Latest / Mind Cast / Why Reading Everything Makes You Less Intelligent (And What AI is Learning from It)

Transcript

Why Reading Everything Makes You Less Intelligent (And What AI is Learning from It) · Mind Cast

• 0:00
  SPEAKER_00: What if I told you that one of the most widely cited weaknesses of the human mind, our inability to process the sheer volume of available information, is actually not a weakness at all?
• 0:11
  SPEAKER_00: What if that so-called limitation is a finely tuned, evolutionarily refined epistemic strategy, one so effective that the world's most advanced AI laboratories are now deliberately engineering their systems to replicate it?
• 0:27
  SPEAKER_00: Here's the counterintuitive truth at the heart of today's episode.
• 0:32
  SPEAKER_00: You don't need to read everything.
• 0:35
  SPEAKER_00: In fact, reading everything might actually make you less intelligent.
• 0:40
  SPEAKER_00: And the research we're diving into today proves exactly that with data, with cognitive science, and with revelatory findings from the AI world.
• 1:08
  SPEAKER_00: Its essential discovery, quality beats quantity.
• 1:12
  SPEAKER_00: AI systems perform better when they learn selectively, focusing on the vital few rather than the trivial many.
• 1:21
  SPEAKER_00: That principle has a name, the 80-20 rule.
• 1:25
  SPEAKER_00: Today, we're going to explore what happens when you apply it not just to productivity, but to the very architecture of human knowledge itself.
• 1:33
  SPEAKER_00: Stay with me.
• 1:48
  SPEAKER_00: Today's episode is one I've genuinely been excited to bring to you.
• 1:52
  SPEAKER_00: We're diving deep into a report titled Epistemic Efficiency and Cognitive Scaffolding, evaluating the Pareto Imperative in Human Cognition and Machine Intelligence.
• 2:05
  SPEAKER_00: Strip away the academic language, and it asks a profoundly simple question.
• 2:10
  SPEAKER_00: How do humans with limited time, attention, and cognitive bandwidth manage to build genuine expertise?
• 2:18
  SPEAKER_00: And how does that compare to the way artificial intelligence builds what looks like knowledge?
• 2:24
  SPEAKER_00: Here's the promise of this episode.
• 2:26
  SPEAKER_00: By the time we're done, you'll understand why reading more is not always better.
• 2:30
  SPEAKER_00: You'll discover that the instinct to focus on a handful of truly great sources is not intellectual laziness, it is cognitive wisdom.
• 2:39
  SPEAKER_00: You'll learn why AI, despite having effectively read the entire public internet, is still missing something fundamental that you carry with you every day.
• 2:49
  SPEAKER_00: And you'll walk away with three concrete actionable habits to sharpen how you think, learn, and engage with AI tools.
• 2:57
  SPEAKER_00: This is not a summary, this is a synthesis, and it starts with a name you may recognize, Herbert Simon.
• 3:06
  SPEAKER_00: Let's get into the first major insight, the 80-20 rule applied to the world of knowledge.
• 3:13
  SPEAKER_00: Most people think information overload is a modern problem caused by smartphones and social media, but humans have always faced far more available information than any one person could absorb.
• 3:27
  SPEAKER_00: The question has never been, can I read everything?
• 3:30
  SPEAKER_00: It's always been, what should I read?
• 3:34
  SPEAKER_00: In 1956, cognitive scientist Herbert Simon introduced a concept foundational to our understanding of human decision-making, bounded rationality.
• 3:46
  SPEAKER_00: The idea is this: human beings don't optimize.
• 3:51
  SPEAKER_00: We have limited time, limited attention, limited mental processing power.
• 3:57
  SPEAKER_00: So instead, we do something far smarter.
• 4:00
  SPEAKER_00: We satisfice.
• 4:02
  SPEAKER_00: Satisfies blends satisfy and suffice.
• 4:06
  SPEAKER_00: We search until we find an option that meets a good enough threshold.
• 4:10
  SPEAKER_00: And then we stop.
• 4:11
  SPEAKER_00: Not out of laziness, but because of an unconscious cost-benefit calculation.
• 4:17
  SPEAKER_00: The moment the cost of continued searching outweighs the potential gain of finding something better, we commit and move on.
• 4:25
  SPEAKER_00: Think about how you chose this podcast.
• 4:27
  SPEAKER_00: You didn't audition every podcast in existence, you used signals, a recommendation, a title that made you curious.
• 4:35
  SPEAKER_00: You made a fast, intelligent judgment call.
• 4:38
  SPEAKER_00: That is bounded rationality in action, not a flaw, a feature.
• 4:44
  SPEAKER_00: Around the same time, researchers Peter Piralee and Stuart Card developed information foraging theory.
• 4:51
  SPEAKER_00: They noticed that humans search for information almost exactly like animals foraging for food.
• 4:57
  SPEAKER_00: Think of a bee moving between flowers or a bear hunting salmon patches in a river.
• 5:03
  SPEAKER_00: Animals don't wander randomly, they read environmental cues and make rapid, efficient decisions about where to invest their energy.
• 5:11
  SPEAKER_00: They move toward high-value patches and abandon low-value ones early.
• 5:16
  SPEAKER_00: Humans do exactly the same.
• 5:19
  SPEAKER_00: We track what Pirelli and Card call information scent, signals like an author's reputation, a citation count, a headline's relevance to our investigation.
• 5:30
  SPEAKER_00: We are, as the researchers put it, inframovores, creatures evolved to maximize knowledge gain per unit of time.
• 5:38
  SPEAKER_00: This brings us to the third pillar, the Paretto principle, the 80-20 rule.
• 5:44
  SPEAKER_00: Vilfredo Pareto noticed that roughly 80% of Italy's land was owned by 20% of the population and found this distribution virtually everywhere.
• 5:56
  SPEAKER_00: A small minority of inputs produce a large majority of outputs.
• 6:01
  SPEAKER_00: The research report makes a compelling case that this applies directly to knowledge acquisition.
• 6:08
  SPEAKER_00: Roughly 20% of core texts and leading experts in any field account for approximately 80% of the structural insights it offers.
• 6:18
  SPEAKER_00: The expert who has read the 30 most important books in their domain likely has deeper understanding than someone who has read 300 at random.
• 6:27
  SPEAKER_00: Focus beats volume.
• 6:30
  SPEAKER_00: Curation beats comprehensiveness.
• 6:34
  SPEAKER_00: The human mind, with its bounded rationality and information foraging instincts, arrived at this conclusion long before any economist put a number on it.
• 6:44
  SPEAKER_00: We have always been, at our cognitive core, Pareto optimizers.
• 6:50
  SPEAKER_00: And that is not a limitation, that is a design.
• 6:54
  SPEAKER_00: This brings us to the second major insight, one that directly challenges a seductive assumption about artificial intelligence, that more data equals more intelligence.
• 7:06
  SPEAKER_00: Let me introduce philosopher Michael Polanyi.
• 7:09
  SPEAKER_00: In the mid-20th century, he made an observation that seems almost obvious once you hear it, yet carries truly radical implications.
• 7:18
  SPEAKER_00: He said, We can know more than we can tell.
• 7:22
  SPEAKER_00: Read that again.
• 7:23
  SPEAKER_00: We can know more than we can tell.
• 7:28
  SPEAKER_00: Polanyi was describing tacit knowledge, the layer of understanding that lives beneath language.
• 7:35
  SPEAKER_00: It's the knowledge in your hands when you ride a bicycle.
• 7:38
  SPEAKER_00: It's the knowledge a master chef has about when a sauce is ready, not from memorizing a recipe, but from making that sauce 500 times, senses calibrated through years of embodied practice.
• 7:52
  SPEAKER_00: Tacit knowledge cannot be fully written down, uploaded, or transmitted through text alone.
• 7:59
  SPEAKER_00: It is built through doing, through practice, failure, risk, mentorship, and the accumulated feedback of real-world experience.
• 8:08
  SPEAKER_00: Here's why this matters enormously for AI.
• 8:11
  SPEAKER_00: Every large language model was trained by ingesting enormous quantities of written text.
• 8:18
  SPEAKER_00: AI learned from the recorded outputs of human thought, but what humans wrote down is only ever a residue of what they actually knew.
• 8:29
  SPEAKER_00: When a Nobel Prize-winning physicist writes a paper, they're translating their deepest intuitions into words.
• 8:36
  SPEAKER_00: The paper is the artifact of the insight.
• 8:38
  SPEAKER_00: It is not the inside itself.
• 8:41
  SPEAKER_00: And AI only has access to the artifact.
• 8:45
  SPEAKER_00: This gap has a name.
• 8:46
  SPEAKER_00: It's the distinction between internalist and externalist epistemic justification.
• 8:52
  SPEAKER_00: A human who has read deeply, traced ideas back to their foundational sources, and wrestled with contradictions has what philosophers call internalist justification.
• 9:03
  SPEAKER_00: They know why they believe what they believe.
• 9:06
  SPEAKER_00: They can challenge it, defend it, and revise it.
• 9:09
  SPEAKER_00: An AI generates statistically plausible text based on patterns detected across massive data sets.
• 9:17
  SPEAKER_00: It operates through externalist reliability, producing outputs that are often correct, but not through genuine understanding, through pattern matching.
• 9:27
  SPEAKER_00: There's another dimension to this gap from German ethologist Jakob von Uxküll, who developed the concept of the Umwelt, the subjective universe of a living creature.
• 9:38
  SPEAKER_00: Every organism experiences reality through its unique sensory apparatus.
• 9:43
  SPEAKER_00: A human experiences it through sight, touch, emotion, social feedback, and critically, through the felt sense of consequence.
• 9:53
  SPEAKER_00: When you make real-world decisions, there are stakes, risk, physical harm, financial loss.
• 10:00
  SPEAKER_00: Your intuitions are forged in the fire of real consequences.
• 10:04
  SPEAKER_00: Touch a hot stove once, and you develop a permanent understanding of heat that no amount of reading about thermodynamics could give you.
• 10:13
  SPEAKER_00: AI has no Umwelt.
• 10:15
  SPEAKER_00: It is never experienced being wrong in a way that actually matters.
• 10:19
  SPEAKER_00: There is a category of wisdom it fundamentally cannot access.
• 10:25
  SPEAKER_00: The implication is profound.
• 10:27
  SPEAKER_00: AI doesn't just capture less knowledge than humans, it captures a fundamentally different kind.
• 10:34
  SPEAKER_00: It captures the visible tip of the iceberg, while the mass of human expertise remains submerged, inaccessible to any system that has never had to live in the world it's trying to understand.
• 10:48
  SPEAKER_00: Now let's turn to the third major insight, containing some of the most surprising twists in the entire report.
• 10:55
  SPEAKER_00: We've established that humans are selective information foragers who build tacit embodied knowledge.
• 11:01
  SPEAKER_00: But what happens when that selectivity goes wrong?
• 11:04
  SPEAKER_00: What happens when the 20% you're relying on turns out to be the wrong 20%?
• 11:10
  SPEAKER_00: And what happens when an AI system starts feeding on its own outputs?
• 11:15
  SPEAKER_00: Researchers have identified a phenomenon called model collapse.
• 11:19
  SPEAKER_00: As AI-generated content proliferates across the Internet, newer AI models begin training on data that increasingly contains AI-generated text rather than original human text.
• 11:30
  SPEAKER_00: They start eating their own tail.
• 11:32
  SPEAKER_00: The degradation happens in three distinct documented stages.
• 11:37
  SPEAKER_00: The first stage is tail density erosion.
• 11:40
  SPEAKER_00: The model begins to lose representation of rare but important ideas.
• 11:46
  SPEAKER_00: The unusual, the counterintuitive, the minority viewpoint start to disappear.
• 11:51
  SPEAKER_00: The rich diversity of human thought begins to flatten.
• 11:55
  SPEAKER_00: The second stage is variance collapse.
• 11:58
  SPEAKER_00: Outputs become increasingly homogeneous.
• 12:01
  SPEAKER_00: Nuance evaporates.
• 12:03
  SPEAKER_00: The model converges on a statistical average of what it thinks people want to hear.
• 12:09
  SPEAKER_00: The third stage, with real-world consequences, is factual failure, hallucinations.
• 12:15
  SPEAKER_00: The model begins confidently asserting things that are simply not true, having lost touch with the diverse, contradictory body of human knowledge that kept it grounded.
• 12:26
  SPEAKER_00: Does this sound familiar from a human context?
• 12:30
  SPEAKER_00: It should.
• 12:31
  SPEAKER_00: This is exactly what happens to a person trapped in an echo chamber.
• 12:36
  SPEAKER_00: When your information diet becomes insular, when you only read sources confirming what you already believe, you undergo your own version of model collapse.
• 12:45
  SPEAKER_00: Your understanding of rare, unexpected events atrophies.
• 12:49
  SPEAKER_00: Your thinking becomes less flexible and more brittle.
• 12:53
  SPEAKER_00: Confidence in your conclusions grows even as their quality quietly deteriorates.
• 12:58
  SPEAKER_00: Confirmation bias is real.
• 13:01
  SPEAKER_00: Insular networks of thinkers who only cite each other are real, and the blind spot for tail risks, the black swans, the once-in-a-generation disruptions may be the most dangerous vulnerability of all.
• 13:15
  SPEAKER_00: Selectivity is a superpower, but a truly sophisticated epistemic strategy must account for its shadow side.
• 13:24
  SPEAKER_00: But here is the most surprising finding in the entire report, from AI research itself.
• 13:31
  SPEAKER_00: In 2022, DeepMind published what became known as the Chinchilla Scaling Laws.
• 13:37
  SPEAKER_00: The dominant assumption had been simple: bigger is better.
• 13:42
  SPEAKER_00: Train on more data with a bigger model and you'll get better results.
• 13:46
  SPEAKER_00: Chinchilla turned that on its head.
• 13:49
  SPEAKER_00: A smaller model, trained on a smaller but far more carefully curated data set, dramatically outperformed much larger models trained on massive unfiltered data.
• 14:01
  SPEAKER_00: The ratio that mattered wasn't raw scale, it was the quality and intentionality of what the model learned from.
• 14:09
  SPEAKER_00: The most sophisticated AI research institution on the planet arrived at a conclusion every great scholar has always known.
• 14:18
  SPEAKER_00: Quality beats quantity.
• 14:20
  SPEAKER_00: AI is learning, slowly, expensively, to do what the human mind has been doing since the Paleolithic era.
• 14:28
  SPEAKER_00: It is learning to filter.
• 14:32
  SPEAKER_00: So where does this leave us?
• 14:34
  SPEAKER_00: Three concrete actionable takeaways you can apply starting today.
• 14:39
  SPEAKER_00: The first, be ruthlessly intentional about your information diet.
• 14:44
  SPEAKER_00: The 80-20 rule is an operational strategy.
• 14:47
  SPEAKER_00: In whatever domain matters most to you, identify the 20% of sources, authors, and thinkers that contain the majority of real insight.
• 14:57
  SPEAKER_00: Invest your time there, deeply and in the right places, not spread uniformly.
• 15:03
  SPEAKER_00: This doesn't mean ignoring breadth entirely.
• 15:06
  SPEAKER_00: You need some exposure to ideas outside your curated core to guard against echo chamber failure, deliberate long tail scanning.
• 15:15
  SPEAKER_00: But your default should be depth over breadth.
• 15:18
  SPEAKER_00: Find the thinkers who influenced the thinkers you already admire.
• 15:22
  SPEAKER_00: Go upstream to the source.
• 15:25
  SPEAKER_00: The second takeaway: protect and deliberately cultivate tacit knowledge.
• 15:30
  SPEAKER_00: In an era where AI can produce polished, confident sounding text on almost any topic in seconds, there is an enormous temptation to outsource the thinking.
• 15:40
  SPEAKER_00: Resisted.
• 15:41
  SPEAKER_00: The kind of knowledge AI cannot replicate, the embodied, judgment-laden, contextually sensitive understanding that comes from doing real work under real conditions, is becoming more valuable, not less.
• 15:54
  SPEAKER_00: It is precisely because AI has flooded the world with surface-level competence that genuine depth now stands out more sharply than ever.
• 16:03
  SPEAKER_00: Seek out mentors, pursue apprenticeship, take on projects where being wrong has actual consequences.
• 16:10
  SPEAKER_00: Allow yourself to be a genuine beginner in domains that matter to you.
• 16:14
  SPEAKER_00: Tacit knowledge is built not by reading about swimming, but by getting in the water.
• 16:20
  SPEAKER_00: The third takeaway, when you use AI tools, maintain internalist epistemic habits.
• 16:28
  SPEAKER_00: Practically, this means never fully outsourcing your judgment.
• 16:32
  SPEAKER_00: When an AI gives you an answer, ask, where did this claim come from?
• 16:37
  SPEAKER_00: Can I trace it to a primary source?
• 16:39
  SPEAKER_00: Does it hold up against what I already know from deep engagement with this domain?
• 16:45
  SPEAKER_00: AI can be an extraordinary tool for retrieval, synthesis, brainstorming, and first draft thinking.
• 16:52
  SPEAKER_00: But it should amplify your judgment, not replace it.
• 16:56
  SPEAKER_00: The moment you stop asking where information comes from and whether it holds up, you've stopped being an epistemic agent and started being a passive consumer.
• 17:06
  SPEAKER_00: The future of intelligence is the hybrid, machine speed retrieval combined with the irreplaceable depth of human tacit wisdom.
• 17:16
  SPEAKER_00: That is the combination neither alone can match.
• 17:21
  SPEAKER_00: And that brings us to the close of today's episode.
• 17:24
  SPEAKER_00: Let me recap.
• 17:26
  SPEAKER_00: We started with a counterintuitive truth.
• 17:29
  SPEAKER_00: The human brain's selective approach to information is not a limitation, it is an architecture, shaped by evolution, refined by real-world consequence, and now validated by cutting-edge AI research.
• 17:44
  SPEAKER_00: We explored bounded rationality and satisficing.
• 17:48
  SPEAKER_00: We followed inframorivores hunting knowledge through high-value patches.
• 17:52
  SPEAKER_00: We applied Pareto's vital few to the structure of expertise.
• 17:57
  SPEAKER_00: We confronted tacit knowledge and the umwelt, naming the gap between what AI captures and what human expertise actually is.
• 18:06
  SPEAKER_00: And we arrived at the chinchilla revelation that even the world's most sophisticated AI systems are learning to do what human minds have always done, filter with intention.
• 18:19
  SPEAKER_00: I want to be transparent before we close.
• 18:22
  SPEAKER_00: The research report at the foundation of this episode is not publicly available.
• 18:26
  SPEAKER_00: What I've brought you is a synthesis of its core ideas.
• 18:30
  SPEAKER_00: If this sparks something, engage directly with the thinkers we discussed: Simon, Polani, Paroli and Card, Von Exko, because their primary works are where the deepest rewards live.
• 18:43
  SPEAKER_00: Go upstream.
• 18:44
  SPEAKER_00: That's the whole point.
• 18:46
  SPEAKER_00: If this episode was valuable, if it shifted something in how you think about information, expertise, or AI, I'd love for you to do three things.
• 18:55
  SPEAKER_00: First, subscribe to Mindcast wherever you listen.
• 18:58
  SPEAKER_00: That single action is the most important thing you can do to keep this show growing.
• 19:03
  SPEAKER_00: Second, leave a review.
• 19:05
  SPEAKER_00: Even a sense or two makes an enormous difference in helping new listeners find us.
• 19:10
  SPEAKER_00: And third, share this episode with someone who needs it, a colleague wrestling with AI, a student overwhelmed by everything they feel obligated to read, or friend curious about how the mind builds knowledge.
• 19:23
  SPEAKER_00: These ideas compound.
• 19:26
  SPEAKER_00: I'm Will.
• 19:27
  SPEAKER_00: This has been Mindcast.
• 19:29
  SPEAKER_00: Thank you genuinely for spending part of your day here and choosing to go deep rather than wide.
• 19:34
  SPEAKER_00: Keep thinking carefully, keep curating fiercely, and I will see you in the next episode.