One of the readings that my students always pick as their favorite lesson about digital media economics comes from the late 1990s when Google engineers accidentally found gold in their trash. When they launched their search engine, it was working perfectly and millions of users were using it every day. Their algorithm was elegant and the results outperformed other competitors in the market. However, their major problem was that no one had figured out how to make profit from the search. A few years later, they discovered that every search query that users typed left behind a trail of behavioral signals. Not only did the search tag the topic users typed in but it showed what they were searching for, how they phrased it, what they clicked on, what they ignored and how long they lingered. In the early years, this data was largely seen as the exhaust of the search engine, and it was essentially discarded as waste. But when the engineers aggregated all of this discarded behavioral data, it revealed things about people’s true desire and intent. So, Google could suddenly sell not just the demographics of their customers to advertisers but also the implied aspirations of all of their search queries.
This story I teach in my media criticism course comes from Shoshana Zuboff’s book “The Age of Surveillance Capitalism.” Zuboff explains that the most valuable asset of the search engine was always the “exhaust fumes” and for the longest time they were hidden in plain sight. Once Google realized that the byproduct of the search process was the thing they could monetize, it propelled the company to become the largest seller of digital advertising in the world.
Currently, higher education might be going through a similar moment. AI has now become a part of everyday life on college campuses and I have embraced it in all my classes at every level. I now encourage my students to use NotebookLM when the text becomes dense or incomprehensible. I encourage the students to upload assigned text to create timelines, flashcards for studying and video summaries of their reading. Students can now process text faster, research more effectively and edit their work without having to go to the writing center. These gains are substantial and any serious conversation about AI in education must begin with acknowledging the benefits of AI.
When Making Learning Easier Makes it Harder
However, there is a counterintuitive aspect to the AI-assisted learning that resembles the situation from the Google’s experience with “digital exhaust.” Just as Google almost discarded its most valuable byproduct, universities now risk bypassing the very thing that makes learning stick. The cognitive effort involved in learning and writing that we willingly outsource to AI may turn out to be the part that matters most. The energy needed to work through a difficult text, the discomfort of holding conflicting ideas in working memory, the pressure of constantly reorganizing one’s own thought process all seem like efforts we are glad to hand over to AI. But these experiences, what psychologist Paul Bloom and his colleagues call friction, are not incidental to learning. They suggest that friction may be the primary mechanism by which learning transforms the human mind. At present, AI is helping to remove this cognitive struggle from the thinking process, treating it as the exhaust of learning.
So now, realizing that friction could be useful, I have to face the dilemma that my use of our campus-approved Copilot AI may be helping students in the short term but hurting them in the long term. Could my encouragement of artificial intelligence as a brainstorming partner, summarizer and comprehension assistant produce net-negative results?
It’s no surprise that people outside of academia have noticed this problem already. Journalist Ezra Klein, a rather enthusiastic user of AI, mentioned earlier this year that the knowledge AI provides for him doesn’t change him in the same way reading does. He says that AI informs him without significant transformation. He ponders the value of the struggle that the demanding encounter with a difficult problem has on people experiencing it. In other words, the struggle is not a sign of faulty process, instead it is the very mechanism by which meaningful learning happens.
I’ve seen this in my classroom when students who have used AI to summarize a reading can identify its main argument, but by the time the discussion progresses beyond the basics, they have little to work with because they haven’t thought it through. The knowledge is there but it did not transform thinking, it just helped with comprehension.
Not All Friction is Created Equal
Psychologists Emily Zohar, Paul Bloom, and Michael Inclicht make a similar argument in a 2026 paper called “Against Frictionless Artificial Intelligence.” They point out two kinds of friction: excess and productive friction. Throughout history, humans invented technology in order to eliminate excess frictions which made life difficult and unproductive. The washing machine is an example of such a tool and today nobody looks back on washing clothes in the river with nostalgia. Productive friction, on the other hand, is the kind of effort that is the basis of real progress and meaningful growth. When we work through a hard problem without knowing the answer and sit in the discomfort of ambiguity, the pressure of thinking benefits us in the long run. Removing it damages the condition under which real learning occurs.
We might currently be approaching education’s digital exhaust moment. The productive friction generated by genuine intellectual struggle could be the very byproduct of the learning process that we might not want to get rid of that easily. The story of Google’s engineers who were too focused on the search to notice the exhaust offers an invaluable lesson. Consequently, the launch of AI into the university classroom serves as opportunity to ask the harder question: What does learning look like in a future that has automated the hardest parts of thinking?
The answer to the question should combine the right amount caution and innovation. When Google discovered the value of its exhaust, it packaged it deliberately, built infrastructure around it, and made it the centerpiece of their new business model. Universities now have same opportunity and should be highly motivated to do something equivalent. Refusing to use AI at this point is as futile and misguided as it was when professors were forbidding students to use the Internet in the 1990s. Recognizing that cognitive friction is the exhaust of the learning process, and building deliberately around is best lesson that can be drawn grom Google’s experience. Teaching students to protect and exercise productive friction should become one of the major pedagogical goals of the university of the future, and the following are three concrete strategies worth taking seriously.
Make Deep-Reading a Learning Goal
Deep-reading or slow and analytical examination of a text has been a proven pedagogical strategy that trains much needed attention and patience through which the human mind develops and transforms. The first task is to reintroduce it as essential practice of analytical thinking beyond basic English or composition classes. Walter J. Ong, who wrote about reading and writing as transformative technologies, suggested that reading throws the psyche back on itself and this introspection is what AI bypasses when it summarizes or ‘reads’ for us. Building structured, deep-reading exercises into introductory-level seminars would allow students to relearn how to pay attention to slow and focused engagement with long texts. In turn, we should change the assessments so that they can reward the quality of a student’s encounter with a text rather than the polish writing at the end of semester. NotebookLM may be used with difficult texts, but intentional exercises in deep reading should become a learning objective in many more courses at all levels. In my own courses, I have purposefully reinserted the encounter with the text as one of the main pedagogical goals of the course, pausing to read with students in class and teaching them to develop reading maps and timelines before moving toward more complex research.
Build Productive Friction Into Argumentation
Furthermore, the Hegelian dialectic should be taught as an explicit form of critical thinking and not as an implied philosophical aspiration. I have been a teacher for three decades and one of the most reliable ways to communicate critical thinking to my students has been through the practice of dialectical thinking. Put simply, for every thesis, students must develop an antithesis. Both must be considered with care before they can be reconciled through synthesis. Students can now easily ask the AI to produce a synthesis without any awareness or effort of what it takes to create such complexly integrated logic. Dialectical thinking should become a part of any argumentation-building assignment; the work should always show the thesis and its genuinely developed opposite. Only after the student has endured the tension of opposing thoughts, should they be permitted to find a way to reconcile those. There are undoubtedly classes that already require students to construct the most persuasive version of the argument through synthesis, but this might be the time to an opportunity to demand that AI help build this essential competency. In these scenarios, AI can generate counterarguments to which students must respond. Students could be encouraged to play a ‘devil’s advocate’ to the arguments built by AI. Or AI can provide models of synthesis itself. But in my classes, I always insist that the synthesis not be rushed, that students learn to embrace the discomfort of holding opposing ideas, and that the effortful navigation of conflicting tension is worth more than a comfortable conclusion.
Give Students More Chances to Think on Their Feet
Finally, interpersonal communication and live discussions should become even more important as the least corruptible formats of modern communication. Live, in-person interaction is one of the last places that AI cannot meaningfully affect. In a Socratic seminar, an oral defense, or a structured debate, students must think with what they actually have in their heads. When forced to respond impromptu to a challenge they did not script in advance, they get to practice real-time thinking. Even in my communication courses, where discussions, debates, and presentations are already central, I have added more low-stakes presentations, unrehearsed reflections, and on-the-spot “vibe checks” to encourage real-time cognitive practice.
The questions of what the future brings and how it impacts college graduates should be central for all universities and all our classrooms. Google captured its exhaust and built entire economy by recognizing that the byproduct was as valuable as the product itself. Universities now face similar choice where AI is the equivalent of the search engine and cognitive friction is our version the exhaust. Those institutions that learn to package and protect that friction at the same as they incorporate AI in educational system will produce graduates who are ready for the new era ahead of them. This approach does not demand entirely new pedagogical practices, but it does require rethinking of a few familiar habits. In the end, universities are expected to be slow to change but I hope more teachers commit to preserving cognitive friction with the same intentionality as they embrace AI.
AI Disclosure: The author used Claude AI in a limited capacity during the editing and proofreading stage. All analysis, conclusions and creative work are the author’s own. The author reviewed and approved all content prior to submission.
Vladimir Bratic, PhD, is a professor of media and communication. He generally takes a techno-optimist approach to the new technologies. He has previously published 25 articles in academic journals, books and online publications, most of them on the subject of media contribution to peace.
