aims to stabilise core content while allowing dynamic updates.

6.3.8 Single purpose & objective alignment

✓ Symbolic AI

Structures LOs around predefined learning objectives, supporting clear instructional targeting (Graesser et al., 2005; Apoki et al., 2019).

✗ Generative AI

Encourages flexible, multipurpose content, which can dilute alignment with explicit goals (Halkiopoulos & Gkintoni, 2024; Brehmer & Buonassisi, 2024).

↺ Hybrid AI

Preserves alignment with defined goals while adapting content to context-sensitive needs (Modran et al., 2024).

✗ Constructivist LO research

Advocates learner-driven goal formation and evolving objectives (Domínguez Romero & Bobkina, 2021; Topali & Mikropoulos, 2023).

Comparative perspective

Symbolic AI strongly supports single-goal clarity. Hybrid AI moderates this stance with adaptive alignment. Generative and constructivist models challenge the one-size-fits-all objective by enabling multipurpose, emergent goal orientation.

6.3.9 Granularity

✓ Symbolic AI

Follows a modular approach with finely segmented content for controlled sequencing (Brusilovsky & Peylo, 2003).

↺ Generative AI

Creates fluid content that may disrupt fixed module structures (Thüs et al., 2024).

✓ Hybrid AI

Preserves modular integrity while dynamically adapting the arrangement of LOs (Lewis et al., 2020).

✗ Constructivist LO research

Prioritises holistic and contextually embedded learning experiences over strict modularity (Vanoostveen et al., 2019; Ilomäki et al., 2006).

Comparative perspective

Symbolic and hybrid AI both support modular granularity, although hybrid systems permit greater flexibility. Generative AI reconfigures content dynamically, while constructivist approaches resist fragmentation in favour of thematic coherence.

6.3.10 Minimised coupling & nonsequentiality

✓ Symbolic AI

Supports modular independence and flexible sequencing through rule-based instructional design (Brusilovsky & Peylo, 2003).

↺ Generative AI

Enables non-linear learning by generating content dynamically, but may introduce unintended dependencies (Halkiopoulos & Gkintoni, 2024; Modran et al., 2024).

✓ Hybrid AI

Preserves modular decoupling while allowing AI to reorganise learning sequences contextually (Thüs et al., 2024).

✗ Constructivist LO research

Fosters interdependent, sequential learning objects tailored to collaborative settings (Ilomäki et al., 2006; Vanoostveen et al., 2019).

Comparative perspective

Symbolic and hybrid AI maintain modular decoupling. Generative AI adds flexibility but risks dependency drift. Constructivist LO research intentionally links LOs in cohesive, sequenced pathways.

6.3.11 Generativity

✗ Symbolic AI

Relies on predefined modules, offering limited recombination and minimal learner-driven creation (Apoki et al., 2019).

✓ Generative AI

Enables real-time recombination and custom content generation using large language models (Banh & Strobel, 2023).

↺ Hybrid AI

Merges structured reuse with dynamic recombination, producing adaptable yet coherent outputs (Gao et al., 2023; Thüs et al., 2024).

✓ Constructivist LO research

Encourages learner-driven creation and reorganisation of learning content (Topali & Mikropoulos, 2023; Gaviria-Chavarro et al., 2023).

Comparative perspective

Generative AI and constructivist approaches promote high generativity—though one is system-driven, the other learner-driven. Hybrid AI blends both. Symbolic AI remains the least generative, anchored in structured reuse.

6.3.12 Modularity & customisability

✓ Symbolic AI

Maintains predefined modularity for consistent reuse and instructional integrity (Brusilovsky & Peylo, 2003; Apoki et al., 2019).

↺ Generative AI

Challenges fixed modules by enabling flexible content generation, but can reorganise and customise existing LOs (Banh & Strobel, 2023; Brehmer & Buonassisi, 2024).

✓ Hybrid AI

Retrieves structured modules while allowing for adaptive refinement based on learner needs (Lewis et al., 2020; Thüs et al., 2024).

✗ Constructivist LO research

Prioritises contextual coherence and pedagogical flexibility over rigid modularity, favouring context-sensitive modification (Ilomäki et al., 2006; Vanoostveen et al., 2019).

Comparative perspective

Symbolic and hybrid AI support modularity, though hybrid systems permit more customisation. Generative AI loosens modular constraints to enhance flexibility. Constructivist research deprioritises strict modularity in favour of situated relevance.

Together, the preceding comparisons point to broader patterns in how the four research directions relate to the traditional LO principles—patterns that emerge more clearly when considered as a whole.

6.3.13 Overarching comparative perspective

The comparative analysis reveals distinct, recurring patterns in how the four research directions relate to traditional learning object (LO) principles.

Symbolic AI aligns most closely with traditional LO principles. Its reliance on predefined structures, metadata, and modular design supports scalability, standardisation, and instructional control—but often at the expense of flexibility and adaptivity. It prioritises reuse, efficiency, and interoperability while resisting learner-driven generativity and contextual adaptation.

Generative AI represents a paradigm shift. It challenges foundational LO principles by replacing pre-authored content with dynamically generated resources. This enables adaptability and personalisation but raises concerns around standardisation, durability, and goal alignment. While it breaks with conventional reuse models, it introduces a new form of adaptive reuse grounded in computational generativity.

Hybrid AI (RAG-based) occupies a middle ground. It retains core LO principles—such as reusability, modularity, and scalability—while extending them through AI-driven adaptation. Hybrid AI maintains the structural integrity of symbolic systems while gaining flexibility from generative approaches. Across most principles, it reinterprets rather than rejects.

Constructivist-oriented LO research consistently challenges traditional LO assumptions. Emphasising situated learning, teacher agency, and learner co-construction, it redefines reuse, modularity, and objective alignment. Rather than standardising content, it prioritises pedagogical relevance, learner engagement, and local adaptability. While technically less scalable or interoperable, it strongly supports learner agency and contextual coherence.

Cross-cutting trends suggest symbolic AI and constructivist approaches represent opposing ends of a spectrum (standardised vs. situated), while hybrid and generative AI reflect attempts to bridge or transcend that divide. The framework also points to a broader shift from content-centric to context-centric design, with newer directions embedding LOs in more adaptive, responsive, and learner-driven environments.

While recent research directions have introduced more flexible, learner-centred approaches to LOs, the way LOs are perceived and applied in practice is also shaped by teachers’ underlying conceptions of knowledge, teaching, and learning. These conceptions are often informed by conceptual metaphors (Lakoff & Johnson, 2003) which structure how teachers understand and interact with LOs. The next section examines how these conceptual metaphors manifest in teachers’ perspectives and influence their use of LOs in teaching.

6.4 What conceptions of knowledge, teaching and learning do teachers display?

How do teachers’ conceptions of teaching and learning correspond with mature understandings of learning according to constructivist learning theories? (RQ3a)

To structure the discussion, we revisit the three categories introduced in Sect. 5: simple, intermediate, and more mature understandings of teaching and learning. These categories were derived from teachers’ responses and serve as a conceptual foundation for analysing how their perspectives relate to different learning metaphors and constructivist principles. In the sections that follow (6.4.1–6.4.3), we examine each category in turn and explore how they align or diverge from the learning metaphors outlined in Sect. 2.2 and the constructivist principles discussed in Sect. 2.1.

6.4.1 Simple understandings

What notions of knowledge, teaching, and learning are implied by the utterances in Sect. 5.1, categorised as simple understandings?

A noticeable pattern is the pervasive use of metaphors. Addressing this question requires identifying these metaphors, unpacking their meanings, and analysing the quality of guidance they provide for conceptions of knowledge, teaching, and learning (see Fig. 10). We argue that these understandings are closely linked to a more primitive view of learning, reflected in the conduit metaphor as well as the metaphors, contexts, processes, and principles commonly associated with traditional LOs.

Fig. 10

Primitive learning metaphors

Metaphors have entailments. For instance, the saying and metaphorical concept ‘time is money’ entails that time ‘is a limited resource’, which entails that time ‘is a valuable commodity’ (Lakoff & Johnson, 2003, pp. 8–9). These metaphoric entailments are not consistent (i.e. they form no single image), but they are coherent (i.e. they fit together; Lakoff and Johnson (2003). This coherence contributes to and reinforces the systematic structures of meaning that highlight or hide some meanings. For instance, the metaphorical concept ‘argument is war’ highlights the battling aspects of arguing rather than the cooperative aspects (Lakoff & Johnson, 2003, p. 5). A key point in Lakoff and Johnson’s (2003) discussion on the conduit metaphor and the other simple metaphors by which we live is that an entailment of a metaphor does not work in isolation but merges with other metaphor entailments to form rich networks of associated meanings that take strong hold of peoples’ thinking and practice. What makes the following utterances simple, therefore, is not based on any single utterance but rather the reinforcing, coherent structure of meaning they form together.

In Fig. 10, we compare extracts from our empirical data with examples of the conduit metaphor reported in the literature (Lakoff & Johnson, 2003; Reddy, 1979; Sfard, 1998). This comparison shows that the teachers’ discussion on knowledge, teaching and learning draws on the following metaphors: commodities, productivity, objects, landscapes, products, manufacturing and containers. In other words, our teachers treat knowledge, teaching and learning as physical objects in one of the aforementioned forms. These metaphors have entailments that can lead us far away from the view on knowledge, teaching and learning emphasised by constructivism and the acquisition, participation and knowledge creation metaphors (cf. Section 6.1, Fig. 8).

This shift in metaphor is not just semantic—it has real pedagogical consequences. These metaphorical framings are not merely linguistic artefacts; they reveal underlying epistemological assumptions that influence how learning is structured and enacted in practice. When knowledge is treated as a thing to be transmitted or stored—rather than constructed, negotiated, or co-created—learning activities risk becoming mechanical and decontextualised. From a design perspective, this suggests that LO systems which mirror or reinforce these metaphors may inadvertently entrench transmissive practices rather than challenge them. Standardisation of learning content, as promoted by traditional LO models, also tends to shape pedagogical processes—encouraging linear instructional sequences and predefined pathways that limit teacher and learner agency.

Looking ahead, an important implication is that professional development efforts must go beyond introducing constructivist principles abstractly. Unless teachers critically examine the metaphors that underlie their everyday language and decisions, attempts at pedagogical innovation may be superficial, short-lived, or reduced to technology-driven reform efforts detached from pedagogical reasoning. Our findings underscore the need for reflective tools and collaborative arenas where such assumptions can be surfaced, questioned, and reframed—especially when working with educational technologies that carry their own implicit pedagogical models.

These findings highlight the importance of addressing entrenched object-based conceptions of learning in both teacher education and technology design. While some teachers already operate with intuitive pedagogical nuance, these insights often remain latent—overshadowed by more transmissive or pragmatic framings. Our study shows that conceptually oriented prompts (e.g., those focused on constructivism and student perspectives) can activate deeper pedagogical reflection, suggesting that such scaffolds may be valuable tools in professional development.

Future teacher education initiatives could benefit from surfacing and critically examining everyday metaphors of teaching and learning, using them as entry points for richer pedagogical dialogue. Similarly, designers of LO systems might consider how interface language, instructional templates, and content structures could be reimagined to support more expansive and learner-centred conceptions of knowledge, learning, and reuse.

6.4.2 Intermediate understandings

The category of intermediate understandings, described in Sect. 5.2, represents perspectives that acknowledge some constructivist principles, yet remain constrained by traditional views and teacher-centred assumptions. Teachers’ recognition of the challenges associated with unmodified reuse and the need to adapt learning resources to particular contexts, topics, and student groups aligns with constructivist principles such as: connecting new and prior knowledge (Cunningham & Duffy, 1996); context dependency (Cunningham & Duffy, 1996; Greeno, 1997; Savery & Duffy, 1995; Sfard, 1998); supporting learners’ thinking (Cunningham & Duffy, 1996; Savery & Duffy, 1995); and providing appropriate scaffolding (Van de Pol et al., 2010; Vygotsky, 1978). However, despite these strengths, intermediate understandings remain limited in their scope and focus. A significant weakness is their predominant emphasis on teachers’ work, adopting the teacher-centric approach traditionally associated with LOs and the conduit metaphor. Limited attention is given to constructivist principles that emphasise the learner’s central role in the learning process. For example, principles such as active learning (Karagiorgi & Symeou, 2005; Mayer, 2002; Papastergiou & Mastrogiannis, 2021; Prince, 2004), reflection and metacognition (Savery & Duffy, 1995; Sawyer, 2005), ownership and students’ voice (Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995), stimulating self-regulated learning (Schraw et al., 2006; Zimmerman, 1990), externalising and articulating ideas (Bransford et al., 2000; Sawyer, 2005), collaborative learning (Cunningham & Duffy, 1996; Honebein, 1996; Karagiorgi & Symeou, 2005; Luckin & Holmes, 2016; Vygotsky, 1978), and epistemic agency (Scardamalia & Bereiter, 2006) are largely neglected. Consequently, intermediate understandings provide inadequate guidance for learning and learners.

These perspectives illustrate a partial pedagogical awareness, but they risk reinforcing a false sense of alignment with constructivist principles. Recognising the need to modify teaching materials is important, yet without deeper reflection on student agency, ownership, the situated nature of teaching and learning, or learning processes, such adaptations may remain superficial. In practice, intermediate understandings may give the appearance of pedagogical responsiveness while still reproducing traditional, teacher-led dynamics.

In this sense, intermediate understandings occupy an ambivalent space: they gesture toward learner-centred pedagogy, but without fully enacting it. This makes them an important turning point in our analysis—not only as a midpoint between simple and mature understandings, but as a space where assumptions can either solidify or shift. The next section explores cases where such a shift does occur, and where teachers begin to articulate richer, more constructivist-aligned perspectives on learning.

From a practical standpoint, intermediate understandings represent a pivotal opportunity for intervention. While they often include tacit awareness of pedagogical complexity—such as the need to adapt materials—they rarely extend to considerations of learner agency, ownership, or epistemic agency. This contrasts with the more developed conceptions observed in responses associated with Prompts C and D, which explicitly introduced themes such as constructivism, 21 st-century skills, and student-centred reuse. These later segments of the interview helped elicit reflections that were more theoretically grounded and pedagogically ambitious.

Targeted reflection tools, peer dialogue, or scaffolded case-based activities could help transform these tentative insights into more robust pedagogical commitments. This also has implications for LO-based technology: systems that foreground student activity, agency, and contextualisation—rather than just content delivery—may better support this pedagogical development.

6.4.3 More mature understandings

Overall, the teachers’ responses in the part of the interview (Sect. 5.3) when the researcher asked about their understandings of constructivism and 21 st-century skills, reflect quite mature views on knowledge, teaching and learning.

First, the teachers critique and distance themselves from various views on learning that are associated with instructionism and shallow learning (Sawyer, 2005):

They criticise the focus on knowledge and skills that are atomistic, out of context, and disconnected from broader concepts, describing such learning as fragmented and superficial (Fig. 7, line 1). They reject the practice of copying dictated knowledge (line 2) and express concerns about approaches that emphasise memorisation and behaviorist techniques, such as rote learning (line 3) or the reliance on fixed answers (line 6). Such methods, they argue, fail to encourage critical thinking or deeper understanding. The teachers also highlight the limitations of instructional approaches centered on factual recall, such as copying knowledge and answering fact-based questions (line 7). They critique an excessive focus on “WHAT questions” (line 9), which they view as inadequate compared to “WHY questions” that foster explanation, understanding, and the ability to connect ideas. Practices like regurgitating content from textbooks (line 17) and exams that prioritise the recitation of facts (line 18) are similarly regarded as insufficient for meaningful learning. Furthermore, they point to strategies such as requiring students to “read this and answer these questions” (line 20) without encouraging discussion or allowing students to make their own choices as examples of shallow learning that stifles autonomy and critical engagement.

Second, the teachers provide detailed descriptions of what they perceive as missing from these instructional and shallow learning approaches, aligning closely with constructivist principles (Sect. 3.1):

One of the teachers’ central concerns is the need for greater emphasis on knowledge integration and learning in context to avoid fragmented knowledge and skills (Fig. 7, line 1). This reflects constructivist principles such as connecting new and prior knowledge (Cunningham & Duffy, 1996); context dependency (Cunningham & Duffy, 1996; Greeno, 1997; Savery & Duffy, 1995; Sfard, 1998), authentic learning (Honebein, 1996; Karagiorgi & Symeou, 2005; Papastergiou & Mastrogiannis, 2021; Savery & Duffy, 1995), and bridging concrete experience and abstract general knowledge (Sawyer, 2005; Vygotsky, 1986).

Another prominent theme in the teachers’ responses is the importance of learning how to learn (line 10). This aligns closely with the constructivist focus on reflection and metacognition, which encourage students to critically assess their knowledge and learning strategies (Savery & Duffy, 1995; Sawyer, 2005))

The teachers also emphasise the significance of qualities often associated with self-regulated learning, such as independence (lines 2 and 19), responsibility for one’s own learning (lines 4 and 7), and effort and commitment (line 4). These qualities resonate with the principles of stimulating self-regulated learning (Schraw et al., 2006; Zimmerman, 1990), ownership and students’ voice (Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995), active learning (Karagiorgi & Symeou, 2005; Mayer, 2002; Papastergiou & Mastrogiannis, 2021; Prince, 2004), and reflection and metacognition (Savery & Duffy, 1995; Sawyer, 2005). These examples reflect a shift toward epistemic agency (Scardamalia & Bereiter, 2006), where learners are seen not only as recipients of knowledge, but as active agents in shaping the learning process itself.

In addition, the teachers advocate for a stronger focus on generic and transferable skills, such as assessing, critically evaluating, and adapting information (line 8); critical thinking skills (line 10); discussing and independent thinking; general thinking skills (line 11); and reusing learning strategies, as well as writing and reading strategies (line 12). These skills align with constructivist principles such as bridging concrete experience and abstract general knowledge (Sawyer, 2005; Vygotsky, 1986), support and challenge learners’ thinking (Cunningham & Duffy, 1996; Savery & Duffy, 1995); multiple perspectives (Cunningham & Duffy, 1996; Honebein, 1996; Karagiorgi & Symeou, 2005; Savery & Duffy, 1995)).

The emphasis on WHY questions and understanding (line 9) further highlights the teachers’ alignment with constructivist principles. These questions encourage reflection and metacognition (Savery & Duffy, 1995; Sawyer, 2005), externalise and articulate (Bransford et al., 2000; Sawyer, 2005), and connecting new and prior knowledge (Cunningham & Duffy, 1996).

Another key area of focus is ownership of problems, tasks, and processes, with examples such as “constructing their own knowledge” (line 9), “own free reasoning” (line 17), “students’ free inquiry” (line 18), and “making their own choices” (line 20). These views are closely linked to the constructivist principles of ownership and students’ voice, which emphasise the importance of empowering students to take control of their learning (Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995).

The teachers also highlight the importance of authenticity in teaching (line 13). This principle emphasises embedding learning in real-world tasks and contexts to enhance its relevance and applicability (Honebein, 1996; Karagiorgi & Symeou, 2005; Papastergiou & Mastrogiannis, 2021; Savery & Duffy, 1995). Related to this is their focus on relevance, as illustrated by the statement: “The teaching must concern the students; if not, the students will feel that the knowledge does not apply to them or their situation” (line 14). This perspective is aligned with principles such as context dependency (Cunningham & Duffy, 1996; Greeno, 1997; Savery & Duffy, 1995; Sfard, 1998), authentic learning (Honebein, 1996; Karagiorgi & Symeou, 2005; Papastergiou & Mastrogiannis, 2021; Savery & Duffy, 1995), and ownership and students’ voice (Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995).

Finally, the teachers advocate for a focus on higher taxonomic levels (line 5) and deep reflection (line 21), which correspond to principles such as reflection and metacognition (Savery & Duffy, 1995; Sawyer, 2005), learning vs. complexity (Barrows & Kelson, 1993; Hmelo-Silver & Barrows, 2006; Savery, 2006; Savery & Duffy, 1995), and externalising and articulating knowledge (Bransford et al., 2000; Sawyer, 2005).

In addition to these concerns, the teachers consider broader didactic challenges (lines 18, 20, and 21) and societal changes (lines 9 and 10) that influence what students should learn and how they should be taught.

Taken together, these responses do more than align with constructivist theory—they illustrate that such understandings are already present, though often latent, in teachers’ reasoning. Importantly, these perspectives emerge not as abstract ideals but as reflections grounded in teachers’ lived experiences, often triggered by critical incidents or frustrations with transmissive approaches. This suggests that pedagogical change is not simply a matter of introducing new frameworks, but of surfacing and strengthening existing but underarticulated beliefs.

Notably, several of these responses emerged in interaction—building on earlier remarks by peers or picking up on ideas introduced earlier in the session. This suggests that what we term “mature understandings” may not always be fully internalised or readily accessible to individual teachers in isolation. Rather, they are often socially scaffolded—articulated and deepened through dialogue, resonance, and collective reflection. This highlights the importance of collaborative professional arenas where such latent perspectives can be surfaced and developed through interaction.

At the same time, these mature understandings remain fragile if they are not meaningfully internalised and reinforced through collective reflection and practice. They coexist with institutional routines, assessment pressures, and platform designs that do not always reward deeper pedagogical reasoning. Without sustained opportunities for reflection and dialogue, such perspectives may struggle to gain traction in daily practice. From a design perspective, and looking ahead, this points to the need for learning technologies that not only accommodate but actively support teacher agency, epistemic reflexivity, and situated pedagogical judgment.

Importantly, the challenge is not the absence of constructivist understanding, but the difficulty of activating and sustaining it in everyday teaching—particularly in environments shaped by LO systems that reward standardisation, efficiency, and teacher control. These deeper pedagogical insights may often remain latent, overshadowed by simpler or more pragmatic framings. This points to a promising, but underexplored, direction for future research and practice: developing reflective infrastructures that help teachers recognise, articulate, and apply their more mature understandings when interacting with LO-based technologies. Rather than only redesigning systems, such efforts aim to shift the pedagogical framing through which these systems are interpreted and used.

6.4.4 How the affordances of LO-based learning technologies affect teachers’ perspectives on teaching and learning (RQ3b)

A notable pattern in the empirical data is how teachers’ reflections on teaching and learning are shaped by the technological and organisational context in which those reflections occur. When reflecting within the constraints and structures of reuse practices—whether embedded in the LMS or more broadly in the school’s digital routines—teachers tend to adopt transmissive or simplified models of teaching and learning. This is particularly evident when they discuss issues related to reuse, curriculum alignment, and assessment—core concerns embedded in the LO logic of digital systems. By contrast, when invited to reflect on pedagogical theory and student learning more broadly, several participants articulate more complex and constructivist-aligned conceptions, such as learning as collaboration, inquiry, and knowledge construction.

This contrast suggests that the affordances and design of reuse-oriented systems may not merely support particular pedagogical practices, but also influence how teachers conceptualise their work. As Lakoff and Johnson (2003) argue, such metaphors are not merely rhetorical devices, but shape how individuals interpret and act within educational contexts. In our data, reuse technologies appear to cue a particular metaphorical framing of teaching—one that aligns closely with the conduit metaphor and the acquisition view of learning. This framing may reduce space for alternative pedagogical perspectives, especially under time and accountability pressures.

At the same time, the interview also demonstrates that teachers do not lack constructivist understanding. Rather, these understandings are often backgrounded in practical contexts, only becoming foregrounded when teachers are prompted to consider abstract or theoretical dimensions of their practice. In this way, reuse systems inspired by LO principles may contribute to what Lakoff and Johnson (2003) describe as the “metaphors we live by”—subtle conceptual frameworks that shape practice without necessarily being made explicit.

These findings reinforce the broader tensions discussed across Sect. 6.1–6.3, suggesting that teacher practice may unconsciously reproduce the very assumptions that LOs were designed to overcome.

7 Conclusions

This study has examined longstanding and emerging tensions between Learning Objects (LOs) and constructivist learning theories by integrating theoretical analyses with empirical data. While traditional LO frameworks prioritise standardisation, reusability, and scalability, they often neglect learner agency, contextual adaptation, and epistemic collaboration—hallmarks of constructivist pedagogy.

This article contributes a novel analytical framework that bridges theoretical, technological, and practical domains. The framework integrates six key positions in the field—traditional LO design principles, constructivist learning theories, symbolic AI, generative AI, hybrid AI (RAG), and constructivist-oriented LO research—and compares them across two analytic dimensions: their alignment with constructivist learning metaphors and their relationship to core LO design principles. This dual-axis comparison enables a fine-grained analysis of tensions, adaptations, and areas of convergence between pedagogical ideals and technological systems. In addition, the integration of four learning metaphors—conduit, acquisition, participation, and knowledge creation—into a single analytical structure provides a nuanced lens for examining the alignment and tensions between LO models and constructivist theory. While subsets of these metaphors have been compared in earlier studies, their combined application to both pedagogical discourse and LO technology design—especially in the context of AI—represents a novel contribution. The framework enables descriptive, explanatory, predictive, and normative analysis of how learning is conceptualised, supported, or constrained across different positions.

By applying constructivist learning metaphors—acquisition, participation, and knowledge creation—we systematically analysed how different research directions relate to these pedagogical ideals. Our review of symbolic AI, generative AI, hybrid AI, and constructivist-oriented LO research revealed that these approaches vary in their support for constructivist principles. Symbolic AI reinforces structured progression but limits collaboration and creativity. Generative AI promotes exploration and dialogue, but raises concerns about reliability and learner agency. Hybrid AI combines structure and flexibility, though often in AI-driven ways. Constructivist-oriented LO research remains the most aligned with constructivist metaphors, emphasising learner control, co-construction, and contextual sensitivity.

These findings suggest that no single approach fully resolves the tensions between LO design and constructivist learning. Rather, they point to the need for pedagogically informed LO frameworks—ones that can accommodate both structured guidance and open-ended inquiry, and that empower teachers and learners to adapt technology to their goals.

Our empirical data confirm this variability. Teachers’ conceptions of learning shift depending on whether they reflect within technological constraints or pedagogical frameworks. When prompted by theory, teachers often articulate rich understandings of learning that resonate with constructivist ideals, even if their everyday practice tends toward transmissive models.

These findings also have forward-looking implications. From a pedagogical perspective, the three-category model—spanning simple, intermediate, and more mature understandings—may support the design of professional development initiatives that scaffold progression toward constructivist-aligned teaching. For example, structured prompts and collaborative reflection formats, like those used in our study, may help surface latent pedagogical insights and promote epistemic agency. From a design standpoint, LO systems could be developed to more explicitly support such transitions, by making visible how different forms of reuse relate to varying conceptions of learning and teaching. In this way, our framework may serve as both an analytical tool and a practical resource for aligning technological design with constructivist pedagogy.

We acknowledge the study’s exploratory and context-specific nature, which limits the generalisability of our empirical findings. Future research could examine how our analytical framework transfers to other educational settings and technological infrastructures. In particular, we see promise in design-based studies that apply constructivist principles in LO development, empirical testing of hybrid AI systems in classrooms, and comparative research on how different LO models affect pedagogical flexibility.