Musings on Gettier and the definition of knowledge

This article first appeared on Scientia Salon

Philosophers have traditionally defined “knowledge” as a belief that is both true and justified, a definition that sufficed until, 50 years ago, Edmund Gettier pointed out that the conditions could be fulfilled by accident, in ways that didn’t amount to what we would intuitively regard as “knowledge”.

Gettier pointed to scenarios such as:

“Smith has applied for a job, but, it is claimed, has a justified belief that “Jones will get the job”. He also has a justified belief that “Jones has 10 coins in his pocket”. Smith therefore (justifiably) concludes that “the man who will get the job has 10 coins in his pocket”. In fact, Jones does not get the job. Instead, Smith does. However, as it happens, Smith (unknowingly and by sheer chance) also had 10 coins in his pocket. So his belief that “the man who will get the job has 10 coins in his pocket” was justified and true. But it does not appear to be knowledge.”

Since Gettier’s paper many attempts have been made to patch up the “justified true belief” definition, often by adding extra conditions aimed at ruling out being right accidentally, though none of the proposed solutions has gained general acceptance and most have been shown not to work. [1]

Analysis of “Gettier problems” usually focuses on the word “justified”, which is often taken to be a binary condition, either fulfilled or not fulfilled. In Gettier’s paper, for instance, beliefs are taken as “justified” if there is “strong evidence” for them, but the idea of justification is not examined further [2]. Yet, in the real world, knowledge, evidence and justifications for claims are always probabilistic.

If “justification” were taken to be infallible — justification so bullet proof that the belief cannot then be in error — then there are no Gettier problems (Smith cannot have a “justified” belief that Jones will get the job unless Jones does get the job).

But, in the absence of human infallibility, “justification” (“the action of showing something to be right or reasonable” — Oxford Dictionaries) should mean “sufficient justification”, which we can take as justification that gives a greater than 90 per cent chance of being correct (or whatever threshold we choose). Now, suppose that Smith is “justified” with 90 per cent reliability in believing that Jones will get the job, and also justified with 90 per cent reliability in believing that Jones has 10 coins in his pocket. Then he has (under these definitions) “knowlege” that Jones will get the job and “knowledge” that Jones has 10 coins in his pocket, but does not have knowledge that “the man who will get the job has 10 coins in his pocket” since the reliability of that claim is 0.9 x 0.9 = 0.8, which falls below the threshold.

That in itself does not avoid Gettier problems (since those probabilities could each have been 0.95, which does multiply to a number above the threshold), but it does suggest that the threshold nature of “justification” is central to the issue.


If we accept that “Fred is 90 per cent justified in believing truth X” counts as “knowledge of X” then we are accepting that the justification could be faulty, and thus be unrelated to the truth of the matter. But, if so, the belief can always be true entirely by accident. Thus it follows that a “justified belief” can be true entirely by accident. There is no way round that, it is inevitable for any threshold for “justification” less than 100 per cent reliability [3].

In response, some have simply declared that, yes, belief that is both accidentally true, and accompanied by an irrelevant and erroneous justification, really is “knowledge”. But then we would be left with a counter-intuitive interpretation of “knowledge” that amounts only to being right, even if accidentally right. No-one would accept that someone knew what the winning lottery numbers would be, just because they happened to have bought the winning ticket.

For that reason, much discussion of Gettier has attempted to add a fourth condition, designed to prevent Gettier problems, to the trinity of justification, truth and belief. These usually try to avoid being accidentally right by bolstering the concept of justification (for example, the justification must not rely on falsehoods; or it must derive from the truth). [4]

None of these solutions solves the issue. To see this simply consider the justification, J, together with the additional condition and call the combination J+. Now either J+ is fallible or it isn’t. If it is fallible then we still have Gettier problems; indeed all we’ve done is re-label J as J+, since we didn’t specify what was necessary for J in the first place, and at most we’ve elucidated aspects of justification. Alternatively, if J+ is infallible then, fine, we’ve avoided Gettier problems, but we might just as well have declared J infallible at the start.

The above has been discussed at length and perhaps a consensus is emerging that there is no way round the Gettier problem, at least not without declaring justification to be infallible [5]. Yet, some are reluctant to accept that, which is perhaps strange since there is no reluctance to require the “truth” part of the definition to be absolute. Indeed, one can regard the extensive literature on Gettier as a search for a wording that stops short of adopting an infallible version of justification, while ensuring that the justification is never erroneous (because if it ever is, then in can leap Gettier).

And thus, it seems to me, that the really crucial part of the “justified true belief” definition of “knowledge” is that it is a belief that is true. This highlights a weird feature of the definition, that in order to know whether something is “knowledge” you first need to know whether it is true, and you can only know that if you already have that knowledge.

Thus the definition is not very useful: we can never use it to determine whether we have knowledge. It relies on having a “god’s eye” view where one already knows the full facts of the matter, else it cannot be applied. But, if one does have that view, then one might as well go the whole hog and declare that only infallible justification counts as justification, and thus solve Gettier’s problem.

Alternatively, if we don’t adopt the god’s-eye view, then we need to accept that, not only will our justifications be fallible, but also that what we regard as “true” might be erroneous, and thus we need reliability thresholds on both the “justified” and the “true” components of the definition.

Of course scientists have long accepted that they can never attain absolute truth about the world, but only approach it with ever-increasing, though not-total, reliability. Philosophy, however, can be considered to be a discussion about a broader “conceptual space”, in which truth could indeed be conceived of as securely known. [6]

So let’s declare this article to be about “knowledge” in and of the real world, and indeed discussion of Gettier is usually in terms of real-world examples (people with coins in their pocket, robotic dogs, red-painted barns, and so forth), however contrived.

On that point, and to keep this article to theme, let’s leave aside axiom-based systems such as mathematics and logic, where one might “know” something owing to it being entailed by axioms. If might be thought that within such systems one can have absolutely reliable knowledge, though even there one would be reliant on mathematicians and logicians having made correct deductions, and, without a proof of human infallibility, that is again not absolutely certain.

Is there an alternative definition of “knowledge”? Oxford Dictionaries defines the word as:

Knowledge: Facts, information, and skills acquired through experience or education; the theoretical or practical understanding of a subject.

That is a bottom-up approach to defining knowledge, starting with information and building on it, rather than the top-down approach of starting with truth. As a scientist, the bottom-up approach seems to me in keeping with how we actually gain knowledge, by acquiring information about the world around us and interpreting it, and then building theories that best model the world. We don’t start with truth, instead we build towards truth — that’s the best we can do — and we acquire knowledge as we go.

To improve the “justified true belief” definition we should thus omit the condition “true”, since without God’s eye we cannot know what is true, and can only evaluate the probability of truth based on the justifications we have. Thus the “true” condition is redundant with the “justification” condition.

That leaves us with knowledge as “justified belief”. We can then ask what counts as “justification”, for which the best answer — if we’re talking about knowledge about the real world — is provided by the methods adopted by science. After all, the methods of science have been developed and honed over the centuries as giving our surest path towards truth about the world around us (demonstrably so, given that the engineering and technology based on science does actually work, thus showing that scientific models are a pretty good match to reality).

Is this definition vulnerable to Gettier? Yes it is, since it accepts that our evidence and justifications are fallible, deriving as they do from only a very limited sampling of the world, and yet a claim might always be true entirely by chance. Indeed, the field of statistics helps us to evaluate what role chance will be playing. The answers it gives are not certain but they are often good enough.

Thus Gettier points to an inevitable limitation on real-world knowledge, deriving from the fact that evidence will never establish anything with absolute reliability (though if one doesn’t like the messiness of the real world one can instead declare oneself omniscient, both in judging truth and in evaluating justifications, and from that pulpit bestow laurels on Gettier-proof knowledge).


[1] Here is the obligatory link to the Stanford Encyclopedia of Philosophy on the topic.

[2] The original paper by Gettier is online here.

[3] This has been argued by several people, for example see Ian M. Church (2013), European Journal of Philosophy, 21 (1):37–49, and citations therein.

[4] A bibliography of Gettier-problem papers is given here.

[5] See, for example, this article by Fred Dretske.

[6] For example Massimo Pigliucci explains that: “You can think of philosophy as an exploration of conceptual, as opposed to empirical, space …” where “conceptual space is much broader than its empirical counterpart”.

3 thoughts on “Musings on Gettier and the definition of knowledge

  1. Anton Szautner

    My mind contains a model that finds the concept of ‘belief’ highly suspicious. I wonder if it is ‘justified’. I don’t understand why I should declare it a ‘belief’ even if I find any validity in it. What advantage or benefit would that confer?

    I must read this post carefully again while setting that distracting objection temporarily aside, and check into the references… 8)

  2. Henrik Dalare

    I understand knowledge to mean that you know something for certain. That doesn’t make knowledge unattainable, because we can know a lot for certain about our uncertainty.

    So this is the knowledge Smith actually had: The probability that the man who gets the job has 10 coins in his pocket is high. And this is the knowledge Smith lacked: The man who gets the job has 10 coins in his pocket.

    Thank you for the interesting read (although it was way above my level)!

  3. verbosestoic

    I probably should read this blog more often, although it’s good to find a blog that updates even less regularly than mine does [grin].

    Thus Gettier points to an inevitable limitation on real-world knowledge, deriving from the fact that evidence will never establish anything with absolute reliability (though if one doesn’t like the messiness of the real world one can instead declare oneself omniscient, both in judging truth and in evaluating justifications, and from that pulpit bestow laurels on Gettier-proof knowledge).

    Actually, that’s not really the issue here, as we already thought that we didn’t need certainty to have knowledge (see reliablism, for example). I originally thought the same thing as you did about the “p is true” part of the definition, but was explained that I was confusing knowledge with knowledge of knowledge; in short, if those conditions are met then I know that p, but that doesn’t itself entail that I know that I know that p, or second order knowledge. To know that I know that p, I’d need to know the proposition “I know that p”, which itself requires justification and to be true. To move up to the next level, I’d have to justify that, and so on. But these are indeed different propositions, so it works out, and so you don’t have to insist that you fill out the whole chain in order to claim to know that p; you can stop at any point higher than that and say that you don’t know if that proposition is true or not, because they are related but independent propositions, and have independent truth values.

    But keeping truth in saves us from a similar problem in the Gettier examples, that I don’t think that you can get away from. No matter what we use for justification, it is possible — if justification is not just “I am logically certain that p” — for me to be justified in believing something that because of other factors turns out to be false. So what I have is a justified, false belief. Under the definition of knowledge that says that to be knowledge we must have a justified true belief, what we’d say in that situation is that I have a belief, the belief was justified, and yet because it was false I didn’t know it. I THOUGHT that I knew it — because of the justification — but I was wrong about that. Under your system, unless you do provide certainty and so never have a case where I can be justified in believing a false system, you’d have no way of saying that I DIDN’T know it then, which leads to the case that I knew at time X something and yet at time Y I know a completely incompatible proposition, which is at least rather odd. And, in fact, depending on how justification shakes out it’s possible, under your definition, to be justified in knowing A and ~A simultaneously, which makes no sense whatsoever … whereas with justified true belief since both couldn’t be true I couldn’t really claim to know both, as at least one of them had to be false.

    So, the truth condition (heh) saves us from these cases. Gettier problems are based around the idea that, yeah, that’s fine when the proposition turns out to be false, but what if it happens to be true? You have a justification, but that justification is irrelevant to the ACTUAL truth value of that proposition; the proposition is true in spite of that justification, not because of it, because the justification fails. So what you have is a justification that doesn’t actually justify your belief that p — because the justification TURNED OUT to be wrong — and yet p is indeed true … and your belief by all reasonable standards was justified, but turned out to be false itself. This means that either you WEREN’T really justified and we have to find a way to explain how, or else you really did know that p despite the fact that your reasons for believing that p turned out to be wrong.

    That’s the issue, and that’s where your view doesn’t really help, because it doesn’t find a way to say that you weren’t justified in that case, and so would have to conclude that you really knew something when your reasoning for why it was true was completely false … but was false for reasons that you couldn’t know at the time … on TOP of now re-opening the issue for PROPOSITIONS that turned out to be false.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s