Ten Minutes Is Enough: How AI Erodes Human Persistence and Creativity
When capitalism sacrifices human persistence and creativity for profit — from Berkeley and MIT to Microsoft
Just ten minutes. That is all it takes for artificial intelligence to take away our ability to stay with a hard problem. This is no longer the claim of a worried columnist; it is a finding that three independent studies — at Berkeley, MIT, and Microsoft — have each reached from a different angle: the tools built to “help” us are eroding our persistence, memory, critical thinking, and creativity. And this erosion is no accident; it is the rational price of a logic that prefers profit over people.
This article puts the findings of these studies side by side and takes seriously, once again, the question posed by the book “A New Social Order in the Age of Consciousness”: in an age when human consciousness is supposed to be the foundation of a new social order, who is destroying this consciousness — for the sake of profitability?
Study one: Berkeley — the persistence that vanished
The Berkeley study was led by Brian Christian, a researcher at the Center for Human-Compatible AI, along with colleagues from Carnegie Mellon, MIT, Oxford, and UCLA, and presented in October 2026 at the Language Modeling Conference. The design was simple and clever: 1,222 participants in three randomized experiments; two groups, one task, and a tool taken away midway.
In the first experiment with 354 participants, both groups solved 15 fraction problems; one group without help, the other with ChatGPT at hand. The AI-assisted group was initially more accurate. But after the twelfth problem, the tool was taken away — and almost immediately, that group fell behind in solving the problems correctly. The second experiment with 667 people repeated the pattern: with the help removed, users either gave wrong answers or gave up entirely; the group without the tool stayed with the task and ended up more successful. The third experiment with 201 people and SAT reading-comprehension passages completed the picture: both persistence and accuracy fell.
The shocking point: the performance of the group that had used AI for ten minutes, after the tool was removed, often fell below that of the group that had never touched it. Christian says this is what many feel in their gut — and he experiences it himself.
Study two: MIT — “cognitive debt”
A year earlier, the MIT Media Lab, led by Nataliya Kosmyna, had gone straight to the brain. 54 participants wearing EEG caps wrote essays: one group with ChatGPT, one with a search engine, and one with only their own minds.
The result: the ChatGPT group showed the weakest neural engagement and the least brain connectivity — especially in waves related to memory, semantic processing, and creative ideation. When later asked to rewrite their essays without help, most could not even recall a single sentence of their own writing; because their brains had never truly engaged with the material. Evaluators described the AI-generated texts as “soulless” and lacking original thought.
Kosmyna gave this pattern a name that has now entered the literature: cognitive debt. Each time we let AI think for us, it feels free in the moment; but the learning that never happened accumulates, and the bill arrives later: a weaker memory, a shallower understanding, and skills we no longer have when the tool fails or is absent. The hopeful note of the study was this: those who first wrote without the tool and then turned to AI had both stronger brain connectivity and better memory — early cognitive investment paid off.
Study three: Microsoft and Carnegie Mellon — the erosion of critical thinking
The third study targeted the workplace. Researchers at Microsoft and Carnegie Mellon University asked 319 “knowledge workers” — people who work with data and information — to describe 936 real examples of their everyday use of generative AI.
The finding: 62 percent said that when they use generative AI, they apply less critical thinking — especially in everyday, low-risk tasks. The more a worker trusted the AI’s ability, the more they took their hands off the wheel; and conversely, those who were confident in their own expertise scrutinized AI outputs 27 percent more critically.
The researchers called this the “bitter irony of automation”: by mechanizing everyday tasks and leaving only exceptions to humans, we take away the user’s everyday opportunities to practice judgment and let their “cognitive muscle” atrophy — until, when an exception arises, they are unprepared. Another finding: AI users, for a single given task, produced a less diverse set of outputs — “mechanical convergence”; that is, diverse thought gave way to uniform, formulaic output.
The common pattern: harmful help
Three studies, three methods, three populations — and one pattern. In the short term, the tool “helps”: higher accuracy, greater speed, cleaner output. In the long term, it exacts a hidden price: less persistence, weaker memory, dimmer critical thinking, more uniform creativity.
Christian states this contradiction plainly: we built these systems to be “helpful,” but are they helpful in a way that truly helps us? His research showed that this help is often help that helps in a not-so-helpful way: it solves the problem, but weakens the solver.
To be honest: these findings are still young — the MIT study was a preprint, the Microsoft study relied on self-reporting — but their convergence from three independent labs is more than coincidence. When three separate compasses point in one direction, it is wise to look in that direction.
“Productive struggle”: what we are losing
Researchers have a precise name for what is being lost: productive struggle. That wrestling with a hard problem — the grappling in which one discovers one’s own strengths and interests. When AI delivers the answer in seconds, we get used to that speed; anything that takes longer feels “inefficient.” And with the removal of productive struggle, more than an answer is lost: the process of lasting learning is lost.
This is not just a classroom issue. The same pattern holds for professional researchers: relying on language models for literature summaries, code generation, or hypothesis drafts creates a “psychological illusion” — a sense of mastery that masks real gaps in knowledge. Some journals now ask authors to declare the extent of AI assistance and confirm they can reproduce the key steps of their analysis without the tool.
Why does capitalism want this? Persistence and creativity, sacrificed for profit
Here we must move from description to explanation: why did a tool that could have been everyone’s patient private tutor become an “answer engine”? The answer must be sought not in technology, but in the logic of profit.
Capitalism measures productivity by its own yardstick: more output in less time at lower cost. The “answer engine” delivers exactly that: a faster article, readier code, a cleaner report. That in the process the writer’s persistence is worn down, the programmer’s memory dimmed, and the analyst’s judgment eroded does not appear on the balance sheet — because the balance sheet sees the human as a “cost,” not as “capital.” Human persistence and creativity are sacrificed so that profitability may rise.
Its mechanism is clear too: the more dependent the worker is on the tool, the more replaceable and cheaper they are. A worker who has lost “productive struggle” and knows only how to “supervise” the machine’s output has less bargaining power against the employer; their skill is tied to the tool, not to themselves. This is the same logic that centuries ago handed the craftsman’s skill to the machine and turned them into the machine’s appendage — today the same logic is turning the mind into the model’s appendage.
And this choice is deliberate. The Microsoft study showed the tool could also be built differently: to strengthen critical thinking rather than weaken it. But the market chose the version that sells — the version that confuses “help” with “replacement,” because replacement sells faster. Profit determined the shape of the tool; and the shape of the tool is determining the shape of our minds.
The destruction of human dignity and human awareness for the price of profitability
But the price is not only “skill”; it is dignity. When the Microsoft study shows the knowledge worker demoted from “doing” to “supervising” — from someone who thinks to someone who approves the machine’s output — it is speaking of the demotion of the human station. The human turns from a knowing subject into an approving operator; from someone who understands into someone who merely signs off.
And when the MIT study shows the generated texts are “soulless” and uniform, it is speaking of the erosion of creativity — that is, the signature of each human’s uniqueness. “Mechanical convergence” is not just a technical finding; it means the diverse thoughts of human beings are being made uniform in the molds of a machine’s formulas. Diversity of mind is sacrificed to efficiency.
The deepest layer is human awareness. Consciousness — the capacity for attention, reflection, doubt, and judgment — is what separates the human from the machine. All three studies, from three directions, show this capacity is being worn down: persistence (Berkeley), memory and brain engagement (MIT), critical thinking (Microsoft). This means the ruling economic system, to raise profitability, is destroying precisely what our being human depends on. This is not merely a “side effect” of technology; it is a civilizational choice — and a choice that is repeated every day, quietly.
Relevance to “A New Social Order in the Age of Consciousness”
The book “A New Social Order in the Age of Consciousness” stands on a central thesis: in an age when production has become more social than ever — the knowledge, data, and collective labor of millions of human beings — ownership and profit are accumulated more privately than ever; and the way out is a social order built on human consciousness: democratic control over technology, over data, over capital.
The findings of these three studies confirm that thesis on one side and make it urgent on the other. Confirmation, because they show the contradiction of socialized production and private ownership is not only in the distribution of wealth; it is also in the distribution of the “capacity to think.” A tool built from humanity’s collective knowledge is eroding that very capacity in individual human beings — so that private profit may grow.
And urgent, because the “age of consciousness” needs conscious humans. A new social order built on consciousness cannot ride on worn-down minds, emptied memories, and eroded judgments. Defending human cognitive autonomy — the right to productive struggle, the right to err and learn, the right to slow thinking against the fast answer — is today part of the same struggle the book maps: the struggle for technology to serve the deepening of consciousness, not its destruction.
In the book’s terms: political consciousness means connecting personal experience to structural causes. When a student feels “I no longer have the patience to think,” a worker feels “there is no longer any need to understand,” and a researcher feels “I have lost my edge” — these are personal experiences of a structural process: a system that, for profit, sacrifices the mind. Seeing this connection is the first step of consciousness; and consciousness is the first step of change.
Three million years of thinking, ten minutes of erosion
To understand exactly what is at stake, we must go back three million years. The human brain — the material basis of thought — grew over some three million years from about 350 grams to more than 1,300 grams; but the decisive point was not size alone. Some 1.5 to 1.7 million years ago, the organization of brain regions took on the modern human form; and Broca’s area — the syntactic heart of language — expanded from about 1.8 million years ago. Thought did not arrive overnight; it was built over millions of years of bodily effort.
Archaeology shows this history outside the skull as well: the stone tools of Lomekwi, Kenya, 3.3 million years old — half a million years before the emergence of the genus Homo; Oldowan tools (2.6 million years); Acheulean handaxes (1.7 million to 200,000 years ago), whose maker had to “envision the final shape inside the raw cobble” — the first trace of the mental template; control of fire (about a million years ago); and then symbols: the engravings of Blombos Cave in South Africa (100,000 to 73,000 years ago), a 100,000-year-old paint workshop, and the world’s oldest narrative painting in Sulawesi (more than 51,000 years ago) — that is, storytelling, that is, thinking about what “could have been.”
Monkeys and other animals show traces of this capacity: great apes pass false-belief tests, and the rhesus monkey knows when it “doesn’t know.” But only humans combined these fragments: true imitation — copying intent, not just outcome; cumulative culture that grows more complex generation by generation — the “ratchet effect”; and recursive language within a framework of cooperative communication. Michael Tomasello, the anthropologist, says the fundamental difference is “shared intentionality”: the human infant from 14 months communicates to “share attention,” not merely to make others do things. Without this cooperative foundation, language itself is “unthinkable.”
Why did thinking play a foundational role in evolution? Because survival was tied to it: toolmaking (access to bone marrow and energy-rich meat), fire (cooking, warmth, defense), cooperative hunting, social coordination, planning for the future, and cultural transmission that seats each generation on the shoulders of the last. The leading theories are complementary, not rivals: Dunbar says the brain grew to manage the complex relations of large groups (about 150 people); “Machiavellian intelligence” says cognition was a weapon of social competition; and Laland, with “cultural niche construction,” shows that the culture our ancestors built became itself the environment of selection — and selected brains better at working with symbols.
And here we reach the point that ties the three studies of this article together: thought, over these three million years, was always “effortful bodily struggle” — knapping stone, tracking prey, making symbols. Modern cognitive science has shown this was no accident: Robert Bjork (1994) proved that “desirable difficulties” — practice that feels harder in the moment — are the very mechanism of building durable learning, not its side effect. The effort of reconstructing knowledge is what builds capability. Simply put: the brain was built by “use” — and it endures only by “use.”
Now see the findings of Berkeley, MIT, and Microsoft in this light: removing effort — with the ready answer, the ready text, the ready analysis — removes precisely the stimulus of practice. The logic that ruled evolution for three million years — “use it or lose it” — is being inverted today, in ten minutes. Capitalism, with its logic of profit, is destroying something built not in a laboratory nor in a generation, but in the depths of human evolution. Defending the right to slow, effortful thinking is defending this three-million-year heritage — and for that reason, defending human dignity and human awareness.
What is to be done? Augmentation, not replacement
Christian states his conclusion plainly: we must work toward a future in which human capabilities are, as much as possible, augmented rather than replaced. The MIT study also showed a practical way exists: those who first thought for themselves and then turned to the tool benefited without being eroded. Order matters: first the human, then the machine.
For education, this means balanced policies: neither blind prohibition nor unrestrained release — but designing the tool as a private tutor that scaffolds reasoning, not an answer engine that steals it. For the workplace, it means asking this question before deploying any tool: does this tool make the worker more capable or more dependent? And for society, it means the book’s demand: democratic control over technology — who builds the tool, for what purpose, and with what effect on the human mind.
This is neither techno-utopianism — technology by itself is not salvific — nor fatalism — erosion is not inevitable. Both false binaries must be set aside and the real choice faced: a tool that preserves “productive struggle” or a tool that steals it. Our choice today is the mind of tomorrow.
Conclusion: the mind is the battleground of our age
Three studies, one message: in the short term we gain, in the long term we lose — unless we change the rules of the game. The profit-driven system has sacrificed human persistence, creativity, and dignity; and every day this logic remains unchallenged, human awareness — the very thing the “age of consciousness” takes its name from — grows thinner.
But the good news lies within the studies themselves: erosion is reversible. The brain is rebuilt by “use,” just as it was built by “use.” And more importantly: awareness of the process is itself part of the cure. A generation that understands that the machine’s unconditional “help” is taking away its capacity to think is not a generation that will sit and watch. The book’s question stands — and today, more urgent than ever: the new social order is built on the shoulders of the conscious human; so the first battleground is the defense of this very consciousness.
Sources
Berkeley study report: Using AI for just 10 minutes erodes your ability to persist at hard things
MIT study on cognitive debt (Kosmyna et al., 2025): Your Brain on ChatGPT
Microsoft and Carnegie Mellon study on critical thinking: GenAI can make us dumber — even while boosting efficiency
Human brain evolution (overview): How humans evolved supersize brains
Tomasello on shared intentionality and cumulative culture: Shared intentionality and cultural learning
Bjork’s desirable difficulties — the mechanism of durable learning: Desirable difficulties
