David Malet Armstrong's book Knowledge, Truth and Belief (1973, pp.150-61) contains an important analysis of the infinite regress of inferences - "reasons behind the reasons" - first noticed by Plato in the Theatetus (200D-201C).

Knowledge traditionally entails true belief, but true belief does not entail knowledge.

Knowledge is true belief plus some justification in the form of reasons or evidence. But that evidence must itself be knowledge, which in turn must be justified, leading to a regress.

Following some unpublished work of Gregory O'Hair, Armstrong identifies and diagrams several possible ways to escape Plato's regress, including:

• Skepticism - knowledge is impossible

• The regress is infinite but virtuous

• The regress is finite, but has no end (Coherence view)

• The regress ends in self-evident truths (Foundationalist view)

• Non-inferential credibility, such as direct sense perceptions

• Externalist theories (O'Hair is the source of the term "externalist")

• Causal view (Ramsey)

• Reliability view (Ramsey)

Armstrong is cited by Hilary Kornblith and other epistemologists as restoring interest in "externalist" justification of knowledge. Since Descartes, epistemology had been focused on "internalist" justifications.

Armstrong does not subscribe to traditional views of justifying true beliefs, but he cited "causal" and "reliabilist" theories as direct non-inferential validation of knowledge. Direct validation or justification avoids the problem of the infinite regress of inferences.

Causality and reliabilism also were not original with Armstrong. He referred to the 1929 work of Frank Ramsey. Today these ideas are primarily associated with the name of Alvin Goldman, who put forward both "causal" and "reliabilist" theories of justification for true beliefs.

Here is how Armstrong described "causal" and "reliabilist" views:

According to "Externalist" accounts of non-inferential knowledge, what makes a true non-inferential belief a case of knowledge is some natural relation which holds between the belief-state, Bap ['a believes p'], and the situation which makes the belief true. It is a matter of a certain relation holding between the believer and the world. It is important to notice that, unlike "Cartesian" and "Initial Credibility" theories, Externalist theories are regularly developed as theories of the nature of knowledge generally and not simply as theories of non-inferential knowledge. But they still have a peculiar importance in the case of non-inferential knowledge because they serve to solve the problem of the infinite regress.

Externalist theories may be further sub-divided into 'Causal' and `Reliability' theories.

6 (i) Causal theories. The central notion in causal theories may be illustrated by the simplest case. The suggestion is that Bap ['a believes p'] is a case of Kap ['a knows p'] if 'p' is true and, furthermore, the situation that makes 'p' true is causally responsible for the existence of the belief-state Bap. I not only believe, but know, that the room is rather hot. Now it is certainly the excessive heat of the room which has caused me to have this belief. This causal relation, it may then be suggested, is what makes my belief a case of knowledge.

the source for causal theories is Frank Ramsey (1929)

Ramsey's brief note on 'Knowledge', to be found among his 'Last Papers' in The Foundations of Mathematics, puts forward a causal view. A sophisticated recent version of a causal theory is to be found in 'A Causal Theory of Knowing' by Alvin I. Goldman (Goldman 1967).

Causal theories face two main types of difficulty. In the first place, even if we restrict ourselves to knowledge of particular matters of fact, not every case of knowledge is a case where the situation known is causally responsible for the existence of the belief. For instance, we appear to have some knowledge of the future. And even if all such knowledge is in practice inferential, non-inferential knowledge of the future (for example, that I will be ill tomorrow) seems to be an intelligible possibility. Yet it could hardly be held that my illness tomorrow causes my belief today that I will be ill tomorrow. Such cases can perhaps be dealt with by sophisticating the Causal analysis. In such a case, one could say, both the illness tomorrow and today's belief that I will be ill tomorrow have a common cause, for instance some condition of my body today which not only leads to illness but casts its shadow before by giving rise to the belief. (An 'early-warning' system.)

In the second place, and much more seriously, cases can be envisaged where the situation that makes 'p' true gives rise to Bap, but we would not want to say that A knew that p. Suppose, for instance, that A is in a hypersensitive and deranged state, so that almost any considerable sensory stimulus causes him to believe that there is a sound of a certain sort in his immediate environment. Now suppose that, on a particular occasion, the considerable sensory stimulus which produces that belief is, in fact, a sound of just that sort in his immediate environment. Here the p-situation produces Bap, but we would not want to say that it was a case of knowledge.

I believe that such cases can be excluded only by filling out the Causal Analysis with a Reliability condition. But once this is done, I think it turns out that the Causal part of the analysis becomes redundant, and that the Reliability condition is sufficient by itself for giving an account of non-inferential (and inferential) knowledge.

6 (ii) Reliability theories. The second 'Externalist' approach is in terms of the empirical reliability of the belief involved. Knowledge is empirically reliable belief. Since the next chapter will be devoted to a defence of a form of the Reliability view, it will be only courteous to indicate the major precursors of this sort of view which I am acquainted with.

Ramsey is the source for reliabilist views as well

Once again, Ramsey is the pioneer. The paper 'Knowledge', already mentioned, combines elements of the Causal and the Reliability view. There followed John Watling's 'Inference from the Known to the Unknown' (Watling 1954), which first converted me to a Reliability view. Since then there has been Brian Skyrms' very difficult paper 'The Explication of "X knows that p" ' (Skyrms 1967), and Peter Unger's 'An Analysis of Factual Knowledge' (Unger 1968), both of which appear to defend versions of the Reliability view. There is also my own first version in Chapter Nine of A Materialist Theory of the Mind. A still more recent paper, which I think can be said to put forward a Reliability view, and which in any case anticipates a number of the results I arrive at in this Part, is Fred Dretske's 'Conclusive Reasons' (Dretske 1971).

The terms "internalism" and "externalism" are used in philosophy in a variety of different senses, but their use in epistemology for anything like the positions which are the focus of this book dates to 1973. More precisely, the word "externalism" was introduced in print by David Armstrong' in his book Belief; Truth and Knowledge' in the following way:

According to "Externalist" accounts of non-inferential knowledge, what makes a true non-inferential belief a case of knowledge is some natural relation which holds between the belief-state, Bap ['a believes p'], and the situation which makes the belief true. It is a matter of a certain relation holding between the believer and the world. It is important to notice that, unlike "Cartesian" and "Initial Credibility" theories, Externalist theories are regularly developed as theories of the nature of knowledge generally and not simply as theories of non-inferential knowledge. (Belief, Truth and Knowledge, p.157)

So in Armstrong's usage, "externalism" is a view about knowledge, and it is the view that when a person knows that a particular claim p is true, there is some sort of "natural relation" which holds between that person's belief that p and the world. One such view, suggested in 1967 by Alvin Goldman, was the Causal Theory of Knowledge. On this view, a person knows that p (for example, that it's raining) when that person's belief that p was caused by the fact that p. A related view, championed by Armstrong and later by Goldman as well, is the a href="/knowledge/reliabilism.html">Reliability Account of Knowledge, according to which a person knows that p when that person's belief is both true and, in some sense, reliable: on some views, the belief must be a reliable indicator that p; on others, the belief must be produced by a reliable process, that is, one that tends to produce true beliefs. Frank Ramsey was a pioneer in defending a reliability account of knowledge. Particularly influential work in developing such an account was also done by Brian Skyrms, Peter Unger, and Fred Dretske.

Accounts of knowledge which are externalist in Armstrong's sense mark an important break with tradition, according to which knowledge is a kind of justified, true belief. On traditional accounts, in part because justification is an essential ingredient in knowledge, a central task of epistemology is to give an account of what justification consists in. And, according to tradition, what is required for a person to be justified in holding a belief is for that person to have a certain justification for the belief, where having a justification is typically identified with being in a position, in some relevant sense, to produce an appropriate argument for the belief in question. What is distinctive about externalist accounts of knowledge, as Armstrong saw it, was that they do not require justification, at least in the traditional sense. Knowledge merely requires having a true belief which is appropriately connected with the world.

But while Armstrong's way of viewing reliability accounts of knowledge has them rejecting the view that knowledge requires justified true belief, Alvin Goldman came to offer quite a different way of viewing the import of reliability theories: in 1979, Goldman suggested that instead of seeing reliability accounts as rejecting the claim that knowledge requires justified true belief, we should instead embrace an account which identifies justified belief with reliably produced belief. Reliability theories of knowledge, on this way of understanding them, offer a non-traditional account of what is required for a belief to be justified. This paper of Goldman's, and his subsequent extended development of the idea, have been at the center of epistemological discussion ever since.