Let me paint you a picture of a typical Tuesday morning. You wake up, you reach for your phone before your feet hit the floor, and within sixty seconds an algorithm has decided what you’re going to think about today. Not what you’re going to think — that’s still (mostly) up to you — but what you’re going to think about. The topics, the frames, the examples, the emotional register. All of it curated by a system whose explicit goal is to maximize your engagement, which is a polite way of saying to keep you agitated and reactive for as long as possible. This is not a conspiracy theory. It’s the documented business model of every major social platform on earth.
Here’s the part that should genuinely disturb you: most people have outsourced not just their information consumption but their actual thinking to these systems. Not just what they read — what they conclude. What they believe. What they’re angry about and what they’re indifferent to. The algorithm doesn’t just show you content. It shapes the questions you ask, the possibilities you consider, the framework through which you evaluate everything else. When you think you’re forming an independent opinion, you’re often assembling pieces that were pre-selected for you by a machine that knows your behavioral patterns better than you know yourself.
This is EP384, and today we’re going after the root of something that I think is the defining cognitive crisis of our time: the systematic erosion of independent thought. Not because people are getting dumber. Because the systems that now surround us are extraordinarily good at doing our thinking for us — and we’ve let them. Let’s talk about how to take it back.
The Architecture of Algorithmic Influence
Before we can talk about solutions, we need to understand the mechanism with some precision. The recommendation algorithms that govern social media, news feeds, video platforms, and increasingly search results aren’t neutral pipes through which information flows. They’re active shaping forces that learn your behavioral patterns and optimize content selection to maximize a specific metric: time on platform, usually measured through engagement signals like clicks, reactions, comments, and shares.
The problem is that the content that maximizes engagement is systematically different from the content that maximizes understanding, accurate belief formation, or wise decision-making. Extreme content gets more engagement than detailed content. Emotionally activating content gets more engagement than calm analysis. Content that confirms your existing beliefs gets more engagement than content that challenges them — because confirmation feels good and challenge feels threatening, and the algorithm has no preference between these two states, only for the engagement signal.
Renée DiResta, one of the foremost researchers on algorithmic amplification, has documented extensively how recommendation systems create what she calls “rabbit holes” — pathways through content that take someone from moderate positions to increasingly extreme ones, not because the algorithm is ideologically motivated but simply because extreme content is engaging content. The journey from curious to radicalized doesn’t require any human actor deliberately pushing anyone anywhere. It requires only an optimization function pointed at engagement, and a human being following the path of least cognitive resistance.
Jonathan Haidt and Jean Twenge’s research on adolescent mental health — summarized in The Anxious Generation — documents the population-level cognitive and emotional effects of growing up in this environment. The decline in measured critical thinking skills in younger cohorts. The increase in anxiety and depression. The narrowing of information sources. These aren’t incidental side effects. They’re the predictable result of environments designed by extremely smart people to exploit human psychological vulnerabilities for profit. We should be clearer-eyed about what we’re dealing with.
The Illusion of Informed Engagement
One of the most pernicious features of algorithmic media consumption is how informed it makes you feel while you’re doing it. You read fifteen articles today. You watched four videos. You had opinions on six different topics. You feel engaged, current, knowledgeable. But here’s the question nobody asks: what was the process by which those fifteen articles were selected? What was the counterfactual — what weren’t you exposed to, and why not?
Information selection is cognition. The question of what evidence gets considered before forming a conclusion is not separate from the thinking process — it is a central part of the thinking process. A lawyer who only reads evidence favorable to their client isn’t doing independent analysis; they’re confabulating a justification for a predetermined conclusion. Most people’s algorithmic media consumption is structurally identical to this, even though it doesn’t feel that way from the inside.
The philosopher of science Karl Popper spent his career arguing that the hallmark of genuine inquiry is that it seeks to falsify its own beliefs, not to confirm them. He called this criterion falsifiability — a good scientific theory is one that makes specific predictions that could, in principle, be proven wrong. The same logic applies to individual belief formation. A thinking process that only seeks confirming evidence isn’t thinking — it’s rationalization. And algorithmically curated feeds are almost perfectly designed to keep you in perpetual rationalization mode.
The most dangerous thought is the thought you’ve already decided is correct before you’ve examined the evidence for it. The algorithm is extraordinarily skilled at delivering exactly that thought to you, packaged as news, at the moment you’re most likely to accept it without scrutiny.
Consider how many of your current strong opinions — political, social, economic, cultural — you would confidently hold if your information environment had been meaningfully different. If you’d grown up reading different sources, following different people, living in a different geographic and social context. The honest answer for almost everyone is: very few of them. Most of our confident beliefs are artifacts of our information environment, not conclusions we’ve reached through rigorous independent analysis. The algorithm didn’t create this problem, but it’s made it dramatically worse and dramatically more invisible.
What Genuine Independent Thinking Actually Looks Like
We throw the phrase “think for yourself” around as if it were simple, as if anyone who isn’t already doing it is just being lazy. It’s not simple. Independent thinking is a skill, and like all skills it requires deliberate practice, good models, and feedback loops that let you know when you’re doing it wrong. Most people were never taught it explicitly, and the environments they now inhabit actively work against it. Let’s be specific about what it actually involves.
First, genuine independent thinking requires going to primary sources. Not articles about studies. The actual studies. Not summaries of books. The actual books. Not analysis of a speech or policy. The actual speech or policy document. This sounds obvious. Almost nobody does it. The gap between what the primary source actually says and what gets reported about it is, in most domains, enormous. Reading the primary source doesn’t guarantee you’ll reach the right conclusion, but it dramatically expands the information you’re actually working with and exposes you to nuances and qualifications that get stripped out in summarization.
Second, it requires steelmanning opposing views. This is a specific cognitive technique that Rationalist community writers like Eliezer Yudkowsky and Scott Alexander have written about extensively. Before you dismiss a position you disagree with, you must be able to articulate the strongest possible version of that position — not the strawman, not the weakest argument you’ve heard for it, but the best argument a genuinely intelligent, well-informed person who holds that view would make. If you can’t do this, you don’t actually understand the disagreement well enough to have a confident opinion about it.
Third, it requires epistemic humility — a real, operational understanding of how often smart, well-intentioned people are wrong, including you. Philip Tetlock spent decades studying expert forecasting and found that in complex domains, expert predictions barely outperformed random chance. He concluded that the key differentiator between better and worse forecasters wasn’t IQ or domain expertise — it was the willingness to hold beliefs tentatively, to update on new evidence, to resist the tribal dynamics that make changing your mind feel like defeat. His work in Superforecasting is essential reading for anyone who wants to think clearly about uncertain questions.
The Specific Cognitive Traps to Watch For
Availability bias on steroids. The availability heuristic is the cognitive shortcut of estimating how likely something is based on how easily you can bring examples to mind. If you can think of many examples quickly, it feels common. If you struggle to think of examples, it feels rare. Algorithms exploit this mercilessly. If your feed is filled with examples of a particular type of crime, you’ll overestimate how common it is, regardless of base rates. If a particular political position is repeatedly associated with embarrassing or extreme examples, you’ll come to see all holders of that position through those examples. Your sense of what’s typical gets warped by what’s algorithmically amplified.
Motte-and-bailey doctrine — a term from philosophy of argument that describes a rhetorical bait-and-switch. The motte is a defensible, narrow position. The bailey is a much broader, more controversial position. The trick is to argue for the bailey, and when challenged, retreat to the motte as if it were what you’d been arguing all along. Social media accelerates this pattern because detailed argument doesn’t travel well. The bailey version of any position — bold, simple, emotionally resonant — spreads. The motte — careful, qualified, conditional — doesn’t. You end up with a discourse full of people arguing past each other because the actual positions are never clearly stated.
The overton window shifts invisibly. The Overton Window is the range of ideas considered acceptable and debatable in a given public conversation. Algorithms shift this window gradually and invisibly by amplifying certain types of content and suppressing others, not through any deliberate editorial decision but through the aggregate effect of engagement optimization. A position that was clearly outside the mainstream five years ago can become the common sense of your algorithmic community without you ever consciously noticing the shift — because the shift was gradual, because the evidence for the shift was selected for you, and because your peer group (also algorithmically curated) is making the same journey.
The Long-Form Reading Practice as Cognitive Defense
There’s a specific cognitive skill that gets built through sustained long-form reading that doesn’t get built any other way — the ability to hold a complex, multi-part argument in working memory and evaluate its components against each other and against your existing beliefs. Short-form content doesn’t build this skill because it never requires it. Algorithmic feeds actively work against it by constantly redirecting your attention to new stimulus before you’ve finished processing the last one. The average social media scroll session involves dozens of topic switches in a span of minutes. This is literally the opposite of the sustained focused engagement that deep thinking requires.
Mortimer Adler’s How to Read a Book, first published in 1940 and still in print, remains the best systematic treatment of how to actually engage with a text analytically rather than just consuming it. Adler distinguishes between levels of reading — elementary, inspectional, analytical, and syntopical (reading multiple books on the same topic in conversation with each other). Most people never get past inspectional reading. Analytical reading — engaging actively with the author’s argument, testing its premises, identifying its gaps — is the practice that builds genuine intellectual independence. It’s hard. It’s slow. It’s also deeply countercultural in 2025, which means the people who do it have a significant and growing cognitive advantage.
The daily independent thinking practice — a minimum viable protocol:
- Thirty minutes of reading from a deliberately chosen source (not algorithmically delivered) before checking any social media or news
- One question per day you don’t know the answer to that you investigate through primary sources
- One piece of writing per week — even a paragraph — that requires you to form and defend a position
- One conversation per week with someone whose starting premises are substantially different from yours
Set a target: two serious books per month minimum. Not business books with one idea padded to 300 pages. Books that make you work. Primary texts in philosophy, history, science, economics. If you’re reading a book that you agree with on every page and never feel challenged by, you’re not reading widely enough. The discomfort of genuine intellectual challenge is the signal that growth is happening.
Rebuilding Your Information Diet from Scratch
I want to be practical here, because the abstract case against algorithmic dependence isn’t enough. You need a concrete alternative, or you’ll just go back to what’s easy. Let me walk you through what a genuinely independent information diet looks like.
Start with a radical audit. For two weeks, track every piece of information you consume. Where did it come from? Who selected it for you, and on what basis? What was the selection mechanism — algorithm, friend recommendation, newsletter, library shelf? At the end of two weeks you’ll have a clear picture of how much of your information environment is algorithmically determined versus actively chosen. For most people, the number is somewhere between seventy and ninety percent algorithmic. That’s a problem.
Replace algorithmic curation with deliberate curation. RSS feeds still exist. Consider building your information diet around these specific source types:
- Direct newsletter subscriptions from specific thinkers and researchers whose work you respect
- RSS feeds from academic preprint servers in fields you care about
- Books — primary texts rather than summaries — at a cadence of at least two per month
- Long-form journalism from publications with documented corrections policies and transparent editorial standards
- Structured conversations with people who hold different premises than yours
You can subscribe directly to the publications, researchers, and writers whose thinking you want to follow, in a format where the selection is yours rather than an algorithm’s. Newsletters from individual thinkers are better than platform-based content because the selection and framing is explicit — you know whose perspective you’re getting and can calibrate accordingly. Email subscriptions to academic preprint servers in fields you care about give you access to the actual research before it’s been filtered through journalism.
Build a practice of scheduled media consumption rather than ambient consumption. Checking your phone or feeds on impulse — when you’re bored, when you’re stressed, when there’s a moment of dead time — means that your information consumption is driven by emotional state rather than deliberate intention. That’s the algorithmic trap. Instead: designate specific times for information consumption. Outside of those times, the feeds are closed. This single change does more to restore cognitive independence than almost anything else I know of, because it breaks the ambient-agitation loop that algorithmic content is designed to maintain.
Writing as a Thinking Tool
When you write something down — really write it, not just note it but actually try to construct an argument on the page — you immediately discover whether you actually understand it. The experience of trying to write clearly about a topic you think you understand, and finding that you can’t, is one of the most valuable epistemic experiences available. It tells you that what you had was the feeling of understanding, not actual understanding. Feynman called this the “illusion of explanatory depth” — we think we understand things far better than we do, and writing exposes the gap ruthlessly.
The form of writing matters too. Writing that tries to persuade someone who disagrees with you — who knows the counterarguments, who will push back on weak reasoning — forces a different and more rigorous process than writing for an audience that already agrees. This is why writing publicly, even if your audience is small, creates more rigorous thinking than journaling. Accountability to a skeptical reader sharpens your reasoning in ways that private writing often doesn’t.
I’d recommend a minimum practice: one substantive piece of writing per week on a topic you care about, structured as an actual argument rather than a list of observations. Not a social media post — that’s too short to require real thinking. An essay-length treatment where you state a claim, defend it with evidence, address the strongest counterarguments, and draw a conclusion. Do this consistently for six months and your thinking will be measurably sharper. I don’t say that lightly.
The Social Dimension: Your Network as a Thinking Environment
We’re social animals and our thinking is profoundly shaped by our social environment — not just through the social media feed but through the actual people we spend time with, talk to, and whose opinions we care about. The people around you determine the questions you consider worth asking, the range of positions that feel defensible, the cognitive norms you’re held to. This is not a small thing.
Nicholas Christakis and James Fowler’s research on social contagion — documented in Connected — showed that attitudes, behaviors, and even mood states spread through social networks in ways that are measurable and significant. Your political beliefs, your risk tolerance, your health behaviors, your financial habits — all of these are significantly predicted by what your close social network believes and does. You didn’t choose your beliefs through pure rational deliberation any more than you chose your accent. You absorbed them from your environment.
This should lead you to think seriously about the epistemic quality of your social environment. Not to replace your friends — social connection has values beyond cognitive performance. But to deliberately seek out people who think differently from you, who hold different premises, who will push back on your ideas rather than affirming them. The most intellectually dangerous thing you can do is to surround yourself entirely with people who agree with you. The most cognitively valuable thing you can do is to maintain genuine friendships with intelligent people who hold fundamentally different worldviews.
This is deeply uncomfortable. Genuine intellectual challenge from someone you respect is harder to dismiss than a contrary social media comment from a stranger. It requires you to actually engage, to defend your positions or update them, to sit with uncertainty when the argument is genuinely close. Most people avoid this discomfort by surrounding themselves with agreement. The few who don’t — who actively cultivate epistemic diversity in their social environments — tend to be the sharpest thinkers I know.
Algorithms in Professional Life: The Hidden Outsourcing
So far we’ve been talking about media consumption and public discourse, but the algorithmic outsourcing of thinking is just as pervasive — and in some ways more consequential — in professional contexts. When you use an AI assistant to draft your analysis before you’ve done your own, when you let your email client’s AI suggest replies before you’ve decided what you think, when you follow the recommendations of an analytics platform without asking whether those recommendations reflect your actual values and goals — you’re doing the same thing, just in a work context.
I want to be clear that I’m not anti-technology. AI tools are genuinely powerful and I use them. But there’s a crucial distinction between using AI to execute decisions you’ve already made through your own thinking, versus using AI to replace the thinking process itself. The former is efficiency. The latter is intellectual atrophy. And like physical atrophy, you often don’t notice it happening until you need the muscle and discover it’s no longer there.
The practice I’d recommend: before using any AI tool for a substantive intellectual task, spend at least fifteen minutes working on the problem yourself. Not to avoid AI assistance — you can use it afterward — but to preserve the cognitive engagement that keeps your own thinking sharp. Form your own hypothesis before looking at the AI’s analysis. Draft your own outline before asking for AI suggestions. When you then compare your thinking to the AI’s output, you get far more value from the comparison — you can see where your reasoning had gaps, where you were on the right track, where the AI is missing something you caught. If you go straight to the AI, you have no basis for that comparison. You’re just accepting output.
The Bigger Picture: What We Lose Collectively
We’re conducting a large-scale natural experiment in what happens to democratic discourse when you route political communication primarily through engagement-optimized platforms. The preliminary results are not encouraging. Political polarization has increased. Trust in institutions has declined. The ability to have productive cross-ideological conversations has declined. The percentage of people who can accurately state the strongest argument for a position they oppose has declined. These are measurable, documented trends. And while there are multiple causes, the algorithmic restructuring of public discourse is among the most significant.
Hannah Arendt wrote in The Origins of Totalitarianism that the most dangerous political development isn’t the arrival of explicit authoritarianism — it’s the prior erosion of the capacity for independent thought that makes populations susceptible to it. A citizenry that has outsourced its thinking to authority (in her time, the authority of ideological movements; in our time, the authority of algorithmic recommendation) loses the cognitive infrastructure for self-governance. That’s a civilizational stake, not just a personal productivity concern. When you do the work of independent thinking, you’re not just doing it for yourself.
Listener FAQ
Q: Isn’t this elitist? Not everyone has time for long-form reading and deliberate information diets.
It’s a fair challenge, and I want to engage with it honestly. The practices I’m describing do require time and privilege that not everyone has equally. At the same time, I’d push back on the framing that algorithmic consumption is somehow more democratic than intentional consumption. Algorithms are built by specific people with specific incentives, and their effects on different populations are not neutral. Working-class communities and communities of color have been targets of algorithmic radicalization and misinformation in documented and disproportionate ways. The solution isn’t to accept algorithmic dependence as the democratic option — it’s to advocate for media literacy education, better algorithmic accountability, and more equitable access to the conditions that make independent thinking possible. And on a personal level, even thirty minutes of intentional reading per day is achievable for more people than tend to prioritize it.
Q: How do you know your own thinking isn’t just as biased as the algorithm’s output?
You’re right that it is. The goal isn’t bias elimination — that’s impossible. The goal is bias awareness and diversity of input. Epistemic humility about your own reasoning is a core part of independent thinking, not a challenge to it. The person who recognizes their own biases and actively seeks out information and perspectives that might correct them is doing something fundamentally different from the person who passively accepts whatever a system selected. Both are biased. Only one is engaging with the problem.
Q: What’s your take on using AI for thinking assistance rather than thinking replacement?
I think it can be genuinely valuable if the distinction is maintained in practice. Using AI as a sparring partner — presenting your argument and asking it to find weaknesses — is excellent. Using AI to generate counterarguments you then evaluate is excellent. Using AI to summarize material you’ve already engaged with seriously is fine. Using AI to do your initial thinking and then editing the output is where I see the atrophy risk. The key question is whether you’re coming to the AI with your own thoughts already formed, or coming empty and leaving with the AI’s thoughts. The former preserves and even sharpens your cognitive engagement. The latter gradually replaces it.
Q: How do you handle the social pressure to have opinions on current events in real time?
By deciding that “I don’t know yet” is a complete and respectable answer. The social pressure to have immediate hot takes on breaking news is one of the most cognitively destructive features of current discourse. The information available in the first 24-48 hours of any significant event is almost always incomplete, often wrong, and systematically skewed toward the most emotionally activating interpretation. Resisting the pressure to form and express immediate opinions is not intellectual cowardice — it’s epistemic hygiene.
Q: Any specific books you’d recommend for building a more rigorous thinking practice?
Several. Mortimer Adler’s How to Read a Book for the foundational practice of analytical reading. Philip Tetlock’s Superforecasting for probabilistic reasoning about uncertain questions. Michael Nielsen’s essay Thought as a Technology for an excellent frame on what it means to build genuine cognitive tools. Annie Duke’s Thinking in Bets for applying decision-quality thinking to everyday choices. And Daniel Kahneman’s Thinking, Fast and Slow — still the most comprehensive treatment of cognitive bias and its interaction with rational deliberation. Read these not to accumulate talking points, but to actually change the way you process information. That requires reading them slowly, with a pencil, taking notes, and then finding places to apply what you’re learning. The reading alone isn’t enough — application is where the integration happens.
The Practice of Adversarial Collaboration
One of the most powerful practices for maintaining genuine independent thought that almost no one does is what researchers call adversarial collaboration — the deliberate practice of working through a question with someone who holds a fundamentally opposed starting position, with both parties committed to following the evidence wherever it leads rather than to winning the argument. Daniel Kahneman has engaged in adversarial collaborations with researchers who disputed his findings, and describes the process as among the most intellectually productive of his career.
You don’t need a formal academic partner to practice this. What you need is the habit of actively seeking out the most intelligent, most informed people who disagree with your most confident positions, and genuinely engaging with their best arguments. Not to defeat them — to understand them. The question is not “how do I refute this?” but “what would have to be true for this person to be right, and how confident am I that those things aren’t true?” This question changes the entire character of the engagement. You’re no longer in combat mode. You’re in inquiry mode. And inquiry mode is where genuine thinking happens.
The adversarial collaboration practice also builds what philosophers call “epistemic humility” in the most practical possible way — not as an abstract virtue you claim to have, but as a lived experience of having been genuinely uncertain about something you previously felt confident about. Each genuine engagement with a strong opposing argument that you can’t immediately refute is a data point about the limits of your current understanding. Those data points, accumulated over time, produce a genuine and calibrated relationship to your own knowledge and its boundaries. That’s not weakness. It’s the beginning of real intellectual maturity.
Historical Perspective: The Permanence of the Challenge
It would be a mistake to frame the challenge of independent thinking as uniquely modern. Every era has had its mechanisms of cognitive conformity — its structures for ensuring that people think within sanctioned frameworks and avoid dangerous independent conclusions. The Catholic Church’s Index of Forbidden Books. The Soviet Union’s control of information about the purges. The social pressure of small communities to maintain orthodoxy on questions of religion, morality, and social organization. The algorithmic mechanisms of today are new in their scale and their precision, but the underlying dynamic — powerful entities shaping the information environment to produce compliant thinking — is ancient.
What’s historically consistent about independent thinkers is that they were almost always inconvenient. Socrates was executed for corrupting the youth with questions. Galileo was placed under house arrest for saying the Earth moved. Darwin spent years delaying publication of the Origin of Species because he anticipated — correctly — the ferocity of the resistance. Every generation has its versions of these figures, and every generation has its versions of the pressure to conform. The lesson is not that independent thinking is impossible — clearly it happens, and clearly it matters. The lesson is that it has always required deliberate effort against real social pressure, and that the pressure today, though different in form, is not different in kind from what previous generations navigated.
What we have now that Galileo didn’t have is access to the entire history of human thought — every major text, every significant philosophical tradition, every scientific revolution — and the tools to trace the original sources rather than relying on secondhand accounts. The information access is unprecedented in its completeness, even as the attention-capture mechanisms are unprecedented in their sophistication. The challenge is to use the first against the second. To use the extraordinary availability of genuine knowledge — of the primary sources, the real research, the actual arguments — to resist the extraordinary pull of the curated, engagement-optimized, algorithmically selected information environment. That’s the work. It’s always been the work. We’re just doing it in a new arena.
That’s EP384. The algorithms aren’t going away. The pressure to let them think for you isn’t going away. But the capacity to think independently — to go to primary sources, to steelman opponents, to hold your beliefs with appropriate uncertainty and update them when the evidence warrants — that capacity is yours to build and yours to protect. Do the work. It matters more than you think.
Editorial StandardsCorrectionsMedical DisclaimerAbout Our ContentAffiliate DisclosureSite Map