---
title: "Oral Fixation"
author: "Geoffrey Challen"
date: "2026-11-12"
description: "Oral exams are back. That's a problem."
canonical: "https://www.geoffreychallen.com/essays/2026-11-12-oral-fixation"
---

# Oral Fixation

One response to AI’s ability to help students get around certain kinds of assessment has been renewed interest in oral exams.
“You won’t be able to AI your way through an oral exam,” a Cornell professor told the [Associated Press](https://www.adn.com/nation-world/2026/04/22/perfect-homework-blank-stares-why-colleges-are-turning-to-oral-exams-to-combat-ai/) in April.
“I want oral exams everywhere now,” said another, at NYU.
Stanford’s academic integrity working group [recommends](https://studentaffairs.stanford.edu/news/academic-integrity-working-group-addresses-generative-ai-and-exam-policies) in-person formats such as oral exams.

I find this incredibly concerning.

I understand the appeal.
An oral exam puts you face to face with a person.
You can watch them, and you hear the words that come out of their mouth, which makes an oral exam very hard to fake.
The examiner can also guide the exam: ask follow-up questions, and probe where a student seems weak.
That last part is more of a mixed bag than it sounds, for reasons I’ll come back to.

But returning to this style of assessment could represent a significant educational setback.
My concerns run along two lines: fairness and frequency.

Fairness first, taken broadly.
An assessment is fair when it is an accurate representation of what a student knows.

For hundreds of years we’ve used written assessments, and I think the reason has to do with the role of writing in our culture.
There is some evidence, though it’s debated, that the importance of the written word is declining[^1].
I just read James Marriott’s [_The New Dark Ages_](https://www.penguinrandomhouse.com/books/829140/the-new-dark-ages-by-james-marriott/), which argues that the written word was essential to the formation of modern society and of certain kinds of government.

Part of the reason is that writing separates an idea from the person espousing it.
If I’m talking to you, I can persuade you in ways that rely on all the other signals humans send each other in conversation.
I can project confidence.
I can talk over you.
I can make eye contact.
And nothing is recorded.
It’s hard for you to remember exactly what was said, and hard to go back over it, alone or with someone else.

If I have to write my argument down and give it to you, your relationship with it is totally different.
You’re a reader.
You read it without me there, usually in private, somewhere else.
You can connect different parts of it, and two things that sounded like they made sense independently may no longer make sense on the page.

All of the arguments for writing as a way of expressing our thoughts clearly apply equally to writing as a mode of assessment.
Writing forces a student to put down something coherent, and it gives the evaluator a different relationship with the evidence the student presents.
It also gives the student time to contemplate.
Think about what you do when you write an essay.
You might write an outline or take some notes.
You might try a few things in a few places, delete what doesn’t work, and rework the rest.
You can read over your answer to make sure it’s correct.
This contemplative approach mirrors how people usually do meaningful things, making oral examination a poor fit for many topics.

Put two people in a room, one examining the other, and a lot of social dynamics come into play.
The student may be nervous[^2].
The exam may not be in their first language, which can inhibit their ability to communicate in a way that writing would not.
They may have manners of speech the examiner doesn’t respond to well.
They may feel the examiner is antagonistic toward them.
People just don’t always get along.
Writing separates the examiner from the student, and that separation makes assessment more impartial.

So an oral exam doesn’t just give up the benefits of writing.
It brings in a lot of noise.
What somebody looks like.
Whether they smell good.
Whether the examiner finds them attractive or unattractive.
I’d encourage anyone excited about oral exams to read [_Noise_](https://www.amazon.com/dp/0316451401) by Daniel Kahneman, Olivier Sibony, and Cass Sunstein, which documents how variable even well-trained human judgment is, and how rarely anyone measures it.
Oral exams open that space wider.
When a group of Swedish engineering courses moved online during the pandemic, [the grade premium for attractive female students in non-quantitative courses went away](https://doi.org/10.1016/j.econlet.2022.110782).
If you’re turning in something on paper, the person grading it isn’t influenced by whether you’re chewing gum, the color of your hair, your gender, what you’re wearing, or whether they find you attractive.

This is also why the examiner’s ability to guide an oral exam is a mixed bag.
Asking good follow-up questions is very difficult to do consistently, across students and across examiners.
The research on oral exams in medical education, where they have a long history, finds that [traditional, unstructured oral exams are much less reliable than structured ones](https://pmc.ncbi.nlm.nih.gov/articles/PMC10369684/).
And the more structured an oral exam becomes, the less room the examiner has to guide it.

If the goal of an oral exam is to help people learn to interview better, or other skills that require face-to-face interaction, then fine.
But most of what people are now proposing to examine orally doesn’t fall into that category.
All of those other signals are noise, and they will bias the results in complicated ways.

There’s a reason we stopped giving oral exams.
Oral exams were the norm in American schools in 1845, when Boston’s School Committee gave its students written exams instead, and written exams spread from there.
Horace Mann, then secretary of the Massachusetts Board of Education, championed the change[^3].
A written exam, he argued, was impartial, and took away “all possibility of favoritism.”
Only the people in the room could see how an oral exam went, but written answers left “a sort of Daguerreotype likeness” of the students’ minds that “remains a permanent record.”
Nearly 200 years later, those same fairness concerns about oral examination remain as valid as ever.

The other big problem is frequency, which incorporates scale, cost, and time.
Oral exams take a lot of human time, and human time is expensive, so you can’t do them very often.
That was part of Boston’s problem in 1845: it had more than 7,000 students to examine.
I ran the numbers for [a talk at Berkeley](https://www.geoffreychallen.com/talks/2026-10-28-computing-education-in-human-language) last month.
This fall CS 124 has 1,081 students.
Each week they spend about an hour in our [computer-based testing facility](https://www.geoffreychallen.com/essays/2021-07-23-computerized-testing-for-cs1) taking a quiz.
Replace that with an hour across from a member of course staff, and that’s 1,081 examiner-hours a week.
Our staff is 84 people, who together work about 570 hours a week.
So we’d need nearly twice the staff we have.
Cross-checking with a second examiner to reduce noise doubles it again: 2,162 hours a week, nearly four times our current staff hours.

And it’s not clear that’s the best use of that time.
It would take my staff away from the valuable work they do now.
Today they do essentially no assessment: all of their time goes to working with students and supporting their learning directly.
With oral examination, they’d do nothing but assessment.

If CS 124 has taught me one thing, it’s [the importance of frequent assessment](https://www.geoffreychallen.com/essays/2022-04-17-eliminate-final-exams).
When you assess students, they learn, and the more often you assess them, the more often they study.
The difference between cramming and studying is just frequency.
Cramming is waiting too long and trying to study too much at once.
Take that same amount and do a little every week, and now you’re studying, and it works.
Oral examination is incompatible with high-frequency assessment, one of the few pedagogical strategies I’m pretty sure would work in any course.

A better answer is some form of computer-based testing.
At Illinois we have the [Computer-Based Testing Facility](https://cbtf.illinois.edu/), which allows proctored, secure examination at high frequency.
We use it [every week in CS 124](https://www.cs124.org/educators#frequent_small_assessment).

There are also exciting opportunities beyond dedicated physical spaces.
Suppose students installed a program on their computers when they took my course.
At the beginning of any class, I could run a short secure assessment: my staff proctor the room, and the software locks down each student’s machine.
That’s an environment like the testing facility that I can set up and tear down in any classroom, for ten minutes or the entire hour.
It’s one more way to control AI usage during assessment so that we can determine what a student actually knows.

And AI itself can help.
With [Craig Zilles](https://siebelschool.illinois.edu/about/people/faculty/zilles), I’ve been working on [conversational assessment](https://conversationalassessment.org/), an exam conducted as a written conversation with an AI interviewer.
Interactive like an oral exam, written so it’s impartial, and cheap enough to give every week.

[^1]: The share of American adults who read for pleasure on a given day fell from 28% to 16% between 2003 and 2023, according to [an analysis of the American Time Use Survey](https://www.ucl.ac.uk/news/2025/aug/proportion-americans-reading-pleasure-fell-40-over-20-years) published in _iScience_.

[^2]: When Panos Ipeirotis ran AI-conducted oral exams for 36 students at NYU, [83% found them more stressful than written tests](https://the-decoder.com/nyu-professor-fights-ai-cheating-with-ai-powered-oral-exams-that-cost-42-cents-per-student/), and only 13% preferred them. That’s with no human examiner in the room at all.

[^3]: From [a history of the period](https://files.eric.ed.gov/fulltext/EJ842865.pdf) by Kenichiro Miyamoto, which quotes Mann and Caldwell and Courtis’s summary of his arguments.