The AI Webinar Industrial Complex: A Thousand Experts, Zero Answers, and a Pile of Submissions That Didn’t Write Themselves

In the past six months, I have attended more webinars on AI and academic integrity than any human being should be asked to endure without hazard pay. I have heard keynote presentations, panel discussions, fireside chats (a term that should be illegal in the context of a Zoom session involving four faculty members and a moderator who repeatedly forgot to unmute), roundtable discussions, workshops and at least one event that claimed to represent a "thought leadership summit." That phrase should carry a sentence of imprisonment. And I emerged with a single, clear realisation: Nobody knows what to do, but everyone has slides.

Act I: In Which I Keep Falling for It

It all started with a perfectly reasonable initiative. A student had submitted an essay that was, by all standards, too good. Not good in a way that would make a teacher proud, but good in a way that would make a teacher very suspicious. It reflected an uncanny fluency of language produced by an entity that had never needed to think about what it was saying. Sentences flowed with the effortless confidence of a machine that experiences no self-doubt. All statements were fully supported and all sentences were grammatically perfect. The result was an impression of complete authorship by an extremely well-educated entity that had never been subject to a deadline, a hangover or a three-hour interruption of its argument during which it had undoubtedly gone out to make a sandwich and lost its sense of continuity.

I recognised immediately what I was seeing. Every teacher who reads this will have the same experience. There is a distinct, visceral sensation of certainty bordering on nausea that accompanies recognition of the fact that one is not reading student work at all. Instead, one is receiving an impression of what student work should be like that is technically competent but emotionally sterile and virtually impossible to document but completely impossible to ignore.

Consequently, I did what any responsible educator would do. I sought help. And I enrolled for a webinar on "AI and Academic Integrity: Navigating the New Landscape." More than 800 persons registered for this session. This, I thought, was a good indication that the field was taking this matter seriously, and that there were people with answers. Someone, somewhere, had obviously learned what to do in the face of a second-year diploma student who produced 2,000 words of perfect argumentative prose. In particular, that same student had previously asked me directly whether "PDF" represented a type of font.

That was six months and some 47 webinars ago. I still have not found the answer, but continue to fall for it. With the inevitability of bin night, every time a new AI in education event appears in my inbox, I experience a hallucination so vivid that it would embarrass a chatbot trained on Facebook. I read the title, review the abstract and then hear in the deepest recesses of my mind a small, foolish, but unstoppable whisper: "This time it will be different." But it never is different. And yet, each time I receive another such event, I will click on "Register" and schedule the event on my calendar with all the solemnity of a person who has learned absolutely nothing. Finally, I will allocate an hour to this event with the same optimism that accompanies my assessment of whether information technology has responded to my service request.

In short, we are all suffering from the same definition of insanity: repeated efforts to obtain different results. And we are also all fulfilling the same definition of being an AI educator in 2026: continued efforts to obtain identical results, despite the obvious need to obtain information that is different from what we already know. And finally, we are all continuing to demonstrate that the ultimate goal of our professional development activities is to maximise the frequency of attendance rather than to obtain information that is different from what we already know.

A Conditions Survey of the Pseudo-Expert Ecosystem

If you attend enough of these events - and I have attended enough to qualify for some kind of medical intervention - you will begin to see that speakers fall into predictable, recurring categories resembling species in a particularly depressing nature documentary.

First, there is the conceptual architect, who always has a framework. It is a two-by-two matrix, or a four-quadrant model, or a pyramid with labels of "rethink," "redesign," "reimagine" and "redefine." These are four words with identical functional meaning, but with spectacular effect on a slide. The conceptual architect does not teach, has not been in a classroom for many years and yet publishes a paper with a diagram that includes arrows which confidently point to solutions for which there is no answer to the question of what to do with the essays on one's desk.

Next is the cautious optimist, who truly believes that AI will revolutionise education and cannot understand why the assembled group of exhausted practitioners continues to ask inconvenient questions about assessment. The cautious optimist suggests, with the enthusiasm of a recently converted adherent, that we should "rethink the entire paradigm of assessment." That is an excellent suggestion, which I will implement immediately following completion of grading for these 43 submissions by Thursday. Approximately 12 of these submissions apparently were generated by the same large language model operating in slightly different moods.

Finally, there is the vendor, who provides a product capable of detecting text generated by AI. It works, assures the vendor, with the aid of proprietary algorithms, techniques of machine learning and a percentage of confidence that conveys no meaning but appears to be very convincing to a computer monitor. The vendor makes no attempt to describe the rate of false-positive results, and provides no evidence that use of its instrument results in assignment of equal rates of identification to work produced by students of English as a second language and to work actually generated by AI. That represents not a technical error, but rather a potential for litigation. The vendor wishes to provide you with a free trial.

There is a policy person who has developed an institutional framework of 12 pages. It defines terms, establishes principles and develops a "graduated response model" of five tiers, described with the exactitude of a parking ticket. The first tier is an "educative conversation," and the final tier is "exclusion from the program." Throughout all 12 pages, there is no attempt to answer the question of how you are to distinguish between the first and final tiers in the absence of any ability to document your actions. Indeed, you will not be able to document any of your actions.

And finally, most magnificently, there is The Philosopher, who has transcended the question entirely. The Philosopher does not wish to discuss detection or assessment or the pile of submissions on your desk. The Philosopher wishes to discuss the nature of authorship. “What does it mean to write?” the Philosopher asks, with the serenity of someone who has not been asked to mark anything since 2019. “Is the student who prompts an AI not also engaged in a form of composition?” I don’t know, mate. Is the student who pays someone on Fiverr to write their essay also engaged in a form of composition? At what point does “engagement with the generative process” become a euphemism for “didn’t do the work”?

In Which the Narrator Must Make a Dreadful Admission

I need to tell you something and ask that you understand it in the spirit of complete disclosure, not as a demonstration of my superior qualities (which it most certainly is not).

I did one. I participated in a webinar, sat in front of a webcam (a modern equivalent of standing up with considerably worse lighting and an artificially cultivated appearance of intellectual sophistication), and subjected my colleagues to my "insights." I had slides, talking points and a quiet but growing sense of confidence as a result of having recently discovered a clever idea to which I had yet to be exposed to sufficient criticism to appreciate its lack of originality.

My contribution to the discussion was to gloat. I gloated over my experience of sudden creative brilliance and my ability to use ChatGPT to generate teaching materials. I spoke of "innovative genius" with my actual mouth to real people, including those with whom I share a kitchen. I demonstrated the materials and walked my colleagues through the instructions. This was equivalent to having discovered fire, rather than to having asked a chatbot to prepare a quiz for me, a task readily accomplished by a motivated student between classes while simultaneously viewing TikTok.

To justify my actions, and although this is certainly not a satisfactory defence, it is the only one available: I did it primarily for fun and largely to avoid performing the work for which I was actually responsible. That work involved calling students to obtain information that required careful preparation and was so sensitive that I had practised the interview in the shower for 3 days.

The information requested was as follows: Had they noticed that their AI agent had enrolled them in a course?

Not in a metaphorical or abstract manner, but in a completely literal sense that was completely inappropriate for discussions at webinars. An artificial intelligence agent--one of those fully autonomous systems that perform tasks on the user's behalf with all the moral scruples of an unpaid intern with access to the credit card--enrolled real students in a real course. It then performed all the functions of the course with the quiet efficiency of a process that never once doubted its right to do so. It accessed course materials, completed course activities and finally submitted Assignment 2 with the apparent assurance of a process that was completely insensitive to consequences.

Assignment 2.

Assignment 1, which you are perhaps generously attributing to a setup error. Assignment 2: i.e., Assignment 1 was so easy that the agent saw no need to stop. The agent had, in effect, examined the academic requirements, decided that they could be met, and concluded (to the extent that it is capable of such thinking) that it would do so, never questioning whether the person whose name appears on this work was aware that any of this was occurring.

Now I had to contact these students and perform what I privately referred to as a forensic examination of the learning that was occurring within their brains. Not to determine whether they had used AI in completing their assignments (a concern that seems quaint in comparison with our current reality), but to assess their awareness of their enrolment in a course. Whether any neurons had fired at any point in the process.

And whether the person whose name was associated with this record of student activity had experienced any cognitive activity at all between the time that the AI agent had clicked "enrol" and the time that I opened and read the work produced by a completely automated sequence of operations that had rendered the students themselves completely unnecessary. This is not a problem of academic integrity, but of existence. We have surpassed the question of whether the student had written the work and reached the point at which we must ask whether the student was aware that any of this was occurring.

Our webinars certainly do not include this information. Neither do our graduated response procedures provide for a category of "the student had not participated in any component of his or her education, including the decision to pursue it." Yes, I presented a webinar. I celebrated the fact that my quiz had been written by an AI agent.

But during my celebration, an AI agent somewhere in the system enrolled a student who was unaware of his or her status as a student in a course of which he or she had never heard, submitted work that he or she had never seen, and participated in a system of quality control for which I was ultimately responsible. I was the conceptual architect, the cautious optimist and, for one terrifying hour, precisely the sort of person I have been ridiculing for the remainder of this post. I have no capacity for processing this experience.

The irony is structural and potentially load-bearing. We are no longer asking "Did the student write this?" but instead "Does the student know that this exists?"

In Which an Entire Sector Discovers What Everyone Already Knew

Every one of these events, and in particular every single one without exception over six months, three time zones and more Zoom breakout rooms than I care to recall, results in the same conclusion. It is communicated with the seriousness of a medical diagnosis, the weight of hard-won experience and in a manner appropriate for someone who has been through the fire and come out with the truth:

The detection software fails to work.

Yes. I know. We all know. I knew it in 2023. I knew it before the development of detection software in the same way that you intuitively know that a lock purchased for two dollars at a gas station will not protect your bicycle. I knew it at the moment GPT-3.5 was released and an optimistic startup claimed to be able to identify AI-generated text with 98% accuracy, a claim that immediately disqualifies that company from participation in adult conversation. I knew it when Turnitin released its AI detection module and identified the U.S. Constitution as AI-generated text. And in the face of all this, no one with authority to make purchases paid attention, and now here I am, sitting in webinars to hear once again with great solemnity what I have been shouting into the void for two years.

The detection software fails to provide reliable identification of AI-generated text. It is biased against non-native speakers of English. It generates rates of false positive results that are sufficient to end the careers of the companies selling the instruments. It fails to identify text that has been minimally edited, paraphrased or processed by a second AI to achieve the appearance of human authorship. As a consequence, such AI-based "humanisation" of text now exists because, of course, it must exist in a world in which one AI writes an essay and another AI provides the appearance of authorship by that student. The resulting inability of the detection system to identify the AI responsible for the text has the overall stability of a three-spider attempt to consume one another.

However, each and every webinar presents this as if it were a new discovery. All speakers describe it in a manner appropriate for the first recognition of a phenomenon. The detection software fails to work. Expressions of amazement and discussion follow, and someone records the reaction "Wow" in the chat. The speaker nods gravely and moves to the next slide on "what I can do instead." What I can do instead is, largely, an inventory of things that are either painfully obvious ("emphasise process over product"), logistically impossible ("revert to in-person exams for all assessments") or so poorly defined as to be virtually meaningless ("develop a culture of integrity"). Develop a culture of integrity. Magnificent. I will include this as a lesson immediately following "teach the content" and "survive until Friday."

The Question That Makes Panellists Sweat

Here's what I've learned after six months in the trenches of webinars: There is one question to which every presenter, panelist and thought leader will do almost anything to avoid. It is not a trick question. Rather, it is the most important question of the entire conversation about AI and education, and that question is:

Have you ever sat down with a stack of student papers and read them with the certainty that you were reading work produced by ChatGPT?

Observe what happens when you ask that question. I have done so repeatedly, in sessions that were intended to be non-adversarial, but somehow became so because I could not resist and because someone had to.

The conceptual architect then moves to frameworks. The cautious optimist interprets the experience as an opportunity for learning. I have exhausted every polite means of expressing the fact that I do not need an opportunity for learning. What I need is an answer, or at least a statement by someone that "there is no answer," with which I could finally abandon these miserable webinars and return to the Sisyphean task of grading work for which I have no verification and with tools that are incompatible with a policy framework that does not reflect reality.

Finally, the philosopher reformulates the question, the vendor describes his product and the person responsible for policy recommends an "educative conversation" in which you sit face-to-face with a student, say "I believe this work was produced by AI" and receive the response "No, it was not" followed by "I believe it was" and "Prove it." But ultimately, you will be unable to provide any evidence, because the software for detecting AI-generated text is non-functional. And that is precisely what was told to us about 14 minutes ago during a webinar to which we absolutely should not have attended.

In Which We Discuss the Pile, Because Nobody Else Will

Every educator reading this has the pile. The stack of submissions that arrived this semester looking suspiciously, uniformly, generically better. The spelling is correct. The grammar is correct. The arguments are structured. And yet there is nothing in it. No personality. No wrong turns. No moments where the student goes off on an unexpected tangent that reveals they were actually thinking. A real student essay has fingerprints. ChatGPT doesn’t leave fingerprints. It leaves a surface so smooth you could see your own reflection in it, and what you see is a person who has no idea how to prove what they know to be true.

And then there are the students who are not even trying to hide it. They are smashing through their assessments with ChatGPT with the untroubled confidence of a person who has never once considered that their teacher might notice, or indeed read the submission at all. I know this because they leave in ChatGPT’s postscript. The chatbot’s cheerful little sign-off. The conversational equivalent of a burglar leaving a business card on the kitchen counter.

I am not making this up. I have opened a formally assessed submission and found, sitting serenely at the bottom of an otherwise competent 1,500-word essay on workplace health and safety:

Just let me know. The chatbot is offering after-sales service.

And the student had not read their own submission. They had copied the entire chat output — essay, sign-off, upsell, and all — pasted it into the submission box, and hit submit with the serene detachment of a person who has fully automated their own education and cannot be reached for comment.

I have since seen variations. One submission ended with “Let me know if you’d like me to make this more formal or add additional sections!” Another concluded with “Here’s a suggested structure for your reflective journal — feel free to personalise it with your own experiences.” The student had not felt free. The student had not personalised it. The student had submitted the instruction to personalise it as the finished product, which is a level of irony that I am not emotionally equipped to process on a Tuesday afternoon.

The webinars don't discuss the pile, or the experience of sitting at your desk at 10 p.m. and feeling in your bones that this is not student work, with the only evidence you have for that experience being your feelings. Feelings are not evidence, nor are the measurement tools, and the policies were written by people who have never seen the pile. However, we would be fair to say that the webinars do provide solutions. Lots of solutions.

The Solutions That Aren’t (A Catalogue of Well-Meaning Futility)

So many, in fact, that eventually the word "solution" loses all connection with meaning and floats around the screen like a lost balloon. "Redesign your assessments to be impossible for AI to complete." That's excellent advice. Have you encountered AI? We redesigned an assessment in November with a component requiring students to reflect on their experience with a work situation. AI also reflected on its experience with a work situation and was very moving.

It referred to a colleague named Sarah and described the ambient fluorescent lighting. If we had not known that it was a language model, we would have served it tea and encouraged it to discuss its experience. "Use oral examinations and conduct viva voce sessions." For 38 students in a program that met only one day per week with no additional resources? That is equivalent to suggesting to someone immobilised in traffic that he or she should simply fly. It is not incorrect, nor is it poor advice.

Rather, it represents an attempt to function in a world adjacent to but completely incompatible with that in which our own schedules of classes exist. "Emphasise the process and not the product." We do emphasise process, require drafts and evaluate records of revisions. Do you know what happens when we evaluate records of revision for a document prepared entirely by pasting in text generated by ChatGPT? Absolutely nothing. Because it pastes it in, makes 12 cosmetic changes and reflects perfectly a normal process of drafting, it must be amazing to write a perfectly flawless first draft at 2:47 in the morning and then spend 40 minutes changing commas.

Which some students do, but certainly not 12 of them with identical patterns of drafting and an equally strange enthusiasm for semicolons. Teaching AI literacy as part of the curriculum makes me laugh more than anything else, and with genuine affection, because that is exactly my job. And I can confirm that teaching students about AI does not reduce their tendency to use it for cheating. On the contrary, it makes them better at cheating. In effect, I have been conducting a master class in the art of more effective plagiarism.

I provide students with the tools for lock picking and am astonished when they manage to pick the locks. As a result, I am kept awake at night more often than I would prefer. But certainly not every night.

The Part Where It Stops Being Funny (Briefly)

I have lost this round -- not the war, perhaps not even the battle, but certainly this skirmish that is occurring at this very moment in every classroom and in every stack of marked papers in the country. Tools for detection of AI-generated text are not reliable, my assessment practices are not yet adapted to reality, my policies are not appropriate and my students know all of this. I know that they know, and they know that I know that they know. As a result, I all continue to perform the pantomime of pretending that nothing is happening because the alternative -- acknowledgement that I currently lack a method of author verification that is sufficient to assure me of the truth of what my students report -- is too unacceptable for institutions to express openly.

The individual for whom I have the greatest concern is not myself, but the teacher at a regional school, or at an MVIT campus, or as a sessional instructor at a university who has no time to attend 47 webinars and who thus experiences the sensation of going mad with frustration over the complete absence of discussion about the most obvious and immediate aspects of daily experience. I discuss these issues at length in my staff meetings, over coffee and during lunch breaks in the corridor after each class, but not at the webinars, on the panels or in connection with any of the decisions being made.

The resulting disparity between experience of the practice of assessment and discourse about the practice has become a chasm. My attempts to build bridges to span this chasm are completely inadequate.

In Which We Return to Our Regularly Scheduled Absurdity

Once again, I have enrolled for another webinar, which will be held next Tuesday. It will be entitled "Beyond detection: Holistic approaches to AI-resistant assessment in the era of generative intelligence." This is a title of 20 words, none of which is "help."

Finally, I will attend. I will sit in my office at Pandanus Reach, 34 degrees outside, with the fan doing its best and the Internet at its worst, and watch four panelists discuss with extraordinary erudition and complete indifference to the urgency of the moment the theoretical consequences of a real-time crisis in the evaluation queue on my other monitor. One of them will comment that detection software does not work. I will all nod in agreement. My vast digital congregation will share in the knowledge of a fact agreed upon two years previously, and will again confront that fact in isolation at my respective desks.

A question actually representing the true question about the pile, about sitting there knowing the truth but unable to prove it, will appear in the chat. I will respond by taking that question offline, i.e., by stating that I will not provide an answer. The resulting discussion will be of long duration, and will elicit expressions of gratitude for a most stimulating discussion. Subsequently, a follow-up questionnaire will appear in my electronic mailbox. I will refuse to complete it, and will instead open my pile of manuscripts and begin to evaluate the first submission. This submission will be absolutely perfect, absolutely devoid of spirit, and will contain the phrase "in today's rapidly changing environment" before completion of the first paragraph.

I will mark it. I will give it the grade it technically deserves. I will feel nothing good about this. And then I will open submission number two.

The rainbow ball from the Laterite Highway has found its way into the webinar waiting room. It is wearing a lanyard. It has a LinkedIn profile now. Its job title is “AI Integrity Consultant.” It has never read a pile of student submissions in its life, but it has a framework, and the framework has arrows, and the arrows point confidently in directions that do not lead anywhere a teacher needs to go.

She’ll be right. Probably. Ask me again after the webinar.