AI Makes Students Dumb and What We Can Do About It

I flunked 40% of my students this semester. Do you know what impossible assignment I gave them?

“Write a 3-pages long Medium-styled article about any AI-related topic you find interesting. Do not use LLMs.”

The results were striking. I did expect some students to cheat and use AI, but I did not expect them to be so bad at it. While the idea was to grade my students based on their ability to write, I found myself grading their (in)ability to read and comprehend the AI-generated text.

To all the people worried about a Terminator-like AI takeover: do not worry. Humans would voluntarily let AI do all the thinking for them.

In this article, I summarize my thoughts and ideas for countermeasures for how to preserve literacy and the ability to think critically in the AI age. While the main purpose of this text is to let me blow off steam after the most frustrating semester as a lecturer, I hope my mitigation strategies might be useful to other teachers and students. Also, I am really interested in other teachers' solutions and thoughts on this topic.

While reading bachelor’s theses, I noticed most students struggled with coherent writing — hardly surprising, given how little they’re asked to write during undergrad. Expecting them to produce a 100-page thesis without prior practice felt unfair. So, I decided to add a low-stakes writing assignment to my course: a Medium-style article on an AI topic. It’s informal enough to ease beginners in, yet structured enough to build critical thinking. Plus, developers often use Medium to showcase skills, so I figured this could double as portfolio buildup for summer internships.

• Another pattern I noticed in the theses was students overusing bullet points, which is also very characteristic of AI-generated responses.
• Even when they wrote the text themselves, their style often mirrored AI’s fragmented, list-heavy approach.
• At first, this struck me as odd, but it made sense: if most of your reading and writing happens within chat-based LLM frameworks, you inevitably absorb their conventions.

This realization solidified my decision to explicitly prohibit any AI use in the assignment. I made it clear that any detected reliance on AI would result in failing the class, not just to test their writing skills, but to push them to think critically without outsourcing their cognition.

Another thing is that if you use AI skillfully, no one will notice that AI is part of the process… Writing with AI may seem like a very trivial process. Apparently, it is not, since Adrian Wallwork wrote an entire book on “AI-Assisted Writing and Presenting in English”!

• weird metaphors: “It’s like having a security analyst that never
  sleeps.”
• using amalgamated styles of writing
• referencing old yearly reports (that actually reveal when the AI model was trained …) e.g., GrandViewReports (2023), McKinsey (2023), Forbes (2023)
• bibliography includes titles translated from another language e.g., an article in Polish references an article with an English title translated to Polish without mentioning that it does so
• overuse of references that do not really refer to anything specific
• a quote as if a student interviews someone, e.g., adding an out-of-context quote from a random individual from another country
• redefining the same acronym on multiple occasions, e.g., halfway through an article about AI, a paragraph begins with “Artificial intelligence (AI) …”
• separating several prompt outputs with a line (a very drastic case, but for real, I had 4 people who did this!).

In each case of a failed article, the AI usage was painfully obvious — visible at first glance. What shocked me most wasn’t just the poor quality of the submissions, but the students’ insistence that they could prove the work was their own. As much as I can understand the lazy approach, after putting effort into defending their work, these students clearly didn’t comprehend the text they’d submitted. Their arguments revealed a fundamental disconnect between the content they’d handed in and their actual understanding of it.

I am not a tech luddite. In fact, I teach only AI courses and co-founded an AI startup. I am not looking at this from the perspective of a person who isn't familiar with these technologies or dismisses them. I am writing from the perspective of a person who uses those tools every day and develops similar ones, yet I still see their harmful aspect when they are used as a crutch to avoid critical thinking.

Critical thinking is a key element of education. When students rely on AI to generate their work, they are essentially outsourcing their cognitive processes. This phenomenon has even received its own scientific term — “cognitive offloading”. For example, writing isn’t just about filling a page with words; it’s about formulating thoughts, organizing ideas, and communicating them effectively. These skills are essential no matter what career you choose. Writing reports, industry articles, and public presentations all require mastery in formulating arguments and skillfully communicating them. The successful completion of these tasks can have a significant impact on career advancement. For this reason, even if a thesis is the only “book” a student ever writes, it is essential for their future careers. This task teaches them how to produce 100 pages of coherent, evidence-based material that proposes arguments and successfully defends them.

Some argue that if AI can complete an assignment, then the assignment itself must be flawed. I disagree. The purpose of an assignment isn’t just to produce a final product; it’s to engage students in the learning process and develop their skills. If we accept the idea that any task AI can do is unnecessary for humans, we forget the purpose of curricula. We learn to solve simple problems first and gradually take on more challenging ones. You can’t jump straight into complex problems without mastering the basics. It’s similar to teaching kids multiplication without a calculator. The goal isn’t for them to manually multiply three-digit numbers, but to develop mathematical thinking. This way, when they eventually use calculators, they won’t just be pressing magical buttons mindlessly; they will understand the underlying concepts.

A common argument in the discussion on the AI usage in education is that teachers who are AI critics refuse to adapt to change and to learn what this technology is about. The “skills of tomorrow” aren’t about teaching people a specific, trendy skill, like once programming or now prompt engineering. It’s about teaching universal knowledge that will help them navigate whatever the future holds. One such skill is critical thinking.

Add recommended and required reading as a prerequisite for your class

I know this is actually mandatory, but many course syllabi just have whatever books from decades ago that are impossible to find. I limit the number of required and recommended readings and focus on specific, relevant chapters. This makes the material less daunting for the students. For some classes, I also assign a short article to read before each session. If you’re interested in my reading lists, I can share them later.

Add reading comprehension exercises

I teach students to read research papers in all the classes that I teach. I take a section of a paper and prepare a list of questions. You can find examples of such assignments in my worksheets. From my experience, students usually initially struggle with these tasks. If we expect them to read a couple of papers while working on their thesis, practicing reading papers in class should help with that.

An example of reasearch paper reading comprehension practice.

Add writing assignments

Standard lab reports often don’t require strong writing skills, so I add a writing assignment that challenges students to practice formulating arguments, such as writing a Medium-style article. This improves their writing skills and encourages critical thinking (or at least it should, you see how it worked out in my case… ).

Grading

Check your students’ understanding orally. Project-based assignments are still cool, but making a student present them afterwards provides clear evidence of their understanding and mastery. In such a scenario, I don’t think it’s hard at all to figure out if the student worked on their own on the assignment or outsourced it to another person or AI.

Lastly, I believe it’s important to encourage student responsibility. As academic lecturers, we teach adults, and while early adulthood can be challenging, it’s time for students to take ownership of their learning journey. Our role is to provide resources to those who want to learn, but whether they engage in the learning process is ultimately their choice. I will help them when asked, but eventually, I will simply grade them and verify whether they have acquired the knowledge and skills the course provides.

I truly believe that writing is an essential skill that can open up opportunities that would otherwise be missed. Your word choice and the ability to construct arguments can help you build a case and create better opportunities for yourself. For example, a well-written motivation letter can set you apart from others with similar skills or experience.

GRE writing prompts

In Poland, at least, we learn to motivate our arguments in writing by using ideas from the required reading. While this may seem challenging, it’s not. You can brute force it by memorizing a few motifs from classic texts like Antigone and Macbeth and pass an exam. I think this is one reason why students struggle to formulate arguments on their own.

If you want to learn how to analyze and formulate arguments independently, I suggest starting with GRE writing prompts. The GRE is a standardized test often required for admission to graduate and business programs, and its analytical writing section is a good place to start. You can find practice prompts online.

How to practice — for each of the writing tasks:

1. Read the task’s description.
2. Select a prompt.
3. Write your answer to the prompt (in either English or, if that’s too hard, start with your first language).
4. Use an LLM to grade your answer. Pass your text and evaluation criteria to an LLM to grade your answer. Specify in your prompt that you want a harsh review.

Here you can find:
- Prompts (topic pools)
- Evaluation criteria

Read a f*$% book

While I am usually all about “we have different learning predispositions, we have visual/auditory/kinesthetic learners, etc. But if you want to learn how to write, you need to first learn how to read, e.g., learn how to identify what argument someone tries to propose by using these specific words, and what explicit and implicit meanings are. Popular science books are great for this.
Here are some of my suggestions for lightweight companion books for a computer vision course:
- “The Worlds I See” by Fei-Fei Li
- “The Alignment Problem” by Brian Christian
- “The Little Book of Deep Learning” by François Fleuret
- “The Deep History of Ourselves” by Joseph Ledoux

Games that teach you how to deliver a structured story.

Play a game

If you want to start with something chill that you might as well treat as leisure, there are multiple games that help to learn how to structure your ideas into a coherent story:
- Story Cubes
- Once Upon a Time
- The Moth Presents: A Storytelling Game (they also have a book on how to tell a story)
- Oh Captain, My Captain

The harsh reality is that nowadays, with the overproduction of university students, to be mediocre, you need to be pretty good, and university education is a prerequisite, not a door-opener. The greatest benefit of attending a university is acquiring the ability to learn fast on your own. It takes a long time to obtain a degree, and unless you learn something in the process, it is a waste of time and money (at least in my opinion).

Lastly, a reminder that good old note-taking is a way of organizing knowledge in your brain.

As much as I love teaching, I refuse to be part of a system that grants degrees to students who lack basic literacy skills. The learning that does happen seems to be focused on finding excuses rather than developing essential skills. Every student who failed had the same ready-made excuse: they “only used ChatGPT to refine their answers and correct grammatical mistakes.” Each of them appealed their grade, claiming they were graded unjustly. These appeals left me baffled. When I review an appeal, I have to write a detailed report explaining that the student cheated. This report stays in their records, and anyone reviewing a future appeal can access all past appeals.

This cycle of appealing decisions and then appealing the appeals seems to stem from a culture where students believe that persistence in challenging outcomes will eventually yield a favorable result. It reflects a deeper issue where the focus is more on avoiding responsibility and seeking shortcuts rather than genuinely engaging with the learning process and accepting the consequences of one’s actions.

In my teaching history, I had one student who successfully appealed and received an opportunity to be re-evaluated through an additional exam. They scored 0% on the exam (0 out of 50 points!). The student’s reaction was a shrug and “no worries, I will try again next year”. This experience perfectly illustrates how the focus on appealing decisions rather than genuinely engaging with the material can lead to a poor understanding of the subject matter. I mean, you really need to do nothing to score a zero…

Anyways, I quit working as an academic lecturer after 4 years. I have lots of teaching materials that I share (and plan to continue to share) online for free with anyone eager to learn (pen-and-paper worksheets, lab exercises, educational videos). I also enrolled as a volunteer teacher for an amazing initiative, Code in Place.

Please, share your thoughts and suggestions in the comments! I really need feedback, this semester left me feeling like I was in the Idiocracy movie!