Testimonial contractarianism: A knowledge-first social epistemology

According to anti-reductionism in the epistemology of testimony, testimonial entitlement is easy to come by: all you need to do is listen to what you are being told. Say you like anti-reductionism; one question that you will need to answer is how come testimonial entitlement comes so cheap; after all, people are free to lie. This paper has two aims: first, it looks at the main anti-reductionist answers to this question and argues that they remain unsatisfactory. Second, it goes on a rescue mission on behalf of anti-reductionism. I put forth a novel, knowledge-first anti-reductionist account, which I dub ‘Testimonial Contractarianism’. According to the view defended here, in virtue of the social contract in play, compliance with the norms governing speech acts is the default position for speakers. Insofar as norm compliance is the default for speakers, I argue, all else equal, entitlement to believe is the default for hearers.


INTRODUCTION
We 1 believe what other people tell us. I trust that you are on the way to the mall if you tell me you are on the way to the mall; I believe that the price of oil has gone down because I heard it on the news. In the summer I will go hiking in the Carpathians; I remember my geography teacher told us that wild life is gorgeous over there. And I believed him.
We believe what other people tell us, and that's all right. We don't need to check on the facts ourselves; nor is elaborated detective work needed to verify the credentials of the testifier. We can just go ahead and believe on mere say-so: it's fine. Call this view Anti-Reductionism (AR) about testimonial entitlement. 2 Now, say Anti-Reductionism is true; one interesting philosophical question that readily arises is how come testimonial entitlement comes so cheap. After all, people are free to lie. Furthermore, they are rational to attend to their own interest, rather than someone else's. It makes sense for you to lie that you're going to the mall if it serves your interests to have me believe that that's the case. Why, then, can I trust that you're telling me the truth? Why am I entitled 3 to believe you? Let us dub this problem the Source Problem for testimonial entitlement (SP): it concerns the source of our warrant to believe what we are being told by other people, on mere say-so.
There are two main anti-reductionist responses to the Source Problem available in the literature. According to Strong Anti-Reductionism (SAR) -championed, most notably, by Tyler Burgetestimonial entitlement is derivable on a priori grounds. Since rational capacities have the function of generating truth, we are prima facie entitled to believe rational speakers on mere say-so. The hearer need not do any work.
More recently, other philosophers defend less ambitious solutions to SP: according to people like Sandy Goldberg (2010Goldberg ( , 2014, Peter Graham (2010Graham ( , 2012 and John Greco (2016Greco ( , 2015, the source of testimonial entitlement is social: there are social grounds for us to trust what other people tell us. Since the proposed answer to SP is more modest, so is the corresponding anti-reductionist condition on testimonial entitlement (henceforth Modest Anti-Reductionism). On Goldberg's view, for instance, hearer's entitlement is restricted to beliefs that were generated by speaker's epistemically trustworthy processes to begin with. According to Graham and Greco, hearers need to do a bit of epistemic work -albeit not much -in order to be entitled to their testimonial beliefs: on Graham's view, they should actively watch for signs of untrustworthiness. Greco's account is even less demanding than that: depending on the social roles of the actors involved in the testimonial exchange, there are, he argues, situations when positive reasons to trust one's speaker are required.
This paper assumes Anti-Reductionism is true, without further defence. The scope of the paper is limited to finding a novel answer to the Source Problem on its behalf.
More precisely, the paper has two aims. The first is negative: I will argue that, on closer scrutiny, we have no good reason to go weak (Section #2); in order to do this, I look at the motivations put forth by champions of Modest Anti-Reductionism in favour of weakening the anti-reductionist condition on testimonial entitlement, and argue that they remain unsatisfactory; as it turns out, SAR does exceptionally well in accommodating the data put forth in support of MAR.
Further, then, I move on to discussing the most notable SAR solution to SP on the market, championed by Tyler Burge. I argue that the a priori derivation of testimonial entitlement fails to go through on function-theoretic grounds (Section #3).
In contrast, my second aim is constructive. In #4 and #5, I will go on a rescue mission on behalf of Anti-Reductionism. First, I take a closer look at the Source Problem and, in line with Faulkner (2011), I argue that, in its most plausible reading, SP amounts to a problem of cooperation. Second, in response to this, I develop a novel, knowledge-first strong anti-reductionist account of testimonial entitlement: Testimonial Contractarianism. I argue that, in virtue of the social contract we are part of, we are prima facie entitled to trust that other people will obey social norms; we don't need positive reasons to believe that they will. Since it is plausible that some epistemic norms -to wit, norms governing speech acts -are also social norms, it follows that in virtue of the social contract we are part of we are prima facie entitled to believe what other people tell us; we don't need positive reasons to trust speakers. The account put forth is strongly anti-reductionist, in that it argues that speakers are prima facie entitled to believe what they are being told. It is knowledge-first in that it grounds entitlement in knowledge norms.

NO REASON TO GO WEAK
The Source Problem is, essentially, a problem of easy entitlement; what the worry amounts to, at first glance, is the following: given that speakers can lie, believing anyone who asserts anything on mere say-so amounts to faulty gullibility: hearers should be epistemically pickier than that. In the light of the easy entitlement worry behind SP, in recent years, several philosophers have offered versions of AR that are proposing some -albeit little -limitations to hearer's entitlement.

Goldberg's socially extended entitlement
According to Sandy Goldberg, our entitlement to believe on mere say so is sourced in the epistemic quality of the belief acquisition processes that our speaker's assertion is based on to begin with. Here is the though: it looks as though we are epistemically entitled to rely on a particular category of belief generating processes, call them 'privileged' processes: perception, memory and inference are such privileged processes. Now, why should we think that we are only entitled to the outputs of our privileged processes and not to the outputs of other people's privileged processes too? After all, it does not look as though the epistemic quality of these processes, what makes them privileged, is in any way dependent of them being hosted by one particular agent or another. Here is Goldberg: [W]hatever meta-epistemological story supports principles of epistemic entitlement will be indifferent to the distinction between the subject to whom the process belongs, and the subject who is relying on that process. Indeed, it is hard to fathom why the metaepistemology of entitlements should care about this distinction (2014,166).
One answer to this question, of course, is suggested by the Source Problem itself: because people can lie. Our access to the outputs of their privileged processes is mediated, and can be obstructed by people's ability to lie. Goldberg grants as much; according to him, however, this fails to satisfactorily answer his question. After all, the question was one concerning epistemic significance: why should we think that the live possibility of lying has such devastating normative import, i.e. it renders someone else's privileged processes completely void of epistemic strength for me? After all, my own processes can also fail me; the mere possibility of failure, however, does not rob them of their privileged epistemic status: Just as one who relies on memory must guard against memory-produced confabulation, so too one who relies on another's assertion must guard against insincerity (as well as other sources of unreliability) in assertion. This consideration is the stuff of defeaters, not something that shows that the source itself cannot be appropriately relied upon (2014,168).
As such, according to Goldberg, our entitlement to believe on mere say-so is sourced in the privileged status of the processes employed by the hearer: "If one observes and understands another speaker's assertion that p, there are no relevant defeaters, and the speaker's assertion that p was itself the result of 'privileged' processes operating in her, (henceforth 'Goldberg's condition'), then one is entitled to believe that p" (2014,174). Now, note that Goldberg's view places no special burden on hearer's shoulders in the testimonial exchange; in virtue of this, it is an anti-reductionism. It is a modest variety thereof, however, because not just any reliance on a say-so will generate epistemic entitlement. It needs be that the assertion at stake is itself properly based, on the 'privileged' epistemic processes of the speaker. Beliefs based on lies or other types of epistemically dubious assertions cannot enjoy entitlement.
Here is the main worry for this view, however: it concerns the generative power of testimony; notably, work by Peter Graham (2006) and Jennifer Lackey (2008) supports the view that testimony can generate knowledge by generating entitlement. In particular, it looks as though weirdly formed assertionsi.e. assertions which are not based on 'privileged' processes operating in the speaker -can also be knowledge generators: take Bertha, Lackey's consistent liar (2008). As a result of a brain lesion, Bertha is prone to lie concerning her perceptual experiences of wild animals. Whenever she sees a lion, she will report she's seeing a deer. Enter Dr. Jones, who identifies the lesion, and, since he cannot fix it, he creates a new one that cancels out its effects: now, every time Bertha sees a deer, she believes it is a lion, and every time she believes there is a lion in front of her, she reports it is a deer. As a result, every time there's a deer in front of Bertha, she reports that there's a deer.
Several epistemologists agree with Lackey that Bertha is perfectly capable of generating testimonial knowledge. Of course, if Lackey is right about Bertha, Goldberg's view is in trouble: after all, the processes underlying Bertha's assertions are clearly too weird to qualify as 'privileged' in the relevant sense. If knowledge is generated in the hearer, however, and since Goldberg accepts that knowing implies having entitlement to believe, it looks as though Goldberg's necessity claim is called into question: testimony generates epistemic entitlement.
Goldberg is, of course, fully aware of this potential worry; what he does in response is to dismiss Bertha-type cases as cases of knowledge generation. Since the intuition is hardly uncontroversial in these cases, he argues, this will not be too big a bullet to bite.
Agreed. Arguably, however, Bertha is just a limit case of a more general worry for Goldberg's condition; i.e., a distinctiveness worry: why should we think that all and only creatures that function biologically just like us -i.e., have the same belief generating processes as ourselves -can be entitlement generating testifiers? To see the worry, consider, first, a variation of a case by Jennifer Lackey: ALIEN: Sam, an average human being, is taking a walk through the forest one sunny morning and, in the distance, he sees some weird creature drop a book. Judging by her strange physical appearance, Sam identifies the creature as an alien from another planet; he has no further knowledge about this particular type of alien or the planet from which she comes. Sam recovers the book that the alien dropped, opens it and reads: 'The President will have blueberry muffins for breakfast tomorrow'. Sam thereby comes to believe that the President will have blueberry muffins for breakfast tomorrow. Unbeknownst to Sam, this particular variety of aliens has a cognitive apparatus that's very similar to ours.
Intuitively, Sam is not entitled to believe that the president will have blueberry muffins for breakfast. More importantly, irrespectively of whether we buy the reported intuition, one thing seems clear: whether Sam is entitled to believe or not does not seem to hinge on the cognitive apparatus of the alien, i.e. on whether the alien is employing a process figuring on Goldberg's 'privileged' list or not.
One reply on Goldberg's view could go along the following lines: In ALIEN, entitlement is absent due to defeat: 4 after all, the case does stipulate that the creature in question is recognisably an alien. If so, Goldberg need not worry about this case to begin with. Consider, though, the converse: ORDINARY SPEAKER: Mary, an average human being, is taking a walk through the forest one sunny morning and, in the distance, she sees a little girl drop a notebook. Mary recovers the notebook, which turns out to be a diary. Nosy as she is, Mary opens it and reads: 'My name is Alice and I'm six years old'. Mary thereby comes to believe that the name of the little girl is Alice and that she is six. Now, my (likely theoretically loaded, anti-reductionist) intuition is that Mary is entitled to believe what she reads in the book. Reductionists, of course, are likely to disagree. 5 Not much hinges on the intuition, however. Say that, unbeknownst to everyone involved, Alice is one of the few people affected by a harmless mutation that renders her belief forming processes completely different than ours. This does not seem to change the intuition in this case -be it for entitlement or against it. 6 If that is the case, however, that is, if my diagnoses of ALIEN and ORDINARY SPEAKER are correct, it looks as though whether the speaker employs a 'privileged' -in the relevant, Goldbergian sense -belief formation process or not is neither necessary nor sufficient for prima facie testimonial entitlement.
One reply to these cases could go along the following lines: 7 maybe instead of the process type itself, what is important is the recipient's entitlement to presume same cognitive processing types. Accordingly, this entitlement is generated for ordinary human beings insofar as we co-evolved with our conspecifics, and evolved as well to be social creatures part of whose sociality lies in our dependence on them for information. In the case of an alien, the human isn't entitled to presume that (even if she is mistaken in supposing that she is; perhaps the alien is cleverly disguised as a human).
I worry, though, that a reply along these lines might face Goldberg's account with a strength dilemma. Here is how: if entitlement is a thick enough epistemic state so as to (at least) entail the possession of the relevant concepts, the account will face over-intellectualization worries. If in order to be entitled to believe that p, one first needs to be entitled to believe that the testifier employs the same cognitive processes as her, the account threatens to become too strong to accommodate unsophisticated cognizers. For instance, small children will not be entitled to believe that there's milk in the fridge based on their mother's testimony because they lack the concept of cognitive process, and thus are not entitled to believe that their mother employs the same cognitive processing types. On the other hand, if one makes the concept of entitlement as thin as to not even presuppose concept possession, it is not clear what work it can still do for the anti-reductionist; after all, the reductionist might well be willing to grant that this thin epistemic state is easily present.
Goldberg is well aware of this problem. In response to ALIEN-type cases, he appeals to evolutionary considerations: […] I want to dig in my heals: insofar as the cognitive-psychology of a species of creatures really is distinct from human psychology, to just that extent I deny that there is any default entitlement to rely on (tokens of) their (apparent) say-so.
[…] As we did not evolve with the aliens, there is no shared species history, and so no default entitlement for us to rely on them (2014,180).
The main worry for this reply concerns the legitimacy of Goldberg's appeal to a specific variety of evolutionary considerations, i.e. social evolutionary considerations, in response to this problem. Recall that Goldberg argues that the reason why we should think that we are not entitled to believe aliens, employing non-privileged belief-formation processes, but we are entitled to believe each other, is because 'we have not evolved with (emphasis added) the aliens ' (2014, 180). While I am ready to grant Goldberg that evolving with the aliens would have made a difference, I deny that evolving like the aliens matters for testimonial entitlement. The former is a consideration pertaining to social evolution. The latter is a consideration pertaining to biological evolution. In imposing a condition pertaining to the type of belief generation process employed by the speaker, rather than pertaining to the social relations between the hearer and the speaker, Goldberg's account can afford relying on the latter, but not the former.
In a nutshell, I agree with Goldberg that whether the hearer and the speaker evolved with each other might matter for testimonial entitlement; as a matter of fact, the last section of this paper will make appeal to just such considerations to answer the Source Problem. What I do not agree with -and what the cases above are meant to disprove -is that the speaker and the hearer need have evolved like each other -i.e. use the same belief formation processes -for success in entitlement.

Graham's filtering requirement
According to Peter Graham, we humans have internalized social norms (Graham, 2010(Graham, , 2015. Furthermore, the motivation provided by internalized social norms frequently takes primacy over motivations that accord with our immediate interest. Speakers have internalized a social norm that prescribes telling the truth informatively, and the motivation provided by this social norm frequently overrides any motivation that rational choice theory predicts we should have: speakers will frequently tell the truth even when it is in their best interest not to (Graham, 2010(Graham, p. 223, 2015(Graham, p. 256, 2012). As such, Graham argues, the threat posed by SP is, to a large extent, averted: […] internalized social norms explain behavior that otherwise would have seemed very puzzling, at least from the point of view of rational choice theory. We have an example of behavior that goes against what rational choice theory would have predicted in the first place. Participants "see" what they are "supposed" to do in such a situation in terms of internalized norms, and those internalized norms then explain, in large part, what they do (2012,112).
Here is, then, a rough sketch of Peter Graham's MAR: the hearer need not do much: all there is needed for (prima facie, pro tanto) testimonial entitlement on the hearer's side is for her to form her beliefs via a properly functioning process of comprehension and filtering that has the function of reliably generating true beliefs. Comprehension will be in charge with uptake, while filtering has the job of detecting indications of untrustworthiness.
Since independent inductive reasons for trusting the speaker are not required, Graham view qualifies as a version of AR. What makes the view modest is the fact that it imposes an active filtering demand on the hearers, which narrows the range of entitlement conferring testimonial exchanges. The thought here is that even if simply taking a speaker's word at face value is tantamount to an objectionable form of gullibility, doing so after having filtered for indications of untrustworthiness isn't. Now, one question that arises about Graham's view is: why should this filtering requirement appease Reductionist worries? Why should we think it satisfactorily answers SP?
In a nutshell, Graham's answer is a reliabilist one. Although social norms internalization affects testimonial exchanges, such that people don't lie as often as it would suit them, testimony is still a considerably less reliable source of information than non-agent-mediated sources, such as perception or inference. 8 The filtering requirement, then, is added in the case of testimony in order to reduce this difference in reliability, and turn testimony into an equally good entitlement-conferring source.
However, empirical results seem to shed doubt on Graham's motivation for going modest. A wide range of studies testing our capacities for deception recognition show that are very bad at it: our prospects of getting it right barely surpass chance (e.g. (Kraut, 1980), (Vrij, 2000) and (Bond & DePaulo, 2006)). To see just how well-established this result is in the relevant psychological literature, consider the following telling passage from Levine et al., 1999: "the belief that deception detection accuracy rates are only slightly better than fifty-fifty is among the most well documented and commonly held conclusions in deception research" (1999,126).
Crucially, it is not hard to see that if these studies are right and we detect deception with an accuracy rate that is barely above chance, the differences in reliability between those who accept testimony without further filtering and those who do make the additional effort of filtering will be negligible. 9 Of course, these results seem to be bad news for Graham. After all, the reason for adding a filtering requirement to AR was to reduce this difference by increasing the reliability of testimony. What the above considerations indicate is that filtering fails to deliver the goods. As a result, it would seem that Graham's motivation for going modest fails on empirical grounds. 10 More recently, though, some voices in the deception detection literature have grown disenchanted with the received view on the issue. In particular, J. Pete Blair, Levine, and Shaw (2010) argue that the past 40 years of research in deception detection have neglected the role of contextual clues in deception detection. According to them, accuracies significantly higher than chance can be consistently achieved when hearers are given access to meaningful contextual information. On the face of it, this seems like it might be the sort of result Graham needs to establish that the filtering requirement makes the needed difference for testimonial entitlement, i.e. by increasing reliability. The contextual information Blair et al. discuss, the thought would go, might be similar to the information hearers would get as a result of the kind of monitoring Graham says hearers must do.
Unfortunately, though, upon closer examination, these results will not do the trick for Graham's Anti-Reductionism. To see why, it is important to look more closely at the type of 'contextual information' that has been given to the subjects for the purposes of this study, and ask the question: 'How plausible is it that this kind of information -i.e. information that is shown to increase reliability in deception detection -is the kind of information that Graham's filtering mechanism can plausibly pick up at the context?' After all, if the study gives information such as, e.g.: 'This is a reliable testifier', this is not the kind of information that Graham's filtering mechanism can plausibly pick up, nor the kind of information that Graham's view can ask for and still count as an Anti-Reductionism about epistemic entitlement. Rather, this is the kind of information that Reductionism asks for: reasons to believe the testifier is telling the truth.
The Blair et al. study identifies three types of what they dub 'contextual content' that raises the success rates for deception detection (2010,(424)(425): (1) Contradictory content: e.g., if a testifier claims to have been at home on a given night, but the hearer was told by a trusted source that she saw the testifier out at a restaurant on the night in question, it is likely that the testifier's statements will be flagged as deceptive.
(2) Statistically normal content: e.g. knowledge about the testifier's normal activities; if the testifier's statements or performance are implausible given this statistically normal information, the statements are more likely to be flagged as potentially deceptive. (3) Information that increases the perceived probability of deceit: e.g. a situation in which a number of shortages have occurred at a bank. The shortages stop when one of the employees goes on vacation and begin again when the employee returns. This information may cause the interviewer to believe that the employee's statements are deceptive.
These results are, of course, hardly surprising, either empirically or epistemologically: it seems trivially true that, if given the right kind of contextual information in advance, most of us should be and are able to go so far as to be impeccable deception detectors: as a limit case, if I know that everybody is lying, for instance, I will likely be very good at detecting deceit. What matters for us here, however, is whether the kind of information that does the trick in the study at hand is the kind of information that Graham's filtering mechanism can plausibly deliver, thereby increasing the general reliability of testimony. The answer, I contend, is clearly 'no.', in virtue of the information in question being too sophisticated in content. Furthermore, interestingly, one out of three Blair et al. experiments failed to confirm their hypothesis (427): this was the experiment that gave participants the most limited and subtle contextual information. Thus, the experiment that most closely resembled a garden-variety testimonial exchange, where the hearer does not have a whole lot of antecedent knowledge about the speaker, failed to deliver high rates of successful deceit detection. This, again, does no look very promising for Graham's filtering requirement.
More importantly: the beef between the Reductionist and the Anti-Reductionist is over what the default epistemic position is in cases of testimony, that is, in the absence of any antecedent knowledge about the speaker or testimony in general: Anti-Reductionism predicts entitlement. Reductionism predicts lack thereof. Compatibly, both views can agree that we shouldn't believe our speakers when we antecedently know that they are likely to be lying. In that, the study at hand has little bearing on the debate.
Compatibly, however, it might still be that further empirical studies will show that garden-variety contextual information that Graham's filtering mechanism could plausibly deliver might do the trick in boosting reliability. If so, of course, my empirically-based argument against the efficiency of Graham's filtering proviso will no longer go through. For a philosophical argument aimed to show that the filtering proviso, even if efficient, is not required for solving the Source Problem, see below in section #4.

Greco's social roles
Just how far does the gullibility worry extend? It's hard to deny that simply taking the speaker's word at face value will amount to objectionable gullibility in some cases. Thus consider: Case 1. An FBI agent questions a suspect in a murder mystery. Case 2. A used car salesman tells you that the vehicle is in mint condition.
By the same token, it's plausible that testimonial entitlement requires independent inductive reason to trust the speakers here.
But what about these cases: Case 3. A teacher tells their pupil that two plus two is four. Case 4. A mother tells her child that they are moving to Norway.
Here it is far from clear that taking the speakers' word at face value will amount to gullibility. On the contrary, John Greco (2015Greco ( , 2016 argues, it is intuitively plausible that, in these cases, the hearers will acquire testimonial entitlement. However, at least on the face of it, neither of the two hearers seems to have access to further reasons to trust the speakers. John Greco's MAR is motivated by these data. The key idea of his modest form of anti-reductionism is that while testimonial entitlement will sometimes require us to have independent inductive reasons to trust the speaker, at other times, we can have it simply by taking the speaker's word at face value. Of course, a key question for Greco is exactly when independent inductive reasons are required and when they aren't. It is here that Greco's specifically social answer to SP comes in. More specifically, he claims that whether or not independent inductive reasons are required for testimonial entitlement depends on the social roles of the participants in the conversation. If they belong to the same community of knowers, testimonial entitlement is easy to come by: all they need to do is take the speaker's word at face value. On the other hand, if they do not belong to the same community of knowers, hearers shoulder a more substantive epistemic burden: they need positive reasons to trust the speaker (2015,292).
According to Greco, participants to testimonial exchanges within a community of knowers engage in a different kind of activity than participants to testimonial exchanges who do not belong to a community of knowers. More specifically, the former centrally involves the distribution of information within a community of knowers, whereas the latter centrally involves the acquisition of information for the community. Crucially, these two activities have different functions, which, in turn, give rise to different normative requirements. Agents who are engaged in acquiring information for their community have a gatekeeping function. They are in charge with letting only genuine information into the system and sifting out misinformation. Note that there is a premium on avoidance of error here. That's why we find demanding normative requirements for testimonial entitlement in this kind of case: in order to ensure avoidance of error, positive reasons for trusting the speaker are required. In contrast, agents who are engaged in distributing information within a community are in charge with efficiently distributing high quality information within a community of knowers. Note that what matters here is productivity in the distribution of truths. That's why the normative requirements for testimonial entitlement are laxer here: allowing hearers to take speakers' words at face value is a highly productive way of distributions information within the community.
Unsurprisingly, on Greco's view, the detective and the salesman cases are cases of information acquisition: the corresponding social roles place the agents in importantly different epistemic communities. As a result, testimonial entitlement will be subject to demanding normative requirements: independent inductive reasons for trusting the speaker are needed. In contrast, children and their parents are a paradigmatic case of epistemic agents belonging to the same epistemic community, as are pupils and their teachers. For that reason, the normative requirements at issue in these exchanges are the more lenient ones: taking the speaker's word at face value is just fine, positive reasons are not needed.
The problem for Greco's MAR is that it is not clear that the data do the needed work of motivating his view. That is because, at least at first glance, a classical, strong variety of anti-reductionism, given that it comes equipped with an anti-defeat condition, will do just fine -or, as I am going to try to argue, even better -in accounting for the reductionist intuitions in the Greco cases.
To see this, note that it is plausible that, in both Cases 1 and 2, the hearers have fairly serious undercutting defeaters for believing what they are told: after all, they know that they are talking to people strongly interested in concealing the truth. As such, it will come as no surprise that in order to acquire testimonial entitlement, our agents need positive reason for thinking that these undercutting defeaters do not obtain (in other words, they need defeater defeaters).
The important point here is, of course, that even if SAR grants a prima facie entitlement to take a speaker's word at face value, this is compatible with SAR requiring positive reason for trusting the speaker in individual cases. In fact, there is reason to believe that SAR's treatment of Cases 1 and 2 promises to be even better than many reductionist alternatives. After all, what many reductionist accounts of testimonial entitlement require in terms of independent inductive reason for trusting the speaker is something like reason to think that she is in general a reliable testifier (perhaps on the topic at hand). The trouble is that this won't do the trick in all cases. To see this, consider Case 1. It may be that the speaker is in general a reliable testifier (perhaps even on the topic of murder cases). But that's not enough for the agent to acquire testimonial entitlement here. The fact that, as a suspect, she has such excellent reason to lie on this particular occasion will constitute an undercutting defeater despite her general reliability (on the topic). What this suggests is, of course, that we not only need positive reasons, we need a particular variety thereof, the kind of considerations that are able to defeat the present defeaters. 11 In other words, an explanation in terms of the anti-defeat condition is the most plausible option here. 12 As a result, there is reason to believe that Greco's MAR fails to improve on SAR on this count. To the contrary: going weak seems counterproductive.

BURGE'S A PRIORI DERIVATION
In the previous section, I have argued that the motivations put forth by champions of Modest Antireductionisms for going weak don't stand up to closer scrutiny. If that is the case, what we are we are left with is SAR; as such, we had better have a good Strong Anti-Reductionist answer to the Source Problem.
According to Tyler Burge (1993Burge ( , 2011, there is reason for optimism: the answer to SP can be identified on purely a priori grounds. In a nutshell, here is the view: rationality has the function of generating true content. According to Burge, functional traits have a particular function only if they fulfil it reliably in normal conditions; take the heart, for instance. Were it not to reliably pump blood in your circulatory system in normal conditions, it would likely not have the corresponding function. Since reason has the function of generating true content, it follows that it will do so reliably in normal conditions. Note that it seems plausible that intelligible propositional expressions indicate a source endowed with rational abilities. If you hear an intelligible message coming your way, it is reasonable to believe that it was a rational source that produced it, rather than, say, that it was randomly generated by some plants in your garden when hit by a gust of wind. You don't need to first gather further positive reasons to attribute rationality to the generating source: that is the default, although defeasible assumption. Since intelligible propositional expressions presuppose rational abilities, intelligible presentations-astrue come prima facie backed by a rational source. Now recall that we have seen that, according to Burge, in virtue of having the function of generating true content, rational sources will reliably do so in normal conditions. If that is the case, and if intelligible propositional presentations-as-true indicate a rational source, it follows that intelligible presentations-as-true indicate a prima facie source of truth. In virtue of this, according to Burge, we are a priori prima facie entitled to take intelligible affirmation at face value (Burge, 1993, 472).
Note that Burge's view correlates a very ambitious answer to the source problem with a particularly strong variety of anti-reductionism, i.e. with placing no burden on the hearer's shoulders; according to him, the following holds: SAR: A person is entitled to accept as true something that is presented as true and that is intelligible to him unless there are stronger reasons not to do so (Burge, 1993, 467).
Absent defeaters, then, according to SAR, if something is presented to one as true, one prima facie entitled to believe it.
Here is my reconstruction of Burge's argument: 13 (1) One is prima facie entitled to believe based on intelligible presentations-as-true if they reliably indicate true content in normal conditions.
(2) Intelligible propositional expressions indicate generation by a source endowed with rational abilities.
(4) Intelligible presentations-as-true indicate generation by a source endowed with rational abilities (from 2 and 3).
(5) Rational abilities have the function of generating true contents.
(6) If something X has the function of phi-ing, then X reliably phi-s in normal conditions.
(7) Rational abilities reliably generate true content in normal conditions (from 5 and 6) (8) Intelligible presentations-as-true indicate generation by a source that reliably generates true content in normal conditions (from 4 and 6).
(9) One is prima facie entitled to believe based on intelligible presentations-as-true (from 1 and 8).
I submit that (6) is in need of a closer look. In particular, what I will argue next is that Burge's derivation has difficulties in answering the Source Problem in virtue of lacking support for (6).
There are two problems with (6) that I would like to point out. The first worry concerns reason's dual function: theoretical and practical. It looks as though the latter is, if not reason's primary function, at least on a par with its theoretical function. If that is the case, however, it is not clear why we should believe that, in cases of conflict, the theoretical function of rationality -that of generating true content -will take primacy. After all, the following picture seems like a plausible one: in most cases, the theoretical function of rationality complements the practical one: true beliefs are good for us, they enable us to find food and avoid predators. On the occasions when that is not the case, the practical function takes precedence over the theoretical one, for reasons pertaining to species preservation. This is just another way to phrase the Source Problem: given reason's dual function, theoretical and practical, why should we believe the former will take precedence over the latter?
Burge is well aware of this problem; his answer comes in two steps. First, he argues that the function of reason of generating truth is transpersonal, that is, independent of the individual's personal interests. Believing against the evidence, for instance, even when it is all things-considered rational to do so, involves a failure of rationality -of theoretical reason. Here is Burge: One of reason's primary functions is that of presenting truth, independently of special personal interests. Lying is sometimes rational in the sense that it is in the liar's best interests. But lying occasions a disunity among functions of reason. It conflicts with one's reason's transpersonal function of presenting the truth, independently of special personal interests (2013,(242)(243).
Second, according to Burge, when a trait has more than one function, normal conditions will be such that the relevant functions will not conflict with each other. In normal conditions, reason will generate both true content and prudential benefits reliably: one will not have to lie in order to promote prudential goods. As such, in normal conditions, rational sources will generate true content: I think that generic rationality has practical and impersonally theoretical dimensions. I think that, prima facie, when a speaker fails to tell the truth because of special interests, the speaker crosses rationality in one significant dimension. Just as a generically rational individual can be irrational on particular occasions, a generically rational individual can be all things considered rational in the particular case, while rationally lying. But the impersonal function of rationality is compromised. So there is some compromise on generic rationality (2013,266).
The claim is that, when a trait has more than one function, in normal conditions, they will not conflict, and thus, in principle, they will all be reliably fulfilled. Now, the worry is that this does not seem to hold for all functional traits: conflicting functions are hardly a rarity in nature. For a straightforward exception to this, consider, for instance human sexual organs, and their functions of excretion and reproduction. Or take the elbow in carnivorans, for instance: cursorial locomotion and manual manipulation are conflicting functions; the price paid by pursuit predators for limbs adapted for speed and locomotor efficiency is the loss of the ability to supinate the forearm, which is essential for grappling with prey. If conflicting functions are not an exception in nature, however, it seems highly unlikely that the Burge derivation will go through on a priori grounds: it looks as though, when a trait has two functions, it need not be the case that, in normal conditions, the two do not come in conflict. To the contrary, conflict might be the default state.
The second worry with (6) concerns the universality of the reliability assumption: Graham (2018) points out that it is not clear that just because a trait T has a function F, it follows that it will reliably fulfil it in normal conditions. Take sperm, for instance: only very few among thousands of sperms actually succeed in fertilizing the egg. Similarly, think of the camouflage mechanism against predators present in many animals, or of danger-detection mechanisms in rabbits. On most occasions, function will not be fulfilled: most animals will not survive predation, and most of the time mechanisms in rabbits will fail to detect real danger. Still, importantly, these traits, while not strictly speaking reliable simpliciter, they are effective: they are reliable enough.
If that is the case, however, it looks as though one cannot establish the reliability of reason in fulfilling its function of generating true content on a priori grounds: rather, one needs to investigate empirically whether reason belongs with hearts, or rather with sperm in this respect.
One thing that one could do in order to save the Burge derivation from the Graham worry would be to check whether the view survives if we substitute reliable enough in the relevant premise. If reason generates true content reliably enough, would that not suffice for epistemic entitlement? Here is the revised Burge argument: (1') One is prima facie entitled to believe based on intelligible presentations-as-true if they reliably enough indicate true content in normal conditions. (2) Intelligible propositional expressions indicate generation by a source endowed with rational abilities.
(4) Intelligible presentations-as-true indicate generation by a source endowed with rational abilities (from 2 and 3). (5) Rational abilities have the function of generating true contents. (6') If something X has the function of phi-ing, then X reliably enough phi-s in normal conditions. (7') Rational abilities reliably enough generate true content in normal conditions (from 5 and 6') (8') Intelligible presentations-as-true indicate generation by a source that reliably enough generates true content in normal conditions (from 4 and 6). (9) One is prima facie entitled to believe based on intelligible presentations-as-true (from 1' and 8').
Think of sperm: what does reliable enough stand for? Well, it looks as though what is at stake will be something like 'enough for reproduction purposes'. So it looks like the 'enough' at stake will be domain specific: enough for the relevant type of effectiveness, in the relevant domain. Couldn't we, then, have a similar picture for rationality, whereby it generates true content reliably enough for our epistemic purposes? And would that not be tantamount to saying that it does so reliably enough for epistemic entitlement?
Unfortunately, things are more complicated than this. And here is why: it looks as though the 'enough' at stake in 'reliable enough' will likely be a biological, prudential one rather than an epistemic one. Why is it, after all, that a trait needs to fulfil its function reliably enough to begin with? Well, one obvious answer is because that contributes to the trait's continuous existence: the heart, in order to persist, needs to pump blood with near maximum reliability. In contrast, sperm need not do more than succeed once in a while, enough to insure continuity of species. If hearts would fail to be reliable enough from a biological, prudential point of view, they will likely be discontinued. Similarly with sperm. But if that is the case, plausibly, in the case of rationality too, the relevant threshold setter will be biological, prudential rather than epistemic: our rational capacities will generate enough true beliefs to keep us alive and well. Of course, the two may coincide. But then again, they might not. It might be that, for mere survival, much less in the way of ratio of true to false beliefs is needed than for epistemic entitlement. True beliefs about the whereabouts of food and approaching predators need figure in the relevant set. Arguably, true beliefs about the year of the fall of Constantinople, or about the highest peak in the Carpathians need not. In any case, the point is that it is not clear -and even less plausibly available on a priori grounds -that the derivation gives us the relevant, epistemic 'enough' when it comes to true beliefs generated by our rational capacities.
If that is the case, though, when it comes to epistemic entitlement, Burge misses support for (1'): if the 'enough' at stake need not be epistemic, one will not be prima facie epistemically entitled to believe based on intelligible presentations-as-true if they reliably enough indicate true content in normal conditions. Rather, the entitlement at stake will be non-epistemic.
To sum up: we have seen that, just because reason has the function of generating true content, it need not follow that (1) it fulfils it reliably enough for epistemic purposes, and (2) that reasons' prudential function does not interfere with the reliability of the theoretical function. If that is the case, though, it looks as though the solution to the Source Problem does not afford Burge's envisaged a priori derivation.

THREE WAYS TO READ THE SOURCE PROBLEM
Now, let's have a closer look at the Source Problem. Recall, the thought was that people's freedom to lie, in conjunction with the rationality of their doing so to further their own best interest was taken to generate difficulties for anti-reductionism. In what way, though? While intuitive enough, SP remains vague about this. Too vague, that is: after all, as Goldberg well puts it, other epistemic sources are fallible too, and this does not rob them of entitlement-generating force.
What I will do next is go through three possible ways to understand SP, in order to better figure out what the task for the anti-reductionist consists in.

The reliability reading
One very straightforward way to read SP is as concerning the reliability of testimony. Roughly, the thought would go as follows: epistemic entitlement requires reliable formation. However, people are free to lie; also, they will do so whenever this furthers their interests. Since the interests of the hearers very rarely align with the interests of the speakers, testimony is not reliable enough to generate epistemic entitlement.
Importantly, though, as straightforward as this reading might be, it cannot be the correct one: SP does not concern the reliability of testimony. The question is not: people lie a lot, therefore why trust them? To see this, note that, after all, testimonial knowledge is extremely ubiquitous, which seems to suggest that testimony is a very reliable source: it is fairly widely accepted that reliable formation is a necessary condition on knowledgeable belief. 14 Second, note that the reliability (or lack thereof) of testimony is an empirical matter: we can go out in the world and measure it. But we would surely not want to say that the beef between reductionists and anti-reductionists in the epistemology of testimony is an entirely empirical issue; that is, it is surely not the case that all we need in order to adjudicate between these two views is some statistics.

Trust vs. Reliance
Here is another way to look at the source problem. Recall that what the reductionist is asking for is reasons to trust the testifier. Maybe the notion of trust itself is one that, independently of how reliable the speaker might be, or of whether she is interested in telling the truth or not, requires epistemic work on the part of the hearer. Anette Baier's classical work on the issue might help here. 15 According to Baier, there is an important distinction between trust and mere reliance: the former requires a leap of faith, and is, as such, a more emotionally charged attitude. "trusting can be betrayed, or at least let down, and not just disappointed" (1986,235). One can rely on inanimate objects, such as alarm clocks; but when they break, one is not betrayed, although one may be disappointed.
Accordingly, maybe one way to frame the source problem is as follows: just as trust and reliance can come apart, so can the corresponding conditions for entitlement. In particular, one could think entitlement to trust, in contrast to entitlement to rely on, requires positive reasons. Testimony is an interpersonal affair; speakers, as opposed to alarm clocks, have the freedom to lie if this furthers their own interest. This, in turn, suggests that one needs to take a leap of faith in testimonial exchanges that is not at stake in relying on instruments. What is at stake, then, in testimony, is entitlement to trust the speaker; as such, the thought goes, positive reasons to this effect are needed. After all, why would I invest a complete stranger with the kind of trust that I only bestow upon my closest and dearest without having any good reason to do so?
Even if we accept this demanding picture for entitlement to trust, however, there is still a long way to the Source Problem. To see why, first, we need to distinguish between trust as a two-place relation -A trusts B -and three-place trust -A trusts B to C. Note that two-place trust is a more demanding affair: I can, of course, trust you to do the dishes, although I don't generally trust you. Relatedly, three-place trust is a less emotionally charged attitude; it is one thing to trust Mary -which might require some happy history between us -and quite another to trust Mary to do the dishes.
Note, too, that three-place trust is still importantly different from mere reliance: I can rely on you to do the dishes -say, because I have no choice -but, at the same time, not trust you to do so. Furthermore, the attitude at stake in in Baier's picture is what Paul Faulkner (2011) calls affective -emotionally charged -trust, as opposed to mere predictive trust. However, the only interest in trust that reductionists have is trust in its predictive dimension: nothing more emotionally sophisticated than that seems to be needed. The latter, however, seems to require much less in the way of support. To see this, note that predictive trust is something that, in one way or another, we manifest all the time in our everyday life, maybe without even giving it much thought: I trust my baker to give me a loaf of bread in return for my 2 pounds, although I know nothing about him. I trust the drivers to stop the cars at the red light, although I have no idea who's driving. I trust the city hall to not have replaced the real oak tree in front of my office window with an undistinguishable fake one. I trust the nanny to take the kids to school, I trust the electrician to repair the circuit. And so on. 16 Predictive trust is everywhere, and no (accessible) positive reasons seem to be required to be entitled to it. To see this, note that, after all, small children do not have the cognitive sophistication to reflect upon the trustworthiness of the drivers, bakers etc, and still, they seem perfectly entitled to place three-place trust in all these people. Paul Faulkner (2011) thinks the Source Problem is essentially a 'problem of cooperation', a variety of the Prisoner's Dilemma for testimonial entitlement:

[S]peakers and audiences have different interests in communication[…] Our interest, qua audience, is learning the truth. Engaging in conversations as to the facts is to our advantage as speakers because it is a means of influencing others: through an audience's acceptance of what we say, we can get an audience to think, feel, and act in specific ways. So our interest, qua speaker, is being believed…because we have a more basic interest in influencing others….
[T]he commitment to telling the truth would not be best for the speaker. The best outcome for a speaker would be to receive an audience's trust and yet have the liberty to tell the truth or not. (2011,(5)(6) According to Faulkner, then, the default position for speakers bears no commitment to telling the truth. If that is the case, the default position for hearers bears no entitlement to believe. Therefore, SP.
It is important to understand the significance of the 'default' position at stake in the argument. Recall that the debate between Reductionism and Anti-reductionism is about prima facie entitlement. Both views will allow, for instance, that if there are defeaters present at the relevant context, hearers are not entitled to believe. This explains the concern with what the default, or starting position is in testimonial exchanges: all else absent, are we entitled to believe what we are being told? Anti-Reductionism says 'yes'; Reductionism says no, you need to go out in the world and look for further support for your belief.
Faulkner's view aims to explain the latter position: all else absent, the default for the speaker is not telling the truth. Therefore, the default for the speaker is not entitlement to believe.
I agree with Faulkner that something in the vicinity of a Problem of Cooperation is the most plausible way to read SP. In what follows, though, I will attempt to reconstruct Faulkner's argument, in order to point out several difficulties with this reading of the Source Problem.
Since the Problem of Cooperation is supposed to be a first-personal decision theoretic problem, modeled on Prisoners' Dilemma, it seems that it needs to be formulated accordingly, as concerning a problem of first-personal perspective. The problem then would take something like the following shape: when the audience knows nothing about the speaker, the audience still knows it is in a speaker's interest not to be constrained in their testimony. The later, in turn, is a reason against believing. If you have a reason to believe that a speaker's interest in the communicative exchange is getting you to believe what they say whether it's true or not, and you have no other information about the speaker, this is in itself a reason to not believe what is said. This, in turn, requires a defeater defeater for entitlement, which is why there is a standing demand that the audience have some positive reason for testimonial uptake. 17 Here is how this version of the argument looks like: (1) Hearers know that they are interested in truth and that speakers are interested in being believed.
(2) Hearers know that the default position for speakers is seeing to their own interests rather than to the interests of the hearers.
(3) Therefore, hearers know that it is not the case that the default position for speakers is telling the truth (from 1 and 2).
(4) The default position for hearers is trust only if they don't know that it is not the case that the default position for speakers is telling the truth.
(5) Therefore it is not the case that the default position for hearers is trust (from 3 and 4).
Note, though that this reading departs significantly from the 'default position' that we took to constitute the beef between Reductionism and Anti-Reductionism to begin with: the hearer, on this reading of SP, has quite a bit of information against believing the speaker's words, at least in the absence of further information. Just like in the cases described in the empirical studies on contextually-informed deception detection, the hearer in this case is in possession of antecedent information undercutting the testimonial source in virtue of speaking against its credentials. There is a question, then, whether this first-personal reading of the Problem of Cooperation is a genuine way of depicting the Source Problem that lies at the heart of the debate between Reductionism and Anti-Reductionism: again, reductionists and anti-reductionists alike agree that, in the presence of defeat, the hearer lacks entitlement to believe.
One way to respond on behalf of the defender of this reading of SP is to claim that this is the default position in testimonial exchanges, in that all cases of testimony exhibit defeat in this way. As such, in all cases of testimony it is the case that one needs positive reasons to believe one's speaker: that is the default position. Two things about this reply, though: first, if this is right, the disagreement between Reductionism and Anti-Reductionism disappears. Since the two views agree that, in cases of defeat, defeater defeaters are needed for entitlement, and since this reply takes the default position to be one in which defeat is present, the debate over entitlement in the default position vanishes.
Second, though, I am skeptical when it comes to the plausibility of premises (1) to (3), for two reasons. First, I'm not convinced that this is the correct utility profile of the case: are all speakers really such that they care about being believed? This seems like a fairly heavy empirical assumption. More importantly, though, are all hearers such that they have all the knowledge ascribed to them in (1) to (3)? After all, these premises assign a whole lot of quite sophisticated knowledge to your everyday hearer. Even if Faulkner is right about the utility profile of the case, the assumption that everyone who ever engages in testimonial exchanges knows that this is the utility profile is certainly quite implausible, since many years of philosophical theorizing have been put into assuming the conjunction of (1), (2) and (3). If all of this is right, though, the first-personal reading of the Problem of Cooperation will not do the work as a proper reading of SP.
Maybe, then, what is needed here is a reading of the Problem of Cooperation that is third-personal, in that it does not presuppose any knowledge on the part of the hearer. Here it is: (1) Hearers are interested in truth; speakers are interested in being believed.
(2) The default position for speakers is seeing to their own interests rather than to the interests of the hearers.
(3) Therefore, it is not the case that the default position for speakers is telling the truth (from 1 and 2).
(4) The default position for hearers is trust only if the default position for speakers is telling the truth.
(5) Therefore it is not the case that the default position for hearers is trust (from 3 and 4).
This way to look at the problem, indeed, does not suffer from any of the drawbacks identified for the first-personal reading, in virtue of not presupposing any knowledge on the hearer's side. Here is the main worry I have with this reading, though: on the reconstruction above, the conclusion fails to follow. In particular, the problem is with premise (3), which is not supported by (1) and (2). That is because being interested in being believed does not exclude also being interested in telling the truth. Speakers might still -by default -also be interested in telling the truth, on independent grounds, that is, independently of their concern (or, rather, lack thereof) with hearer's interests. If that is the case, telling the truth will be default for hearers, therefore trust will be default for hearers. In other words, for hearers' entitlement, speakers do not need to provide the goods from altruistic motivations; they just need to provide the goods.
If this is so, for validity, we need to reformulate (3) as follows: (3)': Therefore, the default position for speakers is interest in being believed.
It is easy to see, though, that (3)' does not do the work needed to generate the SP conclusion (5). If anything, this reading of SP amounts to a a challenge to Anti Reductionism rather than a problem for the view; a challenge to explain why the speakers would, by default, tell the truth rather than not, given that they are rational, self-interested agents, and independently of their lack of concern for the interests of the hearer. If a good answer to this question can be found, we will get the game theoretic picture that we wanted to begin with: the default position for hearers will be entitlement to believe, in virtue of the default position for hearers being truth telling. 18 The next section answers the SP challenge thus construed: as a third-personal challenge of cooperation. In a nutshell, the thought I aim to develop in the rest of the paper is a fairly simple one: the existence of the social contract sets the default position for agents to social norm compliance -i.e., all else absent, it is in their best interest to comply. As such, the default position for receivers of their services within the social contract is entitlement to trust that they will comply. We are entitled to trust the baker and the driver to respect the social norms in virtue of the fact that the social contract they are acting in makes it such that it is in their best interest to do so; similarly, I will argue, we are entitled to believe what we are being told in virtue of the fact that the default position for speakers -within the social contract -is epistemic norm compliance.
To do that, in what follows, I will propose a novel, knowledge-first anti-reductionist account, which I dub Testimonial Contractarianism. The account put forth is strongly anti-reductionist, in that it argues that speakers are prima facie entitled to believe what they are being told. It is a knowledge-first antireductionism, in that it takes the source of such entitlement to be sourced in the knowledge norms present in the social contract. 19

Enlightened self-interest
Socrates is first to make contractarian remarks in Crito: according to him, members within a society implicitly agree to the terms of a kind of social contract by their choice to stay within the society. Social and political norms, then, receive their legitimacy from this implicit 'signing' of the social contract by all the members of society.
Contractarianism is a normative view that draws its resources from the social contract. Thomas Hobbes (1985Hobbes ( , 1651 is the most notable historical figure explicitly defending such a view. In a nutshell, the Hobbesian story goes as follows: in a (actual or hypothetical) primitive unstructured social order, what he dubs a 'state of nature', individuals have unlimited natural freedoms to seek and further their self-interest. However, this also includes the freedom to harm all who threaten one's own selfpreservation. This, in turn, likely leads to a general conflictual, unsafe environment, unfriendly to life preservation and nurture. It is therefore, he argues, in an individual's rational self-interest to voluntarily give up some of his freedom of action in order to obtain the benefits provided by the formation of social structures and civil rights. In a nutshell, the social contract binds individual liberties, with an aim to an increased security of the participants in it.
One can distinguish between two contractarian claims: a strong and a weak one. The strong claim takes contractarian explanations to account for the very legitimacy of moral, political or social norms. I will put this claim aside here. The (weak) contractarian claim that I am interested in merely concerns motivational aspects: the social contract explains the fact that free, self interested agents comply with extant moral, political or social norms, whichever their source, rather than defect when it suits them.
Let us dub this 'Compliance Contractarianism'. Again, according to this view, the social contract explains why people comply with norms when it looks as though doing so goes against their immediate interest. This view says nothing about the source of the relevant normativity, and certainly does not take it to reduce to social normativity. Now, the reason why Compliance Contractarianism is particularly interesting here, given the concerns of this paper, is the fact that it historically concerns itself with answering the problem of cooperation: note that, for each and every agent within the social contract, it looks as though it is, as a matter of fact, in her best interest to defect. For our agent, there are two possible scenarios: either everyone else complies with the norms, or they don't. If they don't, she's better off not complying either. If they do, it is best for her if she defects. In short, though rational individuals might consider it in their best interest to enter the contract in order to ensure a great level of security and welfare for themselves, once inside, they will find that complying with the norms might not be the most profitable way of handling things.
And still, just like with testimonial knowledge, we can see with our own eyes that the social contract 20 does, in fact, in some way, overcome this problem; Amartya Sen's (1970) famous "stranger and the letter" example nicely illustrates this. Suppose you are headed for the post office to deliver a letter and a stranger approaches you and asks for directions to get to the train station. Your immediate self-interest should then guide you to lie to the poor man that the train station is right by the post office, and ask him, while he is heading towards it, to also deliver your letter. This will certainly save you a trip. Though, I suppose everybody would agree, no one would do such a thing, but they would rather give good directions to the stranger and deliver the letter themselves.
The contractarian literature features a fairly widely endorsed explanation for this datum, first defended by David Gauthier (1987). In a nutshell, according to Gauthier, what does the trick is enlightened self-interest: he argues that rational individuals will choose to dispose themselves to be constrained self-interest maximizers rather than straightforward self-interest maximizers. When rational, people will choose to disregard their immediate gain for future, more substantive gain. To see the plausibility of this claim, take some straightforward examples of enlightened self-interest maximization our society is involved in on an everyday basis: spending one's youth working through long years of education, rather than lying on the beach playing Tetris; going to work everyday, rather than staying at home watching a season of one's favorite TV show; buying a house rather than taking a luxury vacation in Hawaii; and so on. 21 Further confirmation notably comes from game theory. According to people like Gary Bolton (1991), Jack Ochs and Alvin Roth (1989), the existence of the social norm affects the utility profile of a particular context. There is future gain to be had form conforming with the social norm -in terms of good reputation, social approval, decreasing risk of being subject to sanctioning. As such, all else absent, in social contexts, norm conformity is the default position, it enjoys default rationality.
Furthermore, leaving the default position requires fairly serious incentive; the stakes need to be rather high for the benefits of defecting to outweigh the benefits of conforming. That is, the expected payoff needs to be as high as to make defecting rational, when combined with the probability of being sanctioned. To see this, consider Ultimatum games: ULTIMATUM: There are two players, a Proposer and a Responder. The Proposer is given a sum of money -say, one hundred pounds. He then must propose a split of the money between the Proposer and the Responder. The Responder's job is to accept or refuse the split. If accepted, both parties receive the amount proposed. If refused, no one gets anything. As such, both parties are better off if the Responder accepts the split.
Note that, as the case is described, absent any further considerations, the Proposer should propose one pound for the Responder and 99 pounds for the Proposer. Also, the Responder should accept. After all, the thought goes, one pound is better than nothing.
Across a very wide variety of human cultures, however, that is not what happens. Instead, the Proposer tends to offer something in the vicinity of a 40/60 split. Furthermore, in cases where the Proposer does offer a much smaller split to the Responder, the Responder tends to refuse.
The explanation of his behavior -which seemingly violates rational choice axioms -is that the presence of the social norm changes the utility profile of the case, such that more in the way of expected payoff is required to outweigh the utility contribution of norm conformity. Studies show that, in higher stakes cases, unsurprisingly, Respondents's behavior changes dramatically: " … among respondents we find a considerable effect of stakes: while at low stakes we observe rejections in the range of the extant literature, in the highest stakes condition we observe only a single rejection out of 24 responders" (Andersen, Ertaç, Gneezy, Hoffman, & List, 2011).
Absent substantial payoff for defecting, however, the default rational position within the social contract is social norm compliance. This, of course, goes a long way in explaining the intuitive easy entitlement to trust that norm conformity will obtain in social settings, such as the baker's, or traffic, or the like. The presence of the social norm will modify the utility profile such that, absent strong incentives to defect, by default, the baker is rational to give me the loaf of bread in exchange of my two pounds, and the drivers are rational to stop at the red light. Conversely, I am entitled to trust that they will. In other words, in virtue of the utility profile generated by the presence of the social norm, the expectation of norm compliance transforms from a merely justified normative expectation (expectation about how the world should be), 22 to a justified predictive expectation (an expectation about how the world will be).

Testimonial contractarianism
By now, it should come as no surprise that I want to propose that a similar picture holds for testimonial exchanges too. Note that it is fairly 23 uncontroversial that the speech act of assertion is governed by an epistemic norm. Note, also, that some epistemic norms are also social norms; plausibly, norms governing speech acts are among the main candidates for this double status. That is not to say that the relevant epistemic normativity reduces to social normativity; rather, the claim is a coincidence claim: some norms are, at the same time, epistemic and social. Plausibly, the norm of assertion will figure among such norms: it is, after all, not only epistemically unacceptable to lie, for instance, but also socially unacceptable.
Several people think knowledge is the norm of assertion: one should only assert that p if one knows that p (KNA, most notably defended in Williamson (2000)). 24 Others disagree and impose weaker conditions (justification, truth), or stronger conditions (certainty), or even contextually variant conditions on permissible assertion. 25 Whatever their preferred norm might be, however, most people in the debate agree upon one minimal, happy Gricean thought: since it should be able to account for assertion's epistemic aim, or function, 26 it features a positive epistemic status. In other words, the norm is not 'Assert that p only if p is false!', nor 'Assert that p only if you have no reason to think p is true!' That this much hearer friendliness exists is accepted by pretty much everyone in the debate. For this reason, the argument to follow can go through by employing any one of the accounts of the normativity of assertion proposed in the literature. Because it is the account I favour, in what follows, I will run with the knowledge norm.
Here is what I will argue next, in a nutshell: given that KNA is also a social norm, all else absent, in virtue of the social contract, conforming to it enjoys default rationality for the speakers. If conforming to KNA enjoys default rationality for the speakers, hearers are by default entitled to believe what they are being told.
To lay my cards right on the table, here is the strong social anti-reductionist view defended in the paper: Testimonial contractarianism: In virtue of social contract, hearers are prima facie entitled to form beliefs based on speaker's assertions.
The view is a Burge-style strongly anti-reductionist view, in that it imposes no epistemic burden on the hearer: absent defeaters, she is entitled to believe on mere say-so. It is a Goldberg, Graham and Greco style anti-reductionism in that it takes the source of this entitlement to be social. It is knowledgefirst rather than truth-first, in that the norm that grounds epistemic entitlement on this view is a knowledge norm.
Testimonial exchanges are distinctively social phenomena; by default, 27 they happen within and as part of the social contract. The fact that a social norm requiring knowledge for permissible assertion is in place, in virtue of the social contract being in place, changes the utility profile of testimonial exchanges. In virtue of its also being a social norm, by contractarian lights, conforming to KNA will enjoy default rationality on the side of the speaker. All else absent, speakers have incentive to assert knowledgeably. Just like stopping at the red light is the default for drivers, and giving me a loaf of bread in exchange for my two pounds is the default for the baker, in virtue of the social contract being in place, the default position for speakers is asserting knowledgeably.
If that is the case, the default position for hearers is entitlement to believe, just like the default position for pedestrians is to cross the street on a green light, and the default position for me is to trust that my baker will give me the loaf of bread in exchange of my two pounds. In other words, in virtue of the utility profile generated by the presence of the knowledge norm, the expectation of norm compliance on hearer's side transforms from a merely justified normative expectation (expectation about how the world should be), to a justified predictive expectation (an expectation about how the world will be).
Of course, just like with any other social norms, strong reasons can come and override the requirements of KNA. If it is urgent to take your wife to the hospital, you might break the traffic norms. If she desperately needs both the bread and my money for her children, the baker might defect and not respond in the expected way. Similarly, if it serves your interests to lie to me, you will. Note, though, crucially, that in all the above cases, the live possibility of norm violation does not affect my default entitlement to trust in norm conformity.
Of course, the discussion here concerns merely the default position for hearers, as sourced in the default position for speakers. As such, all we get from this normative story is prima facie entitlement to believe. Defeaters might and will affect this position. If I know your wife needs to get to the hospital urgently, or that you stand to gain from lying to me, I am not entitled to expect norm conformity anymore. You will have exited the all-else-absent default position, in virtue of further incentives overriding the ones sourced in conforming to the relevant norm. Correspondingly, I will not enjoy default entitlement to trust you.

CONCLUSION
I have argued that you can trust me to tell you the truth. You don't need to know who I am, whether I have a good track record in these respects, or whether I might have good reason to lie or not. Nor do you need to have reason to believe in the reliability of testimony in general. You don't even have to give all these things any thought. All you need is to comprehend my assertion; next thing you know, you're entitled to believe it. Such are the workings of the social contract in conjunction with the knowledge norm of assertion.

ENDNOTES
1 The research for this paper was done with funding form ConceptLab, University of Oslo. I am extremely indebted for extensive comments on this paper to Peter Graham, Sandy Goldberg, Paul Faulkner, Chris Kelp, and two anonymous referees for Nous. I am very grateful for all the helpful discussion to Jessica Brown, J. Adam Carter, Olav Gjelsvik, Andy Peet, Alessandra Tanesini, and Tim Williamson, and the audiences at the Leuven Epistemology Conference 2016, the CSMN Colloquium University of Oslo, Bled Philosophical Conferences 2017, the Pavia New Trends in Epistemology Graduate Conference, the Social Epistemology Sumer School at the Autonomous University of Madrid, the New Trends in Social Epistemology Seminar Series at the University of Glasgow, the Cardiff University Philosophy Seminars and the Epistemology Seminar Series, University of Edinburgh.
2 I do not mean to suggest that anti-reductionisms and reductionisms in the epistemology of testimony are two uniform, clearly delineated bunches. To the contrary, champions of both views make very distinct claims, concerning very distinct issues related to testimonial entitlement. See Lackey (2008) for a very useful taxonomy. Following champions of the views I discuss here, however, for the purposes of this paper, I am going to focus on the particular difference between reductionism and anti-reductionism when it comes to how heavy an epistemic burden they lay on hearer's shoulders in the testimonial exchange: while the boundaries between the two camps are, by no means, clear, reductionists tend to require hearers to have some variety of access to their reasons to trust their testifiers, while anti-reductionism tends to deny such access is necessary. All this is still pretty vague, but one useful way to see the distinction that we care about here is to think of Pritchard's (2004) taxonomy, distinguishing between what he dubs Credulism, on one hand, and Reductionism on the other. Champions of reductionism include Adler (1994), Audi (1997Audi ( , 2004Audi ( , 2006, Fricker (1994Fricker ( , 1995Fricker ( , 2016Fricker ( , 2017, Hume (1739), Lipton (1998), Lyons (1997. For defenses of anti-reductionist (credulist) views, see, e.g. (Burge, 1993(Burge, , 1997(Burge, , 1999, (Coady, 1973(Coady, , 1992, (Goldberg, 2006(Goldberg, , 2010(Goldberg, , 2014, (Goldman, 1999), (Graham, 2010(Graham, , 2012(Graham, , 2015, (Greco 2016(Greco , 2015, (Green, 2016), (Reid, 1764), (Simion & Kelp, 2018). For hybrid views, see e.g. (Faulkner, 2011), (Lackey, 2003(Lackey, , 2008, (Pritchard, 2004). 3 For the purposes of this paper, I use justified, warranted and entitled interchangeably. Nothing hinges on this. 4 Thanks to an anonymous referee for pressing this point. 5 Thanks to Paul Faulkner (p.c.) for pointing this out. 6 An anonymous referee worries that this point will hinge on the precise way in which the case is spelled out: If the difference in the belief-forming process is completely fortuitous -the child is a Swampman who has spontaneously arisen with a belief-forming process that differs from ours but gives similar results, -then Goldberg ought to deny that the speaker's testimony gives entitlement, and say that the appearance of entitlement only arises because we have very good reason to think that the speaker's belief-forming process is like ours. If the mutation is more systematically harmless (say, there is a mutation that affects people's belief-forming process, but the belief-forming process that results was enough like ours that the people with the mutation didn't have enough false beliefs to kill them off when the mutation arose, and so it remains in the gene pool to this day), then it is not clear that it should count as a different belief-forming process at all -after all, it yields similar results because of similar evolutionary pressures.
About the first option: I worry about such error-theoretic reply a bit: if the intuition of entitlement is so stably driven by us having good reason to think that the speaker's belief-forming process is like ours, this seems to be evidence that the correct view of entitlement rests on non-testimonial reasons we have to trust the speaker (i.e. evidence for Reductionism) rather than evidence that our intuitions are problematically going astray. About the second option: the case I had in mind was, indeed, closer to this way to spell it out. That being said, note that Goldberg cannot afford this reply: that is because Goldberg does not offer a view of what all these 'privileged processes' have in commonlike, for instance, that they all share a belief generating function acquired evolutionary; rather, he remains theoretically neutral on the issue by offering a precise list of 'privileged processes;' perception, memory etc. All I need for this case, then, is to stipulate that the little girl's belief formation process is not on the list, although it's well selected evolutionary for its function etc. 8 Graham,personal communication. 9 See Michaelian (2010) and Faulkner (2011, 86-87) for a similar point. See Fricker (2017) for skepticism about the empirical results. 10 One way to go in support of the Graham view at this stage is to argue that, even though we are very bad at identifying classical signs of deception, we are good at tracking potential motivations to deceive (see e.g. Sperber et al., 2010); thanks to Andy Peet for bringing this to my attention and for very useful comments on several drafts of this paper. Of course, one question that arises at this stage is whether this is actually an actively filtering mechanism or rather just an anti-defeat sensibility. See below for an account that makes sense of the latter. 11 See Fricker (1994Fricker ( , 1995 for discussion. 12 Note, too, that any variety of reductionism that attempts to vindicate this thought would likely turn out to be implausibly strong; basically, the view would have to demand that hearers have positive reasons to believe every particular speaker is telling the truth on every particular matter of fact. This would eliminate many of the testimonial sources we usually rely upon from the entitlement game. 13 If I am anywhere close to Burge's line of thought at all, I am highly indebted to Peter Graham's great work on the topic, as well as to his patience and extraordinary clarity in conversation. 14 Couldn't it be that testimonial knowledge is ubiquitous in spite of there being a lot of lying, just in virtue of the fact of there being a lot of testimonial exchanges? The answer is 'no:' reliability matters for knowledge. Processes that only deliver truths half of the time are not knowledge generating processes. If Mary lies to me half of the time, I can't come to know via her testimony: at best, I can hope for a bunch of luckily true beliefs. How about if half of the testimony I get is true, and, on top of this, many of these pieces of testimony cohere? Can't I get knowledge like that? The answer, again, is 'no:' if Mary lies to me half of the time, I don't come to know based on her testimony in the cases in which she speaks truthfully, even if there is no incoherence. Compatibly, I might come to know based on her testimony plus some further, confirmatory evidence form other sources. 15 See, also, Faulkner (2017). 16 These statements can all be recast as belief statements: I believe that the baker will give me a loaf of bread in return for my 2 pounds etc. As such, arguably, it is not clear that the attitude at stake is even trust proper (Paul Faulkner, p.c.). 17 Thanks to an anonymous referee for pressing me on this point. 18 Faulkner (2011) himself offers a reductionist (in the sense restricted employed by this paper, i.e. non-credulist) variety of a social contract -based account. In a nutshell, on Faulkner's view (2011, chapter 7), the problem of cooperation is solved by our intrinsically valuing trust, which is the result of social normativity. On this view, we see another needing information as in itself a reason to give them the information they need. This norm sets up our concept and practice of trust. And it is this practice that ensures that our trusting others means we have a reason to believe them. Trust supplies a reason -and so satisfies the reductive demand for a reason -only because and insofar as there is a practice of trust.
Since this paper's focus is restricted to anti-reductionist responses to SP, I will not engage with Faulkner's view here. Further research will focus on comparing reductionist to anti-reductionist social normativity-based accounts. See, however, Graham (2012) for discussion. 19 Several contractarians (e.g. Gaulthier, 1987) attempt to argue for a purely rational, a priori basis of the social contract itself. I will not take a stance on this issue here. Importantly, though, if they are right, this paper's answer to the Source Problem is a priori, in virtue of its being grounded in the social contract which, in turn, on this picture, would be grounded in a priori norms of rationality. I must confess that I am rather pessimistic that this picture will go through, however: it seems to me as though it is only rational to enter the social contract given the kind of vulnerable creatures we are, which seems a contingent matter of fact. Thanks to Olav Gjelsvik for illuminating discussions on this issue and many others in this paper. 20 Here and below the social contract refers to society in general, not to a particular country or other such explicitly delineated social group. I take it to be intuitive that non-members of our society -such as Lackey's Alien -will fail to be motivated by social norms, and that we lack entitlement to believe they will comply with them.
to be something pertaining to prima facie entitlement, in an empty utility room, as it were? The answer is 'no'. Some information about utilities needs to be allowed into the default position in order to get the Source Problem off the ground to begin with. After all, the fact that speakers are interested in being believed rather than spreading truths, for instance, is a distinctively social phenomenon, pertaining to how our societies happen to function -i.e., in this case, pertaining to the fact that influencing others tends to be to one's one advantage. Building this into the default position presupposes the existence of the social contract in play, at least in the weak variety thereof which is at stake in this paper. Many thanks to Paul Faulkner for pressing me on this one.
Note, also, that without such assumptions in play, the Source Problem vanishes. After all, what we get in an empty utility room is merely a speaker trying to waist as little utility for no gain whatsoever. It is easily arguable that in such a scenario, the rational choice is to remain silent or, alternatively to tell the truth, just in virtue of the fact that it's easier, it takes upless cognitive resources.