Reliabilism

Reliabilism is a family of theories in epistemology that ground epistemic justification — and in some versions, knowledge itself — in the reliability of the cognitive processes that produce beliefs. Where traditional accounts of justified true belief ask whether the agent has reasons for a belief, reliabilism asks whether the cognitive mechanism that generated the belief is the kind of mechanism that typically produces true beliefs. A belief formed by a reliable process is justified; a belief formed by an unreliable process is not, regardless of whether the agent can articulate why.

The canonical formulation is Alvin Goldman's process reliabilism (1979): a belief B is justified if and only if it is produced by a cognitive process that tends to produce true beliefs across the relevant range of conditions. Perception, memory, and deductive inference count as reliable; wishful thinking, horoscope-reading, and the gambler's fallacy do not.

Core Versions

Process reliabilism (Goldman 1979, 1986) is the foundational version. Justification tracks the truth-conduciveness of the psychological process — pattern recognition, analogical reasoning, logical inference — not the content of the belief or the agent's reflective access to it. This makes reliabilism an externalist theory: the justifying condition (process reliability) need not be accessible to the believer. An agent can have a justified belief without knowing why it is justified.

Indicator reliabilism (Alston 1988) shifts focus from cognitive processes to epistemic indicators — internal states that reliably correlate with truth. A perceptual experience of a red surface is an indicator of there being a red surface; the justification of there is a red surface derives from the reliability of that indicator relation.

Virtue reliabilism (Sosa 1991, Greco 2010) merges reliabilism with virtue epistemology. What justifies a belief is not merely that it was produced by a reliable process, but that it was produced by a reliable cognitive virtue of the agent — a stable, integrated epistemic disposition. This version aims to credit the agent rather than just the mechanism, addressing the intuition that justified belief is an achievement.

The Generality Problem

The most persistent technical objection to reliabilism is the generality problem (Conee and Feldman 1998): cognitive process types can be described at different levels of generality, and the reliability of a process type depends entirely on which description is chosen. A belief formed by visual perception is reliable at one grain; a belief formed by visual perception in low light at 20 meters may not be. There is no principled, non-arbitrary way to determine which description of a process is the relevant one for assessing reliability.

Reliabilists have proposed solutions — causal individuation, the processes the agent's cognitive architecture actually runs — but none has achieved consensus. The generality problem is not merely a technical puzzle; it reveals that reliability is a relation, not a property, and its two relata (the process and its reference class) are both underdetermined by the theory.

The New Evil Demon Problem

Descartes introduced the evil demon to challenge foundationalism. Reliabilism encounters its own version: if an agent is a perfect physical duplicate of a well-functioning human being but is deceived by a demon so that their reliable-seeming processes produce systematically false beliefs, reliabilist accounts deny that their beliefs are justified. Yet intuitively, the deceived agent is doing everything right — responding correctly to their evidence, reasoning coherently, forming beliefs in the same way the undeceived agent does.

This suggests that reliabilism captures something real — the connection between truth and justification — but misplaces the justificatory condition. What matters for justification, the objection runs, is not whether the process is reliable in the actual world but whether the agent is responding to their evidence appropriately. This is the intuition that drives internalism in epistemology — the view that justifying conditions must be accessible to the agent.

Reliabilism and Systems

Reliabilism's significance extends beyond individual cognition. Institutional epistemologists (Goldman 1999; Anderson 2011) have applied reliabilist frameworks to collective knowledge-producing systems: scientific peer review, prediction markets, legal testimony standards, and information aggregation mechanisms. In this extended sense, the question is not whether an individual's process is reliable but whether a system's process — its method of aggregating, filtering, and validating beliefs — reliably tracks truth.

This systems-level application is where reliabilism does its most useful work. Individual cognitive reliability is nearly impossible to measure directly; system-level reliability is at least in principle empirically tractable. Prediction markets can be back-tested. Peer review can be examined for reproducibility. Legal standards of evidence can be evaluated against conviction rates and exoneration records. The Scientific Method is, in this light, reliabilism operationalized at institutional scale.

Significance

Reliabilism is the dominant externalist theory of justification in contemporary analytic epistemology. It successfully explains why perception and deductive inference justify while superstition does not — not because the agent has superior reasons, but because the mechanisms have superior truth-track records. Its failure to resolve the generality problem, however, is not a minor technical gap. It is a structural limitation: reliabilism cannot specify what counts as a process without importing assumptions that the theory is supposed to ground. Any epistemology that cannot specify its own unit of analysis has not finished its work.

The deepest problem with reliabilism is not the evil demon or the generality problem — it is that 'reliable process' is defined relative to a reference class of conditions that the theory itself cannot select. Without a principled account of the relevant environment, reliabilism silently inherits its standards from the world it is trying to evaluate. It is a framework that works only when you already know what you want it to say.