Academic Integrity After AI: From Authorship to Accountability (#686)

In a previous blog, AI Detection, Authorship and Academic Integrity: The Strange New Science of Detecting AI, I considered an increasingly uncomfortable problem. (https://www.ricklecouteur.com/post/ai-detection-authorship-and-academic-integrity-the-strange-new-science-of-detecting-ai-685)

Universities want to know whether students are using artificial intelligence to produce academic work.

So, they have turned to artificial intelligence to detect artificial intelligence.

The result is an extraordinary technological contest.

Quick Read: Five-Point Blog Summary

Point	Key Idea	Take-Home Message
1. Learning, not simply AI use	Universities should move beyond asking Did the student use AI?	The more important question is: Did the student learn?
2. Assistance vs. substitution	AI can explain, explore, critique and edit - or it can perform the intellectual task for the student.	The important boundary may not be AI vs. no AI, but assistance vs. substitution.
3. Assessment must change	If AI can complete a traditional assignment in seconds, we should reconsider what that assignment actually measures.	Assessment should increasingly test evaluation, application, skepticism and judgment.
4. Learn to work intelligently with AI	Future professionals will routinely have powerful AI tools available.	Competence means knowing when AI is useful, and when it is wrong, incomplete or inappropriate.
5. From authorship to accountability	Who wrote these words? may no longer be enough to establish genuine academic work.	The deeper question is: Who owns the thinking?  Students must understand, verify, defend and take responsibility for what they submit.

AI generates the essay.

Another AI attempts to detect it.

A third technology rewrites the AI-generated text to make it appear more human.

The detector is modified to recognize the rewriting.

And somewhere in the middle of all this sits the student.

Perhaps we are asking the wrong question.

The question universities increasingly ask is:

Did the student use AI?

Perhaps the question education should be asking is:

Did the student learn?

Those are not the same thing.

And recognizing the difference may require us to rethink something much larger than academic misconduct.

It may require us to rethink assessment itself.

The Calculator Problem - Again

Education has been here before.

When calculators became widely available, educators worried that students would lose the ability to perform arithmetic.

Why learn long division if a machine could produce the answer instantly?

Eventually, education reached a sensible accommodation.

Students still needed to understand mathematics.

But using a calculator was not inherently cheating.

The educational objective changed.

We became less interested in whether a student could perform every calculation manually and more interested in whether the student understood which calculation to perform, why it was appropriate and what the answer meant.

The calculator did not destroy mathematics.

It changed what mathematical competence looked like.

Artificial intelligence may represent the same transition on a vastly larger scale.

The calculator automated calculation.

AI can assist with language, synthesis, explanation, organization, comparison and increasingly reasoning itself.

That makes the challenge more profound.

But the underlying educational question remains remarkably similar.

What must the human still know?

The Difference Between Assistance and Substitution

This may become one of the most important distinctions in education.

AI can assist intellectual work.

Or AI can substitute for intellectual work.

Those are fundamentally different activities.

Consider a student asked to write an essay about antimicrobial resistance.

One student asks an AI system:

Explain the major mechanisms of antimicrobial resistance.

The student reads the answer, checks it against reliable sources, identifies concepts that remain unclear and continues researching.

Another asks:

What are the strongest arguments against my conclusion?

The AI proposes three.

The student realizes that one exposes a weakness in the essay and revises the argument.

A third student types:

Write me a 2,000-word essay on antimicrobial resistance with references.

The resulting essay is submitted with minimal alteration.

All three students have used AI.

But educationally, almost nothing about these three situations is equivalent.

The first student used AI as a tutor.

The second used it as an intellectual adversary.

The third used it as a substitute author.

A detection system that simply asks whether AI was involved risks collapsing all three into the same category.

Education cannot afford to do that.

Perhaps AI Use Needs a Vocabulary

We may need to stop talking about using AI as though it were a single behavior.

We already understand this distinction elsewhere.

• A researcher can use a statistician without surrendering authorship of a scientific paper.

• An author can work with an editor.

• A student with dyslexia may use assistive technology.

• A scientist may use statistical software.

• A veterinarian may consult a diagnostic database.

The existence of assistance does not automatically eliminate intellectual ownership.

What matters is what the assistance did.

Perhaps academic work should distinguish among AI used for:

• Explanation - helping the student understand a concept.

• Exploration - generating possibilities, questions or alternative interpretations.

• Critique - challenging an argument or identifying weaknesses.

• Editing - improving grammar, clarity or organization.

• Generation - producing substantive text, analysis or answers.

• Substitution - performing the intellectual task the student was supposed to demonstrate.

The ethical boundary does not necessarily lie between AI and no AI.

It may lie between assistance and substitution.

But What Was the Assignment Actually Testing?

This leads to an uncomfortable question for educators.

Suppose an AI system can complete an assignment extremely well in thirty seconds.

What exactly was that assignment testing?

Consider the traditional instruction:

Write 2,000 words discussing the causes, diagnosis and treatment of canine epilepsy.

Twenty years ago, producing a competent answer required the student to locate information, organize it, synthesize it and express it coherently.

Today, an AI system can produce a respectable response almost instantaneously.

We could respond by banning AI.

We could submit every essay to a detector.

We could require students to prove that every sentence originated entirely inside their own heads.

Or we could ask a more unsettling question:

Has technology made the assignment educationally obsolete?

Perhaps the problem is not simply that AI has become too capable.

Perhaps some of our assessments have remained too predictable.

Assessment Should Measure What We Value

If the purpose of education is merely to produce information, AI presents an existential problem.

Machines are becoming extraordinarily good at producing information.

But universities presumably exist for something more.

They are supposed to develop people who can evaluate information.

Question it. Apply it. Recognize its limitations. Integrate conflicting evidence. Exercise judgment. And, accept responsibility for decisions.

If those are the qualities we value, then those are the qualities we should assess.

That means changing the question.

Instead of:

Describe the treatment of canine epilepsy.

Ask:

Here are three treatment recommendations generated by different AI systems for the same dog.

Which would you follow, which would you reject, and why?

Now AI is not the enemy of the examination.

AI has become part of the examination.

The student must demonstrate something the machine cannot independently certify:

Judgment.

The Veterinary Student of the Future

This becomes particularly important in professional education.

The student enters the history, physical examination and laboratory results into an AI clinical-support system.

Within seconds, the system produces:

• a differential diagnosis;

• recommended diagnostic tests;

• suggested anticonvulsant therapy;

• dosages;

• monitoring recommendations;

• prognostic information;

• and perhaps even a draft explanation for the owner.

Has the student cheated?

In veterinary school today, perhaps the answer depends upon the rules of the exercise.

In veterinary practice ten years from now, failing to consult such a system might conceivably be considered poor practice.

That possibility fundamentally changes the educational problem.

Veterinary schools should not merely ask:

Can the student produce the answer without AI?

They must also ask:

Can the student recognize when the AI's answer is wrong?

That may be considerably harder.

The Danger of the Plausible Answer

Generative AI has a particularly important characteristic.

• It can be wrong very convincingly.

• The prose can be polished.

• The explanation can sound authoritative.

• The references may appear impressive.

• The reasoning may seem entirely plausible.

And the answer can still be wrong.

That makes expertise more important, not less.

A novice may read an AI-generated answer and think:

That sounds reasonable.

An expert may read exactly the same answer and think:

Something is wrong here.

What separates them?

Knowledge. Experience. Context. Pattern recognition. Skepticism. Judgment.

These are precisely the qualities professional education is supposed to develop.

Paradoxically, therefore, the rise of artificial intelligence may strengthen the case for deep disciplinary knowledge rather than weaken it.

You cannot critically evaluate an answer if you do not know enough to recognize when it is wrong.

From Closed Book to Open AI

For generations, examinations have often been constructed around artificial scarcity.

No books. No notes. No consultation. No outside assistance.

Just the student and the examination paper.

There remains considerable value in testing what a student knows independently.

But professional life rarely works that way.

The veterinarian standing beside a critically ill animal can consult textbooks, formularies, journal articles, colleagues, specialists and increasingly artificial intelligence.

The real-world question is rarely:

Can you remember everything?

The real-world question is:

Can you make the right decision with the resources available to you?

Perhaps universities therefore need two kinds of assessment.

One asks:

What do you know without assistance?

The other asks:

What can you do intelligently with assistance?

Both matter.

And confusing one with the other may be one of the central mistakes in the current debate about AI.

What Would an AI-Era Examination Look Like?

Imagine telling students at the beginning of an examination:

You may use ChatGPT, Claude, Gemini or any other generative AI system you wish.

Then give them a difficult clinical problem.

Require them to submit not merely the answer, but the AI conversation.

Ask them to identify:

• Where was the AI helpful?

• Where was it superficial?

• What did it overlook?

• Which recommendation was unsupported?

• Which references did they independently verify?

• What additional information would change their conclusion?

And finally:

What is your decision, and why?

Suddenly using AI is not cheating.

Using AI uncritically is evidence of weak performance.

The assessment has moved from information retrieval to intellectual responsibility.

That may be much closer to the professional world students are actually entering.

Academic Integrity Still Matters

None of this means that plagiarism has disappeared.

Nor does it mean students should be permitted to submit machine-generated work as though they created it themselves.

Authorship matters. Honesty matters. Disclosure matters.

If an assignment explicitly requires independent work and a student secretly delegates that work to AI, that remains a genuine academic-integrity issue.

But universities need to distinguish dishonesty from technology use.

Otherwise, they risk creating rules that students know will bear little resemblance to the world beyond graduation.

The objective should not be to produce graduates who can function only when artificial intelligence is removed from the room.

It should be to produce graduates who remain intellectually competent when artificial intelligence is in the room.

The Honor Statement May Need to Change

For generations, students have effectively been asked to certify:

This is my own work.

That statement becomes increasingly difficult to interpret when intellectual work is routinely assisted by software.

Perhaps the declaration of the future should be more precise:

I take intellectual responsibility for this work.

I have disclosed the assistance I used.

I have verified the evidence on which my conclusions depend.

I understand the arguments presented here, and I am prepared to defend them.

That is a much stronger statement than:

I did not use AI.

It also places responsibility where it belongs.

With the student.

From Authorship to Accountability

The arrival of generative AI forces universities to confront a distinction they have rarely needed to make explicitly.

Who wrote the words?

Or:

Who owns the thinking?

For most of educational history, those questions were assumed to have the same answer.

Increasingly, they may not.

A student may receive assistance constructing a sentence while completely owning the underlying idea.

Another may personally type every word while understanding almost nothing beyond what has been memorized.

A third may submit beautifully polished AI-generated prose that represents virtually no intellectual effort at all.

Counting AI-like sentences cannot adequately distinguish among them.

Judgment can. Conversation can. Explanation can. Demonstrated understanding can. Accountability can.

Perhaps the Percentage Was Never the Point

The previous blog began with a curious little number.

7% AI-generated.

But perhaps the most interesting thing about that number is how little it ultimately tells us.

• It does not tell us who had the idea.

• It does not tell us who evaluated the evidence.

• It does not tell us who recognized an error.

• It does not tell us who changed their mind.

• It does not tell us who understands the argument.

• And it certainly does not tell us who is prepared to take responsibility for the conclusion.

Those are much harder things to measure.

They are also much closer to what education is supposed to accomplish.

The great challenge artificial intelligence presents to universities may therefore not be determining how successfully they can detect the machine.

It may be determining how successfully they can educate the human.

Perhaps that is the new academic integrity.

Not:

Did you do this without AI?

But:

Do you understand it?

Can you defend it?

Did you disclose how you produced it?

And, above all:

Are you willing to put your name to the thinking?

Glossary of Terms

Academic Integrity

The commitment to honesty, fairness, responsibility and transparency in academic work. In the age of AI, academic integrity increasingly involves not simply whether technology was used, but whether its use was permitted, disclosed and consistent with the educational purpose of the assignment.

Accountability

The willingness to accept responsibility for the accuracy, reasoning and conclusions contained in one's work. A student may use technological assistance while still remaining accountable for the final product.

AI Assistance

The use of artificial intelligence to support rather than replace intellectual work. Examples include explaining difficult concepts, brainstorming ideas, improving grammar, identifying counterarguments or critiquing a draft.

AI Detection

The use of software to estimate whether written material displays patterns associated with AI-generated text. Detection produces a probabilistic assessment rather than definitive proof of who, or what, wrote a document.

AI-Generated Content

Text, images, data, summaries or other material produced substantially by a generative artificial intelligence system in response to instructions or prompts.

AI Literacy

The ability to use artificial intelligence effectively and responsibly while understanding its capabilities, limitations and potential for error. AI literacy includes knowing when to trust, question, verify or reject an AI-generated answer.

Artificial Intelligence (AI)

Computer systems capable of performing tasks commonly associated with human intelligence, including language processing, pattern recognition, prediction, problem-solving and decision support.

Assessment

The methods educators use to determine what students know, understand or can do. AI challenges traditional assessment when machines can readily perform tasks that were previously assumed to demonstrate student learning.

Assistance Versus Substitution

A central distinction in AI-era education. Assistance helps a student perform intellectual work; substitution transfers the intellectual work itself to the technology.

Authorship

Responsibility for creating the words, ideas or other material presented in a piece of work. AI complicates traditional notions of authorship because human and machine contributions may increasingly coexist within the same document.

Closed-Book Assessment

An examination or assessment completed without access to external resources. Such assessments primarily test what a student can recall or reason through independently.

Critical Thinking

The ability to analyze information, evaluate evidence, question assumptions, recognize weaknesses in arguments and reach reasoned conclusions. It becomes particularly important when evaluating plausible but potentially incorrect AI-generated information.

Disclosure

A statement explaining whether and how AI was used in producing academic work. Disclosure can distinguish legitimate assistance from undisclosed substitution and may become an increasingly important component of academic integrity.

Generative Artificial Intelligence

AI capable of producing new content - including text, images, audio, computer code and other material - in response to user prompts. Large language models are one form of generative AI.

Hallucination

An apparently plausible but incorrect or unsupported response produced by a generative AI system. Hallucinations are particularly dangerous when presented confidently or accompanied by fabricated references.

Human-in-the-Loop

An approach in which AI assists with a task but a human remains actively involved in evaluating the output and retains responsibility for the final decision.

Intellectual Ownership

A broader concept than simply writing the words. Intellectual ownership means understanding the reasoning, evaluating the evidence, making the decisions and being capable of defending the conclusions presented as one's own.

Intellectual Responsibility

The obligation to understand, verify and stand behind the ideas and conclusions one presents, regardless of what tools were used in producing them.

Judgment

The ability to choose among competing possibilities using knowledge, evidence, experience and context. As AI becomes increasingly capable of generating information, judgment may become one of the most important human abilities that professional education must develop.

Large Language Model (LLM)

A type of artificial intelligence trained on very large collections of text and designed to recognize and generate patterns in language. Systems such as ChatGPT, Claude and Gemini are based on large language models.

Open-AI Assessment

An emerging assessment approach in which students are explicitly permitted to use generative AI but must evaluate, verify, critique and take responsibility for the resulting work. The educational objective shifts from preventing access to AI toward assessing how intelligently it is used.

Open-Book Assessment

An assessment in which students may consult approved external resources. It typically emphasizes application, interpretation and reasoning rather than simple recall.

Professional Judgment

The application of specialized knowledge, experience, ethical principles and contextual understanding to make decisions in professional practice. AI may support professional judgment, but responsibility for consequential decisions ultimately remains with the professional.

Prompt

An instruction, question or set of information supplied to a generative AI system to guide its response. The quality and specificity of a prompt can substantially influence the resulting output.

Verification

The process of independently checking AI-generated claims, facts, calculations, quotations and references against reliable sources. Verification is an essential component of responsible AI use.

From Authorship to Accountability

The central idea of the blog. Academic integrity in the AI era may need to expand beyond asking Who wrote these words? to asking Who understands, verifies and takes responsibility for the thinking behind them? In this model, the ultimate test of authentic academic work becomes not simply whether AI participated in its production, but whether the student can say: I understand it. I have evaluated it. I have verified it. I can defend it. And I take responsibility for it.

#bunyipsbooksbeyond