Philosopher Jennifer Nagel (University of Toronto) argues that universal skepticism—the idea that we can’t truly know anything—is itself a cognitive illusion, and that knowledge is best understood not as a constructed proof but as a primitive, factive mental state: a state of mind you can only have toward the truth.
The Nature of Knowledge
Knowledge is a factive attitude: if someone knows something, it must be true. This distinguishes it from non-factive attitudes like believing, hoping, or being certain, which can be held toward falsehoods.
Example: “Roger knows he’s being followed” logically entails he actually is being followed; “Jane thinks she’s being followed” does not.
Knowledge is the “mother of all factive attitudes”—seeing, realizing, noticing, and remembering are all ways of knowing, and they all entail truth.
Humans are deeply sensitive to the knowing/not-knowing distinction in everyday life: we navigate it constantly in conversation, games, and social interaction. Natural languages universally mark this distinction with separate verbs for knowing versus thinking or believing.
The Skeptical Illusion
When people become hyper-self-conscious about knowledge—asking “but how do you really know?”—their ordinary confidence in what they know can evaporate. This is the engine of skepticism.
Skeptical arguments (e.g., “you could be dreaming,” “it might be a lookalike”) are dialectically powerful because they force the knower into an endless justificatory game. The skeptic never asserts anything, only questions, making them impossible to defeat in argument.
Nagel’s key insight: this skeptical doubt is a cognitive illusion, generated by specific psychological mechanisms, not a genuine insight into the impossibility of knowledge.
It arises especially under social pressure: when someone resists or challenges your claim, you instinctively shift into argument mode, trying to prove what you already know. If the challenger keeps resisting, you start to feel like you don’t know—even though you do.
This is a heuristic that usually serves us well (updating beliefs when others resist), but it can be hijacked by skeptical scenarios to produce spurious doubt.
Why Traditional Analyses of Knowledge Fail
Philosophers since Plato have tried to analyze knowledge as justified true belief (JTB) plus further conditions. This project has largely failed.
Gettier cases (1963) showed that justified true belief isn’t sufficient: a person can have a justified true belief by sheer luck (e.g., glancing at a broken clock that happens to show the right time) without genuinely knowing.
Proposed fixes—causal theories, reliabilism—have all faced counterexamples (e.g., the “fake barn” case: someone in fake barn county correctly identifies a real barn, but their belief-forming process is unreliable in that environment).
Contextualism (e.g., Gail Stein) tries to resolve the tension between everyday knowledge attributions and skeptical doubt by claiming “knows” shifts meaning across contexts—low standards on the street, high standards in the philosophy seminar. Nagel rejects this because knowledge should be portable across contexts; contextualism also can’t handle cross-context evaluation of knowledge claims.
Nagel follows Timothy Williamson’s knowledge-first approach: knowledge is primitive and bedrock, not reducible to simpler components. We should build our understanding of belief and other states out of knowledge, not the reverse.
Knowledge Attribution and Social Cognition
Our sense of who knows what is constantly being updated through conversational interaction. Sociologist John Heritage’s work shows that every conversational turn presupposes an epistemic imbalance: questions posit the other person knows something you don’t; statements invite the other to accept and update.
Backchannel responses (“oh,” “uh-huh,” “really?”) signal acceptance and mark the pooling of knowledge.
Resistance triggers argumentation—a universal human response to challenged claims.
This social machinery is what skeptics exploit: by refusing to accept even the simplest claim, they push your buttons and generate the feeling of not knowing.
Knowledge attribution is also shaped by epistemic territory: we accept claims from others when they fall within domains we recognize them as knowledgeable (e.g., you trust a waiter’s directions to the restroom without proof).
Knowledge Without Knowing That You Know
You can know something without knowing that you know it. This avoids an infinite regress (knowing → knowing that you know → knowing that you know that you know…).
Timothy Williamson’s anti-luminosity argument shows there are borderline cases where you know but aren’t in a position to know that you know, based on the intransitivity of indistinguishability (like color patches so close you can’t tell them apart, but over a series the difference becomes clear).
Example: you may know the easier route into a building (acting on that knowledge) without being able to articulate why you take that route.
The Self, Memory, and Knowledge
Nagel is skeptical of claims (e.g., Thomas Metzinger) that the self is an illusion. She sees the self as a real agent embedded in reality, with a self-conception that serves as a “dashboard” for navigating the world—useful but incomplete.
The self-other distinction is fundamental to sentient life: any creature capable of self-movement must distinguish self-caused from environment-caused sensory changes.
Social cognition is uniquely human in its strategic sophistication: we deliberately show things to others and ask for their commentary, actively pooling intelligence in ways no other species does.
Free Will
Nagel considers free will a profound mystery (not yet a solvable puzzle). She believes free will must involve knowing what you’re doing—mere randomness isn’t freedom worth having.
Understanding knowledge better may help illuminate free will, but the connection remains tenuous.
AI and Knowledge
Large language models exhibit behaviors that look like understanding and knowledge (e.g., grammar, reading comprehension). Nagel is open to attributing knowledge to them but notes a problem: our intuitive knowledge-attribution systems evolved for interacting with unified biological agents, while LLMs are better understood as instantiating shifting simulacra (personas) depending on context and prompting.
Maurice Shanahan’s view: an LLM conversation involves a superposition of possible characters that gets disambiguated over time. There may be no single fact of the matter about what the model “knows” in the way there is for a human.
Nagel is confident that LLMs genuinely know grammar and can conjugate verbs across languages—this is uncontroversial knowledge.
Truth
Nagel endorses a deflationary/disquotational theory of truth (following Tarski): “P is true” adds nothing beyond P itself. Truth is not a deep property to be reduced (like salt = sodium chloride); it’s a logical device for generalization.
There are no “cheesy hacks” for determining truth (e.g., consensus doesn’t make something true). You have to do the first-order research.
What Nagel Wants Students to Learn
Not certainty (a subjective feeling) but knowledge—including propositional knowledge of philosophical history and the ability to think independently.
Understanding is a species of knowledge: knowing why something is the case, not just that it is.
She values students finding their own philosophical voice, especially as AI presents genuinely new challenges that earlier generations couldn’t have anticipated.
Current Work
Nagel is finishing a book called Recognizing Knowledge: Intuitive and Reflective Epistemology, focusing on how knowledge functions in live conversational interaction and how big data from linguistics and sociology can illuminate the mechanics of knowledge attribution.
