First published on my legacy blog Principia Politica, based on an exam essay awarded full marks in the Cambridge Philosophy Pre-U (A Level equivalent) in 2018, ultimately contributing to highest mark in the world for the Philosophy & Theology Pre-U that year (losing only one mark in an essay on free will).
Are there secure foundations to knowledge? ‘Foundationalism’ is a theory of knowledge, or how we justify our beliefs. Foundationalists think all beliefs ultimately get their justification from a set of very basic beliefs or sensory states which are the ‘foundations’ of knowledge. Otherwise, there’d be no foundation, and to fully justify any claim we’d have to say ‘A is the case, because B, because C, because D’, and so on until infinity! Foundationalists think infinite justification is absurd, so they prefer to base knowledge claims upon a secure foundation – a bit like building a house on rock, not on sand…
So there ought to be secure foundations for knowledge. But it is difficult to see what these foundations really are. The general form of foundationalism seems to be acceptable, but it is difficult to point to the content of any particular theory. There are secure foundations out there, but we don’t really know what they are! Perhaps a moderate solution is needed: we might admit that Susan Haack is right in embracing ‘foundherentism’, which combines the ‘foundationalist’ insight with the ‘coherentist’ view that beliefs can justify each other without resting solely on a bedrock of ‘basic’ beliefs which don’t need further justification. Foundationalists think knowledge is like a pyramid. Coherentists think it’s like a web: not only does basic belief B justify derivative belief D, but D also can justify B and vice versa – within webs of reinforcing beliefs. Perhaps by ‘secure’ foundations we don’t mean an indubitable foundation but some kind of sound basis for knowledge which can be strengthened by coherence relations and, perhaps, reliable processes too. Contrary to Descartes’ assertion that we can all be dead certain of certain beliefs – e.g., ‘I exist’ and ‘God exists’ – maybe knowledge has somewhat weaker, but still significant, foundations. The alternative is the scepticism which results from rejecting basic beliefs.
To start with, let’s consider epistemic foundationalism – the view that there must be some foundations to knowledge, whatever they are. I embrace epistemic foundationalism because the alternatives are pretty bad. The argument for epistemic foundationalism is the so-called ‘regress argument’. Richard Fumerton uses the Principle of Inferential Justification, which states that: S’s belief that P on basis E is justified if (1) S’s belief that E is justified and (2) S’s belief that E makes probable P is itself justified. Let’s take (1) first, and return to (2) later.
For S’s belief that E to be justified, S must believe some other proposition – E1, say – which must also be justified. And so on. So we can see that part (1) of Fumerton’s simple principle leads us to regress: the justified belief that E1 depends on another proposition E2, then another proposition E3, and so on until infinity. According to the regress argument for the view that knowledge has foundations, a contradiction is generated. On the one hand, regress leads us into there being an infinity of prior justifications for any given belief. On the other hand, it seems implausible that there can exist such an infinite regress! How can finite minds ever comprehend infinite chains of reasons? What’s more, if the infinite chain were completed, it would no longer be infinite! The obvious way of resolving this dilemma is by embracing the claim that some knowledge is non-inferential – i.e., not inferred from any other belief. This non-inferential knowledge is therefore foundational. It either justifies itself or doesn’t need justification: these are the foundations you’re looking for! We don’t know exactly what they are, mind you. All that Fumerton’s argument proves is that there must be foundations of some kind – otherwise knowledge as we understand it is impossible.
But infinitists such as Peter Klein think infinite regress is perfectly reasonable: we can have infinite chains of justification that enable us to really have ‘knowledge’. He rejects one alternative – ‘coherentism’ – on the grounds that it leads to circularity. Coherentists say that S’s belief that P is justified if S’s whole system of beliefs is justified by the coherence of the relations between the beliefs in that system – like a spider’s web. But this suggests that it’s possible for P to justify itself – albeit indirectly – by a chain of reasons which loops back on P. But Klein thinks we should all accept the Principle of Avoiding Circularity: circular reasoning is unreasonable. So we should reject pure coherentism. (I’ll return to Haack’s modified version – ‘foundherentism’ – later.) Also, Klein rejects foundationalism, because he thinks we should all accept the Principle of Avoiding Arbitrariness, which stops knowledge from having arbitrary foundations that have no justification. In Klein’s view, to argue for the view that knowledge has secure foundations is like arguing for the view that one can simply believe something without having any good reason for believing it – so one foundationalist cannot challenge another if their arbitrary foundations differ! For instance, foundationalist Bertrand Russell thinks sense data – or sensory states, like a perception of a chair – are basic reasons for believing propositions such as ‘There’s a chair over there!’. But religious foundationalists like Alvin Plantinga think that a religious believer’s belief is God is foundational, or basic. But both these foundations might be arbitrary – after all, a religious believer’s so-called ‘basic’ belief that God exists is not likely to convince an atheist who has different ‘foundations’ to their knowledge. So if you want to avoid arbitrariness and circularity, Klein wants you to embrace infinitism!
Infinitism, however, isn’t really that convincing. Look back at condition (2) of Fumerton’s Principle of Inferential Justification: S’s first belief that P on basis E is justified IF S’s further belief, ‘E makes probable P’, is itself justified. Like (1), this leads to regress. If one’s belief that it will rain tomorrow (P) is justified by your belief that the Met Office said so (E) and your further belief that what the Met Office says is reliable (i.e., E makes probable P), one would also need a reason for thinking that reliability itself is a good reason for believing reports! You would need the further belief, which also requires justification, such as ‘The Met Office’s reliability is a justification for my believing its predictions’. But what is this the case? And so on, and so on. So (1) leads to regress, (2) leads to regress, and (2) also leads to an infinite number of infinitely long chains of reasons! Even if infinitism is not incoherent, it’s surely absurd. As Carl Ginet points out, ‘[i]nference cannot originate justification, it can only transfer it’ from one belief to the next. And as Jonathan Dancy argues, ‘[j]ustification by inference is conditional justification only’. To argue on the basis of inference from one belief to another presupposes that the inferential relation itself is justified – which at some point must require some kind of foundation, such as the basic belief that the inductive principle (inferring generalisations from observations) is valid. If one doesn’t have any foundations, one doesn’t have justification – infinitism cannot originate justification, it can only transfer it.
So what are the foundations? Laurence BonJour thinks logical beliefs like ‘anything deduced from a true proposition is true’ (Bertrand Russell’s example) are foundational. They are the basis for inference. This seems reasonable – without certain basic logical beliefs, in maths or in scientific principles, we can’t really go about making justifiable inferences about, well, anything. But what makes these beliefs justified? Well, if they’re basic, they don’t require further justification! But perhaps what we mean is not ‘they don’t require further justification at all‘, but ‘they don’t require further justification of the usual kind‘. They require special justification, which might still allow them to be foundational – or not requiring normal justification. They might be arrived at by reliable processes. As Carruthers (1992) argues, the means by which ‘one’s belief comes to be innate is reliable’ – or generated by a process which tends to generate true beliefs. Our cognitive processes must be reliable for epistemic foundations to be secure. So it seems even self-evidence principles like ‘The rules of addition are valid’ need to be based on reliable cognitive processes. They are still strictly foundational, but they’re not as secure as we once thought. Their security depends on processes outside of them. Beliefs derive their full warrant from external (cognitive) processes, not just internal justifications. Underneath the superficial security of foundational beliefs lie important reliable processes.
Another possible foundation is sense-data: e.g., the sense-datum of a table. We base our knowledge on such sense-data, which don’t need justifying. They’re just experiences which form the basis of our justified beliefs. But as BonJour notes, this kind of foundationalism is problematic. Either sense-data assert that something exists or they don’t. If they don’t, then how could they justify beliefs which do make such assertions? On the other hand, if a sense-datum of a table can assert that the table exists, then what makes this assertion justified? The assertive representational content itself – not the sense-data – is doing the justification! Sense-data can’t justify beliefs, because either they don’t have the necessary conceptual content to do so, or they implicitly contain beliefs about existence which are the real sources of justification. These implicit beliefs (e.g., the belief that ‘There’s a table there’ whenever we see, or have a sense-datum of, a table) surely need further justification. But this justification needs yet another justification, and so on, until we get to something basic. The alternative is infinite regress. What’s certain is that sense data are not the secure foundations we’re looking for.
Susan Haack wants us to embrace a synthesis. Foundations can’t be indubitable – or beyond doubt. But they can be secure enough to provide reasons for believing non-basic beliefs. We may not have indubitable knowledge in sense-datum-like ‘S-states’. But we can have some kind of knowledge in such states, mutually supported by cohering ‘C-evidence’ which ascribes S-states to a subject. Basic beliefs can be justified not just by themselves but at least in part by their coherence relations with other beliefs. Justification goes up (from basic beliefs to non-basic beliefs) and back the way down again (as non-basic beliefs cohere with one another, and support the basic beliefs on which they rely). Basic beliefs justify non-basic beliefs, but then non-basic beliefs justify each other through their coherence (consistency or mutual derivability). These non-basic beliefs can then lend further support to the basic beliefs – in a virtuous circle of foundherentist justification! A problem with Haack’s account is how basic beliefs can derive any justification at all from coherence with non-basic beliefs independently of the justification originated from basic beliefs. If basic beliefs have already done the justifying of non-basic beliefs, isn’t non-basic beliefs’ supporting basic beliefs just circular? Maybe, but maybe not: as Peirce once argued, these special chains of justification might be relations of ‘mutual support’, not simple ‘circularity’. But as an alternative to the extremes of logical and sense-datum foundationalism, Haack’s foundherentism is at least a promising alternative. As Haack puts it: foundherentism, ‘without sacrificing objectivity, acknowledges something of how complex and confusing evidence can be’. Foundherentism might someday solve the dilemma of foundationalism’s good form but poor content.
So it seems there must be some foundations for knowledge if our beliefs are to be justified. But these foundations might not be indubitable – they might rely on non-basic beliefs (as Haack claims) or on reliable cognitive processes (as Carruthers thinks). We can’t definitely point to particular foundations, and we can’t claim these foundations are totally secure. We can’t be certain that knowledge has secure foundations. But in order to have justification at all, it seems there must be some kind of foundation – however secure that foundation may be.
Image sourced from Creative Commons.