The Case for Take-At-Home Exams

Take-At-Home exams aren’t particularly common. Students have no supervision. Students can cheat. With LLMs, in particular, students are just bound to cheat. They are going to find all sorts of information, and it’s already pre-modified so that they don’t even need to properly read it to pass a rudimentary plagiarism check.

Take-At-Home exams also suffer from making the student feel safe in not having complete coverage over the syllabus: they set the topic. They can afford to drop a few points.

I believe these two points to be as well-argued as they are made irrelevant by further scrutiny.

Let’s start from the bottom: “students can afford to skip parts of the syllabus”. I’ll start off by saying that they already do. Even model students, at times, are forced by the ludicrous schedules of asinine departments to place their bets on which topics are going to come up at the particular sitting they’re on. Even assuming perfect randomisation, however, the only guarantor of the student’s complete preparation is an assumed fear that dominates the student’s interest. I hope that the horror that this assumption represents is immediately visible. Fear is a strong, but largely ineffective incentive. It leads to chasing outcomes, rather than process. Using fear as a tool only leads students into the embrace of Goodhart’s law, as they try to signal competence in a game (the exam), rather than acquiring it[1], [2]. This leads to memorisation, trying to outfox the professor by guessing the question, mechanical exercise, cheating. Fear is also linked, unsurprisingly, with stress. Stress is consistently linked to poor learning outcomes in a plethora of state-of-the-art pedagogical research[3], [4].

A strong patch to this kind of exam is the oral exam. By negating the efficacy of many of these “tricks”, and adding some element of human flexibility and presence into the loop, it can stem some of the fear, in favor of a desire to acquire competence. When met with flexibility and understanding, the mechanics of the exam play second fiddle to the student’s ability. Unfortunately, modern University has an obsession with speed and costs. It’s borne out of the fact that it needs money, and it’s given more money relating to its throughput. Therefore, more speed more money more university.

This is where take-at-home exams can shine. They outsource a large part of what would have to be in the oral exam to the student. They allow for revisions. They allow for cheap retries. They allow for flexibility, and the oral component, i.e. the discussion, can be very short. We use this already in dissertations and PhD defenses. It’s the standard mode of interaction in the research world, where research proposals are presented in short meetings, papers are exposed in short slots at conferences, and questions follow. This is a subset of competence that’s entirely neglected in written examination especially: creativity in looking for questions, synthesis, clarity. Instant correctness, frankly, is secondary in the modern world. We have instant access to information. Instant access to data. Instant access to methods. Simple natural language compilers^{1 1}I’m referring to LLMs. For most simple practical purposes in data analysis, they are natural language compilers. Their points of failure come at higher levels of complexity in a codebase. Data analysis is mostly library calling, and they’re pretty good at that. to examine that limitless data, and parse that information. No good researcher is measured on their recall anymore. They weren’t in the past either, for the most part, but now more than at any time before, there’s no point in testing recall. It’s a pointless exercise in endurance-checking, more than anything, and it’s why students subconsciously reject it more than they did in the past. It’s harder to churn through material when you’re convinced with excellent reason that it’s largely useless.

A Take-At-Home, therefore, is no less valuable as an evaluation tool than the written exam. Actually, the former’s strengths are better aligned with the needs of a modern world than those of the latter.

Now onto the cheating component. There’s no doubt that students have more access to external help in take at homes than in written or oral exams. I argue that students are less likely to want to cheat, and that this means that a comparatively lower amount of friction is needed to get them to not do so.

First and foremost: cheating in a take-at-home is hard. Not because of supervision, but because cheating requires some degree of creativity: when you’re already allowed to go on the internet to find your sources, it takes a special kind of effort to cheat. No self-respecting student would simply copy an existing article as even in a vacuum plagiarism software would catch it trivially. Before LLMs, that’d largely have been it. Now with LLMs, every student has access to a clean-room machine that can “rewrite” their sources with some added glue and stylistic changes. This is where the discussion, or defense, of the product comes in. If a student mindlessly copies from the output of an LLM, it immediately shows. If they put in the effort to understand what the machine wrote, frankly, they did their job, because they understood the subject, unless the product was either off-base, or sub-par. In that case, the grading will already reflect the lack of quality.

This hybrid formula’s advantage is in its speed on the faculty’s side, the lack of useless stress on the student’s side, and the superior representation of the student’s competence on the institutional side. What’s more, it aligns a student’s incentives with those of the examiner by allowing a modicum of personal creativity into the process. A well made take-at-home, truth be told, asks more of the student. It asks them to understand the subject enough to be creative within it. It asks them to find it within them to seek unanswered questions, and try their hand at finding answers, rather than being passive observers in someone else’s research, waiting patiently for their turn, that’s to come five, six, ten years down the line. It asks more, indeed, but it demands less. It’s in this echo of the recognition made by the European Charter for Researchers that sits the value of take-at-homes: knowledge is borne out of freedom and curiosity[5]. In other words, if you wish for knowledge to thrive, you need to nurture it, not squeeze it out of warm bodies.