This is a pre-publication draft of "Prosumer Approaches to AI: Techno-Social Moments and a Model for Literacy," forthcoming in Social Research. The piece was co-written with Matthew Duncan.
In a recent meeting to showcase a new artificial intelligence (AI) tool rolled out by our university library, the conversation turned to data protection and student agency. The tool solves several problems for researchers using AI: It allows users to upload and query with prompts their own text files, protecting the documents behind a firewall. It also enables users to create agents, instances of AI tools that have been customized with prompts and parameters that make them better at performing and repeating specific tasks. The presenters were reluctant when asked a question by the audience: Could the agents be “published”? What followed made sense: The agent would be trained in part by the uploaded texts, so releasing the agent to the public meant losing the protective firewall around the data. What stood out, however, was the comfort level in the group with the terminology. It was understood that “publish” meant customizing and releasing an AI agent publicly, beyond the confines of the university. The popular AI customizing environment Microsoft Copilot Studio (like many other AI creation platforms) urges users to share their creations with the prominent display of a publish button. The demonstration made clear that universities are making progress in responding to AI: Protecting data and configuring AI tools for specific purposes are foundational for academic uses of AI. At the same time, it revealed room to grow with our understanding of AI literacy. What does it mean when customizing an AI tool and releasing it to the public is seen as an act of publishing?
Anyone familiar with developing computational systems will note that “publish” is often used as a common shorthand for releasing something technical. It implies a vetting that conveys trust and promotes uptake by users, as when software companies publish an update. As scholars of composition, however, we recognize another set of associations like purpose, audience, exigence, medium, and genre. In short, we notice the link between creative technology use and a host of rhetorical, intellectual concerns. Early fear related to AI centered on the relationship between writing and thinking: If AI does the writing, we may well lose the thinking. This is a valid and still pressing concern. But the emergence of low-threshold, customizable AI systems adds another element to the calculation. What if the thinking can also play out through the tool building?
In this article, we will look at the connections between writing, thinking, and computational tools in order to better understand possibilities for integrating AI into composition instruction. We’ll look at three technical moments, the emergence of word processors in the late 1980s and early 1990s, the development of prosumer video in the early 2000s, and the rise of algorithms in the 2010s. Each of these technical developments shifted our understanding of human thinking, a pattern that can actually be traced all the way back to the invention of the technology of writing itself. To make sense of these developments, we offer the model of prosumer literacy. Broadly, we propose that the most successful approaches to literacy in writing instruction aim to move students from consumers to producers of knowledge using composition activities and technical tools. We also suggest that readers can engage with texts in both passive and active modes and that texts and technologies facilitate engagements that range from more passive to more active. We suggest that active, hands-on work with writing and technical tools creates access to and increases literacy. We argue that a continuum approach makes sense for understanding passive-active reading modes and consumer-producer engagements with technologies. We conclude by reviewing three tools that illuminate a prosumer continuum for AI literacy.
There are historical precedents for viewing emerging computational technologies through the lens of writing. When personal computers and word processors made their way into education in the late 1980s and early 1990s, scholars offered insights that were eerily similar to current responses to AI. “What is at issue,” Myron Tuman wrote in Word Perfect, “is how the computer, like the printing press, will create entirely new forms of communication and subtly alter the development of related communication skills, all in the name of literacy” (1992, 4; emphasis in original). At that moment, there were worries about the fluid nature of word-processed text replacing our ability to structure and organize thought. The ability to delete sentences (at the time) without a trace, to bring the words of others quickly into one’s own writing, even to move large chunks of written material within a document shifted the mental operations that had been associated with more linear type- or handwritten intellectual work. At the same time, scholars like Jay David Bolter saw promise in the fact that “in using these facilities, the writer is thinking and writing in terms of verbal units or topics, whose meaning transcends their constituent words” (1991, 16). Fluid text challenged existing relationships between thought and language even as it added new affordances for writers.
When one writing technology replaces another, we tend to charge the newest tool with limiting or extending thinking—actually, both. Consider the early 1990s hype around hypertext. In Writing Space, Bolter looked back at the history of writing while proclaiming “the computer as a new writing space” (1991, 15). The links between writing and thinking are central to emerging technologies. “In the act of writing,” Bolter explains, “the writer externalizes his or her thoughts.” Technologies of writing are linked with thinking to such an extent that “it becomes difficult to say where thinking ends and writing begins, where the mind ends and the writing begins” (1991, 11). This hypertext moment illustrates an early response to shifts in the links between writing, thinking, and technologies that can inform current discussions of AI. In a classic utopian vision of emerging technology, Bolter proclaims that “instant access is an aspect of the electronic utopia of literacy, in which the barrier between writing and thinking dissolves and all symbolic information, anywhere in the world, is as immediately available to the writer/reader as his or her own thoughts.” The loss of boundaries between human thought and computer writing spaces is cast not as problem but solution because “our thinking is a network of interrelated signs that can connect to the network inside the machine” (1991, 217).
Early theorizing about hypertext demonstrates the allure of mapping emerging technologies over cognition. The link is already established in the foundational essay “As We May Think,” in which engineer and inventor Vannevar Bush (1945) longs for nonlinear, computational tools because “the human mind . . . operates by association. With one item in its grasp, it snaps instantly to the next that is suggested by the association of thoughts, in accordance with some intricate web of trails carried by the cells of the brain.” And responses quickly arise to temper enthusiasm and complicate initial conclusions about shifting thoughts and literacies. Amid the hype about hypertext supporting and extending thinking, the edited collection Literacy and Computers presented counterarguments from writing scholars who noted that “there is no evidence that readers can grasp information more easily when it is presented in a network rather than in hierarchical and linear form” (Charney 1994, 242). In a similar edited collection titled Hypertext and Cognition, Andrew Dillon concluded that “the practice of hypertextual design has been accompanied by an uncritical acceptance of a host of quasi-psychological notions of reading and cognition” (Dillon 1996, 27). Advocates and skeptics alike offer generalizations about thinking that are for the most part speculative.
What was made clear by the hypertext moment was that writing tools had significantly increased opportunities for creative production. And simultaneously, responses to hypertext taught us that these tools might also decrease agency. Michael Joyce, the author of the early and canonical hypertext afternoon, a story, recognized and made clear this tension by distinguishing exploratory and constructive hypertexts. The distinction was driven by the emergence of platforms that allowed readers to actively reshape the text using built-in authoring tools. Solely clicking on links might create new ways of moving through texts but still results in passive reading. Alternatively, constructive hypertexts allow readers to create new associations by taking on author roles. Joyce warns that it would be unwise “to suggest that students could gain [powers of associative thinking] by simply exploring a scholar’s representation of it” (1995, 45). Even if hypertexts operate in ways similar to the associative structure of the mind, without the act of construction they will never significantly shift our thinking in ways that further our own abilities to make those connections. If the new technology were to have an impact, it would do so by “enabl[ing] audiences of expert and novice readers alike to act as scriptors and to focus upon the discovery of coherent structures and linkages, and, most important, to use a full range of cognitive skills, especially visual ones, to discover new structures and linkages” (46–47). This is the core tenet of literacy models based on production over consumption: To promote literacy, we should move from passive consumption to active construction when presented with new writing tools.
Of course, readers always have the ability to move between passive and active modes as they engage with texts, and the active, reader-author paradigm is not limited to computational technologies. The parallel is the readerly versus writerly text described by semiotician Roland Barthes. Encountering a readerly text, we’re passive. Underwhelmed by the simple demands of the “classic” reading relationship, we’re cast as the “user” or “customer” of the text. In terms of literacy, we’re deciphering denotative text with a “kind of idleness” (2000, 4). With writerly reading, we move into interpretation, taking on more agency with every constructive act. The writerly text creates a space “in which meaning proliferates by layering; and an agglomerative space, certain areas of the text correlating other meanings outside the material text, and with them, forming ‘nebulae’ of signifieds” (8). Here, the ability of readers to operate with either an active or passive stance is complemented by the relative complexity and openness of the texts they encounter. Thus, the aim is to create writerly texts that engage readers with active construction of meaning since, as Barthes explains, “the goal of literary work (of literature as work) is to make the reader no longer a consumer, but a producer of text” (4). Our engagements with language sound much like those with hypertext. In both, we find a dynamic where readers can range from passive to active stances in their engagements and where texts themselves can favor either the consumption or production of knowledge.
This brings us to tools and materials. If agency is key, and passive or active possibilities appear in most intellectual encounters with texts, then turning to writing technologies helps us recognize their potential value for facilitating active literacy engagements that produce new knowledge. The lessons from hypertext are helpful here as well. First, the detailed descriptions of tools in books about hypertext represent the interpretive thinking that is a first step on the way toward literate agency. Reflection and speculation produce knowledge using written language as materials. Beyond this step, however, is the use of the tools. Here, the functions of the software stand out, what writing scholars call “affordances,” opportunities and apparatus for engaging with and creating texts. Tools like Hypercard and StorySpace allowed readers to add not only comments but also materials, to reshape and extend the project. Key here was ease of use, and a large reduction in the programming demands required to create complex computer-based projects. It is clear how these technologies increased agency for readers and writers. These programs required no or little programming knowledge. They exemplify early computational tools that make it much easier to move from consumer to producer.
Looking through the 1990s hypertext moment, we also discover tool building. StorySpace is a poster program for working with computers and emerging technologies to create tools that extend thinking, reading, and writing. It is also a clear example of the ways tool construction (and literacy) operates on levels that are both technical and social. The project brought computer scientists and humanists together. The team of humanities scholars and technical experts engaged materials and concepts alike, writing books and building software. Bolter’s foundational book Writing Space is in many ways an accounting of the social and conceptual knowledge that emerged from participating in the project. And Joyce’s afternoon, a story exemplifies producer literacy since he used the tool being built by the team to create a new genre of fiction. George Landow’s book Hypertext (1992) is itself filled with reflections on composing through the similar hypertext-authoring platform Intermedia. We see how creating and working with these tools yields all kinds of adjacent literate knowledge. These were large projects with university support, edge cases at the far end of a literacy model emphasizing creating knowledge through construction. At the same time, both tools were used by writers, students, and researchers, and they lowered technical thresholds, creating affordances for making knowledge. Productive approaches are everywhere, and more frequent, the closer we get to constructive tool use. In many ways, these technologies prefigured the prosumer video movement that would come some 10 years later.
In fact, the prosumer is one of two key lenses we will use to make sense of our current responses to AI. Original uses of the term can be traced to Alvin Toffler’s The Third Wave. In a preindustrial model, Toffler explains, “the vast bulk of all the food, goods, and services produced by the human race was consumed by the producers themselves” (1980, 53). Toffler points us toward the prosumer’s economic roots, which space permits us to only briefly weave into this article. Responding to industrial separation between consumer and producer, he posits a third-wave economic paradigm where we “see a progressive blurring of the line that separates producer from consumer. We see the rising significance of the prosumer” (284). Toffler’s economic focus is one of many that come together in a prosumer model of literacy. There is a related technical dynamic very much intertwined with access to resources. Then there is a compositional element, reliant on technical affordances and economic access. In bringing together the producer and consumer, these layers become blurred as well, overlapping and shaping one another, which gives us a complex continuum along which we find literate activity informed by technical and social layers.
To help us make sense of this model, we will introduce our second lens: the mode of looking both at and through the materials we study. While Bolter and others were creating authoring tools and writing about hypertexts, literary scholar Richard Lanham emphasized the difference between looking at and looking through as a means for making sense of the emerging technologies (1993, 5). For materials like writing, we don’t look at a text. Instead, we see through the letters to their meaning. “Hypertext” on the surface is nine letters. Looking through the letters, we find reading, writing, technology, agency, and culture. What makes Lanham’s approach helpful is the call to switch between views, leaving us with at/through lenses and a final understanding based on the oscillation between modes. Applied to our discussion so far, we see the written word or hypertexts, and through them understand that readers can operate in modes that can be more or less passive, depending on their approach and the texts they engage. We find tools that can increase affordances, and through them see agency for shaping and responding to materials. We see readers and writers building their own tools and find humanists and programmers composing together. Looking both at and through hypertext and modes of reading, we realize that construction has both technical and social layers and that literacy moves from the passive to the constructive, with compositional acts and materials located everywhere in between; we recognize a continuum from consumer to producer.
This lets us now take up a second technical moment: the explosion of user-created video in the early 2000s. Tool developments lowered barriers for video production, extending agency. iMovie for Macintosh and MovieMaker for Windows PCs reduced the technical knowledge required to create video compositions. At the same time, new cameras emerged that occupied the sweet spot between bare-bones consumer devices and high-end professional tools. The impact was enough to launch a new category: the prosumer camera. Prosumer cameras extend our thinking about the consumer-producer continuum through technical layers. They included affordances like connections for external microphones, more capable sensors, and image stabilization. This technical layer developed among a social one as well. Combined with easy-to-use media transfer and low-threshold video editors, prosumer cameras quickly shifted video creation from consumption toward production. Media literacy work followed, with knowledge of video composing no longer gained by simply looking at but also by creating through the medium of video. Once prosumer production was in place, the move to online distribution was not far behind. And with it came the algorithmic technical moment that plays out now more than ever through websites like YouTube.
We’re moving quickly, from 1990s hypertext to video composing in the 2000s to contemporary social algorithms, so we should pause for a moment to address technical determinism. We’re outlining prosumer models of literacy and, as a thought exercise, linking them to tool affordances and constructive compositional moves. We understand, however, that there is always tension and that boundaries are blurred between human, technical, and systems agencies. The obvious danger is failing to recognize the shaping influences of technologies and economic actors and systems. A similar concern is the ability of technologies (and perspectives on technologies) to alienate and disempower humans. In a classic discussion of digital literacy, Stuart Selber worries that technologies more often limit human agency: “Constructed as consumers, users consider their available options and select the tool that seems to meet the technical and functional requirements of the task at hand, which are outlined in advance” (2004, 38). The level of agency drops when the tool itself “constructs” not only the available responses but also the identities of readers and writers.
So we need to complement our view of tools as opening affordances for prosumer literacy with the understanding that the tools themselves limit agency and creative possibilities and that these limitations are linked with power. We also need to recognize that tools alone don’t determine the outcomes. Literacy plays out through a continuum of both technical and social agency. Here, we need a clarification of our literacy model. The prosumer continuum is helpful because it implies a continuum where agency can increase or decrease in concert with a range of factors—the tools involved, modes of engagement, social contexts, and so on. In fact, any one of these complicating concerns can also be viewed using a continuum; technologies can limit or increase access, contexts can be enabling or constraining; engagements can be passive or active. It can be helpful to view agency linked with literacy and these complicating concerns as layered and made up of both technical and social elements. Agency, affordances, and access all play out through technical and social layers that overlap.
We must be open to the elements and layers having fluid boundaries and co-informing one another. As sociologist Bruno Latour puts it, “to avoid the threat of ‘technical determinism,’ it is tempting to defend adamantly ‘social determinism,’ which in turn becomes so extreme (the steam engine becoming, for instance, the ‘mere reflection’ of ‘English capitalism’)” (2005, 84). Bolter and Richard Grusin similarly encourage us to “let go of the premise of cause and effect” (2000, 77) as we analyze the relationships between technologies and society. Selber concurs that we need to “[reject] theories claiming that technology alone creates educational change, [instead locating] the potential for such change in a nexus of social forces” (2004, 8). Latour explains that the “social is not a place, a thing, a domain, or a kind of stuff, but a provisional movement of new associations” (2005, 238). All of these thinkers help us move beyond “new tools create more agency” conceptions. Instead, we place consumption and production in relation with each other, but rather than cause and effect, we have back-and-forth movements ongoing in between, and we have concurrent technical and social layers with fluid boundaries. This is a classic actor-network circulation model operating as an expansion of posthuman thinking that grants agency to nonhuman entities.
The prosumer mode relies on a sense that humans and technologies exist in codeterminate tension. Tools, agency, groups, or any other elements exist in relation with countless other elements. Tension between items becomes a system of stabilization when multiple elements are included. All of this is linked with the posthuman move of granting agency to things. Discussing groups, for instance, Latour explains that if “durability, range, solidity, commitment, loyalty, adhesion, etc. have to be accounted for, this cannot be done without looking for vehicles, tools, instruments, and materials able to provide such stability” (2005, 35). And the key is that these elements inform and stabilize one another through ongoing activity. This is construction in event mode. Latour calls it “a performative definition [that] draws attention to the means necessary to ceaselessly upkeep the groups” (35). Prosumer literacy is emergent.
As we think about this performative dimension of prosumer literacy, sociologist and historian of science Andrew Pickering offers insights. For Pickering, agency becomes a kind of fluid that moves between the nodes of the actor-network, keeping it in motion and stabilizing it through a process of co-emergence: “One can start from the idea that the world is filled not . . . with facts and observations, but with agency. The world . . . is continually doing things, things that bear upon us not as observation statements upon disembodied intellects, but as forces upon material beings” (1995, 6). These forces play out in relationships between instruments, material objects, human actors, and literate acts. As Latour puts it, “there is no society, no social realm, and no social ties, but there exist translations between mediators that may generate traceable associations” (2005, 108). For us, these are the agencies, affordances, tools, concerns, and so on that exist in layers on a literacy continuum. And we now have tensions that construct stability in the emergent system.
Pickering explains that the processes that play out in these relations are characterized by resistance and accommodation, drawing our attention to “the point of intersection of human and material agency” and to “emergent resistances and accommodations at the interface of these heterogeneous realms” (1995, 168). Regarding the development of instruments, Pickering posits “a tuning process that delicately positions [an instrument] within the flow of material agency . . . [like] a finely engineered valve that both regulates and directs the flow of water from a pipe” (144–45). Viewed as tuning in response to material agencies, the model now can be layered over the tension inherent in working with prosumer tools. As ease of use increases, material tool resistance goes down. At the same time, compositional possibilities increase. And engagement with the material entanglements bundled into less friendly tools decreases. Again, we have sliding scales, but this time over a layer of ease of use leading toward agency in composition we must add another layer allowing for ease of use limiting engagement with larger systems and material constraints.
With the understanding that we’re not naive to the dangers of technical determinism, and with the clarification that prosumer tools at the same time increase agency and decrease knowledge of larger systems, we can continue our historical tracing on the way toward prosumer AI literacy. The next step brings prediction algorithms into view, a moment marked by the advancement of rating and recommendation systems constructed as black box procedures, making it more difficult to see through these technical surfaces to uncover their systemic operations, not to mention their biases and agencies. Looking at a recommendation algorithm directly is largely impossible. As digital media and communications scholars Kevin Brock and Dawn Shepherd explain, popular search and social media content algorithms such as those of Google and Facebook are obscured as proprietary technology (2016, 21). As a work-around, we can investigate recommendation algorithms by focusing on their outputs and then looking through those materials to make sense of the operations in the black box.
Again, we arrive at a moment where human, tool, and systems agencies become necessarily blurred. To investigate an algorithm requires privileged technical knowledge. Or it calls for the kinds of at/through technical reading that Brock and Shepherd demonstrate. This inferring of procedures from effects on technical surfaces gives at/through a more speculative inflection. It is like the way media studies scholar and video game designer Ian Bogost describes our ability to know the experience of objects in posthuman conceptions of the world. We can know the experience of things “only by tracing the exhaust of their effects on the surrounding world” (2012, 100). Given the environmental impacts of technical systems, exhaust is an apt image for the effects of algorithms as they appear on the surfaces of our lives. It is apt as well that Bogost follows by delineating the stakes of knowing computational things with constructive language. “Understanding a computer platform,” he explains, “is vital to a critique of the particular works, genres, or categories of creative production built on top of it” (100). Posthuman and prosumer approaches can help. Digital rhetorician Casey Boyle (2016), for instance, uses posthuman theories to show how rhetorical acts are reproduced through habits and take place in ecosystems. Abigail Bakke (2020) and John Gallagher (2017) discuss habits and ecologies as well. These habits and ecosystems support prosumer literacy. Consider websites like YouTube. Using smooth affordances, we upload our videos, which appear on the surface of our screens. Behind the scenes, information is weighted, parameters are set, predictions are made, and machine learning takes place as recommendations play out in an ecosystem of materials and agency.
Boyle’s posthumanism is offered as a way to make sense of “rhetorical training in an age of constant connections to digital media and networks.” We need to reconfigure writing education, he says, in response to shifts in how students use technology to compose. Students need to learn that posthuman systems tend “to produce greater capacities within any given ecology” (2016, 547), meaning that the environments in which students work, including their technological elements, shape potential responses. For Boyle, literate acts in these situations occur through repetition, iteration, habit, and practice. The notion that “practice makes persuasion,” echoed in Brock and Shepherd’s work, evokes a sense of mutual involvement and cooperation between human and machine, user and tool, and surface and algorithm as participants engage in “an ongoing series of mediated encounters” (Boyle 2016, 534). Boyle’s posthumanism, however, does not take up the computational procedures in play during human and machine encounters. For those, we can turn to scholars looking at and through algorithms.
Brock and Shepherd’s theory of procedural enthymemes stands out. An enthymeme is a persuasive argument in which underlying assumptions go unstated. A statement like students who use AI while writing are cheating, for instance, relies on the unstated premise that all uses of AI while writing constitute cheating. Bringing these hidden premises to the surface allows us to question the grounds and persuasiveness of the unstated assumption itself. Does all AI use while writing constitute cheating? We can see the intellectual power of the move as we pursue the question; what about spell-checking or assignments that use writing as a vehicle for learning AI functionality? For a contemporary example, consider all the unstated assumptions in the simple statement that guides millions of AI users as they encounter ChatGPT’s prompt to “ask anything.” The suggestion that the chatbot can respond to “anything” is premised on the (false) assumption that the knowledge captured in the language model is comprehensive. There is an assumption (accurate) that a response will be provided regardless of the level of certainty in the model data. There is even an assumption of personal connection with the AI tool, as when a good friend in conversation might encourage us to “ask anything.”
The enthymeme resonates with an at/through approach to understanding technologies. And Brock and Shepherd extend its value to algorithms with the “procedural enthymeme,” a product of the rules of a system persuading users to act in a certain way. Put simply, if an enthymeme is an open proposition in language, a procedural enthymeme is an open proposition in system design. They cite tic-tac-toe boards or traffic lights as examples shaping the corresponding behavior of players or drivers respectively. Of note is habituation, which eventually shifts agency so that participants in the system embody the rules in their own behaviors. Habituation, it turns out, both shapes and is shaped by underlying procedures. Looking at curated advertisements, for instance, sheds light on Google’s mysterious PageRank algorithm. And adjusting preference settings for Match.com can change profile visibility and activity on the site. Brock and Shepherd find that the recommendation algorithms can be discerned through their effects and can even be controlled with adjustments to behavior. Of course, looking at recommendation algorithms via the results of their operations represents further abstraction from the computational processes of production. Fine-tuning activity produces a response from an algorithmic system (and can always be built on through composition that constructs knowledge), but the degree of control at the machine level operates on the consumer end of the spectrum.
So we should take a closer look at algorithms and agency, and to do so we can turn to the classroom. John Gallagher, who studies writing technologies, offers the idea of algorithmic audiences. To name algorithms as an audience is to grant agency to an automated part of an information system. It reminds us that we’re operating in human-machine ecosystems. Gallagher claims that algorithms have “something akin to agency” without having “intention” (2017, 26). Here, we have agency and intention in tension, allowing us to factor in the deliberate aspects of agency; we recognize an agency layer that runs from mechanical action to thoughtful intention. Gallagher’s approach promotes awareness and critical literacy of algorithmic operations as well as conceptual concerns linked with what it means to be an agent. “Algorithmic audiences can assist teachers and students” (33; emphasis added), Gallagher reports, because they provide working examples of ready-to-respond audiences, and that begins with granting agency to algorithms. And this overlaps with the classroom, a space with its own procedural enthymemes. The embodied habits of education are algorithmic in their own way. Procedures flow through the teaching space, with weighted agency, operational moves, and predictable outcomes. If the analogy is apt, we can learn more about all three by studying the relationships between algorithms, agency, and classrooms.
Abigail Bakke, scholar of technical communication and digital rhetoric, sheds more light on these relationships. She looks at how students are influenced by search results, asking students to experiment with a “turning the dials” approach to take advantage of existing tools (2020, 14). To this, Bakke adds reflection that centers student agency in relation to recommendation algorithms. Bakke urges caution. It is easy for Google to become an agent in student research, even a trickster. So she demystifies search processes by having students construct search queries. Agency flows in both directions, the algorithm performing its operations and the students composing their queries meant to engage with the underlying operations. Bakke eschews checklist models of literacy. Instead, she asks students to look at search results. Further, she then extends this process by asking students to reflect on the engagements. She deploys composition that constructs knowledge through reflection, a classic prosumer model using low-threshold means for engaging with the technologies of algorithms.
We’re almost ready to apply prosumer literacy models to the current state of AI. But first we need to think a bit about interfaces. Again, we’ll step back in time. In 1994 Cynthia Selfe and Richard Selfe showed how interfaces present us with not only technical but also social concerns. Looking at the intersections of computers and writing, they pointed out that the operations understood through the metaphor of files and folders could have been easily represented with other motifs—a kitchen, for instance. When they created the first graphical user interfaces for computers, developers made cultural decisions. A kitchen paradigm would have created a very different computing experience for the world given that the desktop interface is perhaps the most prevalent prosumer technology ever developed, and it generally goes unnoticed—another unstated assumption. The larger point is that interfaces both expand and limit access and agency, and this happens at both technical and social levels.
We can now use the interface, our at/through approach, and the prosumer lens to take up AI tools. Let’s start by comparing two options accessed through OpenAI’s ChatGPT website. The first is the standard chatbot with its ease-of-use represented by the capacious invitation in the prompt field to “ask anything.” The interface offers little friction, and correspondingly there are few affordances for user configuration. Socially and technically the chat window configures users as consumers. At the same time, the affordances of writing and the agencies conveyed through language are clearly present in the ability to prompt. We can look through this simple “window” and see movement along the prosumer continuum as users produce knowledge through the prompts that they compose. Given the consumer orientation of the tool, a good deal of AI literacy instruction here focuses on increasing agency through “engineering” prompts. Practitioners have developed some highly sophisticated approaches to prompting in the chatbot text field. And with prosumer literacy, we’re still able to look through this simple tool to discover the techno-social concerns associated with the interface—ethical issues related to tool training, environmental damage, and disruptions to behaviors and institutions linked with AI. Prosumer AI literacy at this point takes up all of these concerns and adds “tuning” the device through deliberate writing using the affordance of prompts.
From the same site, we can shift AI literacy possibilities by leveling up from the generic chatbot to a custom GPT (generative pretrained transformer). In the OpenAI ecosystem of early 2026, GPTs enable users to shape interactions by configuring a reusable set of parameters. The GPT nomenclature translates into the term “agent” used by other AI tools, so agency is in play when creating a GPT. And economic factors are now key components of the literacy possibilities. Users have no options to create a GPT without a paid subscription. For premium accounts, the interface offers a + Create button, hinting at the productive possibilities to come. Choosing to create an agent, however, opens another chatbot field. Tool building is subsumed under prompting as the user chats with a language model to set up the agent. The session guides the user through the creation of a system prompt (priming instructions that guide an AI agent’s behavior) and an avatar. Next, users can select a Configure button opening a screen that reveals the prompt (editable) they have created for the system and options for setting default language models. Users can also open advanced options that will link the GPT through interfaces to other tools. Because the initial creation continues the prompt-based interactions of the standard chatbot, it limits tool customizing and agency. At the same time, it opens new affordances, primarily in the form of the customized configurations and a system prompt that shapes any interaction with the tool. The literacy includes all the social concerns linked with AI use and new knowledge of (1) methods of working with repeat-use, custom AI tools, (2) the agency and outcomes involved in shifting between language models, and (3) the difference between session and system prompts.
We should think a bit more about the paid subscription required to create custom GPTs. On one end of the spectrum, we have the consumer model offered by a basic ChatGPT account. Yes, users can operate in productive modes through the written word as they prompt. At the same time, they produce more knowledge for the system as they pay for their “free” account with their data. Behind paywalls like the one blocking the ability to create custom GPTs we find more productive affordances and agency for AI users. But the paywalls limit access and enclose expertise. Let’s focus on the tensions in this economic layer by bringing in two additional possibilities for developing AI tools. Paid models both limit and make available constructive affordances. Open-source tools solve some of these economic access issues and generally create more compositional opportunities. Still, open-source tools have large infrastructure needs, learning costs, and so on. For students and educators, institutionally supported tools are the most common, with their associated educational and economic layers. For literacy, the key is seeing how consumption and production overlap with economics and agency as well as with technical limitations and affordances. For prosumer AI literacy, we discover these techno-social concerns by looking at and working through a range of AI systems.
Let’s look at another technology system, LM Studio, to see these moving parts in action. In terms of technical affordances, LM Studio sits toward the far end of the continuum. It essentially extends a developer environment with an interface that handles functions and operations using buttons, pull-down menus, and graphic user tools. It can load any publicly available language model, allow users to configure parameters, and provide access to the tools that can be used to create agents. Looking at affordances for constructing with AI, one might expect a high cost of entry. At first glance this is not the case. Its free-to-use status means most anyone can access this construction tool. And its interface is relatively easy to use. Its economic position on a scale of access, however, is more in line with expectations. LM Studio does require a newer personal computer with significant memory and processing resources if users are to run more advanced models. And looking through the seemingly open-source surface, we find that LM Studio has been built with venture funding and that paid tiers offer more affordances for those who can afford them. And because interfaces reveal both the technical and the social, we recognize a larger AI ecosystem, the kind of system that demands extensive natural resources and that makes it harder to afford the computational resources that bring a personal computer within reach.
In terms of AI construction at the local level, however, LM Studio is definitely a prosumer tool. It can operate on machines with eight gigabytes of memory and a dedicated graphics card from the previous few generations, so the many affordances are quite high in proportion to the costs of access. And the tool is a boon for creators. After loading and providing access, it easily lets users adjust the operation of the language model for more or less probability in predictions (temperature) and for controlling the pool of available choices. Moving the top-k slider, for instance, allows fewer or more choices in the pool of tokens for prediction. Similar sliders, buttons, and switches control how the pool of choices is constituted, the length of chatbot responses, and options for the format of the output. Many tools allow for some of this control, but LM Studio makes it a central part of engaging with AI as users switch between and make adjustments to language models. This is the interface instantiation of the tuning that Pickering describes as we engage with the material world. It is not surprising either that language resurfaces as the technical and the social combine in the name LM Studio, a place for writing and creativity. Nor that the key act associated with customizing the tool brings together the technical and the rhetorical through the “system prompt.” All of this concludes far down any continuum scale as we look through the convergences and realize that in many ways readers and writers can now easily construct and configure their own AI systems.
So the affordances push the composing tool toward production. And a host of AI literacy outcomes follow: knowledge of the computational resources involved in setting up and running AI systems, access to and knowledge about a range of language models, customizing settings to influence operations and output, and composing system prompts to shape interactions. But moving to local operations creates an entire new set of affordances linked with the ability to manage and maintain control of data. Because LM Studio runs on a personal computer rather than on a platform in the cloud, perhaps its most important function is its local hosting capacity, which creates affordances for data security and protection of intellectual property. Language models are notorious for playing fast and loose with data to the extent that they can be said to plagiarize and violate copyright. These legal concerns are also part of prosumer AI literacy. So are, of course, concerns of data security. So looking through LM Studio provides a case study in the ways that tools can be used to explore privacy and intellectual property concerns in order to promote a techno-social understanding of AI.
But we should look at one more model. The tool under discussion in the anecdote that opens this piece is called PromptLab. On our campus, the libraries have developed PromptLab using an open-source system called LibreChat. PromptLab lets users easily access the dial-turning approach that leads to opportunities for AI literacy and user agency. The interface for PromptLab looks, at first, a lot like every other commercial chatbot. The construction-focused approach, however, is quickly revealed by the tools in the sidebar on the right with the label Agent Builder above a large + sign. Here readers and writers create reusable agents that allow them to compose system prompts and tune model parameters. Rollover tool-tip explanations offer lessons on how the parameters work. These instructive affordances make explicit for prosumer users the functions of tools typically understood by the technically savvy. Unlike GPTs made through the ChatGPT interface, control over agent creation is far more transparent. Once created, agents can also be shared through an institutional commons. This sharing is one remove from the publishing we might find on public sites and is bounded by academic audiences and concerns.
Among these key concerns is integrity in handling intellectual property and data. We’ve seen how LM Studio offers affordances for this kind of literacy with its local instantiation on a personal computer. For PromptLab, however, the institution stabilizes and enforces the boundaries (thus the decision by designers to limit full public release of PromptLab creations). In prosumer fashion, however, these more stringent boundaries translate into more constructive affordances for students. While building an agent, users have the option to upload up to 50 text documents. These can be added to prime the system using a simple Context button that includes a tool tip. Providing context documents for the agent takes configuring the system to another level. Users are also met with a File search button and tool tip that gives them the ability to easily add texts that can be queried for information. The context and file search functions afford users the ability to further configure the system and to work with documents without ceding agency over intellectual property rights or data to commercial interests. The affordances in turn expand possibilities for prosumer AI literacy to now include data and intellectual property concerns, system priming via documents, retrieval-augmented text generation (RAG), and institutional aspects of AI use.
This quick tour of ChatGPT, LM Studio, and PromptLab combined with our historical tracing of hypertext, prosumer video, and algorithms offers insights for anyone interested in computers and writing. Countless decisions await as emerging AI technologies shape culture and make their way into classrooms. AI systems are following those that came before, offering more affordances with lower thresholds. Looking both at and through these tools shows how AI literacy can operate through layers that are both technical and social. Through tool knowledge, we can also develop understandings of thinking, economic forces, environmental impacts, and cultural concerns. These tools both expand and limit agency. Consumption and production play out on a continuum, one that includes sliding scales of access, agency, and affordances operating in fluid techno-social layers. We know that prosumer technologies expand affordances and that configuring these tools moves readers and writers closer to the role of constructors. And so we’ll close with a final point about technical moments. Hypertext technologies did not cease to co-influence writing and literacy as the century turned over. And knowledge circulates today in video as much as in written form. These moments don’t pass. They move forward. At this point, we are no longer at the policy level for programmatic responses to AI. And we’re certainly beyond the option of avoiding AI as we chart the way for writing. We need approaches that understand that it is already a prosumer moment for AI. Our response if we wish to stay not only consumers but also producers is to push prosumer AI literacy.
References
- Bakke, Abigail. 2020. “Everyday Googling: Results of an Observational Study and Applications for Teaching Algorithmic Literacy.” Computers and Composition 57: art. 102577. https://doi.org/10.1016/j.compcom.2020.102577.
- Barthes, Roland. 2000. S/Z: An Essay. Trans. Richard Miller. Hill and Wang.
- Bogost, Ian. 2012. Alien Phenomenology, or What It’s Like to Be a Thing. University of Minnesota Press.
- Bolter, Jay David. 1991. Writing Space: The Computer, Hypertext, and the History of Writing. Lawrence Erlbaum.
- Bolter, Jay David, and Richard Grusin. 2000. Remediation: Understanding New Media. MIT Press.
- Boyle, Casey. 2016. “Writing and Rhetoric and/as Posthuman Practice.” College English 78 (6): 532–54. https://doi.org/10.58680/ce201628626.
- Brock, Kevin, and Dawn Shepherd. 2016. “Understanding How Algorithms Work Persuasively Through the Procedural Enthymeme.” Computers and Composition 42: 17–27. https://doi.org/10.1016/j.compcom.2016.08.007.
- Bush, Vannevar. 1945. “As We May Think.” The Atlantic, July. https://www.theatlantic.com/magazine/archive/1945/07/as-we-may-think/303881/.
- Charney, Davida. 1994. “The Effects of Hypertext on Processes of Reading and Writing.” In Literacy and Computers: The Complications of Teaching and Learning with Technology, edited by Cynthia L. Selfe and Susan Hilligoss. Modern Language Association of America.
- Dillon, Andrew. 1996. “Myths, Misconceptions, and an Alternative Perspective on Information Usage and the Electronic Medium.” In Hypertext and Cognition, edited by Jean-FranVois Rouet, Jarmo J. Levonen, Andrew Dillon, and Rand J. Spiro. Mahwah: Lawrence Erlbaum Associates.
- Gallagher, John R. 2017. “Writing for Algorithmic Audiences.” Computers and Composition 45: 25–35. https://doi.org/10.1016/j.compcom.2017.06.002.
- Joyce, Michael. 1995. “Siren Shapes: Exploratory and Constructive Hypertexts.” In Of Two Minds: Hypertext Pedagogy and Poetics. University of Michigan Press.
- Landow, George P. 1992. Hypertext: The Convergence of Contemporary Critical Theory and Technology. Johns Hopkins University Press.
- Lanham, Richard. 1993. The Electronic Word: Democracy, Technology, and the Arts. The University of Chicago Press.
- Latour, Bruno. 2005. Reassembling the Social: An Introduction to Actor-Network-Theory. Oxford University Press.
- Pickering, Andrew. 1995. The Mangle of Practice: Time, Agency, and Science. University of Chicago Press.
- Selber, Stuart A. 2004. Multiliteracies for a Digital Age. Southern Illinois University Press.
- Selfe, Cynthia L., and Richard J. Selfe. 1994. “The Politics of the Interface: Power and Its Exercise in Electronic Contact Zones.” College Composition and Communication 45 (4): 480–504. https://doi.org/10.58680/ccc19948766.
- Toffler, Alvin. 1980. The Third Wave. Bantam Books.
- Tuman, Myron C. 1992. Word Perfect: Literacy in the Computer Age. University of Pittsburgh Press.