The Modern Paradox of Freedom
Note: the following is adapted from an essay for my “Tyranny and Totalitarianism” course, taken in spring 2026. Some views are no longer held strongly, though I still weigh the overarching thesis highly.
Will the advent of artificial intelligence give rise to a freer society? A cursory glance signals yes. It is true that — global outliers notwithstanding — the traditional Arendtian conception of totalitarianism is all but impossible in today’s digitized, fragmented world. However, this does not necessarily imply that a high “floor” of individual freedom has now been securely established. Two new technologies burgeoning in tandem with the digitization of the world — artificial intelligence and social media — are epistemically unprecedented. That is, they offer novel methods of organizing belief that were not previously available. I argue that 1) these new methods exist and are near-ubiquitously utilized, 2) their degree of proper truth-tracking is not correspondingly adequate, and 3) they therefore give rise to a modern “paradox of freedom” which threatens individual liberty in similarly unprecedented ways and merits extensive consideration to mitigate.
By “individual liberty” I mean specifically intellectual liberty — the capacity for autonomous belief formation — rather than the political, economic, or bodily freedoms with which it is often conflated. And by “knowledge” I assume the classical epistemological account of “justified true belief,” where mere belief is elevated to knowledge only when it is both accurate and held for the right kinds of reasons. I commit to this particular framing because the threat I will describe is not one of forbidden information or restricted speech, but of a quieter corrosion at the level of justification itself.
The artificial intelligence chatbot is quickly replacing web browsing as the layman’s go-to method of satiating an intellectual curiosity. From summarizing dense historical events to diagnosing medical ailments to explaining the nuances of contemporary political legislation, the epistemic use cases for generative AI are vast and universal. We are witnessing a paradigm shift in how the average citizen interacts with information. What makes this method of knowledge acquisition profoundly novel is the notion of an authoritative synthesis dictating epistemic flow. Unlike traditional search engines, which present a curated menu of disparate sources requiring the user to triangulate and evaluate the information, the large language model (and its ilk) digests the internet and regurgitates a definitive answer. The epistemic labor is entirely outsourced. It replaces the dialectic of research with the passive monologue of an oracle, presenting a seamless narrative devoid of the friction that traditionally accompanies the search for justified belief, which I have defined knowledge in terms of.
The rise of social media yields a similar story. The social dissemination of knowledge is now more potent than it ever was before. Some may contend that social media is just one more step in the continuous and gradual socialization of knowledge. But crucially, what is novel today is the notion that for a given subject, this social transfer of knowledge is often done via parties with no nominal relationship to the individual other than their role in the context of that very transfer. That is, content consumers are primarily learning from complete strangers. This collapse of context matters deeply because the authority of the speaker is no longer derived from their proximity to the individual (like the trusted word of a friend or family member), but from their proximity to the algorithm, which does not cleanly latch onto objective expertise. This constitutes a fundamentally new epistemic method: a peer-to-peer knowledge transfer where validation is determined by an algorithm optimizing for aesthetics and relatability. The medium replaces the expert and the trusted peer.
One might object that this charge is itself ancient. In the Phaedrus, Plato’s Socrates famously worried that the written word would hollow out memory and confer the mere appearance of wisdom, producing students with “the show of wisdom without the reality.” Every subsequent epistemic technology — the codex, the printing press, the encyclopedia, the search engine — has provoked some version of this lament, and the world has nevertheless gotten on. So is the present case genuinely new, or merely the latest installment in the classic story of humanity? Three features distinguish it. The static text Socrates feared was at least uniform, public, and inert: every reader received the same words and was forced to do the interpretive labor themselves. AI and algorithmic media, by contrast, are individuated, dynamic, and opaque. The output is tailored per user on-demand and produced by a process whose internal workings are largely proprietary or inscrutable. Writing replaced one moment of dialectic with a fixed artifact; the chatbot replaces dialectic with the simulation of dialectic itself.
A naïve reading of the above would be thrilled at the prospect of acquiring knowledge so easily with the aforementioned methods. But knowledge is not belief: knowledge necessitates justification and a tendency toward truth. Artificial intelligence and social media do not offer a tendency toward truth. I commit, here, to a deflationary account of truth loosely tracking correspondence theory (we should label statements corresponding with external reality as true, and false otherwise). Some other accounts (coherentist, pragmatist, consensus-based) are actually vulnerable to distortion by emerging technology: a system that produces internally coherent outputs, or that “works” by some pragmatic metric, or that manufactures consensus, can satisfy such accounts while still being epistemically catastrophic. Correspondence has the virtue of stubbornness; it cannot be wholly captured by an algorithm whose primary feedback signal is human attention.
At their core, modern generative AI systems do not understand the world; they map the statistical probability of word sequences based on the massive corpuses of text they were trained on. This mechanism inherently prioritizes linguistic plausibility over fact. Furthermore, they are strictly bound by the quality of their training material. If the dataset from which an AI draws its probabilistic weights is rife with human biases, historical inaccuracies, and internet-born fabrications, the model’s output will reliably mirror these flaws out-of-the-box. Because it lacks an internal empirical mechanism to verify its claims against physical reality, the model is blind to truth; it only recognizes and replicates patterns.
Similarly, the architecture of social media is fundamentally indifferent to objective reality. The epistemic engine of the modern platform is the recommendation algorithm, a mechanism engineered with the overriding objective to maximize retention. Decades of data have repeatedly demonstrated that content which elicits high-arousal emotions (like moral outrage, fear, and tribal validation) generates the most potent engagement. Consequently, the algorithm systematically amplifies sensationalism and polarization while burying unprovocative truths. When the primary metric of systemic success is human attention rather than accuracy, the epistemic environment naturally and inevitably degrades into a marketplace of compelling fictions.
I have established that two of the most powerful, widely-used methods of affirming one’s beliefs today are fundamentally detached from the concept of truth. This is an environment primed for abuse. Herein lies the modern “paradox of freedom.” While the digitized individual appears to possess an unprecedented degree of epistemic choice — the entire breadth of human knowledge is ostensibly at their fingertips — this extreme freedom is largely illusory. The information we consume, the answers we are readily given, and the communities we form on social networks are covertly curated by opaque algorithms. This hidden architecture is highly vulnerable to manipulation by bad actors. State-sponsored disinformation campaigns, corporate interests, and ideological extremists no longer need to resort to the Arendtian methods of terror and physical coercion to control a population. They merely need to manipulate the abstract algorithmic levers that govern our daily information diet, quietly manufacturing a fractured consensus while maintaining the psychological façade of individual intellectual liberty.
It is worth being explicit about why this should trouble us in moral, and not merely epistemic, terms. I am not claiming that truth itself is a moral property — that would be a stronger metaphysical commitment than my argument requires. Rather, the capacity to track truth is constitutive of meaningful intellectual liberty. A self that cannot reliably distinguish reality from fiction is not freely choosing its beliefs. Liberty in any robust sense requires more than the absence of overt coercion: it requires an epistemic environment in which real, conscious deliberation is at least possible. When that environment is engineered to optimize for anything other than accuracy, the very preconditions of autonomous belief are subtly evacuated, even as the surface experience of choice remains intact.
So while the overt, heavy-handed totalitarianism of the twentieth century may seem like an anachronism in much of the modern democratic world, the threat to individual freedom has not disappeared; it has merely been sublimated into our technology. By delegating our epistemic responsibilities to artificial intelligence and algorithmic social networks, we have constructed a global infrastructure of belief formation that is structurally divorced from the pursuit of truth. To mitigate this unprecedented threat, we must foster a radical shift in our societal epistemic habits. Explicating possible solutions is beyond scope for now, but the bottom line is as follows. If we fail to recognize and confront this paradox of freedom, we risk wandering voluntarily into a digital subjugation of our own making: a world in which we live comfortably entertained and endlessly validated, yet profoundly misinformed and deeply vulnerable to exploitation.