2. An analogy that might support an analogue of Sosa's view

There are, of course, different ways of relating concepts of justification. Consider first an analogy. I have argued elsewhere (Fumerton Reference Fumerton1990) that one might do well to recognize different concepts of what one ought to do based on the fact that prior to action we have a limited amount of time to even consider all of the reasons there might be for us to choose from among alternatives open to us. So consider first a crude version of what we might call actual consequence act consequentialism. On this view one ought to do whatever maximizes value.Footnote ³ A natural worry about such a view is that if it were true, one could never figure out what course of action one ought to take. The consequences of actions go on and on and however prescient one might be, the distant future will be hidden in epistemic darkness. There are, I have argued, other good reasons to reject a view that defines what one ought to do in terms of actual consequences, but I'm not sure that the epistemic concerns are among them. And that's because one can develop derivative concepts of what one ought to do from our base concept. One can predict at least some of the consequences of alternatives open to one, and based on the evidence one has been able to consider, one of the alternatives can take the “lead” in terms of what one is justified in believing to be the value produced. Assuming that we have no justified beliefs about other consequences, that alternative will have the “best chance” (relative to our evidence) of being what we ought to do in our primary sense. In general, among alternative hypotheses, one hypothesis can have the best chance of being true, without its having a very good chance of being true. Indeed, it might have the best chance of being true even if it is still exceedingly unlikely that it is true. Still, isn't it plausible to suppose that if among the alternatives open to me one alternative has the best chance of being what I ought to do, that is what I ought to do? If we reach that conclusion, must we abandon our initial effort at capturing the truth conditions for a claim about what I ought to do? Perhaps. But there is an alternative. One might instead simply make a distinction between “ought”s. There is what I ought¹ to do – that's the ought defined in terms of maximizing value. And then there is what I ought² to do – that's an ought defined in terms of what has the best chance of being what I ought¹ to do.

Once we go down this road there is no reason to stop. There are indefinitely many meta-levels to which one might move to create both potentially weaker and potentially stronger concepts of what I ought to do. It seems at least possible to think about what has the best chance of being what I ought² to do. And that will allow us to recognize what I ought³ to do. And so on ad infinitum. One would rarely, if ever, deploy anything beyond the second level “ought” judgment, but there is nothing to stop us from recognizing the possibility of doing so.

I tried to develop the idea with the crudest account of what one ought¹ to do. But I suspect that the problem to which I am suggesting a solution will develop on virtually any account of what the fundamental “ought” is. So I'm personally sympathetic to the view that what one ought to do in the most fundamental sense is that which maximizes expected value. And I have argued that the probabilities that such an account should utilize are epistemic probabilities – probabilities relativized to an agent's evidence.Footnote ⁴ But it is by no means easy to calculate expected value (on any plausible view of expected value), and one will almost always be faced with making decisions prior to an exhaustive consideration of the relevant values and probabilities of all of the possible consequences of all of the relevant alternatives. But again, one will often be able to do some calculations, enough to put oneself in a position where relative to those calculations one of the alternatives will emerge as having the best chance of being what maximizes expected value. And that will again yield the plausible suggestion that we recognize another sense in which the person who does what has the best chance of being what ought to be done is still doing what ought to be done. Again, one natural way of making sense of this is to recognize a second ought (an ought²) defined in terms of what has the best chance of being what one ought¹ to do.

3. Epistemic rationality

Foley (Reference Foley1990) suggested that the puzzle discussed above generalizes to epistemic rationality or justification. Of course, we don't want to argue that if a belief in a given proposition has the best chance of being justified (among the contraries of that proposition), the belief in question would in some sense be epistemically justified. After all, unlike action, where there is no opting out of decision (inaction is just one among other alternatives), one doesn't need to believe one of a number of alternative hypotheses. One has the option of withholding belief. Foley, however, raises the question of whether one can have a justified but false belief that one has a justified belief. Put again in terms of an “ought” judgment, the question is whether one can have a justified but false belief that one epistemically ought to believe P. On many views of justification, it is hard to see why one couldn't be rational but mistaken about such matters. But isn't there a strong pull to conclude that if one really does have a justified belief that one epistemically ought to believe P then one epistemically ought to believe P. To avoid incoherence, one might again distinguish different senses of the epistemic “ought”. Since, by hypothesis, you have a justified but false belief that you ought to believe P, it is true in one sense (ought¹ let's call it) that you ought not believe P. But in another sense (ought²), you ought to believe P. There is no contradiction. Moreover these different sense of “ought” are clearly connected. In effect, we are defining the second “ought” partially in terms of our understanding of the first “ought”.

When I first explored the idea of recognizing higher level justified belief about what we ought to do or what we ought to believe, I was thinking primarily of a less robust sort of justification we might have for believing a given proposition. As I indicated above, I was primarily thinking of someone who has a justified but false belief that he ought to do X or epistemically ought to believe a given proposition – something that wasn't quite as good as having first-level justification for action or belief. But if we do have these different levels of the epistemic “ought”, I suppose one might argue that the more “ought” concepts we satisfy, the better off we are. So on such a view, it is an epistemically good state of affairs if we believe what we ought¹ to believe, but better still if we both believe what we ought¹ to believe, but also believe what we ought² to believe. Perhaps it would be better still if we believe what we ought¹ to believe, what we ought² to believe, and what we ought³ to believe. This sort of view might remind one, at least superficially, of Sosa's idea that we get a more intellectually satisfying sort of knowledge when we move up a level and not only know that P, but know that we know that P. In a number of places, Sosa suggests that first-level knowledge (animal knowledge) has a foundational structure. And the way to move to the next level is to rely more on coherence. But in conversation, he has suggested to me that one can still ascend to reflective knowledge if one has at one's disposal some other way of arriving at an apt belief that one has an apt belief. Perhaps if there is a God and God inspires us to believe that we know that P, that is another way of moving to the next level of knowledge.

4. To add or to subtract

We just looked at one way to develop levels or layers of justification. One starts with some concept of justified belief, and then allows that one can not only have a justified belief that P, but also have a justified belief that one has a justified belief. To avoid the problem of easy justification, one might insist that to get more satisfying justification, the source of the second-level justified belief must be different from the source of the first-level justified belief. So if one has justification based on visual experience that there is a roundish object before one, one is fighting a long tradition in epistemology if one attempts a track-record vindication of visual experience. One doesn't want to attempt an inductive argument of the following form:

1. (1) I seem to see something round before me and (based on that appearance I conclude) there is something round before me.

2. (2) Repeat with another visual experience as of an object's having a shape and conclude based on that appearance that the object has the shape.

3. (3) Repeat indefinitely.

4. (4) Conclude on the basis of inductive reasoning: Visual experience is highly reliable as an indicator of an object's shape.

I won't argue here that this attempt to vindicate the reliability of visual experience fails. Although I believe Sosa thinks that one can use fundamental belief-forming processes as part of an attempt to vindicate the legitimacy of such processes, I also think that he would agree that it would be more satisfying if we could vindicate the reliability of visual experience by relying, say, on tactile experience (Sosa Reference Sosa, Coppenger and Bergmann2016: 234). Such an attempt faces all of the very significant questions that have arisen in the history of metaphysics and epistemology concerning whether our conclusions about shape based on visual experience have the same subject matter as our conclusions about shape based on tactile experience. But setting such issues aside, the general point stands that one has a better chance of getting a more satisfying justification concerning, say, visual experience as a source of justification if we can vindicate its legitimacy with a different source of justification.

Above we considered a way of adding further conditions to first-level justification in order to get a stronger, more satisfying sort of justification. I have long argued (Fumerton Reference Fumerton and Greco2008, Reference Fumerton and Dougherty2010, Reference Fumerton, Bergmann and Coppenger2016a, Reference Fumerton, McCain and Poston2018) that there is a quite different way of recognizing levels of justified belief. Rather than start by developing a concept of justification that requires minimal conditions to be satisfied – conditions that can be satisfied by children, perhaps animals, perhaps even an automatic doorFootnote ⁵ – we could instead start by developing the concept of a highly sophisticated, ideal justification satisfaction of which brings philosophical assurance. We can then strip away conditions necessary for this ideal justification in order to recognize various sorts of “degenerate” justification – justification that falls short of the ideal. The term “degenerate” is, perhaps, a bit strong. It is only designed to emphasize that these other sorts of justification do, indeed, fall short of an ideal. I have my own suggestion as to what ideal non-inferential justification and inferential justification might look like (Fumerton Reference Fumerton1996, Reference Fumerton2006, Reference Fumerton, Bergmann and Coppenger2016a). And I have further suggestions as to how to recognize various sorts of justification that fall short of the ideal – particularly with respect to inferential justification (see most recently Fumerton Reference Fumerton, McCain and Poston2018). But here I am mainly interested in exploring the structural differences between the two approaches to recognizing levels of justification.

5. Structural differences

There are, of course, structural differences. On the view discussed above, if we start with first-level justification and recognize “better” sorts of justification in terms of satisfying higher-level conditions for justified belief about lower-level justification, there is no theoretical level to the number of levels of justified belief we can recognize. In practice, one will never even ask questions about justified belief beyond the first two or possibly three levels, but the meta-questions will never be unintelligible or illegitimate, and there is no reason in principle why they can't be answered.Footnote ⁶ Some internalists hope that that having a first-level justification will automatically “bring with it” at least the possibility of having indefinitely more complex meta-level justification (see, for example, Fales Reference Fales1996). But there is no reason why the idea behind the approach of “adding” levels of justification to get better justification couldn't also be accepted by externalists who don't think that the meta-level justifications come “for free” along with first-level justification. Wouldn't the reliabilist think that it is, in some sense, better to have a reliably produced belief that P while one also has a reliably produced belief that one has a reliably produced belief that P? Might it not be better still if one has a reliably produced belief that one has a reliably produced belief that one has a reliably produced belief that P? As I have argued elsewhere (Fumerton Reference Fumerton1996), it might not even be that hard for the reliabilist to move up levels provided that the reliabilist's first-level conditions for justified belief are satisfied. Track-record arguments are tailor-made for reliabilists interested in moving up levels.Footnote ⁷ Of course, none of this would get at the heart of the debate between internalists and externalists. Internalists think that layering externalist justification on lower-level externally understood justification gets one no closer to the justification they seek – justification that carries with it philosophical assurance. But here we are concerned primarily with the fact that externalists can recognize higher-level justification understood in terms of the kind of “addition” of which we have been talking. And they might concede that satisfying the higher-level requirements leaves one in a better epistemic position.

6. Problems with subtraction

While levels of justification recognized by addition need posit no upper limit to higher-level justification, levels of justification recognized by subtraction are going to run out of kinds of lower-level justified belief that fall short of an ideal. A simple analogy makes the structural point. We can start with the number 1 and recursively define infinitely many other larger positive natural numbers. But if we start with some arbitrarily large positive number and recursively define positive numbers lower than it, we are going to run out of positive natural numbers. This crude analogy also illustrates a potentially very powerful objection to the idea that we should start with a concept of ideal justification and recognize derivative, degenerate sorts of justification that fall short of the ideal. One can't define all of the natural numbers by starting with the “largest” natural number and by recursively defining smaller numbers. One will necessarily leave out of one's definition infinitely many natural numbers. The problem, of course, is that there is no such thing as the largest natural number. The idea that there is a largest natural number is unintelligible.

Sosa (Reference Sosa, Coppenger and Bergmann2016: 232) similarly argues that there is not and could not be such a thing as ideal justification, at least if ideal justification is anything like the object of the internalist's metaepistemological search.Footnote ⁸ Whether or not this is right depends critically on what the holy grail of internalist justification is. If the search is for a kind of justification the existence of which can be established without employing any legitimate sources of justified belief, the search is, indeed, doomed to failure. One can't step outside of all ways we have of rationally reaching conclusions, and rationally reach a conclusion about which ways of forming belief are legitimate. As Sosa (Reference Sosa, Coppenger and Bergmann2016: 232) points out, even God couldn't reach a conclusion about which ways of forming beliefs are legitimate without using one of those methods that are under investigation. There is a moral here. Internalists had better be careful in characterizing ideal justification. In particular, the internalist should shy away from defending a concept of ideal justification that would satisfy the kind of skeptic who wants a non-question begging answer to all epistemic questions. As I indicated in an earlier footnote, Peter Markie (Reference Markie2005) makes a compelling distinction between an appeal to a source of justification that begs the question against an interlocutor who is questioning the legitimacy of that source, and the use of that source to certify its own legitimacy – to achieve meta-justification. If one is a foundationalist who has an analysis of foundational justification, one needn't be embarrassed or dismayed at the realization that one can't vindicate the legitimacy of that foundational justification without relying on that foundational justification. The view itself entails that one must rely on foundational justification (however that is understood) in order to get oneself a justified belief – and that includes a justified belief that one has a non-inferentially justified belief or that there is such a thing as non-inferential justification.

The above observation might be thought to provide aid and comfort to externalists who are constantly badgered by internalists about the need to provide non-question begging justification that we have justification for our everyday beliefs. So internalists haven't been shy about admonishing the reliabilist for being so sanguine that we have non-inferentially, reliably produced beliefs. Even if we were to have reliably produced beliefs in sundry propositions, how would such “justification” satisfy the philosopher seeking assurance that the propositions believed are true? One might just as well identify justified belief with true belief for all the relevance externally defined justification would have to the traditional philosophical search for epistemologically satisfying justification.

There is a lesson to be learned from the above observations. One needs to be careful how one frames one's criticism of externalism. The problem is not that on the externalist's view one would be able to get meta-level justification employing sources that satisfy the externalist's conditions on justified belief. The problem, rather, is that the externalist's account of first-level justification is implausible. A symptom of its implausibility is that we realize we are not making any progress when we move up levels getting more of the same sort of externally understood justification. As I have noted elsewhere (1996), even some externalists seem to get cold feet in trying to establish that we have reliably produced belief employing track-record reasoning. But given their views, they shouldn't have any such reservations.

So what is ideal epistemic justification? The short answer is it that it is justification that obviates the need for any meta-level justified beliefs to make it “stronger”. Consider, for example, ideal non-inferential justification. It is justification that is “as good as it gets”. It may be fallible – that's a different question. But even if it is fallible, it doesn't get any better if we ask and answer the meta-questions – What justifies you in thinking that this is the correct understanding of non-inferential justification? What justifies you in thinking that this is a good example of a situation in which you enjoy non-inferential justification for your belief? Are you justified in believing that you have non-inferential justification for your belief? These are all perfectly legitimate questions. The answers to them are philosophically important and interesting. But answering them does not improve one's epistemic position vis-à-vis the proposition one is non-inferentially justified in believing. So on my view (a view that I have stated far too many times), one is justified in believing that one is in searing pain when one is directly acquainted with the correspondence or fit between one's thought that one has the pain and the pain itself. This justified belief, I think, is precisely what Descartes searched for as an ideal foundation upon which to build an edifice of knowledge. And again, one doesn't need an answer to the interesting philosophical questions discussed above to enjoy that justification. Children and probably some animals know that they are in pain when they are. And they possess that knowledge even if they have never engaged in philosophical analysis or posed meta-questions concerning how they would know that they have such knowledge.

It is harder to identify ideal inferential justification. If we allow that one can have inferentially justified but false beliefs,Footnote ⁹ one can't expect that ideal inferential justification will survive Descartes’ method of doubt. There is a clear sense in which one can't even expect that ideal inferential justification couldn't get better. After all, if one non-deductively infers P from E1, it is almost always possible to imagine a situation in which one gains additional evidence E2 that further confirms P. One might have more justification for believing P based on the fact that one is justified in believing both E1 and E2. So what would make one's original inferential justification for believing P on the basis of E1 ideal? Again, my thought is that one doesn't get more or better justification for believing P on the basis of E1 by moving up a level to form a justified belief that one has justification for believing P on the basis of E1. In the case of a conclusion formed as a result of employing a belief-dependent, conditionally reliable process whose input beliefs are justified, it seems almost obvious that one's epistemic situation is improved if one has some reason to believe that the way belief is formed is, in fact, reliable. But if the internalist is right, the conditions necessary and sufficient for a belief's inferential justification carry with them all that is required for the kind of assurance that the philosopher seeks when the philosopher aims at justified belief.

Nothing said here is supposed to be an argument for the view that there is ideal non-inferential or inferential justification of the sort to which I have tried to point. Successfully or not, I have tried to develop conceptions of ideal justification elsewhere. I don't have legions of converts. Although he is far too nice to put it quite this way, Sosa, for one, doesn't know what I'm talking about – he thinks that the idea of justification that brings with it the sort of assurance I have in mind is incoherent. I concede that if these and other attempts to defend the existence, or at least the possibility, of ideal justification fail, there is considerable force behind Sosa's suggestion that the internalist's search for such justification is a fool's errand. And, of course, it would be equally foolish to start with a conception of ideal epistemic justification and strip away conditions from ideal justification in an effort to acknowledge less demanding conceptions of epistemic justification.

7. Motivations to subtract rather than add

One might legitimately wonder whether this supposed controversy about how to distinguish levels of justification is substantive. After all, whether we start with minimal requirements for justification and build upon these to develop more demanding (I would argue more satisfying) concepts of justification, or start with ideal justification and strip away from such justification conditions that allow us to recognize less demanding justification, aren't we both trying to accommodate the same data. In the history of philosophy, particularly in the modern era, prominent epistemologists have been accused of over-intellectualizing requirements for both knowledge and justified belief. To take just one example, for a very long time it was taken to be almost obvious that we justify our beliefs about an objective, perceiver-independent world, only by inferring truths about the objects in such a world from what we know about our subjective and fleeting experience. The project of bridging this appearance/reality gap became the focus of much epistemology in the 17th, 18th, 19th, and 20th century. By contrast, many, probably most, contemporary philosophers take it to be almost equally obvious that there is nothing like an inference involved when we are caused to believe truths about our immediate physical environment by perceptual experience. Indeed, it takes considerable effort and a certain sort of skill to “turn one's attention inward” and notice the ways in which experience constantly shifts as our relations to objects change.

Still, disjunctivsim aside, there is a distinction between appearance and reality. Eddington was right when he observed that the reality posited by theoretical physics is almost startlingly different from anything we “naturally” believed based on subjective experience. And once one acknowledges the existence of the distinction, it really is hard for some of us to ignore the epistemic question of how we can legitimately reach conclusion about objective reality based on subjective appearance.

So we have two pieces of data that pull us in opposite directions. On the one hand, we are perfectly sanguine in attributing to all manner of conscious beings justified belief and knowledge – and that includes children and, perhaps, even certain sorts of non-human animals. On the other hand, reflection, particularly philosophical reflection, tells us that there really is a gap we must bridge between appearance and reality. The epistemologist's preoccupation with bridging that gap didn't involve some horrible mistake about the nature of perception and its relation to the physical world. No doubt much of the processing that takes experiential input and churns out beliefs about our physical environment, takes place at the subconscious level, but there is no putting the genie back in the bottle once the philosopher has let it out. It is enormously tempting to deal with the two data points by introducing different levels of justification – either by adding to minimal requirements or subtracting from some ideal.

Aren't we inevitably going to meet in the middle? It depends on what we bring to the epistemological table. I've argued in many papers that the pivotal issue on which the contemporary internalism/externalism debates hinge on the new evil demon problem (and variations upon it).Footnote ¹⁰ I won't rehearse the debate here. I'll only say that I think that only certain versions of internalism get the right answer to the new evil demon problem. And that's because only certain versions of internalism have the insight (its critics would say make the mistake) of thinking that when an evidential connection holds between E and P it holds necessarily (see Fumerton Reference Fumerton and Dougherty2010). In an ideal epistemic world (not the world we live in), to have ideal inferentially justified beliefs we need to uncover the probabilistic connections between our total body of evidence (a body of evidence that is vast) and what we are induced to believe by various stimuli to which we are subjected against that enormously complex set of background beliefs. Ordinary beliefs of animals, children, the vast majority of people, and the vast majority of philosophers aren't accompanied by awareness of the relevant evidential connections. In large part that might be because so much of what causes us to form beliefs stays at the level of dispositional belief or dispositions to believe.Footnote ¹¹ Of course, we can't know that ordinary beliefs satisfy a concept that has stripped away conditions from the ideal until we know what the ideal is, and whether the world is even theoretically consistent with the existence of ideal justification.

By contrast, if we start with an externally defined, minimal set of conditions, that yield justified beliefs, and we add to those conditions externally defined meta-level requirements in an attempt to develop more intellectually demanding conceptions of justification, we'll never get an answer to the question of why the victims of demonic machination have the justified beliefs we intuitively think they do. At least we won't get an answer that satisfies the philosophical yearning for a justification that brings with it assurance.

To add or to subtract – that is the question. The answer to that question raises some of the most fundamental questions in epistemology.