1. ARGUMENTS FOR ELIMINATIONISM

Kornblith presents a variety of arguments on behalf of eliminationism. The first two are directed at the claim that reasons are essential to knowledge, while the final three are meant to show that inquiry proceeds best when we do not try to rely on reasons.

According to the first argument, we are capable of acquiring knowledge even in the absence of reasons. This occurs in, e.g., proprioception, when we form beliefs about the relative position and movement of our own bodies, without (apparently) relying on sensory experience.Footnote ⁴

The second argument relies on the so-called Sellarsian dilemma.Footnote ⁵ Many philosophers have been convinced by the regress argument that all knowledge that is inferential in nature must ultimately be grounded in something that is basic (i.e., non-inferential). Perceptual experience and rational intuition have been the most common suggestions for what can serve as the justifier that does not itself need to be justified. But, Sellars argued, we face a dilemma here: either the perceptual experience or rational intuition has propositional content – in which case it is itself in need of justification – or it does not have propositional content – in which case it is incapable of justifying something else. In either case, there is nothing that is capable of playing the foundational role needed to ground inferential knowledge.Footnote ⁶

The first two arguments, if successful, show that reasons are not essential to knowledge. This is an interesting conclusion but perhaps less powerful than one might expect. It is compatible with claiming that much, or even most, knowledge involves a response to reasons and that reasons still play a crucial role in epistemology. Kornblith's remaining three arguments target this latter claim, and so I shall focus my attention primarily on them.

These arguments, drawn from recent work in psychology, consist of three closely related considerations. First, facts about belief fixation are crucial to justification, but they are often missed by deliberation. Studies have shown, for example, that people tend to believe objects on the right side of a display case are more valuable than those on the left, but they do not cite this as one of their reasons for rating the objects as they do.Footnote ⁷ Second, when people are asked to call to mind their reasons for those beliefs, they come up with only a small, unrepresentative set of considerations. They might mention the color or shape of the object, but this cannot be a full explanation of their belief, given that the side of the display also plays a causal role in their belief fixation. And, third, it is far too easy for people who deliberate on the reasons for their beliefs to manufacture reasons; any set of considerations bearing on the belief in question can be made into a deductively valid basis for the belief through the addition of the relevant conditional, which links those considerations as antecedent with the belief as consequent. Where people are inclined to think that the underlying considerations are good reasons for the belief, they will also be inclined to think the conditional is true. In conjunction with that conditional, then, the underlying considerations will look like better reasons than is otherwise warranted.

Taken together, these three arguments show that deliberation about reasons for belief is not merely unhelpful but misguided and misleading. If we have knowledge, it is not because we are making good use of reasons, and, if we want to have knowledge, we would be well advised not to deliberate on what we take our reasons to be.

Kornblith's argumentative approach here is perhaps more pessimistic than is warranted. If our belief-forming processes were typically shaped by irrational or irrelevant influences, as in the cases he mentions, this would enjoin upon us not merely skepticism about reasons but skepticism about our cognitive abilities in general. Our belief-forming processes would be too unreliable to permit either knowledge or justified belief. But it doesn't look as though the failings identified by psychologists are so widespread as to have such dire implications.Footnote ⁸ There is room, then, to continue thinking that we are capable of knowledge and justification, and there is also room to continue thinking – if we find it serves some purpose – that reasons are an important part of human cognition.

2. REASONS, PRACTICAL INTERESTS, AND SOCIAL INTERACTION

Let us suppose that there is a machine that has various highly reliable sensors, allowing it to form mostly true representations of the world around it. This machine, which we may call a pure cognizer, has been rooted to a fixed location, where it passively takes in information about the surrounding environment. Is the pure cognizer a good model for human cognition (or for animal cognition more broadly)?

The answer, I think, must be no. Where the pure cognizer is utterly indifferent to what it represents, genuine organisms are worried about the many disasters – starvation, predation, hypothermia, etc. – to which they might fall victim. If it is going to endure, the cognizer will need to be able to attend to its practical interests. This modification of the machine will bring with it various complications, including a need to build in motivational states, but I will here focus on its epistemic requirements.

In pursuing its goals, this practical cognizer will want to be able to acquire relevant information, which means that it will need to be able to focus its cognitive endeavors where they will be most effective. The most successful practical cognizers will also be able to acquire multiple streams of information about the same state of affairs (as, for example, humans can do in receiving both visual and tactile information about the shape of objects). Inevitably, though, these different streams of information will support conflicting representations (like the stick that feels straight but looks bent when half-submerged in water). For both of these reasons – to focus attention and to resolve conflicts – the practical cognizer will need to have some idea of how its cognitive faculties work. At a minimum, the cognizer will need to be able to recognize when it is receiving information that is better than what it currently has, and it will need to be able to tell when the information it has is good enough for its present purposes. Insofar as the cognizer is going to pursue better information as a practical goal, it will also need to be able to mesh together its motivational states and these evaluations of its informational states.

Pursuing practical ends thus brings with it a need for something like epistemic evaluation. One might object, however, that the evaluation in question is tied so closely to practical agency that it threatens to conflate the epistemic with the useful. Although having true beliefs will typically be most helpful in getting what we want, it is clear that there are many practically useful shortcuts in belief formation, where survival is best served by pursuing safe, rather than correct, beliefs. Perhaps the most obvious example here is the way we are inclined to believe that strangers are dangerous, even when we have no evidence about their intentions or character. An excess of caution in such cases worked well for our hunter/gatherer ancestors (even if it is now an impediment to successful life in modern democracies).

The practical cognizer, then, will need to engage in quasi-epistemic evaluation of its informational states. This is not yet a defense of full-blooded epistemic reasons. Stated differently, fully epistemic reasons are not reducible to practical reasons; they are not simply practical reasons directed toward one's epistemic needs. But, it should also be clear, the practical cognizer is not an adequate model for human cognition.

As Spinoza noted, “nothing is more advantageous to man than man” (Ethics 4p18 s). The most successful practical cognizers would be the ones that learn to make use of one another as sources of information, both to confirm what each already thinks and to acquire vastly larger bodies of information. This can be done in a limited way through simple observation (in the way, for example, that chimpanzees learn from watching others) and even through crude forms of proto-communication (like the waggle dance of honey bees). But a genuinely social cognizer benefits from passing back and forth much more complex information, as well as information about what is not (and may not even in principle be) directly perceptible.

With increasingly complex and sophisticated information being shared among social cognizers, there will also be increasingly complex disagreements between them. In the case of the individual, non-social cognizers considered above, it was noted that there can exist disagreements between different streams of information – e.g., coming from different senses. In that sort of case, it is possible that the disagreement can be resolved at the sub-personal level. It may well be easier and more reliable for there to be an automatic system that weighs the strength of the different pieces of information and comes to favor one over the other. But this sort of automatic, sub-personal correction is not possible for disagreements between social cognizers. There simply are no sub-personal processes shared across cognizers that are capable of transmitting such complex information in the first place, let alone capable of recognizing, analyzing, and resolving conflicts between those complex bodies of information. Social cognizers need to be able to articulate what they think and why it would be good to think it.Footnote ⁹ To the extent that they are incapable of doing this, they can expect to have their own view of the world regularly overridden and to be pushed into a subordinate social role, as they add relatively little of cognitive value to the community.

This capacity for social coordination will naturally be built out of the individual cognizer's ability to recognize the good-making features of its own representations.Footnote ¹⁰ It will be able, for example, to tell when it has better or worse information from one of its own sense modalities. In the very simplest cases, cognizers will direct each other's attention to the relevant part of the observable environment. Each will be able to assess how things are through use of its own cognitive capacities. In some cases, disagreement might persist, even when both cognizers have their attention trained on the same thing. The simplest solution will be for each to simply default to relying on its own capacities. A more successful approach, however, would have each cognizer track the reliability of the other cognizers with whom it interacts. If one of the other cognizers is better at, say, detecting movement in the distance, the cognizer who learns to trust it will gain the benefit of its superior capability. This adaptation, then, stems from an ability to extend to others the same tracking of good-making features a cognizer applies to itself.

A further gain in cognitive power arises when the cognizers can articulate to each other both their representations of the world and the good-making features of those representations. Cognizers who are able to articulate their cognitive positions in this way are no longer limited by what they can merely point to in the surrounding environment. It also permits them to begin engaging in increasingly sophisticated reasoning, which takes as inputs their representations (weighted in accordance with the strength of their good-making features) and outputs new representations. Even more powerfully, some social cognizers will be able to engage in this sort of reasoning collaboratively, guiding one another to novel conclusions and correcting one another's errors.

In order to make themselves intelligible to one another, the social cognizers will have to articulate their representations in rule-governed ways. Most obviously, there will be rules of grammar, structuring at the sentence-level the content of what each cognizer says. But there will also be rules of inference, structuring the content of what is said in broader ways, dictating how one claim may serve as an answer or a challenge to another. These rules must be generally acceptable, if social coordination among large populations is to occur. For this reason, the rules of inference must embody a standard of correctness that is universal rather than in service to any one particular point of view. That standard of correctness is truth.Footnote ¹¹ The upshot, then, is that epistemic reasons function as our attempt to articulate what we believe and why it would be good to believe it, where its goodness is gauged in accordance with truth as the standard of correctness.

This completes a full defense of epistemic reasons. Several comments on the argument are in order. First, although approaching the defense by considering how to construct cognizers that are increasingly sophisticated (rather than merely beginning with a consideration of human cognition) may seem to be unnecessarily artificial, it does allow us to avoid the entrenched language of epistemology and to take a fresh look at the situation in the way an engineer or an evolutionary biologist might. Second, this approach allows us to grant that there may be instances of knowledge that arise without being grounded in reasons while also demonstrating that reasons are vitally important parts of our epistemic lives.Footnote ¹² Third, it makes clear that the need for epistemic reasons does not derive simply from our nature as believers but rather from our nature as believers who have practical goals and who live in a social setting. For this reason, epistemology – at least as applied to humans – is unavoidably social. Even individual epistemic agents who are not presently interacting with others are nevertheless making use of an epistemic inheritance that has a social origin. And, fourth, it reveals the limitations of modeling human epistemology on cognitive ethology.Footnote ¹³ The intellectual demands we face as sophisticated social beings can make our epistemic lives different in kind, not merely in degree, from those of simpler organisms.Footnote ¹⁴

3. WHEN REASONS FAIL

In closing, let me address a final objection. Kornblith argues that someone who pays close attention to his reasons may follow them impeccably yet still be performing cognitively in a very problematic way. Suppose, for example, that Jim is on a hiring committee for a faculty position. He reads all of the applications carefully and ranks the candidates on the basis of factors like the strength of their letters of recommendation, the quality of their writing samples, etc. When he presents his hiring preferences to the department as a whole, he has many reasons why he has chosen John Smith as the best candidate; if we limit our focus simply to those reasons, it looks like his belief is very reasonable. However, some of Jim's colleagues have noticed that, across many job searches, Jim has always preferred white male applicants over women and minorities who are just as highly qualified. They further point out that the implicit bias that seems to be operative in his thinking has been recognized by psychologists as a significant problem – one that afflicts most people. This general information, in combination with the specific patterns Jim has already shown, are very good reason to think that his beliefs about whom to hire on this occasion are cognitively flawed, no matter how well supported by reasons they may be.

There are two responses that can be made. First, after Jim's colleagues have told him about the troubling pattern of biased hiring beliefs he has displayed, it is no longer correct to say that his belief that John Smith is the top candidate is rational. He now has a new reason to re-evaluate his belief, and the totality of reasons no longer supports the belief he started with.

This answer, although correct as far as it goes, doesn't go far enough. Prior to hearing from his colleagues, Jim's belief was supported by the reasons he offered on its behalf. Absent the new, countervailing reason from his colleagues, those initial reasons were compelling. Nevertheless, his cognitive performance was flawed, even then (though, of course, it was less flawed than it would be if he persisted in holding the belief after getting his colleagues' feedback).

The second response is more concessive. It is undeniable that humans are fallible through and through; there is no reason to think that we wouldn't be vulnerable to errors and flaws in our practice of giving and receiving epistemic reasons. It is also true, however, that we are capable of improving our cognitive performance. The history of philosophy has its share of works that take up this project.Footnote ¹⁵ It is clear that we should also take whatever assistance psychology can offer. But this concession should be accompanied by the recognition that psychology aids us by providing us with further reasons for or against our beliefs. Psychology does not subvert, reduce, or replace our practice of using epistemic reasons – it helps us to perfect it.