Honestly, I think this a far more important conversation to have than about writing, and not just because there are risks to AI proliferation but it's the one area where I see room for AI to act as a useful prosthesis of sorts. We're actually deeply familiar with guides of various kinds intended to facilitate reading or to accelerate transmission of information. Some of them are inside of texts--table of contents, index, glossary, chapters, subheadings, abstracts, annotations, footnotes, etc. Some of them are texts that help with a text--concordances, summaries, commentaries, book reviews. We're familiar with the advantages and disadvantages of using something like "Cliff's Notes" or "For Dummies" guides prior to trying to read a text, but we also acknowledge the necessity of needing help simply by the fact that we teach classes *about* a particular text, that many things cannot be read in isolation or in solitude. For some texts, no one would argue that you have to make sense of it without assistance--Finnegan's Wake, for example. Many of us are operating with vast systems of paratextual knowledge--e.g., we know about the content of readings that we haven't done, and our knowledge is often basically correct. We know there's a value to reading a text deeply and well but we know it's not an infinite value--that there are trade-offs (of time, of efficiency, of necessity, and of affinity--no one wants to struggle profoundly with a reading that they have come to hate or feel alienated by).
What I find strange is precisely what you start with here in your title: literally no one is talking about this aspect of AI. But that is perhaps partly because we oddly treat reading as a lower-level pedagogical job, to be attended to in secondary education. There is very little conversation generally about how to read past high school, only a lot of moral panic about *whether* people are reading enough or reading the right things.
Between the submission of the first draft of my book manuscript (about gen AI and writing) and the final revision, I added two chapters on reading because I had a similar epiphany to what you articulate here, that students were being encouraged to avoid reading, denying them that critical thinking experience, where the intersection of reader and text allows a new idea not in the text to form. Without that process, those Ideas don't come into existence.
To me, it looked a lot like what's happened with writing instruction, allowing a simulation to substitute for the genuine article because that simulation can pass muster for the purposes of schooling. This attitude pre-dates gen AI, also, but gen AI accelerates things in a way that makes me scared we'll abandon some of the important experiences of reading the same way writing is at risk.
I appreciate your insights on this. The title popped up in my email today and I had to read it immediately. At my high school site we have been developing grade level teams to determine a focus language domain (ELD standards) for each grade. Each of our teams independently and overwhelmingly agreed that B.6: Reading Closely, was the most important skill that our students needed and are currently under prepared for. I'm sharing your article with all of them now. Thank you.
In the spirit of John Warner's argument that ChatGPT can't kill anything worth saving, I want to echo your point that generative AI as it currently exists is only good at summary and does not perform close reading and analysis well. That would mean the value of teaching practices that focuses on developing close reading skills should increase at all levels of instruction. If ChatGPT can do your homework, that means your homework has little educational value. In terms of reading, it doesn't seem that generative AI is providing anything substantively different than SparkNotes. The utility of close reading is that thinking deeply about a passage leads to insights about language and its relation to experiences grounded in social contexts. I find that a valuable skill, one that AI in its current forms has little ability to augment. If our response to generative AI is to refocus instructional time and effort on helping our students develop skills to closely analyze text and cultural artifacts, we will accomplish something that needed doing anyway.
If students use AI to skip reading a text their professor considers central to attainment, that is a problem; if they use it to improve their understanding and focus their rereading, that is....awesome. Our underfunded school doesn't provide anything like that kind of support, and most of our students need more help with college-level reading comprehension than our faculty can provide.
I think that's ultimately the tension here. Students can use AI reading assistants in ways that support learning, not offload it. But doing so takes structure and intentional pedagogy. I worry that many underfunded schools won't be able to provide that structure in the first place and AI reading will become a sort of default that some educators expect, like spell check on Word Processors.
I am hoping an entity like Khan Academy (for altruistic reasons) or Pearson (for profit motives) will provide a purpose-built AI collaborative reading assistant. Ideally, it would also help the professor assess the reading assignments and make course corrections before students do poorly on the tests/essays/speeches the readings were to have informed.
Good points, all of them. I think using Hypothes.is this way is a great idea. While I am an instructional technologist now, I did my undergrad and graduate work in history. One of the pivotal moments of my undergraduate career was when a grad student taught me how to approach a book, H.M.D. Parker's Roman Legions. It had never occurred to me to look at the bibliography after reading the preface to get an idea of the depth and scope of the research. It was a valuable lesson, and the scene is still fresh in my memory. The point, of course, is that this is another thing that having an AI summary defeats. It wasn't that someone told me to do this, they sat down and showed me by example. That was valuable. I also wonder how much would have been lost in the summary of Parker's arguments about the Antesignani, which I recall puzzling me at first but were worth working through to understand something about how think about what sources tell us about Roman legions.
Thanks, Guy! I, too, read History as an undergrad and could have never understood Procopius's Secret History of Justinian if I didn't conduct close reading on my own and look at the work of other scholars and how they complicated/ illuminated aspects.
"Without skilled critical readers, we risk producing minds that are primed to accept neatly summarized information at face value rather than question it, analyze it from multiple angles, and draw their own nuanced conclusions. This hollowing out of one of the cornerstones of higher-order reasoning could severely inhibit innovation, civic discourse, and our collective capacity to thoughtfully navigate complex issues." We are already seeing the impacts! People who don't or can't read are easy to manipulate. Those promoting these tools are not doing so with an interest in developing critical thinking voters and citizens, or innovative contributors to society.
You might be interested in this series on originality:
If you substitute ‘walking’ for ‘reading’ in this piece, it becomes clearer how deskilling AI summaries are. If I never walked, how could I learn to actually walk? If you use a tool to do a task, the skill you learn is how to use the tool not the task. If the value of the tool use skill is greater than the skill it is a fair trade off; if not, you have lost out
Great article! I agree that losing our skill to summarize and understand texts is a big problem. I am disappointed that these AI labs did not have the foresight or care to understand how this would be used by students to avoid doing homework.
A counter argument might be to double down on tools or capture objectives achieved by them AI or LLMs. And in the process preserve the qualities of close reading you are trying to achieve?
I somehow see exponential potential in enhancing their learning ability with these new tools. But done proactively, not passively to speed read or summarize without comprehension. Hear me out (these are personal views without prejudice or conflict, based on my own teaching journey done in practice/industry).
On the assumption close reading is trying to achieve some of these:
- deepen students' understanding of a text,
- promote critical thinking, and
- develop analytical skills
E.g.
1. For AI-generated questions and prompts: Tools like TextHelp's Read&Write or Quillionz can automatically generate comprehension questions based on the text, encouraging readers to think critically about key concepts and themes.
2. Personalized recommendations: Platforms like Goodreads or Amazon's Kindle use AI algorithms to recommend books based on a reader's past preferences and reading habits, guiding them towards texts that align with their interests and reading level. Or just ask ChatGPT. Feynman's how do I understand this like I'm 5?
3. Interactive elements: Apps like Epic! or Novel Effect incorporate interactive features such as quizzes, annotations, and multimedia enhancements e.g. animations, fostering engagement and deeper exploration of the text.
4. Collaborative discussions: Platforms like Perusall or Padlet enable readers to annotate and discuss texts collaboratively in real-time, facilitating peer learning and the exchange of diverse perspectives.
5. Diverse perspectives: Tools like Microsoft's Immersive Reader or Project Naptha utilize AI to translate text into different languages, adjust reading levels, or provide alternative formats such as audio or braille, making texts more accessible to a wider range of readers and promoting inclusivity.
Now the cool stuff - To encourage critical thinking and analysis in AI-assisted reading:
6. Text analysis platforms: Tools like Voyant Tools (used your chapter- https://voyant-tools.org/?corpus=f31d7f49a48f6b8ffa04018c316e9795) or Lexos allow users to perform text analysis, including word frequency analysis, topic modeling, and sentiment analysis, providing insights into the text's themes, language use, and emotional tone.
7. Argument mapping software: Platforms such as Argumentative or MindMup enable users to visually map out arguments and counterarguments within a text, helping readers identify key points, evaluate evidence, and construct reasoned responses.
8. Annotation and note-taking tools: Apps like Hypothesis or Notion offer features for annotating texts and taking structured notes, facilitating active reading and the organization of thoughts and observations.
9. Socratic questioning bots: AI chatbots like Socrates (Socratic Lab) or IBM Watson's Debater (just sunsetted 1st May 2024) use natural language processing to engage users in Socratic questioning, prompting critical reflection and deeper exploration of the text's meaning and implications. There are others - Debate Devil, Debate AI, etc. 70+ References included here: https://topai.tools/s/debate-tool
10. AI writing assistants: Tools such as Grammarly or Hemingway Editor provide feedback on writing style, grammar, and clarity, encouraging writers to revise and refine their arguments and interpretations.
You dont need all. But the spectrum is indeed wide. Failing which many of what these tools do (objectives wise) can be prompted or programed into function via APIs e.g. Chat GPT 4, Gemini's, Claude equivalents.
While close reading emphasizes deep engagement with the text, the integration of AI-assisted reading tools can enhance this process and potentially lead to outcomes surpassing those of traditional close reading in several ways e.g.:
A. Efficiency: AI tools can quickly analyze large volumes of text, identify patterns, and generate insights that may not be immediately apparent through manual close reading alone, saving time and enabling readers to delve deeper into complex texts.
B. Personalization: AI algorithms can tailor recommendations, questions, and annotations to individual readers' preferences, learning styles, and comprehension levels, providing a more personalized and adaptive reading experience that meets the diverse needs of learners.
C. Accessibility: AI-driven features such as text-to-speech, translation, and adaptive formatting make texts more accessible to a wider range of readers, including those with visual impairments, language barriers, or learning disabilities, promoting inclusivity and equity in education.
D. Collaboration: AI-powered collaborative tools facilitate peer learning, discussion, and knowledge sharing among readers, creating opportunities for collaborative sense-making, collective problem-solving, and the co-construction of meaning beyond the limitations of solitary close reading.
E. Enhanced analysis: Text analysis and argument mapping tools enable readers to explore texts from multiple perspectives, uncover hidden connections, and critically evaluate arguments, fostering higher-order thinking skills such as synthesis, evaluation, and creativity that go beyond the scope of traditional close reading.
Not really Lynn. Just thinking no one has definitive answers/solutions (not looking for any either). So anything goes. I believe in possibilities too. Or leverage, which comes in varying forms. P.s. also been on both (1st) the sell side, then buy sides. Imperfect to the hilt! Guessing you probably know better too. What If this is the Wild Wild West? 😄
Sounds like we have had similar experiences and perspectives that only come from observing, learning, failing, and then adapting. Without uncertainty there are no possibilities, so it is more fun to think of it as the Wild West than trying to control things we can’t control. It is fun to explore the different points of view in these discussions. 🙏😊
Some great points made here, thank you, Marc. First, I want to state that this is another clear reason why we must teach AI Literacy to all students. I express AI Literacy as being comprised of Awareness, Capability, Knowledge, and Critical Thinking. Within Critical Thinking is the vital aspect of avoiding overreliance. "Overreliance" is a negative thing if we unwittingly lose skills and capabilities to a system such as generative AI. This is bad thing if it is not done purposefully and is done without taking second and third order effects into consideration. An example would be cursive writing. Anyone that is close to my age remembers have to take many classes devoted specifically to cursive writing. For a long time, even dissertations HAD to be hand written, because good handwriting was an important professional skill. Then, thanks to the proliferation of type writers, computers, and printers, that must have skill was "let go." - I'm not saying good reading ability is something to be let go. I am simply giving an example of overreliance and the importance of being aware of it so that it is purposefully addressed and considered.
We do need to inform student of the importance to good reading, but at the same time we in academia need to also be realistic with ourselves. I remember being a student and having too much reading to do. It was impossible. I often had to skip some sleep, skip meals, or workouts, or something else because I had too much reading to do. I dissentingly remember thinking to myself, "I wish I could read a summarized version of this with the specifics of what I need for the assignment." AI can help with that, but as you expertly pointed out, it comes at a price, meaning the student doesn't gain a deeper understanding. - Yet this is the reality of higher education and it has been for a long time. As with everything a proper middle ground is needed.
Thanks for these fine thoughts! I appreciate that you shift the focus from writing to reading, since neither is possible without the other and the skills go hand-in-hand. My favorite passage (of many): "Unchecked, the lure of frictionless reading could produce a profoundly atrophied culture of surface-level ideas rather than exploring them in depth. In such a world, I shudder to think how blind obedience to authoritative-seeming AI outputs could allow misinformation and weaponized narratives to proliferate unabated." We might already be in those places, so the friction of applied thinking is more necessary than ever. Good words.
This is exactly what I have been preaching in my workshops since the outbreak of ChatGPT. Teachers and research students come to me for AI tool recommendations given that I spend a lot of time excavating for these tools, and I always warn that when it comes to reading, AI tools, especially in academic and scholarly endeavours, can cause more harm than good.
At least before you generate a summary, do the first or even the second reading (if the paper is a seminal work or of direct relevance to your research) yourself. Besides developing your 'expert knowledge' and analytical skills, reading the paper yourself and understanding what it is about can help you judge whether the AI generated summary captures the nuance of the paper and spot inaccuracies when they happen, and they do.
Thank you for this interesting piece. Looking forward to read the chapter you published as well.
Honestly, I think this a far more important conversation to have than about writing, and not just because there are risks to AI proliferation but it's the one area where I see room for AI to act as a useful prosthesis of sorts. We're actually deeply familiar with guides of various kinds intended to facilitate reading or to accelerate transmission of information. Some of them are inside of texts--table of contents, index, glossary, chapters, subheadings, abstracts, annotations, footnotes, etc. Some of them are texts that help with a text--concordances, summaries, commentaries, book reviews. We're familiar with the advantages and disadvantages of using something like "Cliff's Notes" or "For Dummies" guides prior to trying to read a text, but we also acknowledge the necessity of needing help simply by the fact that we teach classes *about* a particular text, that many things cannot be read in isolation or in solitude. For some texts, no one would argue that you have to make sense of it without assistance--Finnegan's Wake, for example. Many of us are operating with vast systems of paratextual knowledge--e.g., we know about the content of readings that we haven't done, and our knowledge is often basically correct. We know there's a value to reading a text deeply and well but we know it's not an infinite value--that there are trade-offs (of time, of efficiency, of necessity, and of affinity--no one wants to struggle profoundly with a reading that they have come to hate or feel alienated by).
What I find strange is precisely what you start with here in your title: literally no one is talking about this aspect of AI. But that is perhaps partly because we oddly treat reading as a lower-level pedagogical job, to be attended to in secondary education. There is very little conversation generally about how to read past high school, only a lot of moral panic about *whether* people are reading enough or reading the right things.
Good work!!! I have a draft on this subject in the works!!! Now I will pick up where you leave off.
Between the submission of the first draft of my book manuscript (about gen AI and writing) and the final revision, I added two chapters on reading because I had a similar epiphany to what you articulate here, that students were being encouraged to avoid reading, denying them that critical thinking experience, where the intersection of reader and text allows a new idea not in the text to form. Without that process, those Ideas don't come into existence.
To me, it looked a lot like what's happened with writing instruction, allowing a simulation to substitute for the genuine article because that simulation can pass muster for the purposes of schooling. This attitude pre-dates gen AI, also, but gen AI accelerates things in a way that makes me scared we'll abandon some of the important experiences of reading the same way writing is at risk.
I appreciate your insights on this. The title popped up in my email today and I had to read it immediately. At my high school site we have been developing grade level teams to determine a focus language domain (ELD standards) for each grade. Each of our teams independently and overwhelmingly agreed that B.6: Reading Closely, was the most important skill that our students needed and are currently under prepared for. I'm sharing your article with all of them now. Thank you.
In the spirit of John Warner's argument that ChatGPT can't kill anything worth saving, I want to echo your point that generative AI as it currently exists is only good at summary and does not perform close reading and analysis well. That would mean the value of teaching practices that focuses on developing close reading skills should increase at all levels of instruction. If ChatGPT can do your homework, that means your homework has little educational value. In terms of reading, it doesn't seem that generative AI is providing anything substantively different than SparkNotes. The utility of close reading is that thinking deeply about a passage leads to insights about language and its relation to experiences grounded in social contexts. I find that a valuable skill, one that AI in its current forms has little ability to augment. If our response to generative AI is to refocus instructional time and effort on helping our students develop skills to closely analyze text and cultural artifacts, we will accomplish something that needed doing anyway.
If students use AI to skip reading a text their professor considers central to attainment, that is a problem; if they use it to improve their understanding and focus their rereading, that is....awesome. Our underfunded school doesn't provide anything like that kind of support, and most of our students need more help with college-level reading comprehension than our faculty can provide.
I think that's ultimately the tension here. Students can use AI reading assistants in ways that support learning, not offload it. But doing so takes structure and intentional pedagogy. I worry that many underfunded schools won't be able to provide that structure in the first place and AI reading will become a sort of default that some educators expect, like spell check on Word Processors.
I am hoping an entity like Khan Academy (for altruistic reasons) or Pearson (for profit motives) will provide a purpose-built AI collaborative reading assistant. Ideally, it would also help the professor assess the reading assignments and make course corrections before students do poorly on the tests/essays/speeches the readings were to have informed.
Good points, all of them. I think using Hypothes.is this way is a great idea. While I am an instructional technologist now, I did my undergrad and graduate work in history. One of the pivotal moments of my undergraduate career was when a grad student taught me how to approach a book, H.M.D. Parker's Roman Legions. It had never occurred to me to look at the bibliography after reading the preface to get an idea of the depth and scope of the research. It was a valuable lesson, and the scene is still fresh in my memory. The point, of course, is that this is another thing that having an AI summary defeats. It wasn't that someone told me to do this, they sat down and showed me by example. That was valuable. I also wonder how much would have been lost in the summary of Parker's arguments about the Antesignani, which I recall puzzling me at first but were worth working through to understand something about how think about what sources tell us about Roman legions.
Thanks, Guy! I, too, read History as an undergrad and could have never understood Procopius's Secret History of Justinian if I didn't conduct close reading on my own and look at the work of other scholars and how they complicated/ illuminated aspects.
From 31 years ago...
https://dailypapert.com/obsolete-skill-set-the-3-rs-literacy-and-letteracy-in-the-media-ages/
Discuss
"Without skilled critical readers, we risk producing minds that are primed to accept neatly summarized information at face value rather than question it, analyze it from multiple angles, and draw their own nuanced conclusions. This hollowing out of one of the cornerstones of higher-order reasoning could severely inhibit innovation, civic discourse, and our collective capacity to thoughtfully navigate complex issues." We are already seeing the impacts! People who don't or can't read are easy to manipulate. Those promoting these tools are not doing so with an interest in developing critical thinking voters and citizens, or innovative contributors to society.
You might be interested in this series on originality:
https://janetsalmons.substack.com/p/encourage-originality-create-a-culture?r=410aa5
If you substitute ‘walking’ for ‘reading’ in this piece, it becomes clearer how deskilling AI summaries are. If I never walked, how could I learn to actually walk? If you use a tool to do a task, the skill you learn is how to use the tool not the task. If the value of the tool use skill is greater than the skill it is a fair trade off; if not, you have lost out
Great article! I agree that losing our skill to summarize and understand texts is a big problem. I am disappointed that these AI labs did not have the foresight or care to understand how this would be used by students to avoid doing homework.
A counter argument might be to double down on tools or capture objectives achieved by them AI or LLMs. And in the process preserve the qualities of close reading you are trying to achieve?
I somehow see exponential potential in enhancing their learning ability with these new tools. But done proactively, not passively to speed read or summarize without comprehension. Hear me out (these are personal views without prejudice or conflict, based on my own teaching journey done in practice/industry).
On the assumption close reading is trying to achieve some of these:
- deepen students' understanding of a text,
- promote critical thinking, and
- develop analytical skills
E.g.
1. For AI-generated questions and prompts: Tools like TextHelp's Read&Write or Quillionz can automatically generate comprehension questions based on the text, encouraging readers to think critically about key concepts and themes.
2. Personalized recommendations: Platforms like Goodreads or Amazon's Kindle use AI algorithms to recommend books based on a reader's past preferences and reading habits, guiding them towards texts that align with their interests and reading level. Or just ask ChatGPT. Feynman's how do I understand this like I'm 5?
3. Interactive elements: Apps like Epic! or Novel Effect incorporate interactive features such as quizzes, annotations, and multimedia enhancements e.g. animations, fostering engagement and deeper exploration of the text.
4. Collaborative discussions: Platforms like Perusall or Padlet enable readers to annotate and discuss texts collaboratively in real-time, facilitating peer learning and the exchange of diverse perspectives.
5. Diverse perspectives: Tools like Microsoft's Immersive Reader or Project Naptha utilize AI to translate text into different languages, adjust reading levels, or provide alternative formats such as audio or braille, making texts more accessible to a wider range of readers and promoting inclusivity.
Now the cool stuff - To encourage critical thinking and analysis in AI-assisted reading:
6. Text analysis platforms: Tools like Voyant Tools (used your chapter- https://voyant-tools.org/?corpus=f31d7f49a48f6b8ffa04018c316e9795) or Lexos allow users to perform text analysis, including word frequency analysis, topic modeling, and sentiment analysis, providing insights into the text's themes, language use, and emotional tone.
7. Argument mapping software: Platforms such as Argumentative or MindMup enable users to visually map out arguments and counterarguments within a text, helping readers identify key points, evaluate evidence, and construct reasoned responses.
8. Annotation and note-taking tools: Apps like Hypothesis or Notion offer features for annotating texts and taking structured notes, facilitating active reading and the organization of thoughts and observations.
9. Socratic questioning bots: AI chatbots like Socrates (Socratic Lab) or IBM Watson's Debater (just sunsetted 1st May 2024) use natural language processing to engage users in Socratic questioning, prompting critical reflection and deeper exploration of the text's meaning and implications. There are others - Debate Devil, Debate AI, etc. 70+ References included here: https://topai.tools/s/debate-tool
10. AI writing assistants: Tools such as Grammarly or Hemingway Editor provide feedback on writing style, grammar, and clarity, encouraging writers to revise and refine their arguments and interpretations.
You dont need all. But the spectrum is indeed wide. Failing which many of what these tools do (objectives wise) can be prompted or programed into function via APIs e.g. Chat GPT 4, Gemini's, Claude equivalents.
While close reading emphasizes deep engagement with the text, the integration of AI-assisted reading tools can enhance this process and potentially lead to outcomes surpassing those of traditional close reading in several ways e.g.:
A. Efficiency: AI tools can quickly analyze large volumes of text, identify patterns, and generate insights that may not be immediately apparent through manual close reading alone, saving time and enabling readers to delve deeper into complex texts.
B. Personalization: AI algorithms can tailor recommendations, questions, and annotations to individual readers' preferences, learning styles, and comprehension levels, providing a more personalized and adaptive reading experience that meets the diverse needs of learners.
C. Accessibility: AI-driven features such as text-to-speech, translation, and adaptive formatting make texts more accessible to a wider range of readers, including those with visual impairments, language barriers, or learning disabilities, promoting inclusivity and equity in education.
D. Collaboration: AI-powered collaborative tools facilitate peer learning, discussion, and knowledge sharing among readers, creating opportunities for collaborative sense-making, collective problem-solving, and the co-construction of meaning beyond the limitations of solitary close reading.
E. Enhanced analysis: Text analysis and argument mapping tools enable readers to explore texts from multiple perspectives, uncover hidden connections, and critically evaluate arguments, fostering higher-order thinking skills such as synthesis, evaluation, and creativity that go beyond the scope of traditional close reading.
Wow. Thanks for sharing your perspective. What a deep and complex thinker you are.
Not really Lynn. Just thinking no one has definitive answers/solutions (not looking for any either). So anything goes. I believe in possibilities too. Or leverage, which comes in varying forms. P.s. also been on both (1st) the sell side, then buy sides. Imperfect to the hilt! Guessing you probably know better too. What If this is the Wild Wild West? 😄
Sounds like we have had similar experiences and perspectives that only come from observing, learning, failing, and then adapting. Without uncertainty there are no possibilities, so it is more fun to think of it as the Wild West than trying to control things we can’t control. It is fun to explore the different points of view in these discussions. 🙏😊
Eye Eye Captain!!
Some great points made here, thank you, Marc. First, I want to state that this is another clear reason why we must teach AI Literacy to all students. I express AI Literacy as being comprised of Awareness, Capability, Knowledge, and Critical Thinking. Within Critical Thinking is the vital aspect of avoiding overreliance. "Overreliance" is a negative thing if we unwittingly lose skills and capabilities to a system such as generative AI. This is bad thing if it is not done purposefully and is done without taking second and third order effects into consideration. An example would be cursive writing. Anyone that is close to my age remembers have to take many classes devoted specifically to cursive writing. For a long time, even dissertations HAD to be hand written, because good handwriting was an important professional skill. Then, thanks to the proliferation of type writers, computers, and printers, that must have skill was "let go." - I'm not saying good reading ability is something to be let go. I am simply giving an example of overreliance and the importance of being aware of it so that it is purposefully addressed and considered.
We do need to inform student of the importance to good reading, but at the same time we in academia need to also be realistic with ourselves. I remember being a student and having too much reading to do. It was impossible. I often had to skip some sleep, skip meals, or workouts, or something else because I had too much reading to do. I dissentingly remember thinking to myself, "I wish I could read a summarized version of this with the specifics of what I need for the assignment." AI can help with that, but as you expertly pointed out, it comes at a price, meaning the student doesn't gain a deeper understanding. - Yet this is the reality of higher education and it has been for a long time. As with everything a proper middle ground is needed.
You’re right — people often talk about AI’s effects on writing or productivity, but miss the deeper risk: how we read.
It might not be just about slowing down or adding friction; it might be that reading as a transformative, risky act is disappearing altogether.
If the reader isn’t challenged, we lose the very subject whose understanding matters.
I tried to articulate this shift here:
https://sites.google.com/view/navimusaget/manifest
Thanks for these fine thoughts! I appreciate that you shift the focus from writing to reading, since neither is possible without the other and the skills go hand-in-hand. My favorite passage (of many): "Unchecked, the lure of frictionless reading could produce a profoundly atrophied culture of surface-level ideas rather than exploring them in depth. In such a world, I shudder to think how blind obedience to authoritative-seeming AI outputs could allow misinformation and weaponized narratives to proliferate unabated." We might already be in those places, so the friction of applied thinking is more necessary than ever. Good words.
This is exactly what I have been preaching in my workshops since the outbreak of ChatGPT. Teachers and research students come to me for AI tool recommendations given that I spend a lot of time excavating for these tools, and I always warn that when it comes to reading, AI tools, especially in academic and scholarly endeavours, can cause more harm than good.
At least before you generate a summary, do the first or even the second reading (if the paper is a seminal work or of direct relevance to your research) yourself. Besides developing your 'expert knowledge' and analytical skills, reading the paper yourself and understanding what it is about can help you judge whether the AI generated summary captures the nuance of the paper and spot inaccuracies when they happen, and they do.
Thank you for this interesting piece. Looking forward to read the chapter you published as well.