Composing with AI

Interface Heuristic Development

In “Articulating Methodology as Praxis,” Sullivan and Porter advocated for a heuristic rather than rules-based methodological approach to writing studies research. Connecting with feminist critical theories, they described a heuristic approach as having a “dynamic and relational focus rather than a static one” (p. 59). To scaffold enduring generative AI (GAI) critical literacies, heuristic approaches to teaching and learning could attune to the “dynamic and relational aspects” of GPT in our classrooms, including the vulnerabilities of our students to the “violence of literacy” (Selfe & Selfe, 1994). This would entail knowing our students more deeply than their requests for accommodations and other visible differences from dominant norms. Heuristic classroom development of critical AI literacies is part of doing and paying attention to “a rhetoric of the everyday” (Grabill, 2003, p. 458). Grabill defined this as “a commitment to the mundane, the obvious, and the unstudied as intellectually and epistemologically equivalent to political and ethical commitments to ‘difference,’ to the ‘oppressed,’ to the technopoor’ and other issues of social justice” (p. 465). Prior to GAI, the field of writing studies has operationalized heuristic interface research in classroom, community, corporate, and technical sectors. Developing flexible, inclusive GAI interface heuristic frameworks may help us build critical literacies that can adapt to changes in public knowledge and attitudes, use contexts and settings, individual positionalities and community values, and iterative interface design and infrastructure.

In a workplace GAI study, Hocutt, Ranade, and Verhulsdonck (2022) identified four “insertion points” for technical communicators to influence chatbot development and performance. This research reminds us that GAI design and its processes is not fixed but pliable through ongoing development that could be more inclusive and participatory: “Without intervention by technical communicators, ML [machine learning] will continue to train on biased discourse, continuing to marginalize those already marginalized by those discourses, or foster black-boxed processes that perpetuate discrimination or lack of explanation or transparency of decision-making to users” (p. 128). While our classrooms may not be able to directly impact corporately designed products like ChatGPT, gaining awareness of interface design and its social power can be the foundational space for critical GAI literacy development. Seeing ChatGPT’s interface through their experiences will help build understanding of how technologies work through designs to empower and constrain certain uses and norms. Although technologies have always been implicated in systems of power, ChatGPT has ushered in the super-human circulatory world-making power of AI, or at least the perception of that power.

Even in interface design that predates chat-based interfaces, Corinne Jones (2021) and others have observed that interfaces mostly allude to our conscious attention—so much so, that "they often seem neutral and become invisible" (p. 2). With this transparency, Jones noted interfaces have an (1) “effect on how people circulate information, (2) affect what people know by shaping what information or content they see through circulated materials, and (3) affect who circulates materials as they shape how people know themselves as circulatory actors” (p. 4). In these ways and more, Jones (2021) stated interfaces reproduce and circulate dominant values and behavior norms, but these can be "unconcealed" through heuristic analysis. Jones (2021) points to Stanfill’s (2015) Discourse Interface Analysis (DIA) as a useful framework for heuristic analysis of interface because it can help build greater awareness and intervention of technical and cultural aspects of its design, use, and circulation. Although user behaviors may not be determined by interface designs, the constraints of these designs are not easy to overcome since interfaces are the entry and establish the rules for users to gain access to systems and information.

Stanfill (2015) categorized interface design features by functional, cognitive, and sensory/aesthetic affordances. These affordances show up materially as interactive elements like menu items, dropdowns, and text boxes (Functional); messages to the user about what can and should be done (Cognitive) and visual/tactile features that attract attention, increase pleasure, and create brand image (Sensory/Aesthetic). These interface affordances are “sites of Foucauldian productive power because they encourage certain practices while hindering others” (Jones, 2021, p. 3). These affordances “create norms and make certain practices and positions seem commonplace” (Jones, 2021, p. 3)

In contrast to industry-oriented interfaces like Nielsen’s, which focus on achieving designer and corporate goals for user behavior, Stanfill (2015) and Jones (2021) offer heuristic frameworks that questions the balance of power in interface design. For Stanfill and Jones, interfaces can be particularly problematic for how they position users to underlying systems, infrastructure, processes, data, and information. For example, Jones used Stanfill’s framework as a heuristic to examine three U.S. Chamber of Commerce website interfaces as they circulated publicly during the COVID pandemic. Pertinently, she observed that interactive features like flyer builders and the informational and promotional materials they generated were secondary interfaces. Jones found all interfaces “produced a normative practice of circulating pre-existing content” (p. 5). Jones observed that while interface features often allowed users some ability to personalize their experience and outputs, these allowances were designed to guide user behaviors toward specific goals established by programmers and stakeholders, transforming users into “prosumers,” or “people who create goods and services without compensation” (p. 6). Jones’ conclusions could've been describing GAI like ChatGPT—applications that seem to give us what we ask while norming us into accepting and consuming their goals and values. The following section will briefly adapt principles from interface research by Selfe and Selfe (1994), Grabill (2003), Stanfill (2014), Jones (2021) to build GAI critical literacies through heuristic development and analysis of the ChatGPT 3.5/4 interface.

Interfacing ChatGPT Heuristically

Based on writing studies research, GAI interface heuristics should direct our attention to formal structures of interface design and advance a materialist analysis of particular interface instantiations like ChatGPT's web-based interface as they change over time and within broader interface genre expectations. Heuristics should push evaluators to center questions of power in these designs, perhaps adapting Stanfill’s (2015) inquiry framework: “what is foregrounded, how it is explained, and how technically possible uses become more or less normative through productive constraint” (p. 1062). Designed elements of the interface are the means by which users, our students, are “normed” to ChatGPT’s values. For inclusive and flexible heuristic development, analysis should combine formal awareness of interface design with questions and activities aimed at “unconcealing” these values so we can better understand the tool and the potential consequences of its use. As writing teachers, these are the GAI literacies we need to include as we strive to get students to trust their writing abilities and empower their writerly agency.

The following section summarizes my recent experience with GAI interface heuristic development and analysis in an undergraduate document design course at East Carolina University. As background, these activities occurred during the semester following ChatGPT’s initial public release. This course fulfilled a writing-intensive requirement for all majors, and it tends to combine English and liberal arts majors with students from design, health sciences, computer science, business, and others looking to fulfill the WI requirement. The IRB-approved pedagogical activities and survey data have been summarized, and I’ve included some anonymized individual responses. All data has been collected and used with informed consent. The following discussion describes classroom experiences to demonstrate the types of participatory, flexible heuristic approaches we can adapt to nurture GAI critical literacies in community.