The main goal of this article is to develop an academically grounded exploration of how knowledge construction takes place and in particular how generative artificial intelligence (AI) shapes knowledge construction from both cognitive and sociocultural perspectives. Its purpose is to articulate the theoretical, practical, and memory-related implications of co-constructing knowledge with AI to support critical reflection and scholarly discussion. In order to lay the foundations for these considerations, I first discuss how knowledge and its construction are conceptualized in various philosophical and psychological traditions. In doing so, I focus primarily on presenting the fundamental principles and processes without going into detail about specific empirical findings (these can be found, e.g., in the publication by Jeong et al. 2019).

The nature of knowledge has long captivated scholars across disciplines, from ancient philosophical inquiry to modern cognitive science. Since the time of Plato, the question of what knowledge truly is has prompted a spectrum of interpretations (Oeberst et al. 2016). Plato’s concept of knowledge centered on the idea of “justified true belief,” meaning that for someone to truly know something, it must be true, they must believe it, and they must have a rational justification for that belief (although Plato himself already pointed out difficulties with this concept; see Fine 2003). Plato also distinguished between the changing world of sensory experience and the unchanging world of Forms, asserting that true knowledge comes from understanding these eternal, abstract Forms rather than relying on perception (Nehamas 1975). In other words, knowledge is considered to be acquired deductively, with absolute truth being revealed through logical thinking.

These considerations later developed into rationalism, which found its most well-known representative in René Descartes. The opposing position to Plato, which Aristotle represented, assumes that there is no a priori knowledge, but rather that sensory experience is the only true source of knowledge. In other words, knowledge is acquired inductively, with insights derived from sensory experiences. Aristotle thus developed an antithesis to rationalism, later termed empiricism, which was most prominently represented by John Locke (Markie and Folescu 2023).

By contrast, cognitive psychology tends to define knowledge more pragmatically, often equating it with mental representations stored in human memory (Anderson 1990; Baddeley 1997; but for a recent conceptualization that integrates learning as information processing and knowledge as justified true belief, see Richter and Tiffin-Richards 2024). Despite this disciplinary divergence, a common thread persists: The view that knowledge was inherently a property of individual minds. Yet this assumption invites further scrutiny. Can knowledge exist solely within individuals, or might it also reside in communities and collective systems? Increasingly, scholars challenge the individualistic model, suggesting that knowledge might also be constructed, maintained, and transformed within social contexts (for an early approach in that direction, see Berger and Luckmann 1966). This broader view opens up avenues for understanding knowledge construction not merely as a cognitive process, but as a socially embedded phenomenon. In the following considerations, I explore the concept of knowledge construction through an interdisciplinary lens, building on its philosophical roots and examining its psychological foundations and sociocultural dimensions.

As described in the previous section, there is no universally accepted definition of “knowledge.” This goes back to the age-old controversy between rationalism and empiricism. These fundamental epistemological strands in the history of philosophy then had a corresponding impact on the various psychological traditions of the twentieth century. The lack of a uniform characterization of knowledge therefore results largely from the fact that the concept of knowledge can be interpreted through various theoretical lenses, each emphasizing different aspects of what it means to know something.

From a behaviorist perspective, which essentially continued the tradition of empiricism, knowledge, as a mental state, is not a legitimate subject of investigation. This school of thought prioritized observable and measurable behavior (Skinner 1953; Watson 1913). According to behaviorists, learning is demonstrated when an individual reliably exhibits a specific behavioral response to a given stimulus. What occurs within the mind of a learner—their thoughts, intentions, or other cognitive processes—is considered irrelevant or at least unknowable, hence the metaphor of the mind as a “black box.” Once a stimulus reliably produces the desired response, learning is considered complete.

By contrast, the cognitive perspective sees knowledge as the result of internal mental processes such as perception, memory, and reasoning. Cognitive theories emphasize how information is processed, organized, and stored in the mind (Barsalou 2014). Information is a key concept in this context. From a cognitive perspective, information is structured data that reduces uncertainty by providing meaning the mind can process; however, unlike knowledge, it does not yet entail internalized or integrated understanding. Learning, from this view, involves the development of mental models, the construction of conceptual frameworks, and the application of cognitive skills. This perspective significantly departs from behaviorism by acknowledging and analyzing the internal workings of the mind (Ertmer and Newby 2013).

These differing conceptualizations of knowledge carry substantial didactic and pedagogical implications. Behaviorist approaches often lead to instructional strategies based on reinforcement, repetition, and conditioning (e.g., drill-and-practice activities). Cognitive approaches, however, favor strategies that support conceptual understanding, problem-solving, and the development of metacognitive skills and critical thinking (Magno 2010). The choice of perspective thus influences not only how educators define learning but also how they structure the teaching and assessment of knowledge. Applying the cognitive theory of multimedia learning (Mayer 2001), digital content should reduce extraneous cognitive load and integrate visual and verbal information effectively. For example, well-designed video lectures or interactive infographics can support deep learning by connecting new content to existing mental models.

While cognitivism emphasizes how individuals process externally provided information, constructivism goes a step further and centers on the active role of the individual in interpreting and making sense of their experiences (Riegler 2012). Constructivist theories assert that individuals play a central role in constructing knowledge based on their unique perceptions of the world around them (Fosnot 2013). According to constructivism, each person forms their own version of reality through their personal experiences, background, and social context. Knowledge, in this view, is not an objective truth handed down from outside sources, but a subjective interpretation that varies from person to person (Bodner 1986).

Memory, from a constructivist standpoint, is not seen as a passive storage system that faithfully retains information. Instead, memory is considered an active, reconstructive process in which past experiences are continuously reshaped and reinterpreted in light of current knowledge, goals, and social context. Rather than retrieving fixed data, individuals reconstruct memories to make sense of new situations, typically without intention or awareness, thereby integrating past and present in a dynamic way. This further supports the idea that knowledge is not static, but a fluid and evolving construct shaped by individual perception and context (Rosenthal 2015).

Constructivism holds that learners actively construct their own understanding through interaction with their environment, building on prior knowledge, experience, and social context. From this perspective, digital media are not mere transmitters of information but environments for exploration, reflection, and meaning making. Memory in constructivist environments is seen as dynamic and reconstructive. Digital storytelling, for instance, allows learners to revisit and reinterpret their experiences, reinforcing learning through personal narrative and reflection.

In more contemporary developments—particularly within the situated cognition framework—knowledge is further conceptualized as being distributed across people, environments, and cultural artifacts. This perspective shifts the emphasis away from defining what knowledge is, and toward understanding how knowledge is shared, developed, and manifested within communities and through interaction with tools, language, and other contextual elements. Learning, therefore, becomes a process deeply embedded in social and physical contexts, where understanding is co-constructed rather than individually acquired in isolation (Lave and Wenger 1991). Well-established approaches in this area are anchored instruction (a method of problem-oriented learning; Bransford et al. 1990) and cognitive apprenticeship (a constructivist teaching model where experts make their mental processes visible to apprentices to facilitate learning abstract skills; Collins et al. 1991).

People frequently engage with digital tools and shared online resources that support communication and teamwork, allowing them to exchange ideas and build communities around shared interests. Building knowledge together is a dynamic and evolving process, often leading to new outcomes that go beyond what any individual initially knew or anticipated (Kimmerle et al. 2010). Collaborative knowledge construction involves more than simply adding individual inputs—it requires participants to engage with one another’s ideas, building upon and challenging them to co-create deeper understanding (Cress and Kimmerle 2018). In a threaded online forum discussion, learners respond to and expand on each other’s posts, leading to a nuanced debate that clarifies and deepens the group’s understanding of a complex topic. Effective collaborative learning depends on individuals sharing their own experiences and viewpoints while also remaining open to, and reflective of, the thoughts of others. In a virtual classroom, for example, students may engage in peer feedback sessions where they present their ideas and give constructive critiques, fostering mutual learning.

The field of computer-supported collaborative learning (CSCL) has emerged as a prominent research domain within the broader discipline of the learning sciences (Cress et al. 2021). This area of inquiry concentrates on the processes of collaborative learning and explores how information and communication technologies can facilitate and enhance the co-construction of new knowledge among learners. Within CSCL, two distinct but complementary research traditions have evolved, each grounded in different theoretical perspectives and methodological approaches. These traditions—relevant to the earlier discussion on differing conceptions of knowledge—are typically referred to as the cognitive and the sociocultural approaches (Matuk et al. 2021). The cognitive approach focuses primarily on individual learners, emphasizing internal cognitive processes such as attention, memory, and reasoning. It seeks to understand how individuals process information and develop understanding, even when these processes occur in collaborative settings. By contrast, the sociocultural approach posits that knowledge is not an individual possession to be acquired, but rather something that emerges through participation in social activities and cultural practices.

3.1 The cognitive approach

The cognitive perspective regards knowledge construction as an activity that occurs primarily within the individual mind. In this view, learning is conceptualized as the internal process of acquiring, extending, and refining mental structures. Knowledge is understood as a set of internal cognitive representations that individuals develop to make sense of their environment—whether these representations concern factual information, events, or social phenomena. While cognition is centered on individual mental activity, the influence of the social environment is not disregarded in knowledge construction research. Cognitive processes are, to varying degrees, shaped by social interactions and contextual factors (Wascher et al. 2018). This is particularly evident in CSCL, where social dynamics—such as group collaboration—are acknowledged as playing a crucial role in learning. Accordingly, this orientation is more aptly characterized as a socio-cognitive approach, recognizing the mutual interdependence of cognitive and social dimensions.

Moreover, cognitive activity is often viewed through a constructivist lens, as articulated in the work of Piaget (1977). According to this framework, individuals interpret their experiences based on existing cognitive schemas. Learning involves the assimilation of new information into these pre-existing structures. However, when new experiences cannot be reconciled with current cognitive schemas, individuals engage in accommodation, that is, adjusting their mental models to integrate unfamiliar information. From this constructivist standpoint, learning is fundamentally understood as an active, individualized process of knowledge construction.

The integration of digital media into educational settings adds important new dimensions to this research tradition. Features such as collaborative documents, learning analytics, and adaptive feedback mechanisms can support the individual construction of knowledge while simultaneously enabling more complex forms of group interaction (Zheng et al. 2022). These technologies not only enhance opportunities for learners to engage in the types of metacognitive and dialogical processes highlighted in earlier research but also invite a reconsideration of how knowledge is distributed, negotiated, and co-constructed across individuals within digital environments. As such, the application of digital media in collaborative learning contexts calls for theoretical and methodological extensions of earlier frameworks.

3.2 The sociocultural approach

The dynamic and emergent nature of collective knowledge construction becomes particularly evident when individuals engage in dialogic interaction—actively referencing, responding to, and building upon one another’s ideas, perspectives, and arguments (for the relationship of argumentation and knowledge construction, see Kimmerle et al. 2021). This process creates a collaborative intellectual space where reasoning is not isolated but co-constructed through social engagement (Chinn and Clark 2013; Stahl 2006). In such contexts, individuals do not merely share information; rather, they co-elaborate ideas, transforming fragmented contributions into cohesive and evolving lines of thought.

This concept lies at the heart of the second major research tradition in the learning sciences: a sociocultural perspective that prioritizes the group—not the individual—as the central unit of analysis. Within this framework, learning and knowledge construction are viewed as distributed phenomena, emerging through the interplay of individuals, cultural tools, discursive practices, physical and virtual settings, and shared activities (Mercer and Howe 2012; Sawyer 2014). This approach builds on the foundational theories of Lev Vygotsky (1978), who argued that cognitive development is deeply rooted in social interactions and mediated by language and cultural artifacts. The sociocultural tradition views knowledge as situated—deeply embedded in the social and material activities of groups. Knowledge, in this sense, is enacted and negotiated through shared cultural practices.

Scardamalia and Bereiter (1994) advanced this perspective by proposing the concept of knowledge building, wherein knowledge is collaboratively generated by the community, rather than individually accumulated. In their view, the learning environment should be designed to support the group in collectively advancing its understanding, fostering a culture where contributions are continuously refined and improved upon. In contemporary contexts, the application of digital media significantly amplifies and transforms these collaborative processes. Online discussion forums, collaborative writing platforms, and social media facilitate real-time and asynchronous interactions that span geographic and temporal boundaries. These tools not only provide spaces for idea exchange but also include features—such as comment threads, hyperlinks, tagging, and multimedia embedding—that scaffold collaborative knowledge construction.

All contemporary theories addressing collective knowledge construction must grapple with the intricate interplay between individual cognitive processes and collective, sociocultural dynamics (Cress and Kimmerle 2018). Effective theorizing in this domain seeks not only to explain how individuals internalize and build knowledge through participation in group activities but also how new knowledge emerges at the collective level, often surpassing the sum of individual contributions. A central challenge for researchers is to develop frameworks that explain how this dual-level development occurs—how individual learning and social knowledge advancement co-evolve in mutual dependence. A comprehensive approach addressing this complexity is the co-evolution model of knowledge construction proposed by Cress and Kimmerle (2008; Kimmerle et al. 2015). This model synthesizes cognitive and sociocultural perspectives, drawing inspiration from Luhmann’s (1995) systems theory, which conceptualizes both individuals and social collectives as self-referential systems—closed in terms of operational logic, yet open to environmental stimuli.

Within this framework, individual cognitive systems and social systems operate through distinct yet interconnected mechanisms: cognition and communication, respectively. Each system constructs meaning autonomously yet interacts with the other through a process of selective appropriation. For example, individuals cognitively process and internalize information from shared digital environments, while simultaneously contributing back to the collective knowledge pool through communication acts—such as commenting, tagging, or posting new content. These contributions, once circulated and referenced by others, become part of the social system’s (collective) memory and are further shaped through iterative discourse (Kimmerle et al. 2011).

The model emphasizes that collective knowledge construction is emergent: Epistemic outcomes are not entirely predictable from individual actions alone. Instead, meaning is constructed dynamically through ongoing exchanges between individual participants and the digital or material environment. Importantly, digital media platforms offer a fertile ground for this interplay, providing both the communication infrastructure and the external memory that facilitate co-evolutionary knowledge processes. For instance, collaborative platforms such as Wikipedia, Reddit, or GitHub support asynchronous and distributed knowledge development, where individual edits, comments, and revisions accumulate and recursively influence the collective artifact.

Social media ecosystems and online learning communities also operate under similar dynamics, enabling rich knowledge ecologies in which individual learning and group cognition mutually reinforce one another. Ultimately, understanding knowledge construction through the co-evolution lens demands a focus on reciprocal interactions between personal meaning-making and collaborative activity—particularly as they unfold in digitally mediated environments. This approach offers a nuanced understanding of learning that transcends the individual and captures the complexity of networked, socially situated cognition.

Generative AI is an area of AI that creates new content, such as text, images, or videos by learning from massive datasets (Buder et al. 2025). The integration of generative AI into educational and professional contexts has introduced new dynamics in how knowledge is constructed, accessed, and shared (Klein and Utz 2024). From a theoretical standpoint, co-constructing knowledge with AI systems such as large language models (Luther et al. 2024) can be examined through both cognitive and sociocultural lenses (Cress and Kimmerle 2023). These frameworks provide complementary insights into how individuals engage with AI tools to extend cognitive capabilities, reshape learning processes, and negotiate meaning within social and cultural contexts. A central concern in this evolving relationship is the role of human memory—its functions, limitations, and transformation in the age of artificial co-participants (Hoskins 2024; Oakley et al. 2025).

From a cognitive perspective, knowledge construction involves internal mental processes such as encoding, storage, and retrieval of information. Generative AI, when used as a tool for idea generation, explanation, or problem-solving, supports these processes by reducing cognitive load, enabling rapid information synthesis, and scaffolding complex reasoning (Cress and Kimmerle 2023). By offloading routine memory functions to AI—such as recalling facts, summarizing texts, or generating drafts—users are enabled to redirect their mental resources toward higher-order thinking and metacognitive regulation. This aligns with theories of extended (Clark and Chalmers 1998) and distributed cognition (Hutchins 1995), which posit that cognitive processes can be distributed across tools and environments. In this view, generative AI becomes an external cognitive aid that, while not sentient, functions as an interactive memory extension and a catalyst for reflective thinking.

However, the use of AI to offload memory also raises concerns about epistemic dependency and a potential weakening of internal memory systems. While human memory is reconstructive and susceptible to decay, it also plays a crucial role in conceptual integration and long-term understanding. Co-construction with AI may inadvertently shift the balance from deep encoding to surface-level interaction, where learners rely on retrieval from external sources rather than internal consolidation. Memory does not only have a storage function but is central to the restructuring of knowledge schemas and the ability to make inferences across contexts (Ghosh and Gilboa 2014). Thus, while AI may facilitate immediate performance, it is vital to consider how these tools impact the development of durable, transferable knowledge structures in the human mind.

Against the backdrop of such considerations, Bauer et al. (2025) also point out that it is essential to examine what knowledge-related effects can be expected from collaboration with AI and what can actually be empirically verified. These authors argue that AI may have positive effects by either replacing existing instruction methods, without changing the nature of the tasks or depth of cognitive processing (substitution), or by supplementing existing instruction, adding support, or enhancing learning beyond what non-AI support would provide (augmentation). Bauer et al. (2025) argue that many claims about AI focus too much on substitution or augmentation, while often ignoring a risk they call inversion, that is, the effect that AI use may lower desirable learning outcomes, often because learners over-rely on AI and reduce their cognitive effort. This may result in learners engaging less in metacognitive monitoring. The effect of AI collaboration that has the most transformative potential is what these authors call “redefinition.” This effect occurs when AI enables new kinds of tasks or learning experiences that were not possible without AI. I concur with this line of argument: What makes knowledge construction with AI particularly interesting from a cognitive perspective is changing the nature of how learners engage in deeper learning. This is the case when generative AI creates opportunities for collaborative construction of knowledge.

As outlined earlier, from a sociocultural perspective, knowledge construction is inherently situated within social practices, cultural tools, and collaborative interactions. Generative AI, in this context, can be seen as a cultural artifact that mediates dialogue and meaning-making between individuals and communities. When users engage with AI models, they are not merely retrieving information but participating in a dialogic process where knowledge is co-constructed through interaction with a responsive system shaped by collective linguistic and cultural data (Cress and Kimmerle 2023). This sociocultural framing highlights the inter-subjective nature of AI-mediated knowledge construction. Unlike traditional tools, generative AI mimics human conversational patterns, enabling users to test ideas, negotiate meanings, and simulate peer dialogue.

In collaborative settings, AI can serve as a shared reference point that facilitates group reflection, argumentation, and consensus-building. The model’s ability to provide multiple perspectives, analogies, or counterarguments can support learners in reaching zones of proximal development (Vygotsky 1978), where they accomplish tasks previously beyond their independent capabilities. By generating diverse explanations and framing concepts in different ways, AI can serve as a flexible and adaptive scaffold. For example, when learners struggle with an abstract concept, AI might supply a concrete analogy that makes the idea more accessible. Similarly, by presenting counterarguments, AI can challenge learners’ assumptions, prompting deeper critical thinking, and encouraging them to refine their reasoning. As a result of this dynamic support, learners are able to not only complete tasks that were previously out of reach but also gradually internalize new skills and ways of thinking, moving closer to independent mastery. However, this also necessitates critical engagement by the users, as they need to be aware that an AI system’s responses reflect not just factual information but the values, biases, and assumptions encoded in its training data (Lermann Henestrosa and Kimmerle 2025).

The role of human memory in this sociocultural context is similarly transformed. From this perspective, memory systems should no longer be understood as static, individually held repositories of past experiences or facts. Instead, human memory is part of a distributed system of cultural memory encompassing digital archives, AI systems, and human networks. Generative AI marks a significant escalation in this evolution. Unlike traditional archives—which are conservative in structure and largely passive in interface—AI systems actively reinterpret the stored past. They generate new representations, recombine fragments of culture, and provide synthesized narratives that shape how individuals and communities come to understand what is worth remembering. Memory becomes less about retention and more about relational access—the ability to prompt, query, remix, and dialogically reconstruct versions of the past on demand. However, this extension of collective memory comes with asymmetries. AI systems do not draw from cultural knowledge neutrally; they privilege the data that has been most digitized, most amplified, or most profitably indexed. Thus, every interaction with generative AI is not only an act of recollection but a negotiation over what counts as truth, relevance, or legacy.

In sum, the integration of generative artificial intelligence (AI) into knowledge construction processes signals a profound shift in how learning, memory, and collaboration are considered. From a cognitive standpoint, AI acts as an extension of the human mind—supporting memory, reducing cognitive load, and scaffolding complex thinking. However, this convenience brings risks of over-reliance and a weakening of deep, internalized knowledge structures essential for long-term learning. From a sociocultural perspective, generative AI functions as a new kind of cultural tool, mediating meaning-making within broader social and historical contexts. It enables dialogic engagement and collaborative exploration but also embeds and reproduces sociocultural assumptions present in its training data.

The role of human memory emerges as a central concern in both frameworks. Memory is not just a repository of facts but a dynamic, reconstructive process deeply tied to identity, reasoning, and meaning. As AI systems increasingly serve as external memory support, the boundaries between internal cognition and distributed, socially mediated knowledge tend to blur increasingly. Ultimately, co-constructing knowledge with AI demands not only technical fluency but also critical and reflective engagement. It calls for a balance between leveraging the capabilities of AI and preserving the uniquely human faculties of memory, judgment, and contextual understanding that ground meaningful knowledge construction.