Composing with AI

Embedded Values and Rhetorical Frames in AI Interfaces

This chapter reflects a specific time in ChatGPT's development, acknowledging that technologies change, while providing an historical description of the interface and a glimpse of student heuristic analysis and impressions of ChatGPT upon its public inception.

Since its release in November 2022, ChatGPT's interface has included a variety of disclaimers and hedges about its accuracy. For example, in 2024, a faint message stated, “ChatGPT can make mistakes. Check important info” (Figure 3). Such disclaimers deny responsibility while continuing to provide responses, creating a circular logic where users receive legitimate-looking content they're told not to fully trust.

The conversational capabilities of ChatGPT-3.5’s interface design and its free access drove widespread adoption. Despite known inconsistencies and potential dangers reported to OpenAI in testing and documentation (OpenAI, March 23, 2023), such as the potential to identify individuals with outside data (p. 3). Additionally, the report stated, “as these models are integrated into society and used to help automate various systems, this tendency to hallucinate is one of the factors that can lead to the degradation of overall information quality and further reduce veracity of and trust in freely available information” (p. 6). Few users consult technical reports before using a new application. As users generated poems in pirate style, recipes from random pantry ingredients, or sought medical advice, GPT’s interface provided personalized responses while also transmitting “embedded values.” Embedded values, Stanfill (2015) wrote, are inherent to interface designs, steering users toward standard interactions that reflect the interface’s intended use (Stanfill, 2015).

GAI applications like ChatGPT have shifted the interface paradigm from the Graphical User Interface (GUIs) of Stanfill’s analysis. This shift creates an opportunity to critically examine embedded values within the materiality of its chat-based interface, even as that interface ground shifts over time. ChatGPT’s immense circulation and use occurs through its interface, which along with promotional materials and everyday interactions, conveys values about AI’s superpowers along with other “rhetorical frames” (Burke, 1984) that promote acceptance. Its presence on our devices creates conversational opportunities where we share our thoughts and questions, casually, intimately at times, as we would with another human. Just as it grasps our attention with the ease of its natural language interactions, its interface commands authority. There are little to no other options: we ask, it answers. Despite disclaimers, ChatGPT gains our trust, yet its underlying technologies and information sources remain opaque. Even while we know its technologies are beyond us, its conversational interface and human-like language responses affectively gain our human attention and trust. We must teach our students to better assess whether this “friend” fosters healthy, transparent information engagement or, as Selfe and Selfe warned nearly thirty years ago, will we and our students perpetuate “small but continuous gestures of domination and colonialism?” transmitted through its interface (p. 486).

Analyzing ChatGPT's Interface: A Heuristic Approach

While the specific mechanisms of ChatGPT’s responses are complex, interacting with it is as simple as using Google’s search or texting. Its ability to interpret human language relies on the LLM using NLP and vectorization for rapid, “human-like” replies. The interface (chat.openai.com) primarily features (1) a message box for user prompts and (2) the much larger window streaming GPT's responses. Secondary features include a chat history window, a “new chat” button, and user settings. A subtle icon allows sharing transcripts, and a question mark icon provides “help” resources, “terms and policies,” and “keyboard shortcuts.” Users can delete or edit conversational directions, but users are prevented from granular sentence-level editing within the interface, positioning the interface as a broadcast system rather than a writing space. These design choices shape our interaction (Selfe and Selfe, 1994) with the underlying AI, enabling ChatGPT to consumer user-generated data.

Usability Heuristics and the ChatGPT Interface

Jakob Nielsen (1992) championed heuristic interface evaluation to identify “usability problems” (p. 373), defined as aspects that hinder ease of understanding, use, or satisfaction (Pribeanu, 2017). Nielsen advocated for evaluators to assess compliance “with recognized usability principles” (p. 373). These principles, along with heuristic and ergonomic evaluation methods, are used in UX/UI to achieve human-centered design. Pribeanu (2017) noted that user guidance, including prompting, feedback, information architecture, and grouping enhances (p. 33). However, this can lead to designers “guiding” users toward the intended uses imagined by programmers and designers. In ChatGPT and other GAI interfaces, typical dropdown and other options are mostly absent, configuring agency firmly within the chat window and the complicated processes obscured by the interface's simple design.

In the 2023 OpenBootcamp UX/UI Challenge, Nielson’s heuristic was applied to ChatGPT. Athul Anil (2023) concluded that ChatGPT complied with most principles, except for a “minor usability problem related to aesthetic and minimalistic design. The interface can be improved by streamlining the chat interface, removing unnecessary elements, and utilizing space effectively” (n.p., bold in original). Anil recommended even greater minimalism, reducing interactive elements as visual clutter while failing to point out this further reduces the functional options available to users. Anil's assessment of ChatGPT's interface seemed to indicate that at least in some UI/UX circle, ChatGPT’s interface is a nearly flawless design. But what and for whom was it designed?