Ask why something is true, and you will receive a reason. Ask why that reason is true, and you will receive another. Keep asking, and one of three things will happen: the reasons will loop back on themselves, or they will go on forever, or they will eventually stop at something that is simply asserted without further justification. None of these outcomes is satisfying. All of them are unavoidable. This is the regress problem, and it has been waiting at the bottom of every confident belief you have ever held.
I was eight or nine years old when I first walked into it, though I would not have been able to name it.
My father had told me something — I no longer remember what, some fact about the world that struck me as requiring explanation — and I asked why it was true. He told me. I asked why that was true. He told me again, and I asked again, and at some point his answers became shorter and his patience became thinner, and then he said something that I have thought about ever since, though I have never been entirely sure he knew what he was saying: "Because that's just how it is."
I remember the feeling of that answer. Not the frustration — I was already past frustration — but something more like vertigo. The chain of reasons had ended. Not because the question had been answered, but because the answering had stopped. There was a place where the explanations ran out, and what lived there was not an explanation at all but an assertion: this is simply so, and we do not go further.
I did not know, at eight or nine, that this is one of the oldest problems in philosophy. I did not know it had a name. I did not know that Aristotle had worried about it, or that a set of arguments attributed to the ancient skeptic Agrippa had formalised it into a trilemma that no one has cleanly escaped in the two thousand years since. I just knew the feeling: the ground giving way beneath the last explanation, and nothing below it but air.
The Trilemma
The Agrippan trilemma — sometimes called the problem of the criterion, sometimes the epistemic regress problem — is simple to state and impossible to resolve cleanly. It begins from a question: what makes a belief justified?
The answer that first suggests itself is: a belief is justified when it is supported by evidence or reasons. This is plausible. It is also the beginning of the problem. Because if justification requires reasons, then those reasons must themselves be justified. And the justification of those reasons requires further reasons. And so on.
At this point, three exits seem available. The first is to follow the chain backward indefinitely — to accept that justification requires an infinite regress of reasons, each supported by further reasons, without any foundation on which the chain rests. The second is to allow the chain to loop — to permit circular justification, where A is justified by B and B is justified by A, and the circle provides the stability that the chain lacks. The third is to stop the chain at some point — to posit a set of beliefs that are justified not by further reasons but by something else: their own self-evidence, or the directness with which they are perceived, or simply their indispensability to any reasoning at all.
These three exits correspond to the three major positions in epistemology: infinitism, coherentism, and foundationalism respectively. None of them is clean. This is not a matter of insufficient philosophical ingenuity. It is, I think, a feature of the problem — a sign that the problem is pointing at something real about the structure of justification rather than a puzzle that awaits the right clever solution.
Foundationalism and Its Difficulties
Foundationalism is the most intuitively appealing of the three positions, and it is the one that most people implicitly hold without knowing they hold it. The idea is that knowledge rests on foundations: a set of basic beliefs that are justified not by further beliefs but by something more immediate — perceptual experience, rational intuition, the self-evident deliverances of consciousness.
Descartes' cogito is the most famous attempt to identify such a foundation: I think, therefore I am. Whatever else might be doubted, Descartes argued, the existence of a doubting subject cannot be doubted by that subject, because the doubting is itself evidence of the existence. Here, he believed, was bedrock: a belief that justified itself by the very act of being held, that could not be undermined by further questioning because the questioning itself confirmed it.
The appeal of this move is real. The difficulty is equally real. Even granting the cogito as bedrock, the structure built on it requires bridging principles — moves from the certainty of my own existence to the reliability of my faculties, to the existence of the external world, to the trustworthiness of my perceptions — that are considerably less certain than the foundation on which they are supposed to rest. Descartes' famous solution to this problem — the proof that God exists and would not deceive us — has not found many defenders in the centuries since, and for reasons that are not difficult to identify.
More generally, foundationalism faces the problem of identifying what the foundations actually are. The candidates that have been proposed — sense-data, incorrigible beliefs about one's own mental states, basic logical axioms, perceptual experiences — have each faced serious objections. Sense-data seem theoretically loaded; what counts as a basic perception is not independent of the framework through which it is observed. Beliefs about one's own mental states are less incorrigible than they appear; we are frequently wrong about what we are feeling, what we intended, what we perceived. Logical axioms seem self-evident until you try to say what makes them so, at which point the same regress threatens to reappear.
Coherentism and Its Difficulties
Coherentism is the position that beliefs are justified not by being grounded in foundations but by their coherence with one another — by the mutual support that members of a belief system provide for each other. No individual belief is justified in isolation; each belief is justified by its fit within the larger web of beliefs of which it is a part.
There is something right about this. The picture of knowledge it offers is organic rather than architectural: not a building resting on foundations, but a web whose strength comes from the density and consistency of its connections, and whose stability does not depend on any single point of anchorage. This seems closer to how knowledge actually works in practice — we do not, in the course of ordinary inquiry, trace each belief back to its foundations. We evaluate beliefs against other beliefs, against our background understanding, against the general coherence of our overall picture of things.
But coherentism faces a problem that is difficult to dismiss: coherence does not guarantee truth. A very coherent system of beliefs can be thoroughly disconnected from reality. A well-constructed novel is internally coherent. A conspiracy theory can be made internally coherent by appropriate selection of evidence. Internal consistency is a constraint on a good belief system, but it is not sufficient for one. Coherentism, taken strictly, seems to permit any sufficiently coherent system to count as knowledge — which is precisely the result we were hoping to avoid.
The coherentist can respond by adding further constraints: the belief system must cohere with perceptual experience, or it must be the most coherent system available given the evidence, or some such. But these additions either reintroduce something like foundations — the deliverances of perception as anchors — or they require further argument about what "most coherent given the evidence" means, which risks sending us back to the regress.
Infinitism and Its Difficulties
Infinitism — the view that justification can proceed through an infinite chain of non-repeating reasons — has fewer defenders than the other two positions, perhaps because it is the most counterintuitive. The idea that any belief I hold could be justified by an infinite series of reasons, none of which I have explicitly articulated, seems to describe a situation no actual reasoner is ever in.
Peter Klein, the philosopher most associated with contemporary infinitism, argues that the potential availability of further reasons is what matters, not the actual articulation of all of them. A belief is justified if there exists an infinite chain of reasons that could, in principle, support it. Whether any reasoner has traced through that chain is irrelevant.
This is a subtle position, but it faces the difficulty that it disconnects justification from anything that actual reasoners can do. If I cannot access the infinite chain — if I have only ever explicitly entertained a finite number of reasons — then the infinite chain is doing the justificatory work without my being able to use it, evaluate it, or in any ordinary sense possess it. Justification becomes a property of belief systems in the abstract rather than a feature of what real epistemic agents actually have when they know things.
Wittgenstein's Different Answer
The most interesting response to the regress problem, I think, is the one that refuses to accept the trilemma's terms entirely. This is Wittgenstein's move in On Certainty, the set of remarks he wrote in the last months of his life, focused explicitly on the question of where justification ends.
Wittgenstein's argument is not that we have foundations, or that we have coherence, or that the regress can go on forever. His argument is that the demand for justification is itself embedded in a practice — in a form of life — and that at some point, the demand for further justification simply does not make sense within that practice. "If I want the door to turn," he writes, "the hinges must stay put." Some beliefs — he calls them hinge propositions — are not justified within a system of beliefs because they are the presuppositions that make the system possible. They are not beliefs we hold because we have reasons for them; they are the background against which reasons are given and evaluated at all.
The earth has existed for a very long time. I have two hands. The world was not created five minutes ago with the appearance of age. These are not things I have evidence for in the normal sense; they are things that function as the hinge on which all evidence-evaluation turns. To demand justification for them is not to ask a harder question than ordinary questions; it is to step outside the practice of asking questions at all.
This is not foundationalism, exactly. Hinge propositions are not justified by their self-evidence; they are not even straightforwardly justified at all. They are held in a different way — more like the way a practice is held than the way a belief is held. And they can shift, over time and between communities, in ways that foundations cannot.
"At the end of reasons," Wittgenstein writes, "comes persuasion." And before that, perhaps most honestly: "If I have exhausted the justifications I have reached bedrock, and my spade is turned. Then I am inclined to say: 'This is simply what I do.'"
I find this answer simultaneously more honest and more unsettling than the alternatives. It does not solve the regress. It relocates it — to the edge of our practices, to the place where what we do simply is what we do, where further reasons are not available because the practice of giving reasons has to start somewhere, and this is where it starts.
A Personal Encounter With the Problem
I teach — or taught, for a period — and I remember a student who had the particular gift, or perhaps affliction, of following arguments all the way down. We were discussing the justification of induction: why we are entitled to assume that the future will resemble the past, that the patterns we observe will continue, that the sun will rise tomorrow because it has risen every previous day.
David Hume, I explained, had shown that we cannot justify this assumption inductively — that would be circular — and cannot justify it deductively either, because the future's resemblance to the past is a contingent fact about the world rather than a logical necessity. So what justifies it?
The student waited a moment. Then she asked: "And what justifies that justification?"
She was not being difficult. She had understood the problem. And I found, to my genuine discomfort, that I did not have a clean answer. I had responses — Wittgenstein's response, pragmatist responses, reliabilist responses — but none of them was a clean answer, in the sense of an account of justification that did not itself require further justification somewhere below.
I told her this. I told her that the problem does not have a clean solution, and that this is worth knowing. She looked at me with the expression of someone who had been promised a destination and been told, instead, that the map was incomplete. It is an expression I have become more sympathetic to over time, because I have felt it on my own face.
Why This Matters
The regress problem is not merely an academic puzzle for epistemologists to debate in journals. It is a feature of every belief you hold, operating at whatever depth you have not yet examined. Most beliefs are never examined to this depth; they function perfectly well without being traced to their foundations, if they have any.
But the regress problem matters in practice in at least two ways.
The first is the way it should affect your confidence. A belief that rests on explicit reasons that rest on explicit reasons is more secure than a belief that rests on a single explicit reason. But even a chain of many reasons eventually rests on something — either an assertion, a circle, or a point where the chain simply stops. Knowing this should produce a specific kind of epistemic humility: not the paralysing doubt that the skeptics who exploited the regress problem were after, but the recognition that even the most carefully reasoned beliefs rest, at some depth, on things that are held rather than justified, that are presupposed rather than proven.
The second is more practical. When someone presents an argument you find compelling, the chain of reasons they are offering has a bottom. The question of what that bottom is, and whether it is one you share, is often more important than whether each step in the chain is valid. Two people can reason correctly from different starting points and reach irreconcilable conclusions without either of them having made an error. This is not relativism; it is the observation that the regress problem applies to everyone, and that the places where justification stops are often not the places where they should be most confident they have stopped correctly.
Living Without a Foundation
There is a particular kind of intellectual comfort that the regress problem threatens, and I think it is worth naming it directly.
The comfort is the sense that your most fundamental beliefs are secure — that they are not merely held but justified, not merely asserted but proven, not merely where the chain happens to stop but where it stops for good reason. The regress problem does not say that this comfort is always misplaced. It says that it requires more scrutiny than it usually receives.
Wittgenstein's spade turns. At some point, reasons run out. What lives below the last reason is not nothing — it is a practice, a form of life, a set of presuppositions that make the practice of reasoning possible. But it is not justification in the normal sense of the word. And knowing this — holding it clearly rather than flinching away from it — changes something about the way you carry your beliefs.
Not toward paralysis. Not toward the corrosive skepticism that says, because every justification eventually runs out, nothing is justified. But toward a specific kind of honesty about where the ground actually is, and what is below it — which is, more often than we admit, something we have not chosen to examine, and would find uncomfortable if we did.
The regress problem is not a problem that philosophy failed to solve. It is a problem that philosophy has spent two thousand years understanding more precisely. The precision is the progress. And the honest relationship to it is not resolution but recognition: every chain of reasons ends somewhere, the ending is not itself a reason, and the willingness to look at where your chains end — rather than stopping before you reach them — is among the most important intellectual practices available.