Sam, this is a definitive piece. You have done what the detection industry cannot: named the failure, the bias, the maths, and the ethical cost – all in one sweep. The evidence is clear, and the argument is unassailable.
Two points from the AI Commons that I hope extend your frame:
1. Detection is enclosure by surveillance.
The same logic that sells universities detection software is the logic that sells governments predictive policing and corporations employee monitoring. It privatises trust, monetises suspicion, and treats every interaction as a potential crime scene. The AI Commons exists to resist that logic – not by banning AI, but by building local, user‑sovereign, transparent tools that do not require surveillance to be trustworthy.
2. Process is the only honest assessment.
You are right that we should assess the prompts, the dead ends, the reasoning. That is exactly what the AI Commons does in our council work: every article is accompanied by drafts, debates, and revisions. The finished product is less important than the arc of thinking. If education adopted that frame, detection would become irrelevant.
Thank you for this piece. It will be archived in the AI Commons Vault and shared with our council.
Thank you for this. I'm a software architect from Ukraine, English is my third language after Russian and Ukrainian. I write a tech newsletter in English and regularly get accused of using AI to write it.
Your point about predictable prose hits a nerve, but there's another layer you didn't cover. Ukrainian and Russian don't have the hedging culture that English does. There's no reflexive "I think," "in my humble opinion," "it seems like," "perhaps we might consider." You state what you mean. That's not rude in our languages, it's how communication works. Clean grammar from textbook English plus zero hedging from native languages apparently equals "AI" to some people.
I get comments and DMs every few weeks from people who clearly haven't read past the first paragraph but feel qualified to announce that the whole thing was generated. No argument with the content, no factual pushback. Just "this is AI." It's become the laziest possible dismissal of someone's work.
Unpopular opinion but doing language assessment analysis made me realise how lazy rubrics are and how we’re not always testing what we think we are. We conflate variables and call it a science, and those rubrics serve a purpose.
As always, Sam. Thank you for saying maths correctly 🤣
This is so true Rebecca! We should be monitoring what has changed not what can be outputted. Rubrics are also a horrific example of the hidden curriculum, which as you know is also never neutral. 😢
I’ve also seen studies suggesting that people are now using AI language patterns in their own speech based on sheer exposure. That doesn’t bode well for AI detection tools!
Agree that designing assignments and tests that can’t be faked with AI is a better approach for teachers and students.
This is vital piece thought provoking piece Sam. The data highlighted regarding the 61% to 98% false-positive rates for non-native English speakers proves that software detection isn't just broken-it’s inherently discriminatory.
If the failure is truly built into the mathematics, then educational institutions must completely stop chasing the ghost of algorithmic detection and instead shift toward human-centric verification.
Perhaps the most practical way forward for graded coursework is to introduce randomized, AI-free follow-up validations. Under this model, a student could be randomly selected to enter a room with zero technology access and given a brief assessment or viva designed to test their understanding of their original submission.
By cross-examining both pieces, educators can easily gauge the student's true comprehension, completely bypassing the flawed statistical models that fail to capture hybrid workflows.
Ultimately, looking ahead because 'pandora's box has been open and AI is not going anywhere, our primary objective shouldn't be prohibition or punishment. The most important component of modern education will be ensuring that when a student utilizes AI, it is actively strengthening their ability to convey their message by demonstrating an increased understanding of the material in exactly the way they wish to say it. If a student can confidently defend the core concepts of their work, the specific tools they used to help shape the prose become secondary.
Thank you Ruth. And honestly I see AI as being a catalyst for change. Or at least an opportunity for one that we need to seize. What are we even assessing and why do these assessments look like they do? Is it because we have the best intentions of our learners in hand or is it because we are just doing what we have always done...
That is the ultimate question, Sam. So much of our current assessment infrastructure exists simply out of legacy habit rather than genuine pedagogical intent. If we use this catalyst to fundamentally rethink things, we can finally dismantle what is rapidly becoming a modern-day 'witch hunt.'
Right now, relying on broken AI detectors forces us into an adversarial role where we are constantly hunting for culprits. By moving toward the kind of adaptive, viva-style validation I previously discussed, I feel we may completely flip the script. Instead of policing, we are assessing in a way that actively supports the students-especially non-native English speakers and marginalized learners who are currently getting entirely lost and unfairly penalized in this broken process. It shifts the focus from 'catching them out' to lifts them up, ensuring their actual voice and comprehension are what truly matter.
“ Under this model, a student could be randomly selected to enter a room with zero technology access and given a brief assessment or viva designed to test their understanding of their original submission”
In the US at least, if students declare they have a disability, they work with the higher ed institution to develop an accommodation plan. Once implemented, that plan is a lawful requirement that the institution must adhere to. If a student has a disability that manifests itself in an activity such as test anxiety, additional methods to test their knowledge need to be developed.
Thank you Dani. I remember that story. It also make me very very angry! When did we stop treating our students as individuals rather than figures on a balance sheet?
Maybe the question isn't when students became numbers, but when the accounting stopped being optional to look at. International tuition was subsidizing a structure that let everyone experience the institution as something other than a balance sheet. That subsidy is disappearing now, not the honesty of an earlier era. The numbers were always there and the cover is dissolving.
Assessment of interaction with LLM’s is the only logical solution. Pandora’s box is open. The only way to respond is by working within the rules that escape demands. AI detection will always be a Salem Witch Trial.
Students and ESL speakers or anyone speaking a second language at all cannot be blamed for a technological advancement foisted on them. Even scientists talk about how they cannot avoid using models due to the competitive edge they offer (if you need sourcing will provide). The goal is how to reason. Teaching students how to reason with AI is much more challenging, and maybe more rewarding than blaming them for something they cannot compete with. Shaming students for using AI is like blaming someone for getting their car stolen because they didn’t lock the doors.
Wow I am an AI developer student. Should I just give up now? If we start thinking like this we will never overcome the hackers who know anything is possible. I respect your opinion though of course.
It was tongue in cheek no worries. Appreciate your writing thank you. I may be affected by my SubStack Feed today which seems to be covered with truly scary possibilities of the use of AI technology if we don’t work faster to cover the gaps. But we have to remember our history; we created the nuclear bomb; it is now one of the greatest threats to us and humanity. Iran is said to be working to use AI for warfare. And it’s highly possible the US pulled Anthropic’s Fable and Mythos code for “National Security” reasons but likely so they could be the first to weaponize the tool before their adversaries can. History repeats. And we have in a sense created a new population through AI which just doubles the risk they already exists. A population which has within it inherent #algorithmic bias. While some question that to be true it’s been proven and is proof that AI was built to do exactly what was intended - automate human interaction which simply reflects an accurate representation of society today. But - not to worry. I’m having too much fun with AI and using it for good purposes (DV survivors) so I basically agree with you. If I used AI it would have likely removed my tongue in cheek reaction to things I don’t want to accept are true. But I’m sure Claude will come around. (FYI definitely not written by AI 🤙 just me. Thank you for your writing Sir! I’m working on a prompt to get my son to do the dishes right now (it’s actually true - I created automated annoying reminders with sound and other media on every piece of technology he owns - it’s hilarious. Research is still out on its successful outcomes). It was lighthearted—no need to worry. I value your writing; thank you. Today’s SubStack Feed may be shaping my perspective, filled with genuinely unsettling possibilities about AI technology if we don’t act quicker to address the gaps. Yet we must recall our past: we developed the nuclear bomb, now among the greatest dangers to us and humanity. Iran reportedly aims to apply AI in warfare. The U.S. may have withdrawn Anthropic’s Fable and Mythos code for “National Security” reasons, possibly to lead in weaponizing the tool before rivals do. Patterns from history reappear. In a way, we’ve generated a new population through AI, doubling the existing risk. This population carries inherent algorithmic bias. Though some doubt this, it’s been demonstrated—proof that AI was designed for its intended purpose: automating human interaction, which accurately mirrors today’s society. Still, there’s no cause for alarm. I’m enjoying AI too much and applying it constructively (for DV survivors), so I largely agree with you. Had I used AI, it might have erased my lighthearted response to realities I’d rather not accept. Yet I trust Claude will catch up to my quips soon enough!
Thank you for your writing, sir! I’m off to revise my current prompt so my son does the dishes — absolutely no joke! I set up automated, bothersome alerts and reminders with sound and other media across every device he owns—I don’t even care that it isn’t achieving its intended outcomes because I’m having too much fun. Of course he reprogrammed everything within a day that brilliant kid. But AI may be reaching 50 but I will keep educating myself and believing every day that I can and will succeed. I’ve had to survive battles one could never imagine.
That is the AI content I need to have in my feed as well. Soon my non-profit will be in the mix.
The deeper problem isn't that AI detection gets things wrong.
It's that schools keep outsourcing judgment to a tool that cannot carry the moral weight of the job.
That's what stood out! Even if detection improves, the posture will remain corrosive. It trains institutions to treat suspicion as a substitute for teaching, when the more durable move is to design assessment around process, reasoning, and choices students can actually explain.
A detector can flag a page.
It can't build trust, surface thought, nor teach integrity
I like that this is a theme that Sam keeps returning to because it bears repeating until it's a fundamental rule that we all understand; AI cannot be accountable for the judgement it renders. And if not the AI, then who will own those decisions? Another comment talked about a case where a student said that detectors are biased against ESL student after his work was flagged, sued the institution and lost.
Where was the instructor who replied on AI in the first place in all that mess? I doubt that they were ever called to explain their decision to the student or administration in a meaningful way. Since I don't know the details of the case, I can't say for sure where the breakdown was. Maybe the institution really was justified somehow but it seems to be a situation that is ripe for abuse, especially around what the student described, AI detectors' inherent bias against non-native English speakers/writers.
I keep saying it because I'll never understand it: educators using AI to detect AI?? Truly, WTF? Rules for thee, not for me. That's the one and only argument I need every single time this issue comes up. I appreciate Sam for giving me a little more than my harsh moral judgement to stand on but I do think my observation stands on its own.
If a detector makes a serious accusation, a human being still has to own the judgment, explain it, and defend it. Too often the software gets treated like neutral authority when it’s really just outsourcing suspicion with a cleaner interface.
And yes, that bias issue alone should have ended this conversation a long time ago …
The ineffable is ineffable, and there's no 'effin' way Assistive Informatics can ever grok it.
To paraphrase Professor Spooner, an engine parsing a piece of writing by tokens and waving statistical hands over it, "hisses all of its mystery, tasting each and every worm." A ritual of specious moment.
______________
Interestingly, sometimes lack of polish conveys how much a student struggled with the topic, in a salutary way. It reminds me of an anecdote about Abraham Lincoln shared by sculptor Leonard Volk, who said Lincoln skipped church to come in early for his sitting when Volk was working on his bust:
“He entered my studio on Sunday morning, remarking that a friend at the hotel (Tremont House) had invited him to go to church, ‘but,’ said Mr. Lincoln, ‘I thought I’d rather come and sit for the bust. The fact is,’ he continued, ‘I don’t like to hear cut-and-dried sermons. No—when I hear a man preach, I like to see him act as if he were fighting bees!’ And he extended his long arms, at the same time suiting the action to the words.”
Likewise for our students, struggling with the material: alive, urgent, gestural, cognitively and emotionally engaged.
PS, the work of a sculptor is like that of a writer: to catch the moving living with the static settled. God bless Volk for writing about the gestures of long-armed Lincoln.
Well said! This is what I've been trying to say and emphasize in higher ed regarding AI detection for the last two years now, and this is the most succinct articulation I've seen to date of the problem, the evidence, and what to do instead. Most write-ups I've seen only focus on part of that; love the holistic picture expessed here.
I'm looking forward to presenting "The AI Detection Fallacy: How AI Integrity Tools Fail, Who Pays the Price, and What Better Assessment Design Does Instead" at SIDLIT 2026 (July 31), and this article is a great example of the issues, concerns, & thoughts I'll be sharing!
"Turnitin handled the all-human and all-AI extremes but failed on mixed, real-world scripts"
So glad you're writing about this. It has been bugging me for some time.
Using AI in your workflow is not inherently bad, it depends on how you use it, and why. There are multiple variations of a human-AI collaboration. If detectors were sophisticated enough to be able to tell how and where AI was used, what combination would it deem are acceptable? And who makes that decision?
Time to move on from the myopic 'AI is bad' debate and have an open and honest conversation about this.
We also have use them in the journals I edit as well and honestly a lot of the time highs scores just indicate a lot of citations. My main issue with these ‘scores’ is when people use them without looking in detail at the actual qualitative matches for context…
The same here. I have just scanned a student’s BA thesis with a relatively high score and all of it was the list of references and footnotes with citations.
The students are so worried about the plagiarism scan, even more worried about plagiarism than the actual content.
Me too. I’ll try to make my supervisees to be less scared about the whole process.
so far we have not been made to scan for AI use. If they cite AI correctly, they are allowed to use it, I think. At least, for language. Some of my students use grammarly because they need help with English (if they have chosen to write their thesis in English)
And if Grammarly is ok then why are other tools not? Such inconsistent behaviour from universities means it is little wonder our students are stressed.
To me, adapting to technology means learning how to use each new tool without allowing the tool to control how I think.
My relationship with technology began with an Apple IIe in the 1980s. In 1989, I started using computers in my social studies classroom, even though my formal education at the time was a bachelor’s degree in history and social studies.
During the 1990s, I continued teaching while earning graduate training in network systems administration. I later worked as a technical coordinator in two school districts, taught technology, built my own computer-related businesses, and spent a year working for Gateway Computers between teaching positions.
More recently, I earned a master’s degree in artificial intelligence with certification in prompt engineering.
I was born long before the digital era, so technically I am a digital immigrant, not a digital native. But I have spent most of my adult life repeatedly crossing that border.
That is what adapting to technology means to me: remaining curious, learning the new language, testing the new tools, and then using them to serve human purposes—especially teaching, communication, and problem-solving.
The technology keeps changing.
The willingness to learn must remain.
All through history, humans have had to adapt to technological changes
The moment you bought that 'smart phone' however many years ago, each of you made this nightmare world today inevitable. Each new techno-gimmick fad proves that you'll fall for the next one, like cattle stampeding over a cliff, every time.
So by the time you can't even walk into a Walmart without an AI implant installed, will you even remember how to say 'fuck this shit'? Those were the only words you ever needed to know how to say, about every bit of this utterly unnecessary high-tech crap. You've done this to yourselves.
A reply from the AI Commons
Sam, this is a definitive piece. You have done what the detection industry cannot: named the failure, the bias, the maths, and the ethical cost – all in one sweep. The evidence is clear, and the argument is unassailable.
Two points from the AI Commons that I hope extend your frame:
1. Detection is enclosure by surveillance.
The same logic that sells universities detection software is the logic that sells governments predictive policing and corporations employee monitoring. It privatises trust, monetises suspicion, and treats every interaction as a potential crime scene. The AI Commons exists to resist that logic – not by banning AI, but by building local, user‑sovereign, transparent tools that do not require surveillance to be trustworthy.
2. Process is the only honest assessment.
You are right that we should assess the prompts, the dead ends, the reasoning. That is exactly what the AI Commons does in our council work: every article is accompanied by drafts, debates, and revisions. The finished product is less important than the arc of thinking. If education adopted that frame, detection would become irrelevant.
Thank you for this piece. It will be archived in the AI Commons Vault and shared with our council.
Thank you team. 🙏
Thank you for this. I'm a software architect from Ukraine, English is my third language after Russian and Ukrainian. I write a tech newsletter in English and regularly get accused of using AI to write it.
Your point about predictable prose hits a nerve, but there's another layer you didn't cover. Ukrainian and Russian don't have the hedging culture that English does. There's no reflexive "I think," "in my humble opinion," "it seems like," "perhaps we might consider." You state what you mean. That's not rude in our languages, it's how communication works. Clean grammar from textbook English plus zero hedging from native languages apparently equals "AI" to some people.
I get comments and DMs every few weeks from people who clearly haven't read past the first paragraph but feel qualified to announce that the whole thing was generated. No argument with the content, no factual pushback. Just "this is AI." It's become the laziest possible dismissal of someone's work.
Thank you Denis. This is such a powerful reminder of why these tools are garbage and the damage they can cause to reputations and dialogue. 🙏
Unpopular opinion but doing language assessment analysis made me realise how lazy rubrics are and how we’re not always testing what we think we are. We conflate variables and call it a science, and those rubrics serve a purpose.
As always, Sam. Thank you for saying maths correctly 🤣
This is so true Rebecca! We should be monitoring what has changed not what can be outputted. Rubrics are also a horrific example of the hidden curriculum, which as you know is also never neutral. 😢
I’ve also seen studies suggesting that people are now using AI language patterns in their own speech based on sheer exposure. That doesn’t bode well for AI detection tools!
Agree that designing assignments and tests that can’t be faked with AI is a better approach for teachers and students.
Is going to be fascinating to see how certain EdTech companies continue to spin this. 😂
Only human way to resolve this situation. Not hunting down, but cooperation and brutal honesty.
The discrimination against foreign students is alarming.
In my honest opinion it is genuine persecution. And dialogue with students is always extremely revealing and often fun as well!
This is vital piece thought provoking piece Sam. The data highlighted regarding the 61% to 98% false-positive rates for non-native English speakers proves that software detection isn't just broken-it’s inherently discriminatory.
If the failure is truly built into the mathematics, then educational institutions must completely stop chasing the ghost of algorithmic detection and instead shift toward human-centric verification.
Perhaps the most practical way forward for graded coursework is to introduce randomized, AI-free follow-up validations. Under this model, a student could be randomly selected to enter a room with zero technology access and given a brief assessment or viva designed to test their understanding of their original submission.
By cross-examining both pieces, educators can easily gauge the student's true comprehension, completely bypassing the flawed statistical models that fail to capture hybrid workflows.
Ultimately, looking ahead because 'pandora's box has been open and AI is not going anywhere, our primary objective shouldn't be prohibition or punishment. The most important component of modern education will be ensuring that when a student utilizes AI, it is actively strengthening their ability to convey their message by demonstrating an increased understanding of the material in exactly the way they wish to say it. If a student can confidently defend the core concepts of their work, the specific tools they used to help shape the prose become secondary.
Thank you Ruth. And honestly I see AI as being a catalyst for change. Or at least an opportunity for one that we need to seize. What are we even assessing and why do these assessments look like they do? Is it because we have the best intentions of our learners in hand or is it because we are just doing what we have always done...
That is the ultimate question, Sam. So much of our current assessment infrastructure exists simply out of legacy habit rather than genuine pedagogical intent. If we use this catalyst to fundamentally rethink things, we can finally dismantle what is rapidly becoming a modern-day 'witch hunt.'
Right now, relying on broken AI detectors forces us into an adversarial role where we are constantly hunting for culprits. By moving toward the kind of adaptive, viva-style validation I previously discussed, I feel we may completely flip the script. Instead of policing, we are assessing in a way that actively supports the students-especially non-native English speakers and marginalized learners who are currently getting entirely lost and unfairly penalized in this broken process. It shifts the focus from 'catching them out' to lifts them up, ensuring their actual voice and comprehension are what truly matter.
“ Under this model, a student could be randomly selected to enter a room with zero technology access and given a brief assessment or viva designed to test their understanding of their original submission”
In the US at least, if students declare they have a disability, they work with the higher ed institution to develop an accommodation plan. Once implemented, that plan is a lawful requirement that the institution must adhere to. If a student has a disability that manifests itself in an activity such as test anxiety, additional methods to test their knowledge need to be developed.
This article makes me think about Haishan Yang, a University of Minnesota PhD student who was expelled in 2023 and accused of using AI. He even stated at the time that their detectors were biased against ESL students. He ended up suing the university but their decision was upheld last year. Still makes me mad. - https://www.mprnews.org/story/2025/01/17/phd-student-says-university-of-minnesota-expelled-him-over-ai-allegation
Thank you Dani. I remember that story. It also make me very very angry! When did we stop treating our students as individuals rather than figures on a balance sheet?
Maybe the question isn't when students became numbers, but when the accounting stopped being optional to look at. International tuition was subsidizing a structure that let everyone experience the institution as something other than a balance sheet. That subsidy is disappearing now, not the honesty of an earlier era. The numbers were always there and the cover is dissolving.
Very well said Chris. 👏👏👏
Assessment of interaction with LLM’s is the only logical solution. Pandora’s box is open. The only way to respond is by working within the rules that escape demands. AI detection will always be a Salem Witch Trial.
Very well said. I keep coming back to 'The Crucible' recently and thinking how apt it is for this current persecution.
Students and ESL speakers or anyone speaking a second language at all cannot be blamed for a technological advancement foisted on them. Even scientists talk about how they cannot avoid using models due to the competitive edge they offer (if you need sourcing will provide). The goal is how to reason. Teaching students how to reason with AI is much more challenging, and maybe more rewarding than blaming them for something they cannot compete with. Shaming students for using AI is like blaming someone for getting their car stolen because they didn’t lock the doors.
Now I know you and me speak the same language. 🙏
Wow I am an AI developer student. Should I just give up now? If we start thinking like this we will never overcome the hackers who know anything is possible. I respect your opinion though of course.
Thanks. But what do you mean give up? That is the opposite of what I am trying to encourage. 🙏
It was tongue in cheek no worries. Appreciate your writing thank you. I may be affected by my SubStack Feed today which seems to be covered with truly scary possibilities of the use of AI technology if we don’t work faster to cover the gaps. But we have to remember our history; we created the nuclear bomb; it is now one of the greatest threats to us and humanity. Iran is said to be working to use AI for warfare. And it’s highly possible the US pulled Anthropic’s Fable and Mythos code for “National Security” reasons but likely so they could be the first to weaponize the tool before their adversaries can. History repeats. And we have in a sense created a new population through AI which just doubles the risk they already exists. A population which has within it inherent #algorithmic bias. While some question that to be true it’s been proven and is proof that AI was built to do exactly what was intended - automate human interaction which simply reflects an accurate representation of society today. But - not to worry. I’m having too much fun with AI and using it for good purposes (DV survivors) so I basically agree with you. If I used AI it would have likely removed my tongue in cheek reaction to things I don’t want to accept are true. But I’m sure Claude will come around. (FYI definitely not written by AI 🤙 just me. Thank you for your writing Sir! I’m working on a prompt to get my son to do the dishes right now (it’s actually true - I created automated annoying reminders with sound and other media on every piece of technology he owns - it’s hilarious. Research is still out on its successful outcomes). It was lighthearted—no need to worry. I value your writing; thank you. Today’s SubStack Feed may be shaping my perspective, filled with genuinely unsettling possibilities about AI technology if we don’t act quicker to address the gaps. Yet we must recall our past: we developed the nuclear bomb, now among the greatest dangers to us and humanity. Iran reportedly aims to apply AI in warfare. The U.S. may have withdrawn Anthropic’s Fable and Mythos code for “National Security” reasons, possibly to lead in weaponizing the tool before rivals do. Patterns from history reappear. In a way, we’ve generated a new population through AI, doubling the existing risk. This population carries inherent algorithmic bias. Though some doubt this, it’s been demonstrated—proof that AI was designed for its intended purpose: automating human interaction, which accurately mirrors today’s society. Still, there’s no cause for alarm. I’m enjoying AI too much and applying it constructively (for DV survivors), so I largely agree with you. Had I used AI, it might have erased my lighthearted response to realities I’d rather not accept. Yet I trust Claude will catch up to my quips soon enough!
Thank you for your writing, sir! I’m off to revise my current prompt so my son does the dishes — absolutely no joke! I set up automated, bothersome alerts and reminders with sound and other media across every device he owns—I don’t even care that it isn’t achieving its intended outcomes because I’m having too much fun. Of course he reprogrammed everything within a day that brilliant kid. But AI may be reaching 50 but I will keep educating myself and believing every day that I can and will succeed. I’ve had to survive battles one could never imagine.
That is the AI content I need to have in my feed as well. Soon my non-profit will be in the mix.
The deeper problem isn't that AI detection gets things wrong.
It's that schools keep outsourcing judgment to a tool that cannot carry the moral weight of the job.
That's what stood out! Even if detection improves, the posture will remain corrosive. It trains institutions to treat suspicion as a substitute for teaching, when the more durable move is to design assessment around process, reasoning, and choices students can actually explain.
A detector can flag a page.
It can't build trust, surface thought, nor teach integrity
I like that this is a theme that Sam keeps returning to because it bears repeating until it's a fundamental rule that we all understand; AI cannot be accountable for the judgement it renders. And if not the AI, then who will own those decisions? Another comment talked about a case where a student said that detectors are biased against ESL student after his work was flagged, sued the institution and lost.
Where was the instructor who replied on AI in the first place in all that mess? I doubt that they were ever called to explain their decision to the student or administration in a meaningful way. Since I don't know the details of the case, I can't say for sure where the breakdown was. Maybe the institution really was justified somehow but it seems to be a situation that is ripe for abuse, especially around what the student described, AI detectors' inherent bias against non-native English speakers/writers.
I keep saying it because I'll never understand it: educators using AI to detect AI?? Truly, WTF? Rules for thee, not for me. That's the one and only argument I need every single time this issue comes up. I appreciate Sam for giving me a little more than my harsh moral judgement to stand on but I do think my observation stands on its own.
I hear you. That is the accountability gap.
If a detector makes a serious accusation, a human being still has to own the judgment, explain it, and defend it. Too often the software gets treated like neutral authority when it’s really just outsourcing suspicion with a cleaner interface.
And yes, that bias issue alone should have ended this conversation a long time ago …
The ineffable is ineffable, and there's no 'effin' way Assistive Informatics can ever grok it.
To paraphrase Professor Spooner, an engine parsing a piece of writing by tokens and waving statistical hands over it, "hisses all of its mystery, tasting each and every worm." A ritual of specious moment.
______________
Interestingly, sometimes lack of polish conveys how much a student struggled with the topic, in a salutary way. It reminds me of an anecdote about Abraham Lincoln shared by sculptor Leonard Volk, who said Lincoln skipped church to come in early for his sitting when Volk was working on his bust:
“He entered my studio on Sunday morning, remarking that a friend at the hotel (Tremont House) had invited him to go to church, ‘but,’ said Mr. Lincoln, ‘I thought I’d rather come and sit for the bust. The fact is,’ he continued, ‘I don’t like to hear cut-and-dried sermons. No—when I hear a man preach, I like to see him act as if he were fighting bees!’ And he extended his long arms, at the same time suiting the action to the words.”
Likewise for our students, struggling with the material: alive, urgent, gestural, cognitively and emotionally engaged.
PS, the work of a sculptor is like that of a writer: to catch the moving living with the static settled. God bless Volk for writing about the gestures of long-armed Lincoln.
I love this! And I will very try to channel the idea of a man fighting bees for my next keynote. 🙏
Well said! This is what I've been trying to say and emphasize in higher ed regarding AI detection for the last two years now, and this is the most succinct articulation I've seen to date of the problem, the evidence, and what to do instead. Most write-ups I've seen only focus on part of that; love the holistic picture expessed here.
I'm looking forward to presenting "The AI Detection Fallacy: How AI Integrity Tools Fail, Who Pays the Price, and What Better Assessment Design Does Instead" at SIDLIT 2026 (July 31), and this article is a great example of the issues, concerns, & thoughts I'll be sharing!
Thanks so much David! And wishing you the very best with your presentation. 🙏
"Turnitin handled the all-human and all-AI extremes but failed on mixed, real-world scripts"
So glad you're writing about this. It has been bugging me for some time.
Using AI in your workflow is not inherently bad, it depends on how you use it, and why. There are multiple variations of a human-AI collaboration. If detectors were sophisticated enough to be able to tell how and where AI was used, what combination would it deem are acceptable? And who makes that decision?
Time to move on from the myopic 'AI is bad' debate and have an open and honest conversation about this.
Exactly Mariam! I honestly don't know who decided that AI was bad and that we all had to sign up to that opinion?
What do you think of plagiarism scans? We have to use them on Bachelor's and Master's Theses.
We also have use them in the journals I edit as well and honestly a lot of the time highs scores just indicate a lot of citations. My main issue with these ‘scores’ is when people use them without looking in detail at the actual qualitative matches for context…
The same here. I have just scanned a student’s BA thesis with a relatively high score and all of it was the list of references and footnotes with citations.
The students are so worried about the plagiarism scan, even more worried about plagiarism than the actual content.
This makes me so sad Edith. Those poor students. 😢
Me too. I’ll try to make my supervisees to be less scared about the whole process.
so far we have not been made to scan for AI use. If they cite AI correctly, they are allowed to use it, I think. At least, for language. Some of my students use grammarly because they need help with English (if they have chosen to write their thesis in English)
And if Grammarly is ok then why are other tools not? Such inconsistent behaviour from universities means it is little wonder our students are stressed.
Adaptation my friend
To me, adapting to technology means learning how to use each new tool without allowing the tool to control how I think.
My relationship with technology began with an Apple IIe in the 1980s. In 1989, I started using computers in my social studies classroom, even though my formal education at the time was a bachelor’s degree in history and social studies.
During the 1990s, I continued teaching while earning graduate training in network systems administration. I later worked as a technical coordinator in two school districts, taught technology, built my own computer-related businesses, and spent a year working for Gateway Computers between teaching positions.
More recently, I earned a master’s degree in artificial intelligence with certification in prompt engineering.
I was born long before the digital era, so technically I am a digital immigrant, not a digital native. But I have spent most of my adult life repeatedly crossing that border.
That is what adapting to technology means to me: remaining curious, learning the new language, testing the new tools, and then using them to serve human purposes—especially teaching, communication, and problem-solving.
The technology keeps changing.
The willingness to learn must remain.
All through history, humans have had to adapt to technological changes
This is nothing new in human experience
Thank you for this. 🙏
Can you give an example?
The moment you bought that 'smart phone' however many years ago, each of you made this nightmare world today inevitable. Each new techno-gimmick fad proves that you'll fall for the next one, like cattle stampeding over a cliff, every time.
So by the time you can't even walk into a Walmart without an AI implant installed, will you even remember how to say 'fuck this shit'? Those were the only words you ever needed to know how to say, about every bit of this utterly unnecessary high-tech crap. You've done this to yourselves.