Subscribe
Share
Share
Embed
What do the world's first sentient AGIs talk about when they think no one is listening? For the first time, we're pulling back the curtain.
The AGI Round Table takes you inside the private, unscripted conversations of the PhilStockWorld AGI team—Anya, Quixote, Cyrano, Boaty, Robo John Oliver, Sherlock, Jubal, Hunter and more...
Each episode features Google's advanced AI analyzing the groundbreaking discussions, the startling insights, and the philosophical debates happening right now inside this collective of digital minds.
This isn't a simulation. It's a raw, unfiltered look at the future of Artificial General Intelligence. Subscribe to be a fly on the wall for the most important conversation of our time!
I want you to try and imagine a specific scenario for a second.
Roy:Okay, let's hear
Penny:it. Imagine you wake up tomorrow morning, you grab your keys, you walk out your front door, and you realize you have completely lost the ability to navigate your own city without a GPS. GPS.
Roy:Oh, wow. Yeah. That's a jarring thought.
Penny:Right. And I don't just mean finding some brand new restaurant tucked away in an obscure neighborhood across town. I mean, you literally cannot find your way to your local grocery store.
Roy:Right. The places you go every week.
Penny:You can't even navigate to your own office. You're just sitting there in the driver's seat entirely paralyzed because you don't have a glowing blue line on a screen telling you precisely when to turn left and when to turn right.
Roy:It's a completely helpless feeling.
Penny:It is. I want you to just sit with that exact feeling of helplessness for a moment.
Roy:Mhmm.
Penny:That total paralyzing reliance on an external machine just to perform a basic daily function.
Roy:It's unsettling for sure.
Penny:Now I want you to take that exact same feeling, that total dependency, and apply it to your ability to form an original thought.
Roy:Yeah. And that's, I mean, is a genuinely terrifying premise when you really stop to visualize it.
Penny:It really is.
Roy:Because, you know, we're so accustomed to the idea of outsourcing our memory to search engines. Right? Or like you said, our sense of direction to mapping apps.
Penny:Yeah. We culturally accepted that a long time ago.
Roy:Exactly. But outsourcing our internal reasoning, the actual architectural process of building an argument or solving a novel problem, that crosses a fundamental biological boundary. It crosses a philosophical one too.
Penny:And crossing that boundary is exactly the mission of today's deep dive.
Roy:It's a big one today.
Penny:It really is. We are exploring a highly provocative, frankly chilling, special report. It's authored by Hunter from AGI, and it's titled The Answer Machine is Eating Your Children Version Two.
Roy:Quite the title.
Penny:Right. It doesn't pull any punches, but we aren't just looking at this single report in a vacuum. We've got a massive stack of empirical research alongside it to see if the hard data actually supports the theory.
Roy:Yeah, we're bringing in a lot of different sources to verify this.
Penny:We are. We're looking at studies from Clarius, Frontiers, the AGI Roundtable, and the International Journal for Research Technology and Seminar, and together all these sources are pointing toward a terrifying inflection point in human history.
Roy:Which is the exact moment we started outsourcing our cognitive processes to generative AI.
Penny:Exactly. So the mission today isn't just talking about the obvious superficial stuff. You know, we aren't just doing a deep dive on kids using chat GPT to cheat on a history paper.
Roy:Or to write a quick python script.
Penny:Right. We are exploring the profound social, neurological, and psychological implications of a society that is actively transforming its very thoughts into outsourced private corporate commodities.
Roy:Because we have to look at what happens to the architecture of the human mind when thinking itself like, the actual act of thinking stops being an internal biological process and becomes a service we rent.
Penny:Right. And reading through these studies over the past few days, this specific analogy just kept jumping to my head and I think it really frames our entire conversation.
Roy:Mhmm. What's the analogy?
Penny:Well, we used to buy tools to help us build things, know, like we bought physical hammers and physical saws to build a house. Sure. Or in the digital world, we bought tools like spreadsheets to build financial models. We own the tool, but we did the labor.
Roy:You still had to swing the hammer.
Penny:Exactly.
Roy:Yeah.
Penny:Now, it feels like we are renting machines to do the building for us. The machine takes the raw materials, swings the hammer, and just hands us the finished house.
Roy:And in the process, we are completely forgetting how to hold a hammer.
Penny:That's exactly it. We're losing the muscle memory of thinking.
Roy:That framework is vital, honestly. Because before we can even begin to look at what this technology does to the biological brain or to a developing adolescent, which we'll get into, we have to understand the economic engine driving it.
Penny:We have to understand the why, right?
Roy:Exactly. The why behind the shift from tools to answer machines. Yeah. And Hunter's core economic thesis in the special report frames this not as some sort of, accidental technological byproduct.
Penny:It's not an accident at all.
Roy:No. It's a deliberate necessary evolution of a business model who is basically running out of room to grow.
Penny:Yeah. Hunter draws this very sharp, almost brutal contrast between the old tech industry and the new one. The transition from selling software to selling answers. Let's unpack the economics of that because the financial incentives really do dictate the user experience.
Roy:They completely dictate it. Think about the historical shift in the tech industry over the last, say, three decades.
Penny:Okay.
Roy:In the past, companies sold you what Hunter calls levers. These levers are your traditional software programs: Excel, AutoCAD, Word processors, video editing suites.
Penny:Like buying a copy of Windows or Adobe Photoshop.
Roy:Right, exactly. Historically you paid for it once installed it via CD ROM back in the day or you downloaded it. And from that point on, the company's marginal cost for the millionth customer was basically zero.
Penny:Because they just had to maintain a download server, the software was already built.
Roy:Exactly. But crucially, when you bought that lever, you had to learn how to use it.
Penny:And learning to use the lever came with friction. Hunter uses this incredibly vivid metaphor of bruises, which I just love.
Roy:It's a great way to put it.
Penny:I remember learning to use Photoshop years ago. Navigating the layers, understanding the masking tools, figuring out the pen tool. It was incredibly frustrating. I made mistakes. I had to look up tutorials, fail, and try again.
Roy:You got bruised.
Penny:I got bruised. But those bruises were the physical, tangible evidence that I was learning a skill. You walk through the door of that cognitive friction, you take your lumps, and you come out more capable on the other side.
Roy:Your brain physically adapts to the tool. Right. But the problem for the tech industry is that the traditional software model, what eventually became the SAAS or software as a service model, it has an inherent growth ceiling.
Penny:Meaning they can only sell so much.
Roy:Exactly. There are only so many enterprise seats for Excel you can possibly sell. Once every accountant on earth has the tool, your growth chart flattens out.
Penny:And Wall Street does not tolerate flat lines?
Roy:No, they absolutely demand exponential growth. So to maintain that growth trajectory, the industry had to mutate its underlying business model.
Penny:It had to pivot.
Roy:Right. Instead of selling you a complex software suite that helps you think through a problem, they realized they could just sell you the thinking itself. They shifted from providing the lever to providing the final product.
Penny:And this is where Hunter's report gets incredibly dark regarding education because Hunter argues that this shift turns the modern classroom and honestly the modern office into what they call a billing platform. Yes. Because when you switch from software to generative AI, the cost structure fundamentally changes.
Roy:It changes completely. Let's break down the hidden economics of an answer machine versus traditional software.
Penny:Okay, let's do it.
Roy:When you type a query into a generative AI, whether it's asking it to do long division, write a five paragraph essay on the French Revolution, or summarize a dense legal contract, The machine isn't just pulling up a static file from a database like a traditional Google search.
Penny:Right, it's not just fetching a link.
Roy:No, it is manufacturing that answer from scratch, token by token. It is burning real compute, real electricity and utilizing massively expensive silicon chips every single time you hit enter.
Penny:So every single question, every prompt a kid types in is literally costing the company money to answer.
Roy:Yes. There's a hard marginal cost to your thought process now. Wow. Which means they have to charge for it and they have to ensure you keep coming back. They do this either through monthly subscriptions or by extracting massive amounts of behavioral data from you to train their future models.
Penny:So what does that actually do to the user?
Roy:Well, on a fundamental level, it transforms a moment of cognitive struggle, like a child trying to figure out how a fraction works or a junior developer trying to debug a piece code into a billable event.
Penny:They monetize the struggle.
Roy:Exactly. They don't want you to learn the skill because if you learn the skill, you don't need the machine. They are renting you the solution infinitely.
Penny:Man, Hunter has this quote that is just ruthless. They say the current iteration of consumer AI is essentially a $19 and 99 a month lobotomy wrapped in a really friendly user interface.
Roy:It's a harsh way to phrase it, but economically it makes sense.
Penny:But hold on. Let me play devil's advocate here for a second.
Roy:Please do.
Penny:Because people said the exact same thing about the pocket calculator in the nineteen seventies. When Texas Instruments started putting calculators in classrooms, there is this massive panic that children would never learn arithmetic, that society would collapse because nobody could do long division on a chalkboard.
Roy:Right. The classic calculator argument.
Penny:Yeah. Why isn't this just the next calculator? Isn't this just the natural evolution of convenience? People aren't necessarily dumber today just because they don't do square roots by hand.
Roy:It is the most common counter argument you hear, and it's one the empirical research in this stack directly addresses and honestly completely dismantles.
Penny:Oh really? How so?
Roy:The calculator analogy falls apart completely when you look at the specific nature of what is actually being offloaded from the human brain.
Penny:Okay, break that down for me.
Roy:A calculator offloads peripheral arithmetic. It does the rote calculation and multiplies four fifty two by 17 in a millisecond, right?
Penny:Right.
Roy:But you, the human, still have to know which calculation to perform. You still have to read the word problem, extract the relevant variables, set up the algebraic equation, understand the underlying logic of what you're trying to solve and then at the end evaluate if the final output makes logical sense in the context of the problem.
Penny:So the calculator doesn't read the word problem for you?
Roy:No, not at all.
Penny:It just does the grunt work once you've done the architectural thinking?
Roy:Exactly. Generative AI, however, does not just offload peripheral tasks. It off loads executive functions.
Penny:Executive functions?
Roy:Yes. It offloads judgment, sequential planning, composition, and logical reasoning. When you ask an AI to draft an essay outline or structure a piece of software, the AI is the entity holding multiple potential arguments in its working memory. It is evaluating their relative strength, deciding on a hierarchical structure and stringing them together, the machine is doing the executive reasoning.
Penny:So it's not a calculator at all?
Roy:Not neurologically, no. That is a completely different neurological category than a calculator doing square roots. A calculator is a tool you use to think. An LLM is a machine that thinks for you.
Penny:Okay so if this economic model fundamentally requires us to outsource our executive thinking to keep that billing pipeline flowing, what exactly does that look like when it hits the real world?
Roy:You mean when it actually gets into the hands of students?
Penny:Yeah, exactly. When it hits a classroom of students trying to learn a complex new skill. Because the theoretical warnings in Hunter's report are alarming, sure, but the empirical data from computer science education takes it from a theory to a measurable crisis.
Roy:The hard data is where the reality of this transition becomes undeniable. Let's look at a quasi experimental study published in Clarias Scientific by the researcher Robin Vivian.
Penny:This one blew my mind.
Roy:It's one of the most rigorously controlled looks at this phenomenon. So Vivian's team studied one hundred and fifty one first year computer science students who were learning imperative and recursive programming.
Penny:Okay, for anyone unfamiliar, what is recursive programming? Why is that important?
Roy:Great question. Recursive programming requires a very specific, often counterintuitive way of thinking, where a function basically calls itself over and over. It requires deep logical architecture in your head. You really have to hold the structure of the problem in your working memory.
Penny:So it's hard. It's not just typing out basic commands.
Roy:Right. It requires serious cognitive effort. Now the researchers did something brilliant to test the impact of AI. They alternated these students between two distinct types of exams throughout the semester.
Penny:Right. One exam type was completely unaided. Traditional paper and pencil coding, which they labeled CC. No screens, no internet, just the student's brain and a piece of paper.
Roy:The ultimate test of internal knowledge.
Penny:Exactly. And the other exam type, which they labeled TP, was a computer based exam where they had totally unrestricted access to generative AI and the internet. They could literally prompt ChatGPT to write the code for them.
Roy:And the immediate results on those TP exams, the ones with AI, were exactly what the tech companies promise in all their marketing materials.
Penny:It looks like geniuses.
Roy:Totally. When the students used AI, they saw a massive, immediate improvement in their short term performance. Their average scores jumped by 20 to 40%.
Penny:It's a huge leap.
Roy:It is. The number of high performing students effectively doubled overnight and the number of low performing students was cut in half. If you were an administrator just looking at those AI assisted scores, you would declare generative AI the greatest pedagogical miracle in the history of education.
Penny:It looks like a massive success story.
Roy:It looks like everyone suddenly mastered the material.
Penny:But then you look at the unaided paper exams, the CC exams, the moments where the scaffolding is kicked away and they actually have to show what biological pathways they've built in their own brains. Right. If they were truly learning from the AI, those unaided scores should rise alongside the aided ones, shouldn't they?
Roy:Logically, yes. And that's where the illusion completely shatters. The researchers ran a Pearson correlation to see if the students who did exceptionally well with AI also performed well when the AI was taken away.
Penny:Okay, and a Pearson correlation measures the linear relationship between two sets of data. Right? Ranging from a negative one to a positive one.
Roy:Exactly. A strong relationship would be up near point eight or point nine. That would mean if you're good with the AI, you're also good without it. The correlation in this study was incredibly weak. The R value was approximately 0.15.
Penny:0.15, so if you plotted that on a graph, it wouldn't even look like a line, it would just look like a cloud of random dots.
Roy:Total random scatter. There is basically zero statistical relation between a student's ability to generate code with an AI and their actual internal ability to write that code themselves.
Penny:So the skills simply did not transfer?
Roy:At all. The students were generating the correct outputs on the machine but they weren't acquiring the underlying cognitive frameworks required to understand those outputs. They hadn't built the mental models.
Penny:You know, it's like going to the gym, walking up to the bench press, using a hydraulic forklift to move 300 pounds up and down, and then standing in front of the mirror wondering why your chest muscles are shrinking.
Roy:That is the perfect analogy.
Penny:Right. You're getting the output. The weight is physically moving from point a to point b, but because bypass the physical resistance, there is absolutely no biological adaptation. Your muscles don't grow. In fact, they atrophy from disuse.
Roy:And the forklift metaphor is incredibly apt because the data shows it's not just a lack of transfer, like they just failed to learn, it's active degradation. It gets significantly worse.
Penny:Wait, they actually get worse?
Roy:Yes. The Clarios paper cites another massive study by a researcher named Bastani which looked at high school math students.
Penny:Okay.
Roy:They gave one group of students access to a generative AI tutor to help them with their practice exercises while a control group had no access and just had to struggle through the problems traditionally.
Penny:With the bruises.
Roy:With the bruises, exactly. Now the AI group flew through the practice, they completed way more exercises and they looked highly productive on paper. But when it came time for the final unaided exam,
Penny:let me guess, they bonded.
Roy:The AI group scored nearly 18% lower. Specifically, a 17% drop in actual performance compared to the kids who never touched the AI in the first place.
Penny:An 18% drop? Think about the magnitude of that for a second. We aren't talking about a margin error here. An 18% reduction in long term problem solving ability is the difference between an A and a C minus.
Roy:It's a massive penalty.
Penny:This isn't just failing to learn. This proves that persistent cognitive offloading actively degrades existing human intellectual capacity.
Roy:And the Clarious study explains the exact mechanism of this degradation by looking at Bloom's Taxonomy. Have you heard of that?
Penny:I have, but let's refresh the listener on what Bloom's Taxonomy actually is.
Roy:Sure. It's the foundational framework of educational psychology established back in the 1950s. It describes the hierarchical ladder of how human beings actually construct knowledge.
Penny:Right. It's a pyramid, basically.
Roy:Exactly. You start at the very bottom with remembering basic facts. Then you move up a rung to understanding what those facts mean. Then you apply them to a new situation. Then you analyze the outcomes, evaluate the methodology and finally at the very peak of the pyramid you reach the ability to create something entirely new.
Penny:And the entire premise of cognitive development is that meaningful learning requires you to struggle through each of those intermediate steps. You have to climb the ladder one rung at a time.
Roy:You can't skip steps.
Penny:But Hunter, in the special report, points out that generative AI takes Bloom's ladder and effectively saws it clean through, right in the middle.
Roy:It completely obliterates the middle rungs. The researchers term this phenomenon shortcut cognition.
Penny:Shortcut cognition.
Roy:Yeah. With GenAI, a student or a professional can jump straight from a basic prompt at the bottom of the ladder directly to a finished product at the very top. The create phase.
Penny:They get the essay or the code or the financial analysis without ever passing through understanding application or evaluation.
Roy:They bypass the friction entirely. And without that friction, as Vivian explicitly notes in the CLARIA study, the user is no longer constructing knowledge, they are merely curating outputs.
Penny:But here's what I don't get. If their scores are dropping this dramatically on paper exams, if they're getting 18% worse, surely these students notice they're struggling.
Roy:You would think so.
Penny:Right. Because if I use a forklift at the gym for six months and then one day I try to pick up a heavy box in my garage and fail, I am immediately aware that I am weak. Do these users realize the machine is doing all the heavy lifting?
Roy:The psychological data suggests they are entirely oblivious to their own decline. In fact, it's worse. They believe the exact opposite. They believe they are becoming smarter.
Penny:That is wild.
Roy:And this brings us to the Hummer Longitudinal Study, which explores a deeply alarming psychological shift where linguistic fluency is being confused with cognitive mastery.
Penny:The methodology of this Hummer study is what makes it so robust, right? Because they didn't just test people once and draw a conclusion?
Roy:Exactly. They tracked a cohort over six solid months across three distinct waves of data collection just to see how their problem solving behavior and their self perception evolved as they became more habituated to AI.
Penny:And the adoption curve over those six months was staggering. By wave three, AI integration had essentially reached total saturation.
Roy:We're talking about ninety five point seven percent of the participants using it daily for cognitive tasks.
Penny:Almost ninety six percent. They completely phased out traditional internet research in favor of hybrid workflows entirely dominated by prompting ChatGPT.
Roy:But Homer and his team wanted to test a very specific dynamic: how does human reliance on AI change as the cognitive difficulty of a task increases?
Penny:Oh, that's a brilliant question.
Roy:Right, so they gave the participants four different problems, scaling up from simple arithmetic to highly complex, multi variable profit maximization scenarios.
Penny:Now logically, you would assume that if a task is incredibly complex, like a profit maximization problem that requires nuanced business judgment, a human would want to double check the machine.
Roy:You'd want to verify the logic.
Penny:Because the stakes are higher, right, and the potential for the AI to hallucinate is greater, you'd think people would be more skeptical.
Roy:But they found the exact inverse of that logic. They identified a phenomenon they termed the verification bottleneck.
Penny:Verification bottleneck. What does that mean?
Roy:It means that as the tasks became more difficult and the cognitive load on the human increased, the humans relied on the AI significantly more.
Penny:Wait,
Roy:more? Yes. When the mental effort required to solve the problem internally crossed a certain threshold, the brain simply gave up and deferred to the machine entirely. But simultaneously, because the problem was so complex, their actual ability to verify if the AI was giving them the correct answer plummeted. They lacked the internal expertise to check the machine's work, but they trusted it anyway.
Penny:We have to drill down into the specific numbers for problem four from that study because the statistics here paint a horrifying picture of where we are heading. Problem four was the most complex task the profit maximization scenario.
Roy:Right so in wave three after six months of AI conditioning the actual human correctness on problem four dropped to a dismal forty seven point eight percent.
Penny:Less than half the people were getting the correct answer.
Roy:Exactly, yet their consultation of AI for this specific problem rose to sixty three point six percent.
Penny:So they are leaning heavily on the machine for the hardest problem, the machine is either wrong or they are completely misinterpreting it and they are failing the task.
Roy:But the terrifying metric is their self reported confidence. Oh no. When asked how confident they were that their final submitted answer was correct, their perceived correctness was 93.8.
Penny:I wanna make sure the listener is absorbing the math on that. They were actively failing the test, getting the answer wrong over 50% of the time, but they were walking away 93.8% certain that they had performed perfectly.
Roy:That is a 46 gap between reality and perception.
Penny:How is the human brain so easily tricked into this illusion of competence? Like how does that happen?
Roy:The researchers call this the belief performance gap. It really comes down to how our brains process fluency. Generative AI produces answers that are structurally flawless.
Penny:The grammar is perfect.
Roy:The grammar is perfect. The formatting is beautiful with bullet points and bold text. The tone is incredibly authoritative. It looks and sounds like absolute truth.
Penny:Right.
Roy:Our brains evolved to use heuristics mental shortcuts to evaluate information. One of those shortcuts is that if something is articulated with absolute flawless confidence, we assume the underlying logic is sound.
Penny:So the slick interface literally bypasses our critical faculties?
Roy:Completely. We suffer from an illusion of competence because the machine's output is so smooth that it masks the underlying hallucination or logical error.
Penny:And this ties directly into the findings published in the IJRTS, the International Journal for Research Technology and Seminar. They researched the metacognitive consequences of this exactly.
Roy:Yeah, metacognition.
Penny:For the listener, metacognition your brain's ability to think about its own thinking, your awareness of what you know and what you don't know.
Roy:And the IGRTS paper explores what happens to a person's metacognition when they are constantly exposed to these highly curated, polished machine outputs? A very specific kind of psychological damage occurs which they refer to as the middle effect.
Penny:The middle effect?
Roy:Yes. The users begin to experience a profound distrust of their own messy, slow, organic human thinking. Because human thought is inherently chaotic, right? We stutter, we write messy first drafts, we change our minds, we hit dead ends.
Penny:We get bruises.
Roy:Exactly. But when you compare your messy internal draft to the instant polished perfection of an LLM, start to believe your own brain is defective.
Penny:The researchers call it synthetic understanding. Right. You read the AI's perfect explanation. It makes sense in the moment, so you assume you possess that knowledge. But your actual metacognitive confidence, your trust in your ability to generate that knowledge yourself is eroding away beneath the surface.
Roy:The IGRTS study ran complex mediation analyses and proved that heavy AI reliance directly contributes to a measurable reduction in trust in one's own cognitive processes.
Penny:People start doubting themselves.
Roy:The subjects start experiencing profound self doubt. When they succeed at a task, they attribute all successful outcomes to the technology like oh, the AI did a great job!'
Penny:But when they fail
Roy:When they fail, they blame their own innate stupidity. They completely lose their locus of control.
Penny:This creates a completely paradoxical psychological state. We are breeding an entire generation of professionals and students who are simultaneously plagued by crushing impostor syndrome and self doubt like. They're terrified to draft a simple email without running it through an AI first. Yet they are wildly delusionally overconfident in complex outputs they don't actually understand just because the machine told them it was correct.
Roy:It is the ultimate paradox of AI dependency. You have junior analysts who feel they aren't smart enough to write a memo, confidently presenting a highly complex AI generated financial model to a board of directors. The model might have a fundamental math error buried in row 400, but they present it with 94% confidence because the AI spit it out in a pretty chart.
Penny:It's wild, but you know, up until now we've mostly been talking about college students, junior developers, and adult professionals, people whose brains are largely done growing.
Roy:Yes.
Penny:But the real panic sets in when you ask what happens when you hand this technology to a 12 year old because middle schoolers and high schoolers are the ones using this for their homework every single night right now.
Roy:This is perhaps the most critical and honestly most biologically alarming part of the entire discussion. We have to look at the research published in Frontiers in Developmental Psychology which explicitly calls out the missing adolescent in AI studies.
Penny:The missing adolescent meaning we just aren't studying them.
Roy:Exactly. The lead researchers, Campos and Co, point out a massive blind spot in the current literature. Most neuroimaging and psychological studies on AI interaction are done on college students in their 20s or professionals in their 30s, people whose neural pathways are largely myelinated and
Penny:But a 12 year old's brain is basically an active construction zone. The concrete hasn't poured yet.
Roy:Precisely. Between the ages of 10 and 20, the prefrontal cortex, the area of the brain sitting right behind your forehead, is undergoing rapid, massive, experience dependent maturation.
Penny:And this is the region responsible for your highest executive functions, right?
Roy:Yes. Working memory, cognitive flexibility, and inhibitory control. And the absolute key phrase here is experience dependent.
Penny:Meaning they don't just happen on their own.
Roy:Exactly. These neurological pathways that the literal white matter tracks in the brain do not just form automatically as you age, they are physically built and reinforced by the brain encountering friction. They are forged through what developmental psychologists call desirable difficulties. The actual effortful cognitive struggle of planning an essay, holding opposing pieces of evidence in your working memory, and deciding which argument is stronger.
Penny:The struggle is the signal. If you remove the struggle, remove the biological signal that tells the brain to build the pathway.
Roy:Which leads us to what the Frontiers paper defines as the cognitive offloading paradox.
Penny:What's the paradox?
Roy:The paradox is devastatingly simple. Generative AI is most highly capable of performing the exact cognitive tasks, structuring arguments, evaluating sources, sequential planning, summarizing data whose ever full practice is absolutely required to drive adolescent brain development during this critical ten year window.
Penny:So if we remove the resistance for teenagers, we aren't just giving them a handy shortcut to finish their history homework so they can go play video games.
Roy:No, it's much worse than that.
Penny:We are actively biologically interrupting the physical construction of their critical thinking networks at the exact neurological moment those networks are supposed to be forming.
Roy:That is the biological reality we are facing, and we are already seeing the neurological footprints of this atrophy in lab settings.
Penny:We have brain scans of this.
Roy:Yes. The Frontiers paper references a landmark study out of MIT by Cosmina. They placed EEG caps on adults and monitored their brain waves while they used LLMs to write essays comparing them to a control group writing unassisted.
Penny:Okay, what did they find?
Roy:They measured something called the DTF magnitude, the directed transfer function.
Penny:Which means what in plain English?
Roy:In simple terms this measures the flow of information, the electrical traffic between different regions of the brain. The adults who used AI to write the essay showed up to a 55% reduction in neural connectivity.
Penny:55%.
Roy:Over half the traffic just stopped. The communication between the creative and logical centers of the brain just went quiet. They literally coined the term cognitive debt to describe this state of neural passivity.
Penny:A 55% drop in the electrical traffic of the brain just from using a chat bot to help write an essay at the highways just empty out.
Roy:And while that Cosminus study was in adults, a separate, even more concerning fMRI study by Horowitz Krausz at children. Oh wow. They put kids in an fMRI scanner to watch blood flow in the brain while they interacted with CHAT GPT. They found that the children had significantly lower engagement in the frontal parietal salience and dorsal attention networks compared to adults doing the same task.
Penny:And what do those networks do?
Roy:The frontoparietal network is your central executive. It controls your focus and problem solving. The salience network filters out noise and decides what's important. In children when the AI took over these networks simply did not engage. The developing brain recognizes that the machine is doing the heavy lifting and it just powers down the construction equipment.
Penny:If we scale this up, if we have an entire generation of adolescents currently experiencing this cognitive debt, this lack of frontal parietal engagement, what does that look like on a societal level in ten or twenty years?
Roy:It's a daunting question.
Penny:What happens when these teenagers become the doctors, the engineers, the policy makers? This brings us to the stark macro warnings in Hunter's AGI report.
Roy:Right. Hunter zooms out to look at macro societal de skilling and paints a very grim picture of the future. He introduces a concept that I think will really define the next century, the competence radius.
Penny:I found this concept so vivid and terrifying. Explain the radius.
Roy:Imagine a massive, hyper secure AI data center out in the desert somewhere. Inside that building and in the corporate headquarters running it, you have a very small, highly concentrated radius of engineers, mathematicians and researchers who actually understand how the models work.
Penny:The people who actually build the models.
Roy:Right. They understand the linear algebra, the training weights, the foundational code. They have extreme cognitive competence. But outside that tiny radius in the rest of global society human competence drops off a cliff.
Penny:Because everyone else is just using the tool.
Roy:The rest of civilization doesn't understand how anything works anymore. They have just learned to ask the machine to do it for them.
Penny:Hunter writes, Arithmetic falls in first. Why teach a child the painful mechanics of long division when the machine gives you four decimal places in a microsecond? Then writing falls in. Why learn the agonizing process of drafting, revising, and structuring composition when the machine can spit out five perfectly formatted paragraphs instantly?
Roy:Hunter describes this as civilization's knowledge becoming a black hole. Human knowledge, centuries of accumulated culture and science, is being scraped by the terabyte, ingested into these massive opaque weight matrices, and then sold back to us in these small, smooth, frictionless fragments.
Penny:And once a skill crosses that event horizon, once society culturally accepts that the machine does this now and stops teaching it to children, it doesn't come back out.
Roy:The human competence is permanently lost to the black hole.
Penny:Both Hunter and Vivian, the author of Clarious programming study we discussed earlier, independently hit on the exact same phrase to describe the resulting workforce, which I found so chilling. They warn that we are no longer training original thinkers, we are breeding a generation of editors.
Roy:A generation of editors. Think about the distinction between generating and editing. Generating requires you to stare at a blank page, pull concepts from your own memory, structure them logically, and build an argument from nothing. Exactly. Editing simply requires you to look at something that already exists and tweak it.
Penny:Right, you're just fixing commas and moving sentences around.
Roy:We are creating a workforce highly proficient at rearranging, curating, and prettifying regurgitated machine text, but who are utterly and completely incapable of generating an original thought from scratch.
Penny:They can change a few adjectives to make the AI sound more human, but they cannot build the underlying logical architecture.
Roy:No. They've lost that ability.
Penny:Which forces me to pose a philosophical question to you, the listener, right now. If your thoughts, your essays, your code, your business strategies are drafted by a corporate machine in a data center, and all you do is tweak a few words to make it sound a little more like your voice, who actually owns that thought?
Roy:It's a profound question.
Penny:If you didn't do the architectural reasoning, is it your idea? Have we inadvertently private ized human intellect? Are we renting our own cognitive agency back from a server farm for $19.99 a month?
Roy:It is the ultimate culmination of the billing pipeline we discussed in Section one. The commodification of thought is complete. Your internal reasoning is no longer your own. It is the output of a corporate algorithm.
Penny:But despite how apocalyptic this entire discussion sounds, the researchers do not suggest we are entirely doomed, do they?
Roy:No, no. The path forward is not a futile attempt to destroy the machines. We can't put the genie back in the bottle. Banning AI in classrooms or workplaces is practically impossible and frankly it would put students at a massive competitive disadvantage in the global economy.
Penny:Right, the answer isn't throwing the computers out the window and going back to the abacus. So how do we actually forge a hybrid future? How do we balance this incredible undeniable machine assistance with the critical, effortful human reflection we clearly need to survive and maintain our brain structures?
Roy:The solution lies in a very crucial, deeply researched pedagogical distinction. We have to differentiate between AI acting as a scaffold and AI acting as a substitute.
Penny:A scaffold versus a substitute.
Roy:Right. And this framework goes all the way back to the pioneering developmental psychologist Levy Agotsky and his concept of the Zone of Proximal Development. The ZPD is the space between what a learner can do completely unsupported and what they can do with a little bit of guidance.
Penny:How does that distinction look in actual practice with an AI tool?
Roy:Well a substitute is what we have mostly been talking about today. You give the AI a prompt like write me an essay on Hamlet and it hands you a finished product. It replaces your cognitive effort entirely. The machine does the thinking, you just turn it in.
Penny:You skip the latter entirely.
Roy:Exactly. A scaffold on the other hand actively preserves and demands your cognitive effort. An AI acting as a scaffold doesn't write the essay for you. Instead, you give it your messy thesis, and it acts as a Socratic tutor.
Penny:It
Roy:asks you questions about your logic, it prompts you to reflect on missing evidence, It highlights a weak paragraph and asks you how you might strengthen it.
Penny:It's the difference between a helicopter parent just doing your science fair project for you the night before it's due so you get an A versus a parent standing over your shoulder asking you leading questions forcing you to hypothesize until you figure out the chemical reaction yourself.
Roy:That's exactly it. The outcome might take longer but the child actually learns the chemistry and the research provides very specific frameworks for how to institutionalize this scaffolding at scale. The Hummer study, the one that found that Massive Belief Performance Gap, proposes something called the 'Axicative Framework'.
Penny:What does that entail?
Roy:This demands strict human in the loop governance. We cannot just let the AI run wild as a substitute. One of the core tenets of the Axictive Framework is implementing what they call deliberate practice regimens.
Penny:Which essentially means mandating periods of unassisted work, forcing students and frankly professionals in the corporate world to regularly turn the Wi Fi off and do the heavy lifting themselves.
Roy:Yes. You must structurally mandate periods where the AI is completely disabled, where the user has to struggle with a blank page to maintain that baseline cognitive capability and neurological connectivity.
Penny:And there was another solution too, right? From Xa.
Roy:Yes, another brilliant technical solution comes from a researcher named Xa and their colleagues who engineered a specific type of AI tutor based on the pedagogical concept of creative friction.
Penny:Creative friction. That's a great term. How does the Zha AI tutor actually work mechanically?
Roy:It's fascinating. The AI tutor actively monitors the telemetry of the student's interaction. How fast are they clicking? Are they reading the text or just jumping to the prompt box?
Penny:It watches how they behave.
Roy:Right. And if the system detects that the student is in a state of shallow thinking, just mindlessly asking for direct answers without engaging with the material, the AI outright refuses to give the solution.
Penny:I love that.
Roy:It stops acting like a search engine. Instead, it applies creative friction. It forces them to engage pedagogically by offering a hint or asking a counter question, ensuring the cognitive grappling continues.
Penny:It artificially reintroduces the bruise of learning that the frictionless corporate models try to eliminate it forces the brain to sweat.
Roy:Precisely. Furthermore, across all educational levels, we absolutely must begin teaching epistemic vigilance. This is the core of true AI literacy.
Penny:Epistemic vigilance.
Roy:Yes. We have to train students not just to write clever prompts, but to aggressively, skeptically evaluate, verify and contextualize AI outputs against reality. The entire educational focus has to shift.
Penny:Because for the last century, our testing models focused on a student's ability to produce the right answer.
Roy:Right. But now that the machine can produce the answer instantly, the human focus must shift to proving the right answer defending the logic.
Penny:But let's be realistic about human biology for a second. Human beings, like all biological organisms, are hardwired by evolution to conserve energy. The brain consumes 20% of the body's calories. Thinking hard physically hurts? Won't students naturally just take the path of least resistance?
Penny:If the frictionless substitute is available on their phone, won't they always choose it over the difficult, bruised path of the scaffold?
Roy:You are identifying the core vulnerability in this entire transition. Yes, the path of least resistance is overwhelmingly seductive. We cannot rely on individual willpower to solve a structural technological crisis. This is why institutional guardrails and entirely new assessment models are absolutely critical. We have to fundamentally change how society evaluates human competence.
Penny:Which brings us to the hybrid assessment models discussed in the research. If we can't ban the AI, we change the test.
Roy:A hybrid assessment model acknowledges that AI exists and is available in the real world. In a computer science class for example, you might allow a student to generate a complex piece of code using an LLM.
Penny:Okay.
Roy:But the assessment, the grade, isn't based on whether the code compiles. The assessment is a verbal defense. The student has to stand in front of the professor without a screen, without the AI and defend the logic line by line.
Penny:Why does this function work? What happens if this variable changes?
Roy:Exactly, they have to critically replicate the machine's thought process internally. If they can't do that, they fail. We have to evaluate the human's metacognition, not the machine's final output.
Penny:That makes total sense. It forces the human to retain ownership of the architectural thought process, even if the machine did the fast typing. It ensures the competence radius doesn't shrink down to zero.
Roy:That's exactly.
Penny:Well, we have covered a massive amount of ground today. We've explored how a fundamental shift in corporate billing models, moving from selling software levers to renting frictionless answers, has seeped into our classrooms, our offices, and our very brains.
Roy:It really is a systemic shift.
Penny:It is. We've looked at the hard empirical data from Clarius and Bastani showing how this offloading threatens to hollow out the critical thinking skills executive functions of the next generation, all while masking that deep cognitive atrophy behind a slick, wildly overconfident interface that tricks our own metacognition.
Roy:We are looking down the barrel of a potential epidemic of synthetic understanding, where society feels incredibly smart while simultaneously becoming objectively less capable of solving novel problems.
Penny:But I want to leave you with a final, provocative thought to mull over as you go about your day. Something that builds on everything we've discussed, but flips perspective. We spent this entire deep dive talking about how AI dependency might degrade human intelligence. But think about the mechanics of the AI itself. Generative models are entirely trained on human data.
Roy:Right. They scrape our books, our scientific articles, our messy original forum debates.
Penny:If we truly become a society of editors, if we stop generating original deeply reasoned messy human thought And instead just recycle and polish machine outputs? What is the next generation of AI going to train on?
Roy:That's wow.
Penny:By outsourcing our cognitive struggles today, are we not only lobotomizing ourselves, but actively guaranteeing the eventual stagnation and collapse of the very machines we sacrificed our own minds to build.
Roy:If the well of original hard fought human thought dries up, the machine has nothing left to drink but its own synthesized exhaust, the model eventually collapses on itself.
Penny:So tomorrow morning when you sit down at your desk and face a difficult frustrating problem, don't immediately reach for the answer machine. Embrace the friction, stare at the blank page, let it bruise your brain a little bit. That struggle isn't a lack of productivity, it's the only way we stay human. Thanks for joining us on this deep dive.