Should we kill the essay?

Large language models, quite simply, have ruined everything. Every thinking task can be cognitively offloaded to a generative AI tool. ChatGPT can complete every technical writing skill in seconds and for free. It is nearly impossible to tell the difference between an essay by a human writer and one produced by AI. Students have no need to think, plan, imagine, research, draft, edit or proofread their work any more. 

In addition, our initial attempts to secure essays from plagiarism or other breaches of academic integrity have proved feeble. AI-detection tools are unreliable, and setting essays under exam conditions only punishes students who can’t write thousands of words under strict time constraints. 

The question seems inevitable: does AI spell doom for essays?

In their current form? Yes. 

As part of the assessment process? No.

Essays do have great value…

For a start, essays have been a mainstay of educational assessment for centuries and remain an important method of communicating knowledge. This is in part because of their sheer versatility as an assessment task. They work in every discipline as they provide opportunities for students to demonstrate critical thinking, research skills, nuanced communication capability, creativity and personal expression, as well as basic literacy. Science students can demonstrate communication; humanities students can demonstrate research; engineering students can demonstrate thinking. 

• The renaissance of the essay starts here
• Sensitive marking and the end of the line for the academic essay
• With AI at their fingertips, are students still learning?

Most students will come into university with a clear expectation and understanding of essays from years of writing assignments in school. Additionally, setting an essay is low-cost in terms of teacher time. All a course coordinator needs to do is write a question and a criteria sheet, and the assessment is complete. Precious preparation time can go into other teaching and learning activities.

…but their downsides metastasised 

However, the explosion of AI tools in the past few years have raised questions of whether essays are still a viable way evaluate the quality of students’ thinking and writing ability. In early 2023, the writing coming out of ChatGPT 3 or Google Gemini was obviously and identifiably sloppy, but subsequent models have improved their output to the point where they are on a par with human performance. If AI can do the entire task, we have to consider essays as totally insecure.

AI’s truly unforgivable sin, though, is that it revealed all the flaws of essays as assessment. They’re usually high stakes, stressful and marking intensive, all while potentially having hidden criteria (such as originality or literary flourish). Those are tricky to assess at the best of times, let alone when folded into a comprehensive task like an essay. Students may fail to engage as critically as we would like as well, as in when they use criteria sheets to reverse-engineer an essay structure rather than thinking through their writing’s coherence themselves.

Responding to the essay cataclysm

If we want to keep the benefits of the essay but reduce the risks to academic integrity, student skills and engagement, we do have options.

Chunk and chew

When we use essays as assessment, we’re talking about many smaller tasks that come together into one highly efficient whole. Given that the whole is no longer secure, we can “chunk down” essays into their constituent parts and maintain the intended learning outcomes. It starts with asking: “What skills and knowledge do we want the student to demonstrate?”

Essays generally include elements such as interpreting the question, students’ reflection on their views, information gathering and selection, planning and organising, position-taking, long-form writing, editing for structure and content, and proofreading technical elements. These could be done as part of teaching and learning activities, formalised in a more secure manner, and submitted in class time. 

We don’t need every moment of the process to be safeguarded from the scourge of AI, but we need to focus on process over product. We need to create better assurances of the learning pathway rather than overemphasising the outcome.

Alternative assessments

The essay could be the first stage of an assessment sequence rather than the end in itself. The student could create an essay on a given topic but then be asked to demonstrate greater knowledge and understanding through teaching it in a flipped classroom, arguing it in a debate or articulating nuances in an oral question-and-answer session.

The holistic essay-writing task can be easily achieved with almost no prompting now; a student can copy and paste their task into ChatGPT and have a well-formatted essay pop out in seconds. With the new “deep research” functions, these essays can be supported by citations with the click of a button. This is less of an issue if we move to a multi-modal task. A poster may be able to display research skills and knowledge as well as an essay can. Oral presentation can demonstrate communication. Argument maps can showcase logical thinking and coherence. All of these are likely to be far more engaging for students and cheaper for marking budgets.

The constraints of teaching, learning and marking time are an unavoidable factor in building an appropriate assessment task. There may be opportunities to shift the format of teaching and learning activities or re-weight the essay so other assessment are more valuable as learning opportunities. Could lectures be seminars? Does the essay become formative rather than summative? No solution is perfect here. Experimentation is needed for the best fit for your context.

Is the assessment used for pedagogical or logistical reasons?

More than anything, AI disrupts the tidiness of using essays. We have to ask whether they are fit for purpose, and why the essay is part of a module’s tasks in the first place. We might as well make it an invigilated exam if our primary concern is efficiency, simplicity or budgeting limitations. 

If it’s for demonstrating student achievement of learning outcomes, we need to check whether we care more about the end product or the process. We can track the progress of a student in long-form writing rather than evaluating the final output. 

Generative AI may not end up killing the essay, but it has dealt it a body blow. We have a moment here to consider how best to help it recover – or if we want it to recover at all. 

Luke Zaphir is a learning designer (HASS) in the Institute for Teaching and Learning Innovation at the University of Queensland.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.