How should we define propositional knowledge?

Imagine you hold a winning lottery ticket in your hand, fully convinced you have won because a psychic told you so. The ticket is indeed a winner. You had a belief, and it was true, yet we would hesitate to say you *knew* you would win. This intuition reveals that propositional knowledge—traditionally defined as knowing *that* something is the case—is far more volatile than mere correct opinion. For centuries, philosophers rested on a deceptively simple tripartite definition inherited from Plato’s dialogues, specifically the *Theaetetus*. This classical analysis posits that propositional knowledge ($S$ knows that $p$) is **Justified True Belief (JTB)**. Under this framework, three individually necessary and jointly sufficient conditions must be met: 1. **Belief**: The subject must actually assent to the proposition ($S$ believes that $p$). 2. **Truth**: The proposition must correspond to objective reality ($p$ is true). 3. **Justification**: The subject must have good reasons or evidence for holding that belief. However, in 1963, a three-page paper by Edmund Gettier shattered this millennia-old consensus. In [Is Justified True Belief Knowledge?](https://analysis.oxfordjournals.org/content/23/6/121), Gettier presented counterexamples where individuals had justified true beliefs that were only true as a matter of sheer coincidence. > "Then, even if Smith is exceptionally strong, there is no sense in which he knows that the man who will get the job has ten coins in his pocket..." > — Edmund Gettier, *Is Justified True Belief Knowledge?* Gettier's cases demonstrated that justification can be decoupled from the actual truth-maker of a situation, rendering the "knowledge" a product of epistemic luck. This sparked a revolution in epistemology, forcing philosophers to search for a fourth condition to "Gettier-proof" the definition. Prominent contemporary attempts to redefine propositional knowledge include: - **No False Lemmas**: Championed by Michael Clark, this view suggests that a subject's justification cannot proceed through any false steps or assumptions. - **Sensitivity and Safety**: Tracking theories, such as those proposed by Robert Nozick in his 1981 book *Philosophical Explanations*, argue that for $S$ to know $p$, $S$'s belief must track the truth across counterfactual scenarios (i.e., if $p$ were not true, $S$ would not believe it). - **Virtue Epistemology**: Thinkers like Ernest Sosa argue that knowledge is "apt belief"—belief that is true *because* of the cognitive competence and intellectual virtues of the agent, rather than external luck. Despite these rigorous formulations, a consensus remains elusive, suggesting that our intuitive grasp of "knowing" resists simple reductionism. To explore this epistemic puzzle further, consider these questions: 1. If we cannot perfectly define knowledge, does the concept still hold practical value in scientific inquiry and legal systems? 2. How does [Virtue Epistemology](https://plato.stanford.edu/entries/epistemology-virtue/) shift our focus from evaluating the abstract properties of a belief to evaluating the character of the knower? 3. Can a machine learning algorithm, operating purely on statistical correlations, ever possess "justification" in the sense required for propositional knowledge?

Could you know something that is completely false, or believe something you have absolutely no reason to trust? For over two millennia, Western philosophy wagered that knowledge is not merely a subjective feeling, but a tripartite structure: for an agent to know a proposition, the proposition must be true, the agent must believe it, and the agent must be justified in holding that belief. This formulation, known as the Justified True Belief (JTB) account of knowledge, traces its pedigree to Plato's dialogue, the [*Theaetetus*](https://plato.stanford.edu/entries/plato-theaetetus/). In this text, Socrates famously distinguishes mere true opinion from knowledge by suggesting that true opinions must be bound "by a chain of explanatory reasoning" before they become stable. ## The Tripartite Blueprint To understand JTB, imagine an archway supported by three distinct pillars. If any pillar crumbles, the arch of knowledge collapses into mere opinion or luck. Under this classical framework, an agent ($S$) knows a proposition ($P$) if and only if: 1. **Truth**: $P$ is factually correct. You cannot "know" that the Earth is flat, because it is not. 2. **Belief**: $S$ actually accepts $P$ as true. You cannot "know" a historical date if you are merely guessing it on a test without believing your own answer. 3. **Justification**: $S$ has sufficient evidence, cognitive grounds, or reasons for holding $P$. If you buy a lottery ticket and firmly believe it will win, and by sheer coincidence it does, you did not *know* you would win. Your belief was true, but it lacked the cognitive pedigree of justification; it was merely a lucky guess. ## The Gettier Earthquake and Internalist Debates For centuries, this elegant definition was accepted as philosophical orthodoxy. However, in 1963, a three-page paper by Edmund Gettier titled [Is Justified True Belief Knowledge?](https://analysis.oxfordjournals.org/content/23/6/121.abstract) shattered this consensus. Gettier presented counterexamples where individuals had justified true beliefs that clearly did not constitute knowledge because the truth of the belief was a matter of pure coincidence. > "I shall begin by presenting two cases in which the conditions stated in (a) are true for some proposition... and yet it is equally clear that the person in question does not know that proposition." — Edmund Gettier, *Is Justified True Belief Knowledge?* Consider a modern variation: you look at a highly reliable clock in a town square that reads 12:00 PM, and you form the justified belief that it is noon. Unbeknownst to you, the clock stopped exactly twenty-four hours ago. By sheer luck, it happens to be exactly 12:00 PM when you look. Your belief is true, and you are justified in trusting the clock, yet intuitive consensus dictates you do not *know* the time. It was a "lucky" truth. ## The Battlegrounds of Justification The fallout from Gettier's paper forced epistemologists to interrogate what "justification" actually means, splitting the field into rival camps: - **Internalism**: Proponents argue that justification depends entirely on factors internal to the subject's mind. As Roderick Chisholm defends in [*Theory of Knowledge*](https://www.pearson.com/), an agent must have conscious access to the reasons that justify their belief. - **Externalism**: Externalists argue that justification depends on factors outside the agent's conscious awareness. Alvin Goldman pioneered [Reliabilism](https://plato.stanford.edu/entries/reliabilism/), suggesting a belief is justified if it is produced by a reliable cognitive process (like vision or memory), regardless of whether the agent can explain how that process works. These debates leave us with a lingering, profound question: can we ever truly close the gap between how we perceive the world and how the world actually is, or is the ideal of secure human knowledge permanently vulnerable to luck?

Can we ever truly close the gap between how we perceive the world and how the world actually is? While naive realism dismisses this question as a linguistic trick, the most sophisticated defense of the "impassable gap" reveals it to be an inescapable consequence of our evolutionary biology and neurocomputational architecture. To steel-man this position is to recognize that the gap is not a failure of human perception, but its very condition. We do not experience the mind-independent world directly; we experience a highly processed, survival-optimized simulation. ## The Neurocomputational Case for the Gap The strongest version of this argument does not rely on radical Cartesian skepticism, but on contemporary cognitive science and evolutionary game theory. In his work [*The Case Against Reality*](https://www.quantamagazine.org/the-evolutionary-argument-against-reality-20160421/), cognitive scientist Donald Hoffman mathematically demonstrates through evolutionary agent-based modeling that organisms tuned to perceive objective reality consistently go extinct when competing against organisms tuned solely for fitness payoff. Hoffman's "Interface Theory of Perception" argues that our senses act as a desktop interface, hiding the complex, underlying causal machinery of the universe behind user-friendly icons. Furthermore, in neuroscience, the paradigm of predictive processing, pioneered by figures like Andy Clark in [*Surfing Uncertainty*](https://academic.oup.com/book/7325), conceptualizes the brain as a "prediction machine." > "Action and perception are deeply integrated, both being driven by a single urge to minimize prediction error." Under this model, what we perceive as the external world is a top-down "controlled hallucination" constrained only by sparse, noisy sensory inputs. We do not passive-receptively "register" the world; we construct a highly mediated model of it. ## Deep Philosophical Foundations and Formidable Advocates This contemporary scientific view reconstructs the classic transcendental idealism of Immanuel Kant. In the [*Critique of Pure Reason*](https://plato.stanford.edu/entries/kant-spacetime/), Kant famously distinguished between the *phenomena* (the world as it appears to our structured minds) and the *noumena* (the "things-in-themselves" which lie forever beyond our sensory and conceptual apparatus). Kant argued that space, time, and causality are not external realities we discover, but the hardwired, intuitive templates of human cognition. To ask what the world is like outside of these templates is a category mistake: > "Thoughts without content are empty, intuitions without concepts are blind." ## Pre-empting the Sophisticated Objections A common objection is the "Success Argument" from scientific realism: if our perception is so radically divorced from reality, why are our scientific technologies so spectacularly successful? The sophisticated defender of the gap responds by separating *instrumental utility* from *ontological representation*. Newtonian physics is highly successful for building bridges, yet its fundamental ontology of absolute space and time is objectively false under Einsteinian relativity. Our perceptions and scientific instruments need only be structurally isomorphic to the world's causal relations to guarantee survival and utility; they do not need to resemble or reveal the intrinsic nature of the world itself. The gap remains structurally unspannable, yet functionally navigable.