Human and Machine
in the Loop
Key Ideas
The loops, human and machine, exist on a spectrum, not a dichotomy. Using AI in writing involves an iterative and multifaceted approach.
Rhetorical Load Sharing - Sharing the different parts of the writing process with AI tools. Some parts may be more optimal with AI and others with human intervention (see Knowles 2024).
Knowles (2024) explained that the five canons of rhetoric can be used to understand the nature of human-AI interactions: the invention task load, arrangement task load, style task load, memory task load, and delivery task load. The intersection of these canons in an AI context highlights reveals that distinguishing human writing from AI-generated content is more complex than it first appears.
.png)
How will AI use affect the meaning that is created?
Where must the human insert themselves in the process to ensure the validity of the ideas being developed?
When using AI for writing, students must assess the appropriate level of AI use for each moment in the knowledge creation process and consider the following:
Choosing the appropriate AI rhetorical helpers (e.g., Claude, ChatGPT, QuillBot, Grammarly) depends on how the human knowledge maker interprets context, audience, and purpose. Making this selection also depends on the knowledge of how that tool will affect the meaning that is created.
Some useful questions to ask when selecting an AI tool to support rhetoric:
Will generative AI enhance the rhetorical approach? Will the rhetorical helper clarify meaning and open pathways to more nuanced discussion? Will the tool obscure meaning or open more space for questions about accuracy?
Knowles’s (2024) delineation of human-in-the-loop writing processes and machine-in-the-loop may help to answer the above questions since “loops” represent two ends of a spectrum along which activities in knowledge making process may lie.
Knowles breaks this writing process into different “loads”:
Invention and arrangement task load: creating ideas, choosing genre and establishing structure to meet genre conventions
Style task load: rewriting or revising work
Memory task load: finding, evaluating, and integrating factual evidence into writing. Assessing generated outputs.
For example: Certain tasks, like copyediting, might rely heavily on AI tools with humans providing oversight, making humans in the loop. Other tasks, like working with sources, demand greater human involvement, placing machines in the loop for a more human-centred task.
*Adapted from Knowles (2024)*
Remember
Joksimovic et al. (2023) outlined that the “division of roles between humans and machines is rather static and is usually performed on arbitrary assumptions about the expected performance and the notion of who does what better” (p. 2).
Both can co-exist in a spectrum where human and AI influence fluctuates from task to task and with each iteration of what is being created.