How I Killed an AI-Generated Essay During Its Defence

Our assignment was to analyse either an online store or a physical retail environment using the theories and methods presented in these books. Each book offered its own angle and its own perspective.

The final report followed a standard academic structure. It had to include theory, methodology, empirical observations, analysis, and conclusions.

After submitting the report, every student was assigned a peer whose work they had to review. We had to read each other’s reports, prepare presentations, and provide feedback during the seminar.

I received a report written by an economics student. I approached it with genuine interest because the subject was relevant to me. The central claim was that the atmosphere and environment of a store create a positive experience for visitors, which ultimately encourages them to spend more money.

This is not an unreasonable argument.

The problem was that, after reading the report, I experienced a familiar feeling. It was the same feeling I often have when reading AI-generated content.

You read the words. The sentences appear logical. The paragraphs are grammatically connected. You follow the text from beginning to end. Yet when the journey is over, your intellectual balance is exactly zero.

You have consumed words, but you have received no information.

The report contained essentially one idea: a good atmosphere improves the customer experience and therefore improves sales. That idea was repeated perhaps thirty or forty times. Each time it was slightly rephrased, attributed to a different author, or presented as if it were a new conclusion.

Specific page references were mostly absent. Some of the authors may indeed have made statements that sounded similar, but the actual arguments of their books were different and considerably more complex.

After reading the report for the second time, I realised that I was spending my time reviewing AI-generated slop.

That was the moment when the situation became more than an ordinary university assignment.

I Use AI Too

I use ChatGPT and other AI tools when writing essays. I do not believe that using AI automatically makes academic work dishonest or intellectually empty.

The difference is in the process.

Fortunately, I completed my master’s degree before the AI boom. I had to write a large number of essays without generative AI, paraphrasing tools, or systems that could polish and restructure my writing within seconds.

That experience matters because it taught me what the writing process is supposed to contain.

For my current assignment, I first read two of the course books. While reading, I took notes and collected quotations that I expected to use later. I gathered the material in one place and began thinking about the structure of the research, the method I wanted to apply, and, most importantly, how I wanted to conduct the analysis.

Only after that did I carry out the empirical work.

When I had observations and conclusions, I dictated my thoughts through an AI assistant. The system helped me organise them, improve their structure, and place them into a clearer academic context. I also used AI to explore possible approaches to the research and to test whether my methodology made sense.

AI assisted my thinking. It did not replace it.

The final report still came from reading, note-taking, observation, interpretation, and judgement. AI helped me express the result more efficiently.

The report written by the student assigned to me for peer review appeared to have followed the opposite process.

It was a strong feeling that the books had either not been opened or had not been seriously engaged with. None of their useful concepts appeared in the report. Some claims attributed to the authors appeared to be entirely hallucinated. I tried to locate several of them in the books and could not find them.

This created an uncomfortable problem.

When There Is Nothing Positive to Say

Normally, when I give feedback, I try to construct what is sometimes called a feedback sandwich. I begin with something positive, introduce the criticism, and finish with an encouraging observation.

Sometimes this is useful. Sometimes it is simply a polite way of wrapping criticism in softer language.

But in this case, I could not do it.

There was no serious engagement with the theory. There was no meaningful empirical analysis. There was no independent interpretation. There was only one generic statement, repeated again and again through AI-generated reformulations.

There is an important difference between a student who tries, makes mistakes, and produces imperfect work, and a student who submits a text generated without thought.

When someone has genuinely tried, the weaknesses can be identified and improved. There is usually something valuable to recognise: an interesting observation, a promising method, a useful question, or at least evidence of intellectual effort.

When the entire work has been outsourced to AI, the appropriate response cannot be the same.

During the seminar, I therefore gave the student feedback that may have been unpleasant to hear. In other words, I killed the essay during its defence.

But pretending that the report was acceptable would have been dishonest. It would also have been unfair to everyone who had actually read the books, conducted the analysis, and attempted to produce original work.

The Real Cost of AI-Generated Work

My first reaction was frustration about the time I had wasted.

I had been asked to read and evaluate another student’s work. Instead, I had spent part of my life analysing a machine-generated imitation of academic writing.

But the larger problem is not my wasted time.

The larger problem is what happens to education when students lose the ability to critically evaluate both AI and the content it produces.

I now understand professors and lecturers who consider leaving education because they are tired of spending their lives providing detailed feedback on texts that no student seriously wrote.

A teacher can help a student improve their reasoning. A teacher can challenge weak assumptions, explain methodological problems, and show how an argument can be developed.

But what exactly is a teacher supposed to do with twenty pages of automated filler?

Who is receiving the feedback?

The student who did not formulate the argument? The machine that generated it? Or the university system that continues to treat the document as evidence of learning?

At some point, the educational process becomes a performance in which no one is actually learning. The student pretends to write. The university pretends to assess. The AI produces the evidence required to maintain the illusion.

We Are Standing at a Fork in the Road

As a society, and perhaps as a species, we are standing at a fork in the road.

One path leads towards something close to a technological utopia. AI takes over routine work, helps us organise information, removes unnecessary friction, and allows us to focus more deeply on knowledge, creativity, judgement, and analysis.

In that future, technology strengthens human thought.

The other path leads towards cognitive delegation.

We stop using AI to support our thinking and begin using it to avoid thinking altogether. We outsource reading, interpretation, writing, analysis, and eventually decision-making.

In that future, we may become the last generation capable of sustained cognitive work.

The question is therefore not simply whether technology will replace certain jobs. The deeper question is what technology is for.

Does technology work for us?

Does it extend our abilities?

Or do we gradually become a weakened, vestigial component of a technological machine that no longer needs us to think?

The most serious threat may not be mass layoffs or corporate restructuring. It may be what AI is already doing to education and to the intellectual development of the next generation.

Education is not only a system for producing reports, grades, and diplomas. It is a process through which people learn to read difficult material, distinguish evidence from claims, tolerate confusion, develop judgement, and construct their own understanding of the world.

When AI replaces that process, it does not merely change how students complete assignments. It changes how they mature intellectually.

The Ethical Boundary

Canvas Assistant began at the grassroots level as a university project.

From the beginning, we faced an obvious temptation. It would be technically possible to create systems that help students deceive the educational system. We could build tools that answer quizzes, complete examinations, or operate directly inside platforms such as Canvas.

There is clearly a market for such products.

But we believe that crossing that boundary would be profoundly wrong.

What we are building is intended to help students extract transcripts, structure lecture content, navigate educational material, and use lectures as searchable sources of information.

The purpose is not to remove the student from the learning process.

The purpose is to make the process of working with information more efficient, so that students can spend more time understanding, comparing, analysing, and questioning it.

AI should help students train their ability to work with data and information. It should not eliminate the need for cognitive effort.

That distinction may sound simple, but it is fundamental.

A calculator can help someone solve a complex mathematical problem after they understand the underlying principles. It can also be used to avoid learning basic arithmetic. The tool is the same. The educational outcome is not.

The Old World Is Ending

The old educational world is disappearing.

The previous approach to writing, research, data, and assessment cannot survive unchanged. A new world is emerging, with new tools and new rules.

Students must learn how to operate inside it.

At the same time, many lecturers and educational institutions are still living in the past. Universities are extremely slow-moving organisations. They struggle to adapt to sudden technological change, especially when that change challenges the basic assumptions behind assignments, examinations, and academic authorship.

But institutional failure does not remove individual responsibility.

Students must decide what they actually want from their education.

Do they want a certificate?

Do they want the appearance of competence?

Or do they want to become more capable than they were before?

AI can help someone complete a course without learning very much. It can also help someone learn faster, explore more deeply, and work more effectively than previous generations could have imagined.

The technology does not make that decision.

The student does.

This is why we ask students to use AI responsibly. Perhaps that appeal sounds naive. But the alternative is to build an educational future based entirely on surveillance, detection, restrictions, and mutual distrust.

That is not a sustainable system either.

The more important goal is to create tools, norms, and educational methods that reward genuine understanding rather than the mechanical production of text.

We need to teach students not merely how to use AI, but when to resist it, when to question it, and when to think without it.

An Invitation

We are still developing our project, and we do not pretend to have solved these problems.

We regularly meet with students from different universities to discuss what can be improved, how AI is changing education, and how access to educational content can become more effective without undermining the learning process itself.

Students, lecturers, researchers, and others who have serious thoughts about these questions are invited to join our advisory board and participate in the discussion.

The future of education will not be determined only by universities, technology companies, or government policies.

It will also be determined by the everyday choices students make when they open an AI tool and decide whether they want assistance with thinking or an escape from it.

Thank you for reading.

Your Henry Fisher.