Beyond Secondary Orality

This conceptual paper argues that generative artificial intelligence requires an extension of Walter Ong's media ecology beyond primary orality, literacy, and secondary orality. Drawing on Ong's account of how communication technologies restructure consciousness itself, the analysis identifies three assumptions embedded in his framework that no longer hold: that humans originate symbolic content, that distribution follows transparent logics, and that consciousness encounters media as an external environment. Prior scholarly attempts to address these limitations through the concept of "tertiary orality" remain insufficient because they retain the assumption of human authorship. Two extensions are proposed to address this. Algorithmic secondary orality names the condition in which humans continue to create symbolic content while algorithms increasingly determine its circulation. Tertiary algorithmicity names the condition in which algorithmic systems both curate and generate symbolic content, rendering human authorship optional at scale. Three defining characteristics of tertiary algorithmicity are developed: noetic displacement, rhetorical saturation, and existential abstraction. This shift is educationally significant because it enables what learning science identifies as unproductive success: correct academic performance without the cognitive struggle required for durable understanding. Synthesizing Ong's work with post-Ong media ecology scholarship, Stiegler's concept of tertiary retention, and research on productive failure, desirable difficulties, and cognitive load, the analysis develops pedagogical friction as a response framework: the intentional preservation of noetic, rhetorical, existential, and infrastructural resistance necessary for genuine learning in an age of frictionless generation. A productive/exclusionary friction distinction prevents the framework from reproducing inequity under the guise of rigor. The central educational problem posed by generative AI is the bypassing of interpretive labor, intellectual accountability, and the developmental conditions on which human learning depends, not academic integrity.

Keywords: tertiary algorithmicity, pedagogical friction, Walter Ong, media ecology, generative AI, unproductive success, secondary orality, productive failure

This paper was developed with the support of generative AI and AI-assisted research tools, including foundation models such as ChatGPT, Claude, and Gemini; research-support platforms such as Elicit; and custom AI-supported applications. These tools were used in a strategic but meaningful support role for brainstorming, organizational planning, literature discovery, source triangulation, summary generation, the identification of potentially relevant quotations for specific lines of argument, preliminary citation support, visualization, and sentence-level revision. They were also used to surface possible connections across sources and to test alternative formulations during the drafting process.

Because this paper argues for the educational importance of preserving human interpretive labor, intellectual accountability, and conceptual judgment, AI use was confined to support functions and did not replace the author’s responsibility for analysis, synthesis, theoretical framing, or conclusion-making. All substantive scholarly decisions, including source selection, interpretation, conceptual development, argumentation, and final written presentation, remained the responsibility of the author. All quotations, citations, and claims included in the final manuscript were reviewed and verified by the author.

Due to the use of generative AI tools, I have provided immersive digital artifacts about the specific areas of the work.

Most educational discussions on generative artificial intelligence remain organized around a binary of ban or embrace, cheat, or innovate, risk or opportunity. This framing is more convenient for policymakers and practitioners, but it obscures a more fundamental question; what changes when the symbolic environment in which people think, learn, and communicate is increasingly organized by algorithmic systems both in how content circulates as well as how content originates? This question is at the core of the present analysis, which is concerned with the pedagogical, epistemological, and policy implications that generative AI raises for cognition, teaching, and learning in K–12 education.

Media scholars have documented a recurring pattern across previous technological transitions: new media are often treated first as tools to be managed through policy compliance rather than as shifts in the conditions of thought itself. When the printing press was being adopted, early debates focused on access and censorship rather than on how print would restructure memory, enable analytical detachment, and make possible forms of reasoning unavailable to oral cultures (Ong, 1982/2002). Print reorganized cognitive operations, enabling forms of analytical reasoning and individual introspection that were structurally unavailable within oral cultures. When broadcast media began to spread there were initial concerns that were raised centered on content quality rather than on how radio and television were reshaping the relationship between individual and communal consciousness (McLuhan, 1964; Ong, 1982/2002). In each case, the deeper transformation, the reorganization of cognition itself, became fully legible to scholars and institutions only after the transition had already begun reshaping everyday life.

The present moment risks repeating this pattern. While educators debate whether students should or did use GenAI tools to write essays, the more consequential question is what happens to the interpretive processes that education depends on as part of human cognitive development when the generation of text, image, and discourse is regularly delegated to systems that operate through statistical prediction rather than experiential understanding. As Bender et al. (2021) and Floridi and Chiriatti (2020) establish, large language models function as syntactic engines that probabilistically sequence tokens without semantic grounding or communicative intent. When learners routinely delegate cognitive labor to these frictionless systems, the result may be cognitive complacency (Riva, 2025) and what learning scientists call unproductive success, correct performance without the cognitive struggle required for durable understanding (Kapur, 2016). Mollick (2024) observes that the use of generative AI in academic work is already widespread and largely undetected, a pattern corroborated by recent survey data on AI use in K–12 schools (Doss et al., 2025). The urgency is intensified by the speed of commercial scaling, since schools are being asked to respond to systems that are developing faster than educational research, policy, and professional norms can respond. The educational question, then, is not whether students are using these tools but what cognitive consequences follow from that use.

Walter Ong's media ecology ideas provide the most helpful theoretical framework for examining this question. Across his major works, Ong demonstrated that communication technologies are forces that restructure consciousness rather than neutral instruments for transmitting pre-formed thoughts. In Orality and Literacy (1982/2002), his arguments traced how the internalization of writing enabled analytical detachment, individual authorship, and abstract reasoning. These are all cognitive operations unavailable to exclusively oral cultures. In his posthumously published Language as Hermeneutic (2017), he extended this analysis to digitization, arguing that as information becomes discrete, computationally managed, and abundant, the interpretive burden on human beings intensifies rather than diminishes. By digitization, Ong means the conversion of words and information into discrete, computationally manageable units. As information becomes more abundant and easier to manipulate, the human task does not disappear; it shifts toward interpretation, evaluation, and meaning-making across an expanding field of data. Each technological transition in Ong's account altered how they organized memory, attention, and understanding, not just what people communicated.

Media ecologists have begun applying these resources to artificial intelligence. The 2024 special issue of Explorations in Media Ecology positioned generative AI tools as media within McLuhan’s extension framework (1964), arguing that reframing large language models as media enables analysis of their environmental effects, or how they reorganize symbolic conditions of thought, redistribute agency in language production, and reshape lived experience beyond questions of accuracy or use (Petricini, 2024). McLuhan’s extension framework matters here because it treats media as environments that extend and reorganize human capacities, not simply as channels that transmit preexisting messages. This 2024 reframing of generative AI tools as media is necessary, but it has not yet been integrated into a developmental account of media consciousness of the kind Ong articulated. Scholars have offered analyses of particular digital phenomena, but the field lacks a theoretical account of whether the current moment represents a further development within secondary orality or a categorically new stage.

Taken together, Ong’s developmental account of orality and literacy offers a strong theoretical position missing from current discussions of generative AI. His framework makes it possible to ask whether they inaugurate new conditions of consciousness by reorganizing memory, authorship, interpretation, and the relationship between language and experience rather than just how new technologies mediate communication. The question, then, is not simply whether generative AI extends secondary orality, but whether the algorithmic production of symbolic content itself constitutes a break significant enough to require a further developmental distinction.

Several scholars have attempted to bridge this gap by proposing the concept of “tertiary orality” to describe conditions that exceed Ong’s broadcast model. Mayer (2009) proposed a tertiary orality shaped by interactive digital networks rather than broadcast media. Heyd (2021) argued that digital linguistic practices involving human-machine interaction dissolve assumed distinctions between human and machine discourse. Ryu (2023) extended the concept into virtual reality environments. These contributions identify real features of the digital environment, including interactivity, multimodality, and human-machine dialogue, that Ong’s broadcast model could not accommodate. However, every account of tertiary orality, including Heyd’s posthumanist formulation, retains one assumption that the present analysis identifies as no longer viable; namely that the symbolic content circulating through digital networks originates with human authors. Tertiary orality describes what happens to orality when the media environment becomes digital and interactive. It does not account for what happens when the content itself is generated by systems that have never had an experience.

This paper argues that generative AI represents a categorically different condition, and that Ong’s framework requires extension to account for two technological shifts he did not anticipate. The first is algorithmic curation, meaning the condition, emerging with social media platforms in the mid-2000s, in which humans still create symbolic content but algorithms determine what content reaches which consciousness. The second is algorithmic generation, which is the condition, intensifying since the public release of large language models in late 2022, in which neural network systems produce the symbolic content itself, making human authorship optional at scale. I term these stages algorithmic secondary orality and tertiary algorithmicity, respectively. The full development of these extensions, including the criteria for treating them as distinct stages and the responses to foreseeable objections, follows in subsequent sections.

The educational stakes of this transformation are significant. If neural networks can produce convincing academic work without the cognitive labor that typically accompanies understanding, and if students present these outputs as demonstrations of their own learning, then schools risk normalizing precisely the unproductive success this paper identifies as the central educational threat. This is a problem of cognitive development rather than a problem of academic integrity. The interpretive labor that education aims to cultivate, the struggle to synthesize, evaluate, and construct meaning, can be bypassed entirely, leaving the outward appearance of learning tasks, assignments, artifacts, and projects intact while the developmental process is short-circuited.

This paper does not argue that generative artificial intelligence should be banned from educational contexts, nor does it assume that all forms of difficulty are educationally valuable. It also does not deny that generative AI can reduce exclusionary barriers, particularly for multilingual learners, students with disabilities, and students who lack access to certain forms of academic support. The argument is narrower. When generative AI allows learners to bypass the interpretive, rhetorical, and reflective labor through which understanding develops, education must intentionally preserve the forms of helpful friction that remain educationally productive, what I call pedagogical friction. In this paper, friction refers to forms of resistance, delay, difficulty, or interruption that prevent immediate completion and require cognitive, social, interpretive, or embodied engagement. By pedagogical friction, I mean the purposeful slowing down of learning at key moments so that students must engage in the interpretive, rhetorical, reflective, and embodied work through which understanding develops. Pedagogical friction therefore should not be understood as a call to make learning unnecessarily difficult. Rather, it names the deliberate protection of those forms of cognitive, dialogic, embodied, and institutional resistance that enable durable understanding while distinguishing them from barriers that reproduce inequity rather than support learning.

This inquiry is guided by three interrelated research questions:

1. What does Ong's media ecology make newly visible about algorithmic curation and generative AI as transformations of the communication environment? (Theoretical positioning)

2. What are the educational consequences when algorithmic systems do more than mediate and actually generate the symbolic environments in which students learn? (Analysis of consequences)

3. What theoretical framework can guide educators in preserving the cognitive, rhetorical, and existential conditions necessary for durable learning in an age of generative AI? (Pedagogical response)

These questions are sequenced deliberately, moving from theoretical positioning through analysis of consequences to pedagogical response.

This paper is a conceptual analysis. It extends Ong’s developmental framework through close reading of his primary texts, engagement with post-Ong scholarship in media ecology and media studies, integration of complementary theoretical vocabulary from Bernard Stiegler’s work on tertiary retention, and synthesis with research from the learning sciences on productive struggle, desirable difficulties, and cognitive load. The paper proposes a framework for understanding the current technological moment and its educational implications rather than presenting empirical findings. This conceptual framework provides the theoretical apparatus for the empirical investigation that will follow in the dissertation. The dissertation study that follows will use a mixed-methods case study design to examine how educators navigate pedagogical friction under conditions of tertiary algorithmicity.

Because this paper is a conceptual analysis rather than an empirical study, it is important to clarify how its sources function within the argument. Ong provides the primary theoretical foundation because his work explains how communication technologies reshape consciousness, memory, interpretation, and symbolic life. McLuhan (1964) and Postman (1985, 1992) represent the broader media ecology tradition within which this paper situates Ong, while Stiegler (1998, 2010, 2016) provides complementary language for understanding externalized memory, technological mediation, and the ambivalent character of technical systems.

The learning-science sources perform a different function. Kapur’s work on productive failure and unproductive success, Bjork and Bjork’s work on desirable difficulties, and Sweller’s work on cognitive load provide the educational warrant for the claim that some forms of difficulty can support durable learning. These sources do not prove the media-ecological claim that generative AI represents a new symbolic environment. Rather, they explain why the bypassing of interpretive and compositional labor matters educationally.

Scholarship on generative AI and tertiary orality performs a third function. Bender et al., Floridi and Chiriatti (2020), Dron (2023), Petricini (2024), and related scholarship help explain the technological and communicative features of the generative AI moment. The tertiary orality literature, including Mayer (2009), Heyd (2021), Turner and Allen (2013), Ángel-Botero and Alvarado-Duque (2016), Soffer (2020), Ryu (2023), and Cordón-García and Muñoz-Rico(2024), establishes that scholars have already recognized the need to extend Ong’s concept of secondary orality in digital, interactive, multimodal, and human-machine contexts. This paper builds from that literature while arguing that existing accounts of tertiary orality do not fully address the algorithmic generation of symbolic content itself.

Finally, practitioner commentary, news sources, company research, and contemporary examples are used more narrowly. They are not treated as primary theoretical warrant. Instead, they document the current technological and institutional context in which the conceptual problem has become urgent for schools. This distinction matters because the paper’s central concepts, algorithmic secondary orality, tertiary algorithmicity, and pedagogical friction, are offered as theoretical extensions grounded in existing scholarship and developed for later empirical investigation.

Epistemological Stance

My epistemological stance is rooted in critical hermeneutics, which holds that interpretation is always situated and that the tools through which we interpret are shaped by historical and technological conditions. This alignment with Ong’s core argument, that technologies of the word are never neutral but both enable and constrain, reveal, and conceal, grounds the analysis in a tradition that treats media as essential forces rather than as transparent channels. As a technoskeptical scholar, I approach this analysis with the insistence on examining what is lost as well as gained in technological transitions, and asking whose interests are served by particular configurations of technology and practice. Technoskepticism, as I apply it, is neither rejection nor uncritical adoption but an insistence on examining both costs and benefits with equal rigor.

Positionality

I write from the position of a district Director of Innovation and Technology who observes the effects of generative AI on student learning daily across five school sites. These observations are not neutral data; they are perspectival knowledge shaped by professional context and theoretical commitments. My district serves a diverse student population that includes significant numbers of English learners and students from refugee and immigrant backgrounds, and this professional context informs the framework developed in this paper. The productive/exclusionary friction distinction that I develop in Section V emerged in part from observing that the same AI tool can simultaneously reduce exclusionary barriers for multilingual students and bypass the cognitive struggle that builds durable understanding. Navigating AI policy implementation across multiple school contexts, each with different demographic compositions and instructional needs, has made visible the inadequacy of uniform approaches to AI in education and the necessity of contextual judgment that no blanket policy can provide. I name this position explicitly because critical hermeneutics demands reflexivity about the conditions of one’s own interpretation. For that reason, reflexivity appears not only as a methodological statement here but as an ongoing check on how I interpret the educational risks and possibilities of generative AI.

The paper proceeds as follows. Section II examines Ong's media ecology framework, identifies the assumptions embedded in his account of secondary orality, and distinguishes his approach from alternative theoretical traditions. Section III proposes the extensions of algorithmic secondary orality and tertiary algorithmicity, develops criteria for their status as distinct stages, and addresses foreseeable objections. Section IV turns to the educational stakes, grounding the concept of unproductive success in learning science research and examining how tertiary algorithmicity threatens the cognitive processes education depends on. Section V develops pedagogical friction as a response framework, articulating its dimensions and connecting it to the concerns of curriculum, advocacy, and policy. Section VI concludes with a synthesis of the argument, acknowledgment of limitations, and identification of questions for the empirical investigation ahead.

A. Ong's Developmental Account

To extend Ong’s framework responsibly requires first identifying what that framework actually claims and what it assumes. Ong’s stages form a theory of how dominant media restructure the noetic world: the processes and structures of knowing and thinking itself (Ong, 1982/2002). They are more than a chronology of communication technologies. Each stage transforms how human beings know, remember, and relate to one another and to knowledge itself.

In primary oral cultures, the noetic world is memory-based, communal, and situational. Knowledge must be preserved through repetition because, without writing, important ideas cannot be stored outside the human mind. This produces reliance on mnemonic devices such as rhythm, rhyme, proverbs, and formulaic expressions (Ong, 1982/2002, p. 24). Ong observes that the “oral noetic world or thought world relied upon the formulaic constitution of thought” because “knowledge, once acquired, had to be constantly repeated or it would be lost” (Ong, 1982/2002, p. 24). This also clarifies why the mechanics of reading and writing matter educationally: once internalized, they support forms of thought that move beyond memorization. Thought tends toward the additive and aggregative rather than the analytic and subordinative. This does not mean that oral cultures lack complexity. Rather, it means that thought is organized in ways suited to memory, performance, and communal transmission.

Within this oral ecology, communication is often agonistic, meaning that knowledge develops through contest, challenge, verbal exchange, and public response. Ideas are spoken, tested, remembered, and defended in the presence of others; they are not treated only as objects to be privately analyzed. Communication is also empathetic and participatory because the spoken word exists in a shared human situation. It is shaped by voice, gesture, tone, audience, and immediate social response. The spoken word exists only in the “real, existential present,” embedded in the total human situation of gesture, tone, and shared physical presence (Ong, 1982/2002, p. 99). Crucially, primary orality is a complete cognitive ecology with its own epistemological logic, not a deficit condition.

Literacy transforms this noetic world at the most fundamental level. Writing externalizes memory, making it explicit and freeing thought from the demands of constant repetition that oral cultures cannot sustain. Once words can be stored on a page, knowledge no longer depends entirely on repetition, performance, or communal memory. This makes possible forms of analytical and abstract thinking, including the detached examination of ideas. Knowledge becomes objectified in texts, meaning that words are made into objects of reflection that can be examined apart from the speaker who first produced them. In this sense, “objective” does not mean neutral or free from interpretation. It means externalized and available for rereading, critique, and revision. Thus, written knowledge can exist independently of any particular knower, enabling critique, revision, and the accumulation of learning across generations. The individual reader, encountering a text in solitude, develops a private interiority that is not possible in an oral culture. “By separating the knower from the known,” Ong writes, “writing makes possible increasingly articulate introspectivity, opening the psyche as never before to the exterior objective world quite distinct from itself” (Ong, 1982/2002, p. 105). This noetic distance is productive and enables the development of formal logic, philosophical inquiry, and scientific thought. But it comes at a cost: the existential immediacy of oral communication, its embeddedness in the human world, gives way to abstraction and distance. Each medium, as Ong traced in Interfaces of the Word (1977), biases consciousness toward particular sensory and cognitive operations while diminishing others.

Secondary orality emerges with electronic broadcast media. Radio and television retrieve characteristics of oral culture, including immediacy, participation, and communal experience, but within a fundamentally literate framework. Ong identifies a renewed “participatory mystique” in mass media audiences (Ong, 1982/2002, p. 136), yet secondary orality depends on literate infrastructure such as scripts, schedules, production notes, institutional routines, and broadcast systems. In this sense, secondary orality contains two related but distinct epistemological positions. For producers, the apparent spontaneity of radio or television is carefully planned through literate and institutional practices. For audiences, however, the experience can feel immediate, communal, and participatory, even though that participation is mediated by industrial systems rather than face-to-face exchange. It is scripted spontaneity, or a produced form of immediacy that retrieves some features of oral presence while remaining dependent on writing, planning, and mass distribution. Postman (1985, 1992) extended this insight, arguing that each medium creates its own epistemology, a way of knowing that shapes what counts as truth, argument, and evidence.

This developmental account provides media ecology’s core insight: each major shift in communication technology restructures consciousness itself. The transitions carry cognitive, social, and epistemological consequences; they are never neutral. Each stage brings capacities the previous stage lacked while diminishing capacities the previous stage sustained. This pattern of gain and loss is essential to Ong’s framework. It means that extending the framework requires asking not only what new capacities emerge but what existing capacities face erosion.

B. Ong on Digitization: Insight and Limits

Ong did not stop at secondary orality. In his posthumously published Language as Hermeneutic (2017), he turned his attention to digitization and its implications for the word. Ong recognized that computing represented something new; namely the reduction of information to discrete, manipulable units. He argued that as information becomes computationally managed and abundant, the interpretive burden on human beings intensifies. The more information available, the greater the need for hermeneutic skill, for the ability to interpret, evaluate, and assign meaning. Digitization, in Ong’s view, amplified rather than reduced the importance of human interpretation.

But Ong's analysis of computing was shaped by the technology available to him. His critique was directed at what artificial intelligence researchers now call Good Old-Fashioned AI, or GOFAI, which is rule-based, deterministic, symbolic systems that process information through explicit logical instructions. These systems, Ong observed, lack what he called "living silence," the existential depths from which human meaning emerges. A computer could manipulate symbols according to formal rules, but it could not inhabit the existential situation that gives language its meaning. Ong was making an argument parallel to Dreyfus’s (1972, 1992) phenomenological critique of artificial intelligence; that formal symbol manipulation cannot replicate the embodied, situated understanding that grounds human cognition.

This critique retains force as a philosophical observation about what computers are. However, it does not fully account for what contemporary artificial neural networks can now do. The distinction between traditional algorithms and neural network algorithms matters here. Traditional algorithms operate through explicit logical instructions written by human programmers. They are deterministic and given the same input, they produce the same output. Traditional algorithms are more transparent in principle, not always in practice. Their operations are, at least in principle, more traceable than those of contemporary neural networks. Neural networks operate on different principles. Through a process called gradient descent, they adjust millions or billions of parameters in response to training data, developing internal representations that no programmer explicitly designed. Their outputs emerge from statistical pattern approximation across a vast data set of human-generated text, image, and audio. Increasingly, however, newer systems are also trained on synthetic or machine-generated content, which further complicates the relationship between algorithmic output and human symbolic experience. The resulting systems produce language that is syntactically fluent, contextually responsive, and, in many cases, indistinguishable from human writing.

This is not a claim that large language models possess consciousness, intentionality, or understanding. The relevant observation is narrower and more consequential for education; the outputs of these systems now make up a significant and growing portion of the symbolic environment in which students learn. What AI-generated content does to human consciousness is a different question from what AI is, and it is the former question that media ecology is equipped to address. Ong’s framework was never primarily concerned with the inner life of media technologies but with how dominant media restructures the cognitive operations available to human consciousness. By that criterion, the relevant fact is not whether neural networks think but that their outputs now saturate the informational environments in which thinking occurs. Yet because these systems function as if they read, write, respond, and reason, human users often engage them as quasi-interlocutors, which creates new pedagogical demands around interpretation, attribution, and trust.

C: Prior Extensions and Their Limits: Tertiary Orality in the Literature

A disparate body of work across linguistics, media ecology, library science, and sociology has proposed what several scholars call tertiary orality, a third stage in the evolution of digitally mediated oral expression and participation; namely the media-conditioned habits of knowing, participation, and expression historically associated with oral forms (Ong, 1982/2002). This literature responds to conditions Ong’s broadcast model did not anticipate, particularly the interactivity and hybridity of digital platforms. Reviewing it is necessary because showing why existing proposals remain insufficient is part of the case for extending the framework further.

Mayer (2009) appears to have been among the first to name the concept. Where secondary orality described the consciousness fostered by radio and television, Mayer’s formulation pointed toward something qualitatively different, which is an orality shaped by the interactive circulation of digital networks. Turner and Allen (2013), working from library and information science, approached tertiary orality through the challenge that nontraditional oral documents, including video blogs, oral histories, and managerial communications, posed to institutional practices of description, storage, and retrieval. Their concern was practical; how should literate institutions accommodate artifacts that are fundamentally oral and dialogical?

Ángel-Botero and Alvarado-Duque (2016) grounded the concept empirically through a phenomenological study of digital radio listeners. Working within the media ecology tradition and drawing on Van Manen's (1990) four existentials, they identified three characteristics of tertiary orality as it operated on web radio platforms like liveliness, transcoding, and aggressiveness. Soffer (2020) offered a complementary account through voice search, arguing that voice querying represents an attempt to discipline oral expression through the cognitive constraints of anticipated textual form. The result, in the analysis, is a hybrid communicative mode that cannot be understood as fully oral or fully literate, and that sits in permanent tension between the two.

Heyd (2021) made the most direct theoretical argument for tertiary orality as a distinct category, observing that oral practices have entered the realm of human-machine interaction. People now talk to machines and not only through them, a posthumanist dimension that, in her account, distinguishes tertiary from secondary orality. Ryu (2023) extended the concept into virtual reality, and Cordón-García and Muñoz-Rico (2024) offered a comprehensive synthesis identifying platforms like TikTok and YouTube as amplifiers of participatory characteristics that dissolve traditional distinctions between sender and receiver.

This body of work demonstrates that the inadequacy of Ong’s secondary orality as a framework for digital media is recognized across multiple scholarly traditions. It identifies real features of the digital environment, including interactivity, multimodality, human-machine dialogue, and platform-mediated participation, that Ong’s broadcast model could not accommodate. Yet the concept of tertiary orality, across all its formulations, maintains one assumption that the present analysis identifies as no longer viable; the symbolic content circulating through digital networks originates with human authors.

Every account reviewed here treats the content as fundamentally human in origin. Users produce posts, videos, and voice queries. Professionals create journalism and entertainment. The machines may respond, retrieve, or transform, but the symbolic content that anchors these interactions originates in human minds. Tertiary orality describes what happens to orality when the media environment becomes digital and interactive. It does not account for what happens when the content itself is generated by systems that have never had an experience. This is the threshold that generative AI crosses, and it requires a reconceptualization of the axis along which the stages are organized. The more fundamental distinction in the current media environment is not between speech and writing, but between human and algorithmic origination of symbolic content.

D. Three Assumptions That No Longer Hold

The case for extending Ong’s framework rests on identifying the assumptions embedded in his account of secondary orality that no longer describe the current media environment. These assumptions function as the analytical hinge of the entire argument. If they hold, then contemporary digital communication is no more than an intensification of secondary orality; if they break, then new conceptual categories are required. Exploring the three assumptions provides enough evidence to argue for an extension of Ong’s framework.

The first assumption is that humans create symbolic content. Across every stage Ong described, the source of symbolic expression remained human consciousness. Oral performers composed and delivered. Authors wrote. Broadcast producers scripted and filmed. The media changed; the human origin of the content did not. Generative AI disrupts this assumption directly. Neural network systems trained on human-generated data can now produce essays, analyses, images, and arguments without human authorship in any traditional sense. The outputs are not transcriptions of human thought; albeit they are trained on vast human-created data sets. They are statistical predictions of what plausible text looks like, generated by systems that have never experienced what they write about. For education, this means that a student can submit work that reads as a demonstration of understanding without any understanding having occurred in the production process.

The second assumption is that distribution follows relatively transparent logics. In primary orality, distribution was face-to-face. In literate cultures, publication processes and editorial decisions operated through identifiable human judgment. In secondary orality, broadcast schedules determined what content reached audiences, and two viewers watching the same channel at the same time saw the same content. Algorithmic curation disrupts this assumption. On social media platforms, distribution is governed by engagement-optimization algorithms that are proprietary, opaque, and dynamically responsive to individual user behavior. Two users opening the same application encounter different content. The shared symbolic environment that characterized broadcast-era secondary orality fragments into individualized information streams. For education, this means that students arrive in classrooms having inhabited different informational worlds, their sense of what makes up common knowledge is shaped by algorithms optimized for engagement rather than accuracy or understanding.

The third assumption is that consciousness encounters media as an external environment. Ong’s framework treats media technologies as environments that consciousness inhabits and is shaped by, but that remain external to consciousness itself. The book sits on the shelf. The broadcast emanates from the television. Readers and viewers certainly select among available texts, programs, courses and communities so media environments have never been neutral nor entirely external. Oral cultures, print cultures, and broadcast cultures all involved patterns of selection, reinforcement, and social belonging.

What changes under algorithmic mediation is the scale, speed, and automation of that selection. Dynamically personalized algorithmic environments complicate this distinction. When the symbolic environment is continuously tailored in response to prior user behavior, consciousness increasingly encounters an environment that reflects its own patterns back to it. The symbolic world a learner encounters is increasingly shaped by systems that infer attention, predict preference, and return those patterns back as a self-reinforcing algorithmic loop that is difficult to break from (Stiegler, 2012; Kaluza, 2022; Rickert, 2024). For education, this creates a challenge; students develop within informational environments that reinforce existing patterns of attention and interest rather than introducing the productive disequilibrium that constructivist learning theory identifies as necessary for cognitive growth (Nikolaidis & Lara-Steidel, 2024; Robson, 2019).

These three assumptions are the analytical hinge of the proposed extensions and have been ruptured. They have been transformed by algorithmic curation and generative AI. If Ong's framework depended on conditions that no longer apply, then the framework requires extension instead of just an application to new circumstances. The next section develops the specific form these extensions may could take. The case for their necessity rests on the claim that the current media environment has transformed the conditions under which Ong’s account of secondary orality remained coherent.

E. Alternative Frameworks and the Case for Ong

The decision to ground this analysis in Ong's media ecology rather than in other available theoretical traditions requires justification. Several frameworks address aspects of the current media environment, and the proposed extensions draw on some complementary resources. But Ong's framework offers something the alternatives do not, which is a developmental account of how communication technologies restructure consciousness, with sufficient historical depth to accommodate the stage extensions this research proposes.

Bolter and Grusin’s (1999) concept of remediation captures how new media incorporate and transform earlier media forms, but its analytical focus remains on media-to-media relations rather than on consciousness transformation. This helps explain how generative AI remediates the essay, the search engine, the tutor, and the encyclopedia. Yet it does not fully explain what happens to student thinking when symbolic production itself becomes algorithmically generated.

Kittler’s (1990, 1999) media archaeology provides a different and important contribution. His work directs attention to the material, technical, and institutional conditions under which discourse is produced. This emphasis is especially relevant to the present study because algorithmic gatekeeping is infrastructural, cognitive, and pedagogical. Platform design, data systems, recommendation algorithms, access controls, and institutional policy shape what learners encounter and what forms of knowledge become visible. In this sense, Kittler helps clarify the dimension of infrastructural friction developed in Section V.

However, the main driving question in this research is concerns how generative AI reshapes the conditions of learning, interpretation, authorship, and consciousness. For that reason, Ong remains the focus of the theoretical anchor. Ong’s account of media as transformations of consciousness provides the conceptual structure needed to ask what happens to learners when symbolic production can be automated. Kittler is important, but as a complementary framework. He helps explain the technical and institutional conditions through which algorithmic environments operate, while Ong helps explain why those environments matter pedagogically for thought, interpretation, and learning.

Hayles’s (2012, 2017) work on technogenesis and cognitive assemblages shares Ong’s concern with how technology reshapes thought, particularly the balance between “deep attention” and “hyper attention.” However, her orientation toward distributed cognition does not support the hermeneutic dimension central to Ong’s project and to the educational questions this paper addresses. In other words, Hayles helps describe how cognition becomes distributed, but she does not center interpretation as accountable for meaning making, whereas the hermeneutic labor this paper argues is at risk of being delegated to generative systems. Since the present analysis is concerned with what happens to interpretive labor when that labor can be delegated to algorithms, Ong’s hermeneutic emphasis provides a better fit.

Stiegler’s (2010) concept of tertiary retention provides the most useful complementary vocabulary. Extending Husserl’s phenomenology, Stiegler described how technical objects serve as external supports for memory, forming a third form of retention alongside primary retention, which is the flow of immediate experience, and secondary retention, or recollection. A book, a photograph, a recording are each a technical object that preserves traces of experience that can be reactivated by consciousness. Stiegler argued that these technical supports are essential parts of the cognitive processes they support more than just supplements to memory.

Contemporary artificial neural networks used in generative AI complicate this relation. Traditional tertiary retentions preserved traces of human or material experience like a book retained the symbolic labor of an author, and a photograph retained an optical registration of an event. Generative AI systems, by contrast, produce outputs that take the form of tertiary retentions, including texts, images, audio, and video, without originating in lived experience. The system simulates the traces of experience by recombining statistical patterns learned from prior human and, increasingly, machine-generated data. This makes algorithmic generation not only phenomenologically different from prior forms of retention, but epistemologically different: it changes the grounds on which a learner can judge authorship, evidence, and warranted meaning.

However, Stiegler’s phenomenological and epistemological approach lacks a strong connection to developmental stages that Ong’s framework provides. The proposed approach, then, is to work within Ong’s developmental framework while drawing on Stiegler’s concept of tertiary retention as a complementary analytical tool. The extensions developed in Section III are extensions of Ong. The analysis of what those extensions mean for the relationship between consciousness and algorithmically generated content draws on Stiegler. The two traditions support distinct aspects of the same transformation. Figure 1 summarizes the developmental sequence used in this paper and identifies where the two proposed extensions enter Ong’s media ecology framework.

Note. The figure represents an analytical sequence rather than a claim that one media condition fully replaces another. These stages coexist unevenly in contemporary classrooms and communication environments.

This comparison clarifies the specific contribution of the present framework. Algorithmic secondary orality is more than digital communication. It names the condition in which human beings continue to produce symbolic content while opaque algorithmic systems increasingly determine what reaches consciousness. Tertiary algorithmicity names a further shift. Algorithmic systems no longer only distribute symbolic content. They also generate it, making human authorship optional at scale. The educational problem therefore shifts from the curation of attention to the bypassing of symbolic labor itself.

A. Algorithmic Secondary Orality

Before addressing generative AI directly, the framework requires an intermediate stage that accounts for the transformation of the media environment between Ong’s broadcast-era secondary orality and the current moment. The rise of social media platforms, recommendation engines, and algorithmic feeds represents a significant restructuring of the relationship between consciousness and the symbolic environment (Stiegler, 2016; Bucher, 2018; van Dijck, 2013), yet scholars have often placed this transformation too quickly into Ong’s existing categories (Foley, 2012). I name this transitional phase algorithmic secondary orality. This phase dates roughly from the mid-2000s through the early 2020s. During this period, humans continued to create symbolic content, but algorithms increasingly determined what content reached which consciousness. This does not mean that algorithms acted apart from human intention or institutional interest. Platform algorithms were designed, owned, tuned, and governed by organizations whose incentives included engagement, retention, advertising value, behavioral prediction, and market advantage. This extends what Zuboff (2019) describes as surveillance capitalism, in which communicative activity becomes both the raw material and the product of an attention-extractive system. This paper acknowledges that political-economic frame but does not make it the primary analytic lens; its focus remains media-ecological and educational. Algorithmic secondary orality, therefore, marks a departure from Ong’s stages because the structuring of forces governing symbolic exchange shifted from shared cultural logics to personalized, opaque, and behaviorally optimized computational processes.

Ong's secondary orality presumed a broadcast architecture. Radio and television operated on a one-to-many logic with shared temporality. Relatively few producers transmitted content to mass audiences through channels whose scheduling and editorial decisions remained broadly visible. Millions watched the same program simultaneously. A teacher could reasonably assume that students in the same community had access to the same evening news, the same cultural programming, the same informational baseline. The shared symbolic environment was imperfect and unevenly accessed, but it was structurally shared. The algorithmic platforms that increasingly organize contemporary symbolic environments were built in the first two decades of the twenty-first century and disrupted Ong’s idea of secondary orality at a fundamental level, leading to a more precise algorithmic secondary orality. Facebook, YouTube, Twitter, Instagram, and TikTok still promoted human-created content, but they placed algorithmic systems between creation and reception (Bucher, 2012). The feed replaced the schedule.

This transition began on September 5, 2006, with the launch of the Facebook News Feed, which reorganized the symbolic environment by using algorithms to curate a centralized stream of information based on user relevance rather than strict chronological publication (boyd & Ellison, 2007). Where broadcast media provided a fixed sequence of content shared across audiences, algorithmic platforms provided dynamically personalized streams in which two users opening the same application at the same moment encountered different content, sequenced according to computational predictions about what would capture and retain their attention (Pariser, 2011). The editorial function shifted from human gatekeepers operating through identifiable judgment to computational systems designed by human institutions but optimized through proprietary and opaque processes for engagement metrics (Tufekci, 2017; Wu, 2016; Zuboff, 2019).

Several features justify this stage’s continued association with the category of secondary orality. Content creation remained human because users produced posts, videos, and commentary, and professional media organizations continued to create journalism and entertainment. The literate substructure of consciousness persisted since users read, wrote, and engaged in textual analysis. And the participatory and communal qualities Ong identified in broadcast audiences found amplified expression in the interactive dynamics of social media, where the “participatory mystique” (Ong, 1982/2002, p. 133) intensified through interaction, response, and community formation at a scale broadcast could not support.

Yet crucial differences require treating this as a distinct stage. The fragmentation of the shared symbolic environment is the most educationally consequential. Research on algorithmic personalization suggests a complex landscape where some users encounter intersecting views (Bakshy et al., 2015; Bruns, 2017), while platform interfaces and cognitive constraints can nudge users toward reinforcement of existing dispositions (González-Bailón et al., 2023; Levy, 2021). The result is a condition in which the informational environment shaping a given consciousness is partially determined by that consciousness’s prior behavior (Nguyen, 2020; Vosoughi et al., 2018). Additionally, engagement-driven ranking can magnify divisive or emotionally charged content (Brady et al., 2017; Vosoughi et al., 2018), and the feedback loops between user behavior and algorithmic ranking create conditions for behavioral shaping that serve commercial rather than developmental objectives (Zuboff, 2019; Stiegler, 2016).

Algorithmic secondary orality, therefore, represents the first necessary extension of Ong’s framework. It names a condition in which the media environment transformed significantly while the human origin of symbolic content remained intact. Understanding this intermediate stage is important for grasping what distinguishes the condition that follows. If algorithmic secondary orality changes who sees human expression, tertiary algorithmicity changes whether human expression is necessary at all.

B. Tertiary Algorithmicity

This paper argues that generative AI introduces a categorically new condition. Large language models, image and video generators, and multimodal systems can now produce text, code, images, video, and analyses that are, in many contexts, indistinguishable from human-authored work (Heyd, 2021; Petricini et al., 2025). I term this condition tertiary algorithmicity, a media environment in which algorithmic systems both create and curate symbolic content, making human authorship optional at scale (Dron, 2023). Humans are not absent from this system; their role shifts from primary producers of symbolic content to prompt initiator, consumer, evaluator, editor, validator, data source, and institutional actor. The human role is reorganized. The learner may still be active in the workflow, but activity is not the same as the generative cognitive labor through which understanding develops. The cognitive and educational consequences of that shift are developed in the discussion of noetic displacement that follows.

The shift from algorithmic secondary orality to tertiary algorithmicity is a qualitative rupture rather than a quantitative intensification. In algorithmic secondary orality, humans created content and algorithms distributed it. In tertiary algorithmicity, algorithms create and humans may consume, evaluate, or simply accept without ever engaging in the generative cognitive labor that producing the content would have required (Dron, 2023; Stiegler, 2016). The cognitive operations that Ong (1982/2002) identified as central to literate consciousness, including synthesis, analysis, composition, and argumentation, can now be performed by systems external to the learner.

The term choice requires justification. I use “tertiary” to indicate a third major transformation in the relationship between consciousness and the symbolic environment. But I choose “algorithmicity” rather than “orality” because the oral-literate continuum no longer captures the primary transformation that is underway. For the developmental analysis advanced here, whether algorithmically generated content takes oral or literate form matters less than whether a human or an algorithm produced it. A podcast episode and a written essay differ in sensory modality, but if both are generated by AI systems, they share a more fundamental characteristic that cuts across the oral-literate distinction. Tertiary algorithmicity names a condition defined by the algorithmic source of the symbolic environment, regardless of the modality through which consciousness encounters it. The term also acknowledges Stiegler’s concept of tertiary retention, introduced in Section II, though generative AI complicates that framework by producing outputs that simulate the form of preserved experience without originating in any experience.

Baudrillard's (1994) account of simulation provides a structural parallel from a different theoretical tradition that clarifies what makes this threshold categorically distinct. Baudrillard proposed four successive phases in the assumed relationship between the image and reality, which are 1) the image as faithful reflection, 2) the image as mask that distorts reality, 3) the image as mask concealing that no reality exists behind it, and 4) finally the image as pure simulacrum bearing no relation to any reality whatsoever (pp. 6–7). This progression tracks a different object than Ong's stages, but the structural logic converges. Where Ong traces the transformation of consciousness through changes in the dominant medium of symbolic production, Baudrillard traces the progressive detachment of symbolic forms from any referential ground. The two frameworks intersect at tertiary algorithmicity, which describes both a new stage in Ong's developmental sequence and a condition that accelerates the fourth phase of Baudrillard's image typology; which is an environment increasingly populated by symbolic forms that function as if they referred to grounded claims, while originating in statistical pattern completion rather than human cognition.

The convergence is instructive but not comprehensive. Baudrillard's analysis ultimately concludes that the real has been entirely supplanted by simulation, leaving no ground from which to mount a critical response (pp. 159–164). The present framework does not follow him to that conclusion. If Baudrillard's diagnosis were accepted without qualification, the pedagogical response developed in Section V would be incoherent, since pedagogical friction presupposes that meaningful cognitive labor remains possible and educationally necessary even within a heavily simulated symbolic environment. The distinction matters because Baudrillard is useful here as someone who has a diagnosis of the structural condition, not as a support for the educational response. The framework draws on his account of how symbolic forms detach from referential grounding while maintaining, with Ong (1982/2002) and Postman (1992), that purposeful educational intervention in media environments remains both possible and necessary.

This distinction sharpens the definition of tertiary algorithmicity itself. If simulation, in Baudrillard's terms, is the generation by models of a real without origin (p. 1), then tertiary algorithmicity names the specific media ecological condition in which that generation becomes infrastructural rather than exceptional. The outputs of generative AI present as knowledge while severing symbolic form from the experiential and interpretive labor that once anchored it. They substitute for the noetic work through which knowers historically produced and warranted meaning rather than just circulate faster or farther than prior media artifacts. This is not a claim that generative AI creates a total hyperreality. It is a claim that the symbolic environment now includes, at increasing scale, texts that function socially as grounded claims even when no consciousness has done the grounding. There are three interrelated characteristics that define tertiary algorithmicity as a distinct media ecological condition. Each connects with one of the three assumptions identified in Section II as no longer holding, and each creates specific consequences for the educational processes examined in Section IV. To start, we will discuss noetic displacement.

Noetic Displacement

Noetic displacement names the condition in which the externalization of cognitive labor moves from storage to generation, and in which generative and interpretive operations that previously had to occur within the learner’s own noetic activity are increasingly offloaded to an external socio-technical system. The externalization that began with literacy, the movement of memory from inside the mind to the surface of the page (Ong, 1982/2002, pp. 78–79), extended through secondary orality’s broadcast storage and algorithmic secondary orality’s curated retrieval. In each case, the externalization involved the preservation and distribution of what humans had already thought, and the human knower still had to do the work of producing and owning meaning. Tertiary algorithmicity crosses a threshold because GenAI platforms and tools produce new combinations of language that take the form of knowledge without the experiential grounding that has historically accompanied symbolic production (Ong, 1982/2002, pp. 43–44). In this sense, displacement is not only the erosion of an internal capacity over time but the structural substitution of an external generator for the internal processes education seeks to cultivate.

For education, noetic displacement poses an immediate practical question. When a student can generate a competent analysis of a poem, a historical event, or a scientific process without having engaged in the interpretive labor that such analysis traditionally required, what has the student learned? The cognitive operations that education aims to develop, including synthesis, argumentation, and the construction of meaning from evidence, can be performed by systems external to the learner. The first assumption identified in Section II, that humans create symbolic content, ruptures here because the cognitive work that would have produced the text is increasingly performed in the model-and-interface stack of the GenAI platform rather than in the student’s own noetic activity. Bozkurt (2025) identifies a related concern, arguing that generative AI risks producing “intellectual modal collapse” in which the range of student cognitive engagement narrows rather than expands. Where Bozkurt identifies this threat from within critical pedagogy, the present analysis locates it within the media ecological transformation of the relationship between consciousness and the symbolic environment.

Rhetorical Saturation

Rhetorical saturation is the condition in which algorithmic systems produce discourse at scale, flooding communicative environments with synthetic text, image, video, and audio that is increasingly indistinguishable from human production. That means the origin of symbolic content becomes structurally uncertain, the markers by which consciousness has historically distinguished human from mechanical expression lose their reliability, and the dialogic exchanges available to participants in such an environment increasingly lack the genuine contestation that Ong (1982/2002, pp. 43–45) identified as characteristic of meaningful discourse. Rhetorical saturation concerns both origin and exchange: uncertainty about where symbolic content comes from and uncertainty about whether one is engaged in genuine intellectual contestation or simulated response. This raises the question, is this interaction meaningful discourse?

Three aspects support this condition and the first is origin uncertainty. Under tertiary algorithmicity, any text, image, video, or audio might be algorithmically generated, and research suggests that humans struggle to distinguish AI-generated text from human writing at rates close to random chance (Jakesch et al., 2023; Clark et al., 2021; Chein et al., 2024). The second is the simulation of expression, wherein generative systems trained on human outputs learn to mimic distinctive style, voice, and apparent personality (Serapio-García et al., 2023), eroding the signals by which consciousness has historically distinguished human from mechanical communication (Jakesch et al., 2019; Mieczkowski et al., 2021). The third is the simulation of dialogue; conversational AI systems now simulate exchange with remarkable fluency (Jones et al., 2024), yet these interactions lack what Ong would recognize as genuine intersubjectivity. The system produces statistically plausible responses rather than grounded communication (Bender et al., 2021; Floridi & Chiriatti, 2020). The result is a communicative environment that can simulate responsiveness without requiring genuine contestation from an accountable interlocutor.

These three features describe what is being produced and circulated; together they also reshape the conditions of exchange itself. When origin uncertainty, simulated expression, and simulated dialogue converge, the result is rhetorical saturation. This is a communicative environment in which the volume of synthetic discourse is high, the ability to distinguish it from human discourse is low, and the quality of available interlocution degrades. The second assumption breaks here. The content being curated and the dialogue being conducted may themselves be algorithmically generated. For education, rhetorical saturation means that the environment in which students develop their capacity for critical evaluation and genuine intellectual exchange is increasingly populated by content and interlocutors whose origins are unknowable. This problem is not limited to uncertainty about the origin of content. It also changes the nature of exchange. Simulated interlocutors can be infinitely patient, immediately responsive, and rarely resistant in the way a real human interlocutor may be. As a result, learners may experience the surface form of dialogue without the productive friction that comes from disagreement, misunderstanding, challenge, or accountability to another person (Turkle, 2011). Without the agonistic contestation of a real human other, learners risk bypassing one form of cognitive and rhetorical struggle required for durable understanding (Kapur, 2016).

Existential Abstraction

Existential abstraction names the condition in which symbolic production is severed from lived experience, situated perspective, and personal accountability. Text generated by an artificial neural network is disconnected from the existential situation that Ong (1982/2002) identified as the ground of genuine communication. It expresses positions without having stakes. It simulates conviction without being committed, operating without the communicative intent that anchors human speech to reality (Bender & Koller, 2020; Frankfurt, 2005). The connection between claim and claimant, which persisted across every stage Ong described, is therefore weakened rather than eliminated. Under tertiary algorithmicity, students may encounter claims that sound authored, situated, and accountable while lacking the human presence that traditionally made those qualities meaningful.

Ong traced a progression from orality’s inseparability of speaker and speech, through literacy’s introduction of authorial distance, to secondary orality’s partial restoration of dialogic qualities. Tertiary algorithmicity extends this trajectory not to distance but to elimination. Text can now be produced without an author in any experiential sense. No consciousness stands behind the output. No person bears the intellectual risk of having committed to the claims the text advances. The third assumption identified in Section II, that consciousness encounters media as an external environment, breaks here because the symbolic environment is now partially populated by expressions that originated in no consciousness, yet take the form of personally invested human communication. Bozkurt (2024) argues that this dissolution of authorship requires fundamental reconsideration of cocreation and ownership in academic contexts. The present framework extends this concern beyond academic integrity. When text can be produced without intellectual commitment, the formation of intellectual character that authorship has historically demanded loses its primary mechanism of accountability.

For education, existential abstraction means that students can produce without investing, performing competence without the vulnerability that accompanies genuine authorship. A student who submits AI-generated work is not accountable for the ideas in the way that a student who composed the work would be. Education depends on students developing the willingness to take intellectual risks and accept responsibility for their thinking. When that commitment becomes optional, one of the primary mechanisms through which intellectual character forms is undermined.

Noetic displacement, rhetorical saturation, and existential abstraction are operationally intertwined, but the aspects can be identified separately. Each names a different way the learner as a person is repositioned under tertiary algorithmicity, including as thinker, as interlocutor, and as accountable claimant. A student who submits AI-generated work experiences all three simultaneously. Cognitive labor is displaced, the output circulates within a saturated environment where its origin is uncertain, and the connection between the text and personal intellectual commitment is abstracted away. Yet each identifies a distinct dimension of the transformation. Noetic displacement concerns cognition; rhetorical saturation concerns the communicative environment, and existential abstraction concerns the person who stands behind, or is separated from, the claim. Together, these three characteristics form the media-ecological condition this framework terms tertiary algorithmicity.

C. Teleology, Counterarguments, and the Case for Intervention

Stage models carry an inherent risk because arranging historical developments in a sequence can imply that each transition is inevitable, represents progress, and supersedes prior stages. This teleological reading must be explicitly resisted (Smith & Marx, 1994). Ong himself resisted it. His account consistently emphasized that each stage brought losses as well as gains. Literacy enabled analytical detachment but diminished the communal and mnemonic capacities of oral cultures (Ong, 1982/2002, pp. 104–105). Secondary orality retrieved communal qualities but within a framework subject to broadcast economics (Ong, 1982/2002, pp. 135–136). The proposed extensions follow this pattern; algorithmic secondary orality enabled participatory expression at unprecedented scale but fragmented the shared symbolic environment (van Dijck, 2013; Zuboff, 2019). Tertiary algorithmicity offers powerful tools for synthesis and production but threatens to bypass the cognitive labor through which understanding develops (Stiegler, 2016). Neither stage is simply better or worse than its predecessor, and as developed further in Section VI, the stages coexist within contemporary communication environments rather than fully replacing one another.

Three objections to the proposed extensions deserve direct response. The first holds that algorithmic secondary orality and tertiary algorithmicity are simply further developments within secondary orality rather than distinct stages. This objection underestimates the significance of the shifts involved. Ong distinguished his stages by qualitative transformations in how consciousness relates to the symbolic environment, not by the introduction of recent technologies within a stable relationship (Ong, 1982/2002, pp. 77–78). The transition from broadcast to algorithmic curation represents such a qualitative transformation that the shared symbolic environment fragments, the editorial function becomes opaque and commercially driven, and feedback loops create a dynamically responsive environment that Ong’s broadcast model did not anticipate. The transition from human creation to algorithmic creation represents a still more fundamental rupture; the human origin of symbolic content, which connected all prior stages in Ong’s account, dissolves. These are structural changes in the conditions under which consciousness is shaped, not incidental variations within a stable paradigm.

The second objection is that the framework may sound technologically deterministic, as if generative AI inevitably produces particular educational outcomes. That is not the claim. Following Ong (1982/2002), McLuhan (1964), and later critiques of determinism (Feenberg, 2002; Smith & Marx, 1994), this framework treats technologies as environments of tendency and affordance rather than as forces that mechanically determine human action. Technologies do not determine outcomes, but they create conditions that make certain outcomes more likely and others less so (Postman, 1992). A student who decides not to use generative AI still inhabits a media environment saturated with algorithmically generated content. A teacher who bans AI tools in the classroom cannot ban the cognitive habits students develop through interactions with these systems outside it. Acknowledging that technologies create conditions under which consciousness develops is not technological determinism. It is the prerequisite for any intervention, because educational responses can only be designed once the conditions they aim to address have been named (Hutchby, 2001)

A third potential objection emerges from the theoretical tradition on which this section has been drawn. If Baudrillard (1994) is correct that simulation has reached a point where "there is no more hope for meaning" (p. 164) and that resistance is always already absorbed by the system it opposes (pp. 163–164), then any proposed educational intervention, including the pedagogical friction framework developed in Section V, is naive at best and counterproductive at worst. On this reading, friction would simply be reabsorbed as another simulation of educational rigor, another set of forms that persist after the cognitive substance has been displaced. Other postmodern and poststructuralist thinkers, notably Haraway (1991), have read the destabilization of fixed referents as opening emancipatory possibilities rather than only loss. The position taken here neither denies nor takes up that reading; it argues that within the specific institutional context of K–12 schooling, the educational case for preserving interpretive labor remains compelling on learning-science grounds.

This objection carries real force, and the framework does not dismiss it entirely. The risk that pedagogical friction could become performative is genuine. Institutions could adopt the language of cognitive resistance while continuing to optimize for efficiency, market alignment, measurable productivity, and consumer satisfaction. This risk is not incidental. It reflects broader political-economic pressures that increasingly position education as a market service and learners as consumers of optimized experiences rather than as participants in a demanding intellectual community. This paper acknowledges that political-economic dimension but does not make it the primary analytic frame. The narrower claim here is educational, namely if schools adopt the rhetoric of friction while removing the conditions for real intellectual struggle, pedagogical friction becomes another managerial slogan rather than a framework for learning.

However, the objection depends on accepting Baudrillard's totalization of the simulacrum, which the present framework does not. The purpose of cultivating such learners is more than academic performance. Schools need learners capable of interpretation, judgment, dialogue, intellectual responsibility, and participation in shared democratic and professional life. Durable understanding matters because students are not only completing tasks; they are becoming people who can evaluate claims, revise their thinking, explain their reasoning, and act responsibly in environments where fluent output can be manufactured without understanding. Learning science research demonstrates that specific interventions, such as desirable difficulties (Bjork & Bjork, 2011), productive failure designs (Kapur, 2016), and deliberate cognitive load management (Sweller, 2011), produce measurable differences in durable understanding. These are more than simulations of learning and are empirically documented mechanisms through which learners construct transferable knowledge. The framework proposed here is grounded in that empirical tradition, not in Baudrillard's philosophical nihilism. It takes his diagnosis seriously while refusing his prognosis.

This observation connects to the next section. If the transitions described here were inevitable progress, educators could only adapt. If they represent contingent shifts with both costs and benefits, then intervention is possible. The question becomes what forms of intervention can preserve and cultivate the cognitive processes that education both depends on and seeks to develop within a media environment increasingly organized around their bypass? This is the question of pedagogical friction.

If tertiary algorithmicity represents the media environment in which contemporary students increasingly operate, educators must confront a specific question; what does learning require in such an environment, and what happens when the environment's default tendencies work against those requirements? This section argues that decades of research in the learning sciences establish cognitive struggle as a form of durable learning, and that tertiary algorithmicity's defining feature, the capacity to generate competent symbolic output without human cognitive labor, threatens the processes through which understanding develops.

A. Learning Science Foundation: The Necessity of Productive Struggle

The constructivist tradition in educational theory, running from Dewey through Piaget to Vygotsky, converges on a finding that carries particular weight for this argument. There is an agreement that durable learning requires enough cognitive struggle to grapple with new topics, content, and skills in order to gain mastery. Growth occurs when existing schemas prove inadequate and must be reconstructed through effortful engagement with problems the learner cannot yet solve. Dewey (1933) argued that reflective thought is impossible without a problematic situation, a condition of uncertainty that arrests automatic action and forces genuine inquiry; the problematic situation cannot be bypassed without also bypassing the thinking it generates. Piaget (1952) identified accommodation as the driving force of cognitive growth, which can occur when assimilation fails and the mind restructures itself to account for information that does not fit existing categories. Vygotsky (1978) located learning at the boundary between what learners can do independently and what they cannot yet do alone, the zone of proximal development, where the learner is stretched beyond current competence but supported by a more knowledgeable other that scaffolds the learner’s own cognitive effort. This synthesis emphasizes cognitive dimensions of the constructivist tradition; the social and cultural dimensions central to Vygotsky's account inform the analysis but are not its primary focus here.

These foundational insights have been operationalized through several related lines of contemporary research. Bjork and Bjork (2011) coined the term "desirable difficulties" to describe instructional conditions that slow initial performance but enhance long-term retention and transfer. The generation effect is particularly relevant here since producing an answer, even an incorrect one, creates stronger memory connections and deeper schema development than receiving a correct answer passively. Karpicke and Blunt (2011) demonstrated through research on retrieval practice that the cognitive effort of recalling information strengthens memory in ways that rereading and passive review cannot replicate. The difficulty of retrieval is part of the mechanism through which learning becomes durable; in other words this is a feature and not an obstacle.

Kapur’s (2008, 2015, 2016) framework of productive failure provides the most directly relevant lens for evaluating the educational implications of generative AI. Students who attempted to generate solutions to novel problems before receiving instruction outperformed those who received direct instruction first, particularly on transfer tasks. The mechanism involves activation of prior knowledge, awareness of knowledge gaps, and the creation of what Kapur calls “preparation for future learning” (Kapur, 2008, p. 380), the cognitive readiness that makes subsequent instruction meaningful. Kapur’s taxonomy distinguishes four quadrants: productive success, meaning correct performance with genuine understanding, productive failure, which is incorrect performance that leads to understanding through struggle, unproductive failure, identified as incorrect performance without learning, typically when the task exceeds the learner’s capacity), and unproductive success, meaning correct performance without underlying understanding. It is this fourth quadrant that names the educational threat of tertiary algorithmicity.

Sweller’s (2011) Cognitive Load Theory provides a complementary analytical framework. Sweller distinguishes extraneous load, meaning where cognitive demands imposed by poor design, which should be minimized; from intrinsic load, or the inherent complexity of material, which cannot be eliminated; and finally germane load where cognitive effort is directed toward schema construction. Effective instruction manages extraneous load while preserving germane load, because germane load is the cognitive work through which learning actually occurs. These findings collectively establish the principle that becomes urgent under conditions of tertiary algorithmicity; namely healthy friction is an essential part of learning. Cognitive effort directed toward understanding is the mechanism through which understanding develops. Remove the effort, and you may remove the development, even when the output looks correct. This concern aligns with Wolf's (2018) account of the deep reading brain, in which the cognitive labor of slow, recursive engagement with text builds capacities that surface-level processing cannot.

B. Unproductive Success in the Context of Generative AI

Generative AI makes Kapur’s concept of unproductive success newly urgent because it can produce fluent, plausible, and often correct academic performances without requiring the learner to engage in the interpretive and compositional processes through which understanding is built. A student may generate the right answer, the polished paragraph, or the coherent summary while bypassing the struggle to organize prior knowledge, test a claim, revise a misconception, or construct a schema that can transfer to future tasks. The educational danger is not only that students may submit AI-generated work as their own. That concern matters, but it frames the problem too narrowly as academic integrity. The deeper problem is developmental. Under conditions of tertiary algorithmicity, the artifact of learning can remain visible while the process of learning becomes optional.

Generative AI complicates the cognitive load balance in a way that the learning science literature has not yet fully addressed. It can reduce extraneous load in areas like formatting citations, organizing notes, cleaning up syntax, and managing the mechanical aspects of academic production. This reduction is legitimate and, in many cases, pedagogically beneficial. But generative AI can also reduce germane load, the cognitive effort directed toward schema construction. When a student prompts a language model to produce an analysis of a historical event or scientific process, the system generates output that takes the form of competent analysis. The student may review, edit, and submit this output. But the cognitive labor that the analysis would have required, the struggle to identify relevant evidence, construct an argument, and synthesize sources into a coherent position, has been performed by the system rather than by the student. The absence of understanding is difficult to detect from the product alone because the product looks like the product of understanding. The traditional educational assumption that quality of product reflects quality of process breaks down when the production process can be delegated to a system that produces quality without cognition.

Kapur's (2016) framework identifies unproductive success as a cognitive phenomenon, a failure of schema construction masked by correct output. But the persistence of unproductive success in educational institutions requires a structural explanation as well. Baudrillard's (1981/1994) analysis of what he called the "phantom scenario of pedagogy" in the university provides one. He described a condition in which the exchange of knowledge between teachers and students had become "nothing but a doubled collusion of bitterness and indifference," where diplomas circulated as academic currency detached from any counterpart in genuine intellectual labor (p. 155). The institutional forms of education, including assignments, evaluations, and credentials, continued to function even after the cognitive substance they were designed to represent had been displaced.

This structural analysis anticipates the specific danger that generative AI introduces at scale. When students can produce competent outputs without cognitive engagement, and when institutional assessment systems accept those outputs as evidence of learning, schools enter what Baudrillard would recognize as a condition in which the simulacrum of education substitutes for its substance. The system does not resist this substitution because the forms of academic exchange remain intact. Grades are assigned, credits accumulate, and diplomas are conferred, all without the schema construction that Kapur's framework identifies as the point of the exercise. This is a structural tendency of institutions to accept the simulation of learning when confronting its absence would require rethinking the assessment architectures on which the institution depends rather than a problem of individual students cheating.

The scope of this problem extends beyond academic integrity, which is how most educational institutions have framed it. Framing generative AI as an integrity problem locates the issue in individual student behavior and misses the media ecological dimension. Tertiary algorithmicity creates an environment in which the path of least resistance leads to unproductive success. The tools are available, the outputs are convincing, and the cognitive bypass with AI tools can be done without actual learning. The environment itself normalizes the bypass of cognitive labor, just as broadcast media normalized passive reception and algorithmic secondary orality normalized engagement-optimized information consumption. The normalization occurs because the media environment’s default tendencies favor frictionless production and not because students are lazy or dishonest, or make bad choices. Tertiary algorithmicity threatens learning because the media environment's default tendencies favor the bypass of cognitive labor. Many educators already engage in deliberate, sustained, and institutionally supported resistance to those tendencies through process-oriented assessment, oral defense, peer critique, and the other practices Section V develops. The framework's contribution is not to invent such practices but to name the conditions under which they become necessary and to provide theoretical language for connecting them to a coherent response.

C. Three Dimensions of Educational Threat

The educational consequences of tertiary algorithmicity can be analyzed through three dimensions. These consequences do not operate uniformly or mechanically. Schools are not identical to social media platforms or algorithmic entertainment environments. They remain structured human environments with teachers, relationships, routines, curriculum, assessment practices, and opportunities for reflection. At their best, schools provide precisely the hooks for change that broader media environments often lack: sustained attention, guided interpretation, accountable dialogue, and pedagogical intervention. Yet students now enter these educational spaces with habits of attention and interpretation increasingly shaped by algorithmically mediated environments. The concern, therefore, is not that schooling is fully determined by the media environment, but that educators must now work against default conditions that often reward rapid consumption over deliberation, emotional response over sustained analysis, and frictionless production over interpretive labor.

The first dimension concerns attention and the sensorium. Drawing on Ong’s (1967) concept of the shifting sensorium, algorithmically mediated environments restructure cognitive processing to prioritize high-stimulus, reactive engagement over sustained analysis. Students do not develop within a single, undifferentiated media ecosystem. Schools, classrooms, families, peer groups, social media platforms, gaming environments, and generative AI tools each shape attention and interpretation differently. The concern here is not that schools and social media are the same environment, but that students move across overlapping media ecologies. School remains a structured human environment with teachers, routines, curriculum, assessment, and opportunities for guided reflection. At the same time, students increasingly enter school with habits of attention shaped by algorithmically mediated environments that reward rapid consumption over deliberation, emotional response over critical evaluation, and confirmatory content over productive dissonance (Metzger et al., 2023). Carr (2011) and Wolf (2018) documented these attentional consequences before generative AI arrived; tertiary algorithmicity intensifies the problem by adding generated content to curated content, increasing the volume and apparent quality of material available for consumption without effort.

The second dimension concerns knowledge and interpretation. The externalization of memory that began with literacy has progressed through distinct phases: writing externalized what individuals knew, print externalized what communities preserved, and digital networks externalized what institutions organized. Each phase kept the generative act within human consciousness. Tertiary algorithmicity crosses a threshold because it externalizes the production of knowledge-like discourse itself. This carries a consequence that Baudrillard (1981/1994) identified in an earlier media context: increasing information may produce less meaning rather than more, as the volume of available content overwhelms the interpretive capacities required to evaluate it (pp. 79–80). Under tertiary algorithmicity, the problem intensifies because algorithmically generated content does not only add volume. It can also scramble the relationship among source, evidence, authorship, and authority by arriving in the form of competent prose without experiential grounding.

The third dimension concerns agency. Creative agency becomes distributed between human and system in ways that complicate traditional notions of authorship. A student who prompts, evaluates, and edits AI-generated text has exercised a form of agency, but it is qualitatively different from composing original work from the encounter between one’s own understanding and the demands of a task? Epistemic agency, the capacity to evaluate knowledge claims and determine what to believe, faces the challenge of origin uncertainty when students cannot reliably determine whether content was human- or machine-generated. Dialogic agency, the capacity for genuine intellectual exchange, encounters the challenge of simulated interlocutors that provide responsiveness without the cognitive demands of real disagreement and negotiation.

These three dimensions come together to form a single educational concern. Tertiary algorithmicity’s default tendency is to produce environments in which students can perform competently without developing the cognitive capacities that mastery and performance traditionally reflected. The questions are broader than just whether educators will design deliberate structures to counteract this tendency, instead it includes whether they can do so given the landscape of constraints under which schooling operates. Counteraction will require a combination of sensibilities, knowledge, ethics, and resources. These actions can be anchored in learning science to support understanding that includes which difficulties are productive, media-ecological awareness of how environments shape cognition, equity-attuned judgment about exclusionary barriers, and institutional capacity to sustain assessment and policy designs that do not default to frictionless generation. This is the question of pedagogical friction, which the next section develops.

If tertiary algorithmicity creates conditions that default toward the bypass of cognitive labor, and if learning science establishes that cognitive labor is constitutive of durable understanding, then the educational question becomes one of design. How can educators, institutions, and policymakers preserve the conditions for learning within a media environment organized around their elimination? This section develops pedagogical friction as a response framework, defines its dimensions, addresses its limits, and connects it to the concerns of curriculum, advocacy, and policy.

A. Defining Pedagogical Friction

Pedagogical friction is the intentional preservation of cognitive, rhetorical, and existential resistance necessary for durable learning, under conditions where frictionless automation has become the default. The concept is not a call to reject technology, ban tools, or return to some imagined pre-digital simplicity. It is a recognition that learning requires certain kinds of difficulty, that those difficulties are what the learning sciences describe as desirable, and that under the conditions of tertiary algorithmicity, those difficulties must be deliberately designed and structurally protected rather than assumed.

The concept draws its necessity from the media ecological analysis developed in the preceding sections. Each transition in Ong’s framework altered what cognitive operations were available and what operations were required. Pedagogical friction names the educational intervention required when the current transition threatens to make the cognitive work of interpretation, composition, and evaluation optional.

The term ‘friction’ is deliberately chosen. In physics, friction is energy loss, a force that slows movement and reduces efficiency; the natural impulse is to minimize it. The educational argument turns counter to this impulse. The friction that pedagogical friction names is not wasted energy but the effort through which cognitive development can occur. The generation effect, retrieval practice, productive failure, and germane cognitive load all describe forms of difficulty that can support learning. However, this does not mean educators can always identify the perfect amount of friction in advance. Learning is complex, contextual, and uneven. What is productive for one learner may be excessive or exclusionary for another. AI may help educators estimate readiness, generate alternative explanations, reduce extraneous load, or provide adaptive scaffolds. But deciding whether a scaffold preserves productive struggle or removes it remains a pedagogical and ethical judgment. The challenge is to protect productive forms of difficulty while reducing genuinely unproductive obstacles.

B. Four Dimensions of Pedagogical Friction

Pedagogical friction operates across four dimensions. The first three correspond to phenomenological dimensions of human engagement with the symbolic environment. The fourth provides the structural conditions that make the other three possible.

Noetic Friction: The Head

Noetic friction is the internal cognitive resistance required to transform external information into internalized understanding as seen in the struggle to comprehend a difficult text, synthesize competing sources, construct an argument from evidence, and revise one’s own thinking when it proves inadequate. In the learning science framework, it encompasses the germane cognitive load that Sweller, Ayres, and Kalyuga (2011) identify as essential for schema construction, the generation effect that Bjork and Bjork (2011) describe, and the productive failure that Kapur (2008, 2015, 2016) documents. Tertiary algorithmicity threatens noetic friction through bypass. When a generative AI system produces a competent analysis upon prompting, the schema construction that would have occurred through the struggle of composition does not occur. The student performs without learning.

Rhetorical Friction: The Room

Rhetorical friction is the social and dialogic struggle involved in discussing and defending ideas against unpredictable human interlocutors. Rhetorical friction occurs in classroom discussion, peer review, oral defense, collaborative inquiry, or in any other context in which a learner must articulate, defend, and revise claims in response to others whose reactions cannot be predicted or controlled. The unpredictability of human conversation is what makes it productive. Encountering genuine disagreement, misunderstanding, or critique forces cognitive adjustment in ways that prepared responses cannot.

In remote and hybrid classrooms, however, ‘the room’ is already mediated differently. Platform interfaces structure turn-taking, mute the micro-signals of embodied presence, and narrow the sensory field through which learners register disagreement, confusion, hesitation, and care (Short et al., 1976; Tomprou et al., 2021). Even before generative AI enters, this mediated sensorium can reduce the immediate dialogic pressure that physical proximity often supplies, making rhetorical friction more fragile and easier to avoid (Bailenson, 2021). In response, educators often shift more weight onto written work as a way to understand the learner’s thinking, voice, and development. Tertiary algorithmicity destabilizes even that compensatory move. When generated text can simulate student voice, the teacher’s effort to know the learner through writing becomes uncertain: the text may reveal the student’s thinking, the system’s prediction, or some hybrid of both. The threat to rhetorical friction, then, is not only that dialogue becomes mediated, but that the signs through which teachers recognize student understanding become harder to trust.

AI systems provide fluent interaction that lacks the genuine authenticity of human dialogue. The system does not misunderstand in productive ways or bring experiences that challenge assumptions from unforeseen directions. When peers also outsource reading and thinking to AI, the collective conditions for classroom dialogue erode further. This is because rhetorical friction requires interlocutors who have done their own cognitive work.

Existential Friction: The World

Existential friction is the experience of being held personally accountable for claims made in physical space. It involves the vulnerability of presenting one’s own thinking to others who can see, hear, and respond to the person behind the ideas. Existential friction encompasses the stakes of authorship and the embodied experience of intellectual engagement in the presence of others. Tertiary algorithmicity threatens existential friction through abstraction. When text is algorithmically generated, the connection between claim and claimant dissolves. The intellectual risk, the possibility of being wrong in front of others and having to reckon with that wrongness, diminishes. For education, which depends on students developing the willingness to take intellectual risks and accept responsibility for their thinking, this represents a loss that extends beyond cognition into the formation of intellectual character.

Infrastructural Friction: The System

Infrastructural friction is the structural and policy condition that makes the other three dimensions possible or impossible to sustain. It includes guidance documents like district and university AI policies, assessment norms, curricular requirements, professional development priorities, and the values communicated through institutional practice. When institutional infrastructure names productive struggle as a protected educational value, individual educators can design the right kind of friction-centered pedagogy with institutional support. For example, a teacher who requires handwritten first drafts or oral defenses of written work can point to institutional norms rather than relying on personal authority alone. When institutional infrastructure is not anchored in learning science, as most current AI policies are, friction preservation becomes an individual burden.

The gap in current institutional AI policy is significant (Kasarci et al., 2025; Malomo et al., 2025; Rice, 2024). Most district and university frameworks address academic integrity, data privacy, bias mitigation, and risk management. These are necessary concerns. But they are silent on learning science and do not articulate what learning requires, what cognitive processes education aims to develop, or how generative AI’s default tendencies relate to those processes. Friction-centered policy would close this gap by naming productive struggle as an institutional value and providing frameworks for distinguishing productive from exclusionary friction.

C. Productive and Exclusionary Friction

The pedagogical friction framework requires a distinction between productive friction and exclusionary friction. Productive friction refers to difficulty that builds capacity. It includes the struggle to retrieve, generate, explain, defend, revise, or apply understanding. Exclusionary friction refers to difficulty that blocks access without contributing meaningfully to cognitive development. This distinction is essential because an argument for preserving friction can easily become an argument for preserving inequity if it does not ask whose learning a given difficulty supports and whose participation it obstructs. This distinction prevents the framework from becoming a simplistic argument for preserving all difficulty. Not all friction serves learning. Some friction is exclusionary, functioning as a barrier that prevents access or participation rather than a condition that promotes cognitive development.

The English Learner Paradox illustrates this distinction. A student who uses generative AI translation tools to convert an essay written in their native language into English is bypassing friction. But the friction bypassed may be exclusionary rather than productive. The language barrier prevents the student from demonstrating content knowledge and participating in academic discourse. Removing this barrier through translation provides access (Ngo Cong-Lem et al., 2025). At the same time, the translation may flatten the student’s distinctive voice, producing grammatically correct but stylistically generic prose that reduces or removes the perspective and phrasing that make writing recognizably individual (Belcher, 2024; Toncelli & Kostka, 2025). The technology may solve an access problem while introducing an authorship problem.

No mechanical resolution is possible for this paradox. Productive friction is cognitive effort that builds schema, develops capacity, and enables transfer like wrestling with difficult texts, defending arguments against genuine critique, composing original work, and revising based on feedback. Exclusionary friction is arbitrary obstruction that prevents participation without building capacity like language barriers imposed on content assessment, inaccessible formats, and procedural requirements unrelated to learning objectives. Generative AI can appropriately reduce exclusionary friction; it should not eliminate productive friction. The challenge is that the same tool use may reduce both simultaneously, and distinguishing the two requires situated professional judgment that no policy document can fully specify. The framework functions as an interpretive lens rather than a decision rule.

This means that determining what counts as productive or exclusionary friction is always more than a technical instructional judgment; it is also a question of standpoint and value. From a standpoint of epistemological perspective, what appears as ‘rigor’ or ‘support’ can look different depending on one’s position in relation to power, language, disability, and institutional norms (Hill Collins, 2000). The pedagogical friction framework therefore invites what Bang and colleagues describe as axiological innovation or the deliberate examination and redesign of the values, ethical commitments, and evaluative criteria that organize a learning environment (Bang et al., 2016). In practice, this pushes the framework beyond teacher-level technique toward collective work among students, educators, leaders, policymakers, and researchers to clarify what forms of struggle are worth preserving, for whom, and for what purposes, and to build the material conditions required to enact those commitments with sincerity.

A more pointed challenge to the pedagogical friction framework comes from disability studies scholarship on academic ableism. Dolmage (2017) argues that what higher education often names "rigor" has functioned historically as gatekeeping, preserving access for students whose embodied relationship to reading, writing, and timed performance already matches institutional expectations. From this vantage, any call to preserve difficulty warrants scrutiny, because the same difficulty that builds schema for some students may operate as structural exclusion for others. Annamma, Connor, and Ferri's (2013) DisCrit framework extends this critique by showing how race and disability intersect in educational categorization, such that what counts as productive struggle for one student may be documented as deficit for another. For students with disabilities, generative AI tools often function as accessibility supports, including speech-to-text, summarization for cognitive load management, and real-time clarification, rather than as cognitive shortcuts. A framework that commits to friction preservation without attending to this distinction risks recoding legitimate accommodations as bypass, which reproduces rather than resists the exclusionary dynamics it claims to address.

The framework developed here does not resolve that tension, and it should not claim to. What it offers is an interpretive structure that requires educators to ask, for each instructional decision, whether the friction in question builds schema the student would otherwise not construct, or whether it imposes barriers that treat a particular cognitive profile as the unmarked norm. This question has no universal answer. AI systems may assist educators by generating alternative explanations, translating language, adjusting reading levels, offering multimodal supports, or estimating where a learner may need scaffolding. In this sense, AI may extend the long-standing promise of educational technology as an equalizer for learners with disabilities. But that promise should not be confused with automatic equity. The claim that AI can provide the “right” amount of cognitive friction for every learner remains a pedagogical and ethical claim, not merely a technical one. Whether a support preserves productive struggle or removes necessary learning work still requires situated judgment by educators, learners, families, and support professionals. That judgment cannot be fully outsourced to policy, rubric, or technological affordance.

This qualification grounds the framework’s equity dimension. Any educational argument that says “preserve struggle” without attending to whose struggles are productive and whose are exclusionary risks reproducing the inequities that already mark educational systems. Friction-centered pedagogy must ask “for whom does this difficulty build capacity, and for whom does it restrict access?” rather than only “does this difficulty build capacity?”

D. Implications for Curriculum, Advocacy, and Policy

The implications of pedagogical friction are curricular, advocacy-oriented, and policy-related because the framework treats generative AI as a learning issue rather than only a technology issue. If tertiary algorithmicity makes cognitive bypass easier, then curriculum must make thinking visible, advocacy must defend productive struggle as an educational value, and policy must create the institutional conditions under which teachers can preserve friction without acting alone. For curriculum, friction-centered design means reconsidering what assignments ask students to do and when AI enters the learning process.

Process-oriented pedagogy, which assesses the quality of thinking rather than the polish of products, becomes more important when products can be algorithmically generated. Sequencing matters because tasks that require noetic struggle before AI assistance preserve the generative cognitive work that Kapur’s research identifies as preparation for future learning. Oral components, peer critique, collaborative inquiry, and written reflection on process all introduce forms of friction that resist algorithmic bypass. The curriculum question is where in the learning sequence AI appears, and what cognitive work must precede its introduction, not whether to allow AI in learning.

For advocacy, the framework identifies a role for educational leaders in naming friction as a protected value. The dominant discourse around educational technology frames innovation in terms of efficiency, personalization, and scalability. But efficiency is often the visible edge of a deeper economic logic. In under-resourced schools, efficient tools may be presented as equity solutions because they offer something where human support is scarce. This makes the argument difficult: the same tool may reduce access barriers while also narrowing the conditions for durable learning. Advocates for necessary friction-centered pedagogy must therefore articulate that productive difficulty is not waste, delay, or resistance to innovation. It is part of how learning develops and must be designed, protected, and defended. This advocacy is particularly important from those who occupy technology leadership positions, because the argument carries different weight when it comes from practitioners who understand both the capabilities and limitations of the technologies in question. However, it cannot belong to technology leaders alone. Teachers, learners, families, and support professionals live with the consequences of these systems and must be involved in deciding which forms of friction are worth preserving, which should be reduced, and which should never have been imposed in the first place.

For policy, the framework identifies a specific gap in current institutional practice. AI policies that address integrity and risk without addressing learning science leave the most important educational question unanswered. Pedagogical friction-centered policy would make explicit what current policies often leave implicit: institutions should value not only academic products but also the cognitive, rhetorical, existential, and affective processes through which those products are developed. Learning is not only cognitive. It is also emotional and relational, involving confidence, anxiety, pride, belonging, vulnerability, and motivation. Policy should therefore provide frameworks for distinguishing productive from exclusionary uses of AI, support professional development that helps educators make these distinctions in practice, and create assessment structures that evaluate process alongside product.

The connection between infrastructural friction and the other three dimensions is the condition of possibility for the entire framework. Without institutional support, the preservation of noetic, rhetorical, and existential friction becomes a matter of individual educator commitment, which is valuable but insufficient at scale. Friction, if it is to survive the default tendencies of tertiary algorithmicity, must become a structural value, not just a personal one.

A. Synthesis

This paper has argued that generative AI cannot be adequately understood through the lens of tool adoption or academic integrity alone, and that a media ecological analysis is required. The specific contribution has been to demonstrate that two intellectual traditions that rarely cite each other, Ong’s media ecology, and the learning sciences of Kapur, Bjork, and Sweller, are combined for a documented gap that the concepts of tertiary algorithmicity and pedagogical friction are designed to fill. Media ecology provides the diagnostic that each major shift in communication technology restructures consciousness, and the current shift renders human authorship of symbolic content optional at scale. Learning science provides the evaluative criteria of cognitive struggle that is fundamental to durable understanding, not an obstacle to it. Neither tradition alone captures the educational problem that generative AI poses. Media ecology without learning science can describe the transformation but cannot specify what is educationally at stake. Learning science without media ecology can identify the threat to cognitive development but cannot explain why the threat is structural rather than individual. The framework developed here, connecting tertiary algorithmicity’s three characteristics to pedagogical friction’s four dimensions, operates at the intersection where both traditions are necessary.

The productive/exclusionary friction distinction prevents this framework from becoming a simplistic argument for preserving all difficulty. By insisting that friction-centered pedagogy ask for whom a given difficulty builds capacity and for whom it restricts access, the framework grounds its normative claims in equity rather than nostalgia. This distinction is not peripheral to the argument but central to its credibility within a Curriculum, Advocacy, and Policy program that demands attention to whose interests are served by particular educational configurations.

B. Limitations

This paper is a work of conceptual analysis. It does not present empirical evidence for the claims made about how tertiary algorithmicity affects student learning, how educators navigate friction preservation in practice, or how institutional policies shape the conditions for pedagogical friction. These are empirical questions the framework identifies but cannot resolve. The stage model, even with the non-teleological framing developed in Section III, risks implying cleaner boundaries between stages than exist in practice. The five stages coexist in the present moment, and many students operate across multiple media environments simultaneously. The stages are analytical tools for identifying dominant tendencies, not descriptions of exclusive conditions. For that reason, the framework should not be read as a clean historical sequence in which one media condition replaces another. Primary orality, literacy, secondary orality, algorithmic secondary orality, and tertiary algorithmicity coexist unevenly in contemporary classrooms, often within the same lesson, assignment, or student workflow.

The framework also operates within a media-ecological and learning-science register rather than a primarily political-economic one. This is a limitation. Economic differences shape how tertiary algorithmicity enters schools since affluent communities may be better positioned to preserve human-rich forms of pedagogical friction through smaller classes, expert teachers, tutoring, discussion-based learning, and careful AI governance, while under-resourced schools may be more likely to receive scalable AI tools framed as efficient substitutes for scarce human support. In this sense, the same technologies that promise personalization and access may also reproduce unequal educational conditions if they are used to compensate for structural underinvestment rather than to support richer forms of teaching and learning. Related political-economic frameworks, including surveillance capitalism (Zuboff, 2019) and Varoufakis’s (2024) account of technofeudalism, address these dynamics more directly. This paper acknowledges those concerns but does not make them its primary analytic frame; its narrower purpose is to theorize how algorithmic curation and generation alter the symbolic conditions of learning and why pedagogical friction remains educationally necessary.

The paper’s educational argument focuses on preserving cognitive struggle, reflecting both a theoretical commitment to cognitive constructivist and learning-science traditions and a technoskeptical disposition rooted in critical hermeneutics. Other scholars, working from different traditions, might emphasize the generative possibilities of human-AI collaboration, the democratization of access to cognitive tools, or the emergence of new forms of literacy that the friction framework does not adequately capture. The argument presented here does not deny these possibilities. It insists that they must be evaluated against the specific question of whether the cognitive processes through which understanding develops are being preserved, bypassed, or transformed in ways that require new pedagogical responses.

Finally, the paper focuses primarily on the risks generative AI poses to interpretive labor, authorship, and durable learning. Although it acknowledges accessibility and equity benefits, it does not fully examine the conditions under which generative AI can enhance learning, support multilingual writers, expand participation, or create new forms of human-AI collaboration. That limitation is intentional but important. A fuller empirical agenda must investigate not only how AI bypasses learning but also when it scaffolds learning, redistributes agency productively, or removes exclusionary barriers without eliminating the cognitive labor through which understanding develops.

C. Reflexivity and Positionality

The positionality statement in Section I named the professional context and epistemological commitments that shape this analysis. What warrants additional reflection here is how those commitments have shaped the framework’s specific characteristics. My dual position as a district technology leader and doctoral candidate creates an analytical disposition that attends closely to what may be lost in technological transitions, particularly when the public discourse surrounding a technology is dominated by promises of efficiency, personalization, and innovation. The phenomena this paper describes, such as students submitting AI-generated work without understanding, teachers navigating policy without learning-science guidance, and institutional frameworks addressing compliance while remaining silent on cognitive development, are conditions I observe in professional practice.

This vantage point gives the analysis specificity, but it also shapes its emphasis. The framework is intentionally designed to name what can become obscured when the benefits of generative AI are already amplified by vendors, platform companies, investors, and institutional narratives of innovation. A different scholar, positioned differently in relation to these technologies, might foreground human-AI collaboration, accessibility, or productivity more centrally. This paper does not deny those possibilities. It argues that they must be evaluated against a question that current public and institutional discourse often underemphasizes: whether the cognitive, rhetorical, existential, and institutional conditions required for durable learning are being preserved, bypassed, or transformed. The value of this framework lies in its capacity to name what is educationally at stake when the default tendency of a media environment is to minimize productive pedagogical struggle.

D. Toward the Dissertation

This qualifying paper constructs the theoretical apparatus that a subsequent dissertation will put to empirical use. The conceptual framework, including the extended Ong stages, the pedagogical friction dimensions, and the productive/exclusionary friction distinction, generates specific questions that cannot be answered through conceptual analysis alone. How do educators in diverse institutional contexts navigate the preservation of pedagogical friction under conditions of tertiary algorithmicity? What forms of infrastructural friction most effectively support friction-centered pedagogy? How do students at different developmental levels experience the tension between frictionless automation and productive struggle? Where does the productive/exclusionary friction boundary fall for specific populations, including English learners, students with disabilities, and students in under-resourced schools?

These questions require the mixed-methods case study investigation that the dissertation will undertake. The qualifying paper’s contribution is to provide the theoretical framework within which such investigation becomes possible and meaningful, to articulate the concepts that give empirical observations analytical purchase, and to ground that framework in both the media ecology tradition and the learning science literature in a way that makes the dissertation’s contribution legible to scholars across these fields. The work that follows will determine whether the framework developed here proves useful in practice, which is the test that any conceptual contribution must eventually face.

Ángel-Botero, A., & Alvarado-Duque, C. F. (2016). Oralidade terciária: Olhar ecológico na rádio digital. Palabra Clave, 19(2). https://doi.org/10.5294/PACLA.2016.19.2.6

Annamma, S. A., Connor, D., & Ferri, B. (2013). Dis/ability critical race studies (DisCrit): Theorizing at the intersections of race and dis/ability. Race Ethnicity and Education, 16(1), 1–31. https://doi.org/10.1080/13613324.2012.730511

Bailenson, J. N. (2021). Nonverbal overload: A theoretical argument for the causes of Zoom fatigue. Technology, Mind, and Behavior, 2(1). https://doi.org/10.1037/tmb0000030

Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. https://doi.org/10.1126/science.aaa1160

Bang, M., Faber, L., Gurneau, J., Marin, A., & Soto, C. (2016). Community-based design research: Learning across generations and strategic transformations of institutional relations toward axiological innovations. Mind, Culture, and Activity, 23(1), 28–41. https://doi.org/10.1080/10749039.2015.1087572

Baudrillard, J. (1994). Simulacra and simulation (S. F. Glaser, Trans.). University of Michigan Press. (Original work published 1981)

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). https://doi.org/10.1145/3442188.3445922

Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On meaning, form, and understanding in the age of data. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (pp. 5185–5198). https://doi.org/10.18653/v1/2020.acl-main.463

Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the real world: Essays illustrating fundamental contributions to society (pp. 56–64). Worth Publishers.

Bolter, J. D., & Grusin, R. (1999). Remediation: Understanding new media. The MIT Press.

boyd, d., & Ellison, N. B. (2007). Social network sites: Definition, history, and scholarship. Journal of Computer-Mediated Communication, 13(1), 210–230. https://doi.org/10.1111/j.1083-6101.2007.00393.x

Bozkurt, A. (2024). GenAI et al.: Cocreation, authorship, ownership, academic ethics and integrity in a time of generative AI. Open Praxis, 16(1), 1–10. https://doi.org/10.55982/openpraxis.16.1.654

Bozkurt, A. (2025). Algorithmically manufactured minds: Generative and agentic AI in a time of post-truth, reconfiguration of student agency and death of critical pedagogy. Open Praxis, 17(2), 206–210. https://doi.org/10.55982/openpraxis.17.2.792

Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the diffusion of moralized content in social networks. Proceedings of the National Academy of Sciences, 114(28), 7313–7318. https://doi.org/10.1073/pnas.1618923114

Bucher, T. (2012). Want to be on the top? Algorithmic power and the threat of invisibility on Facebook. New Media & Society, 14(7), 1164–1180. https://doi.org/10.1177/1461444812440159

Bucher, T. (2018). If…then: Algorithmic power and politics. Oxford University Press.

Carr, N. G. (2011). The shallows: What the Internet is doing to our brains. W. W. Norton.

Chein, J. M., Martinez, S. A., & Barone, A. R. (2024). Human intelligence can safeguard against artificial intelligence: Individual differences in the discernment of human from AI texts. Scientific Reports, 14(1), 25989. https://doi.org/10.1038/s41598-024-76218-y

Clark, E., August, T., Serber, S., Haber, N., Celikyilmaz, A., & Smith, N. A. (2021). All that’s ‘human’ is not gold: Evaluating human evaluation of generated text. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (pp. 7282–7296). https://doi.org/10.18653/v1/2021.acl-long.565

Cordón-García, J.-A., & Muñoz-Rico, M. (2024). De la oralidad primaria a la tercera oralidad: Transformaciones culturales y cognitivas en la era digital. Anuario ThinkEPI, 18. https://doi.org/10.3145/thinkepi.2024.e18a38

Dewey, J. (1933). How we think: A restatement of the relation of reflective thinking to the educative process. D. C. Heath.

Dolmage, J. T. (2017). Academic ableism: Disability and higher education. University of Michigan Press.

Doss, C. J., Bozick, R., Schwartz, H. L., Chu, L., Rainey, L. R., Woo, A., Reich, J., & Dukes, J. (2025). AI use in schools is quickly increasing but guidance lags behind: Findings from the RAND Survey Panels (RR-A4180-1). RAND Corporation. https://www.rand.org/pubs/research_reports/RRA4180-1.html

Dreyfus, H. L. (1972). What computers can’t do: The limits of artificial intelligence. Harper & Row.

Dreyfus, H. L. (1992). What computers still can’t do: A critique of artificial reason. MIT Press.

Dron, J. (2023). The human nature of generative AIs and the technological nature of humanity: Implications for education. Digital, 3(4), 319–335. https://doi.org/10.3390/digital3040020

Feenberg, A. (2002). Transforming technology: A critical theory revisited. Oxford University Press.

Floridi, L., & Chiriatti, M. (2020). GPT-3: Its nature, scope, limits, and consequences. Minds and Machines, 30(4), 681–694. https://doi.org/10.1007/s11023-020-09548-1

Foley, J. M. (2012). Oral tradition and the Internet: Pathways of the mind. University of Illinois Press.

Frankfurt, H. G. (2005). On bullshit. Princeton University Press.

Gencarelli, T. (2000). The intellectual roots of media ecology in the work and thought of Neil Postman. New Jersey Journal of Communication, 8(1). https://doi.org/10.1080/15456870009367381

González-Bailón, S., Lazer, D., Barberá, P., Zhang, M., Allcott, H., Brown, T., Crespo-Tenorio, A., Freelon, D., Gentzkow, M., Guess, A. M., Iyengar, S., Kim, Y. M., Malhotra, N., Moehler, D., Nyhan, B., Pan, J., Rivera, C., Settle, J. E., Thorson, E. A., … Tucker, J. A. (2023). Asymmetric ideological segregation in exposure to political news on Facebook. Science, 381(6656), 392–398. https://doi.org/10.1126/science.ade7138

Haraway, D. J. (1991). Simians, cyborgs, and women: The reinvention of nature. New York, NY: Routledge.

Hayles, N. K. (2012). How we think: Digital media and contemporary technogenesis. University of Chicago Press.

Hayles, N. K. (2017). Cognitive assemblages: Technical bodies and virtual minds. Critical Inquiry, 43(1), 54–69.

Heyd, T. (2021). Tertiary orality? New approaches to spoken CMC. Anglistik: International Journal of English Studies, 32(2), 131–147. https://doi.org/10.33675/ANGL/2021/2/10

Hill Collins, P. (2000). Black feminist thought: Knowledge, consciousness, and the politics of empowerment (2nd ed.). Routledge.

Hutchby, I. (2001). Technologies, texts and affordances. Sociology, 35(2), 441–456. https://doi.org/10.1177/S0038038501000219

Jakesch, M., French, M., Ma, X., Hancock, J. T., & Naaman, M. (2019). AI-mediated communication: How the perception that profile text was written by AI affects trustworthiness. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1–13). https://doi.org/10.1145/3290605.3300469

Jakesch, M., Hancock, J. T., & Naaman, M. (2023). Human heuristics for AI-generated language are flawed. Proceedings of the National Academy of Sciences, 120(11), e2208839120. https://doi.org/10.1073/pnas.2208839120

Jones, C. R., Rathi, I., Taylor, S., & Bergen, B. K. (2024). People cannot distinguish GPT-4 from a human in a Turing test. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3715275.3732108

Kaluža, J. (2022). Habitual generation of filter bubbles: Why is algorithmic personalisation problematic for the democratic public sphere? Javnost – The Public, 29(1). https://doi.org/10.1080/13183222.2021.2003052

Kapur, M. (2008). Productive failure. Cognition and Instruction, 26(3), 379–424. https://doi.org/10.1080/07370000802212669

Kapur, M. (2015). Learning from productive failure. Learning: Research and Practice, 1(1), 51–65. https://doi.org/10.1080/23735082.2015.1002195

Kapur, M. (2016). Examining productive failure, productive success, unproductive failure, and unproductive success in learning. Educational Psychologist, 51(2), 289–299. https://doi.org/10.1080/00461520.2016.1155457

Karpicke, J. D., & Blunt, J. R. (2011). Retrieval practice produces more learning than elaborative studying with concept mapping. Science, 331(6018), 772–775. https://doi.org/10.1126/science.1199327

Kasarci, I., Akın Demircan, Z., Çeliker Ercan, G., & İnci, T. (2025). Managing artificial intelligence ethics in higher education: A systematic framework for issues and policy recommendations. International Journal of Current Educational Studies. https://doi.org/10.46328/ijces.223

Kittler, F. (1990). Discourse networks 1800/1900 (M. Metteer & C. Cullens, Trans.). Stanford University Press. (Original work published 1985)

Kittler, F. A. (1999). Gramophone, film, typewriter (G. Winthrop-Young & M. Wutz, Trans.). Stanford University Press. (Original work published 1986)

Levy, R. (2021). Social media, news consumption, and polarization: Evidence from a field experiment. American Economic Review, 111(3), 831–870. https://doi.org/10.1257/aer.20191777

Malomo, O., Adekoya, A., Donald, A. M., Eyob, E., Omojokun, E., Mummalaneni, V., & Garuba, M. (2025). AI as asset and liability: A dual-use dilemma in higher education and the SPARKE Framework for institutional AI governance. Online Journal of Applied Knowledge Management, 13(2), 57–76. https://doi.org/10.36965/ojakm.2025.13(2)57-76

Mayer, R. (2009). L’oralité tertiaire: Positionnement, statut, modalités. Pratiques Audiovisuelles. https://doi.org/10.4000/PA.186

McLuhan, M. (1964). Understanding media: The extensions of man. McGraw-Hill.

Metzger, S. A., & Krutka, D. G. (2023). Interrogating the smartphone: Teaching through technoskeptical questions. Social Education, 87(5), 313–318.

Mieczkowski, H., Hancock, J. T., Naaman, M., Jung, M., & Hohenstein, J. (2021). AI-mediated communication: Language use and interpersonal effects in a referential communication task. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), 1–14. https://doi.org/10.1145/3449091

Mollick, E. (2024). Co-intelligence: Living and working with AI. Portfolio.

Nikolaidis, A., & Lara-Steidel, H. (2024). White ignorance and attention in the age of digital technologies. Philosophy and Education, 80(2), 131. https://doi.org/10.47925/80.2.131

Nguyen, C. T. (2020). Echo chambers and epistemic bubbles. Episteme, 17(2), 141–161. https://doi.org/10.1017/epi.2018.32

Ong, W. J. (1967). The presence of the word: Some prolegomena for cultural and religious history. Yale University Press.

Ong, W. J. (1977). Interfaces of the word: Studies in the evolution of consciousness and culture. Cornell University Press.

Ong, W. J. (2002). Orality and literacy: The technologizing of the word (2nd ed.). Routledge. (Original work published 1982). https://doi.org/10.4324/9780203426258

Ong, W. J. (2017). Language as hermeneutic: A primer on the word and digitization (T. D. Zlatic & S. van den Berg, Eds.). Cornell University Press.

Pariser, E. (2011). The filter bubble: What the internet is hiding from you. Penguin Press.

Petricini, T. (2024). Special issue introduction: AI and media ecology. Explorations in Media Ecology, 23(2), 93–103.

Piaget, J. (1952). The origins of intelligence in children (M. Cook, Trans.). International Universities Press.

Postman, N. (1985). Amusing ourselves to death: Public discourse in the age of show business. Viking.

Postman, N. (1992). Technopoly: The surrender of culture to technology. Knopf.

Rice, M. (2024). The micropolitical landscape of publicly discoverable policies for generative AI in large US school districts. Technology, Pedagogy and Education. https://doi.org/10.1080/1475939X.2024.2421494

Rickert, T. (2024). Ambient engineering: Hyper-nudging, hyper-relevance, and rhetorics of nearness and farness in a post-AI algorithmic world. Rhetoric Society Quarterly. https://doi.org/10.1080/02773945.2024.2407263

Riva, G. (2025). The architecture of cognitive amplification: Enhanced cognitive scaffolding as a resolution to the comfort-growth paradox in human-AI cognitive integration (arXiv:2507.19483). arXiv. https://arxiv.org/abs/2507.19483

Robson, S. (2019). Theories of cognitive development. In Developing thinking and understanding in young children. Routledge. https://doi.org/10.4324/9780429469442-2

Ryu, S. (2023). Exploring tertiary orality in virtual reality. In HCI International 2023 (pp. 287–296). Springer. https://doi.org/10.1007/978-3-031-36004-6_39

Short, J., Williams, E., & Christie, B. (1976). The social psychology of telecommunications. Wiley.

Sligo, F. (2020). Literacy and orality at work. Peter Lang. https://doi.org/10.3726/B17476

Smith, M. R., & Marx, L. (Eds.). (1994). Does technology drive history? The dilemma of technological determinism. The MIT Press.

Soffer, O. (2020). From textual orality to oral textuality: The case of voice queries. Convergence, 26(4), 927–941. https://doi.org/10.1177/1354856519825773

Stiegler, B. (2010). Technics and time, 3: Cinematic time and the question of the new sophist (S. Barker, Trans.). Stanford University Press.

Stiegler, B. (2012). Relational ecology and the digital pharmakon. Culture Machine, 13. https://culturemachine.net

Stiegler, B. (2016). Automatic society: Volume 1. The future of work (A. Brown, Trans.). Polity.

Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive load theory. Springer. https://doi.org/10.1007/978-1-4419-8126-4

Tomprou, M., Kim, Y. J., Chikersal, P., Woolley, A. W., & Dabbish, L. A. (2021). Speaking out of turn: How video conferencing reduces vocal synchrony and collective intelligence. PLOS ONE, 16(3), e0247655. https://doi.org/10.1371/journal.pone.0247655

Tufekci, Z. (2017). Twitter and tear gas: The power and fragility of networked protest. Yale University Press.

Turkle, S. (2011). Alone together: Why we expect more from technology and less from each other. Basic Books.

Turner, D., & Allen, W. S. (2013). Documents, dialogue and the emergence of tertiary orality. Information Research, 18(3).

van Dijck, J. (2013). The culture of connectivity: A critical history of social media. Oxford University Press.

van Manen, M. (1990). Researching lived experience: Human science for an action sensitive pedagogy. State University of New York Press.

Varoufakis, Y. (2024). Technofeudalism: What killed capitalism. Melville House.

Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146–1151. https://doi.org/10.1126/science.aap9559

Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes. Harvard University Press.

Wolf, M. (2018). Reader, come home: The reading brain in a digital world. HarperCollins.

Wu, A. X., & Taneja, H. (2021). Platform enclosure of human behavior and its measurement: Using behavioral trace data against platform episteme. New Media & Society, 23(9), 2650–2667. https://doi.org/10.1177/1461444820933547

Wu, T. (2016). The attention merchants: The epic scramble to get inside our heads. Knopf.

Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power. PublicAffairs.