Beth McMurtrie, a senior writer at the Chronicle of Higher Ed who focuses on teaching, initiated a lively conversation this summer about the problem of student reading: “Is This the End of Reading?” McMurtrie explores a number of potential reasons for the apparent diminishment in student ability and willingness to handle reading assignments in all disciplines. These reasons include social media, the impact of COVID, and limited experience with reading in secondary schools. Soon after, the media scholar Eileen G’Sell published a review essay, “Algorithms and the Problem of Intellectual Passivity.” G’Sell’s analysis of the impact of algorithmic feeds and filtering on students, particularly on their reading and analytical skills, is insightful. G’Sell draws extensively upon the recent book by Kyle Chayka, Filterworld: How Algorithms Flattened Culture.
These conversations have not been surprising in identifying the concerns or potential causes of the reading problem. Instead, I have found them useful in providing an opportunity for some exploration of reading pedagogy that I believe is needed since we tend to take it for granted. We give a fair amount of attention to how we teach writing, and that’s a good thing as far as I’m concerned; it results in better student engagement and learning as writers. But what are our assumptions with regard to the reading we assign and expect from our students? Is there a pedagogy of reading for the age of algorithms that we might develop or derive from what seem to be our better practices in the teaching of writing?
In my continuing interrogation of the potential uses and limitations of Generative AI tools for learning writing, I have had my eye on the difference between algorithmic and heuristic tools when used for learning. (See my article, for example, “Why Aren’t We Asking Questions of AI?”). I am also interested in what cognitive psychological perspectives on writing and literacy, studied since the 1980s, might tell us about the new impact of AI in education. For some further reading, see Alice S. Horning, “Neuroscience of Reading: Developing Expertise in Reading and Writing.”
G’Sell and Chayka remind us that our students have had algorithms filtering and determining a great deal of the information they engage and the culture they live in: recommendations for what they read, watch, listen to. The passivity that G’Sell observes is an outcome of algorithmic analysis: the issue is not that artificial tools are playing a role in our thinking, it is that the algorithms running the tools, procedures for problem solving that lead to definite solutions, are too good at providing answers.
Algorithms and algorithmic thinking are not foreign to us, either. But even if we are not old enough to remember listening to an album before streaming services or even compact discs, I think as educators we expect the learning process–at least the process of reading a text while researching a topic–to proceed more effectively and appropriately through heuristics rather than algorithms: “apply or approximate this concept with this new material,” rather than “follow these steps exactly to get a specific result.” (At least, that’s a claim I’m working on. My engagement with the generative AI chatbot Claude 3.5 Sonnet defines the differences and confirms some of my assumptions, but also challenges the view that heuristic tools and skills should be prioritized over algorithmic skills in learning. You can view the chat here.)
Here’s the definition from Claude: “An algorithm is a step-by-step procedure or set of rules for solving a problem or accomplishing a task, typically in a finite number of steps and with a specific outcome.” Heuristics are problem-solving tools, but unlike algorithms they are open-ended, trial-and-error rubrics and often experienced-based schematics for approximating and anticipating possible outcomes rather than executing a definite solution. They tend to raise more questions than they answer, which is what makes heuristics effective for a key cognitive element in the learning process, metacognition. It also makes heuristics quite familiar to my area of study, the teaching of rhetoric. In Greek, the word for discovery or invention, which is the first canon of rhetorical study, is heurisis.
In fact, one of the heuristics I have been developing in my classes to teach writing as well as critical reading and information literacy is the question-based, rhetorical heuristic known as stasis theory. The heuristic leads the user through 4 categories of questions that could be asked in any topic or case as a way to develop the stance. Having a handle on the right question is much more important for persuasion than the right answer; the wrong question means no audience. This is shorthand I am using as I explore certain AI tools: does the tool provide answers only, or invite algorithmic passivity? Can I use the tool, instead, as I use the stasis heuristic, to develop better questions for my research, better articulate the problems that will lead me to more research? I’ll be exploring with my English methods seminar this fall the use of Google’s NotebookLM, which assists with reading, annotating, and organizing research sources that I (not the AI) provide. There are, it seems to me, some crucial heuristic elements to this tool that are effective and appropriate and worth consideration. I’ll let you know how the experiment goes, and welcome feedback from your experiments with this or other digital tools in teaching reading and research. You can also take a look at a different, short assignment I have developed that integrates AI and the use of stasis questions. “Generative AI and Stasis Theory” was recently published by the AI Pedagogy Project at Harvard. [For more on the ways I address potential student AI use in my courses and focus on acknowledgment rather than prohibition or detection, revisit this post “Ethical AI: Educational Perspectives.”
I am currently developing some articles and presentations that organize and extend some of these strategies for thinking about heuristics and hope to share them in the coming year. As part of that, knowing that the concept of heuristics is interdisciplinary, showing up in education, psychology and cognitive science, behavioral economics, and computer science as well as ancient rhetoric, I would very much welcome further conversation and complication of my initial thinking.
In the meantime, one way I think we all might approach the issue of student reading more heuristically is to engage in some metacognition of our own. When we assign reading in our courses, what kind of work are we expecting students to do as readers? How much time should we reasonably expect them to give to the assignment? What role does reading play in the course learning goals, in the overall work of the semester?
I have clear answers to those questions with regard to writing assignments in my courses. I am less clear and confident regarding my reading assignments. And I am even less confident that we, as a faculty, have a good understanding of the reading experiences of our students across their courses in a given semester, in terms of both quality and quantity.
The issue of credit hours has prompted us to define and articulate more clearly the workload in our courses, within and beyond the class period. Wake Forest’s course workload estimator can help do that, prompting further thinking about the quality and quantity of assignments, reading as well as writing. It quantifies assignment time differently, for example, depending on different goals for reading: page density, difficulty, and purpose (survey, understand, engage). Like a good heuristic, this estimator prompts metacognition about the purpose of workload in a course; it doesn’t promise to do the teaching or deal with that workload for us. That’s probably good for our sake.
Post by Dr. Sean Meehan, Director of Writing and Co-Director, Cromwell CTL, Washington College
If you have a pedagogical topic or issue you’d like to share in a blog post on The Catalyst, or in another venue on campus, please contact the Cromwell CTL.