The Regress Problem: Formal Analysis

Core Idea

The regress problem asks: what makes a belief justified? If justification requires further justified beliefs, and those require further justification, a vicious infinite regress threatens. Formal analysis models justification as a relation between beliefs in a graph: either beliefs form infinite chains (unjustified regress), cycles (coherentism), or ground out in fundamental beliefs (foundationalism). Each structure corresponds to a different epistemological theory.

Explainer

You know from studying justified true belief (JTB) that justification is one of the central requirements for knowledge. The regress problem emerges as soon as you ask what justification is made of. Suppose you believe that it will rain (B1). What justifies B1? Your belief that the forecast says rain (B2). What justifies B2? Your belief that you reliably read weather apps (B3). What justifies B3? And so on. Using the tools of first-order logic and graph theory, we can model this as a directed graph where nodes are beliefs and edges represent "is justified by." The regress problem is a structural question about what shapes this graph can and cannot take.

There are exactly four possible structures, each corresponding to a major epistemological position. First: the graph has an infinite chain — every belief is justified by some further belief, indefinitely. This is infinitism, defended by philosophers like Peter Klein. It is consistent but counterintuitive: it seems to require that you actually possess infinitely many justifying beliefs, which seems humanly impossible. Second: the graph contains cycles — belief A is justified by B, which is justified by C, which is justified by A. This is the structure of coherentism: your belief set is justified as a mutually supporting whole rather than by an external anchor. The challenge is that any sufficiently large coherent set, including false ones, can close into cycles, raising the isolation objection: a coherent web of fiction seems no more justified than a coherent web of truths.

Third: the graph grounds out — some nodes have no incoming edges, meaning some beliefs are basic beliefs that are justified without depending on other beliefs. This is foundationalism, the dominant view in the Western tradition from Descartes onward. The challenge is to explain what makes basic beliefs justified if not further beliefs. Foundationalists typically appeal to self-evidence (the belief that p is justified by the very content of p, as in "2+2=4") or to direct experience (perceptual beliefs are justified by the experience itself, not by inferential reasoning about it). Formal analysis clarifies what foundationalism requires: the basic beliefs must have *non-inferential* justification, meaning their node in the graph has no incoming justification edges but still has positive epistemic status.

The formal framing makes clear why the regress problem is structural rather than psychological. It's not about whether people actually trace all their justifications — of course they don't. It's about what the *logical structure* of a fully justified belief system would have to look like. Any such structure must either terminate, cycle, or extend infinitely; there are no other options. The question "what is the architecture of knowledge?" reduces to asking which of these three graph structures is epistemically legitimate. Each answer generates a different theory with different commitments about what occupies the terminal nodes, how coherence generates justification, or what infinite justification chains would require.