Testimony is not always reliable: people can make sincere errors (such as falsely remembering or misinterpreting their observations) and even intentionally mislead other. One does not need to observe human behavior very long to see how false rumors and conspiracy theories, fake news, myths, and libels can spread through social transmission (McBrayer, 2020). Researchers often lament the poor state of popular- level knowledge and how difficult it can be to correct. For example, scholars have long since debunked the myth that the Medieval Church taught a flat Earth, and that the Church opposed Christopher Columbus’s voyage for this reason. Yet the myth persists in popular imagination, and even gets repeated by scholars in other fields (Russell, 1991; Hannam, 2023).

One might thus argue that the unreliability of testimony is precisely the reason why we need science – and that the scientific process itself thus demonstrates the lesser epistemic status of religious beliefs grounded in testimony. As Petteri Nieminen and colleagues have recently put this point, referencing the example of testimony in medicine, “From the viewpoint of science, testimonials lack statistical power and cannot be generalized to be valid for large populations nor to construct viable theories without lots of additional research” (Nieminen et al., 2021, p. 459). In evidence-based medicine, patient testimonials, case reports and even expert testimony are regarded as far inferior to good statistical, randomized trials using proper scientific methodology (e.g. Sutter, 2006). Thus, instead of relying on testimonials uncritically, scientific inquiry in the field of medicine seeks to test and validate its conclusions statistically. In contrast to this, Nieminen and colleagues list examples of religious beliefs based on testimony, such as Christians’ belief in Jesus’s resurrection and Mormons’ belief in the revelations to Joseph Smith, and argue that these should not be regarded as rational by scientifically minded people: “For a scientist, none of the above examples would be adequate as proof, because without other documentation they would still be regarded as hearsay” (Nieminen et al., 2021, p. 455). They also point out that the testimony of researchers about the conclusions of their research can at least in principle be reproduced and fact checked by other researchers, in contrast to many cases of religious testimony. They conclude that religious reliance on testimony represents a comparatively unreliable “experiential” mode of rationality.

To be charitable, I will assume that Nieminen and colleagues do not mean to dismiss testimony as a source of knowledge in general. Rather, I will take their point to be simply that beliefs based on scientific processes have a much higher epistemic status than beliefs based on religious testimony. Scientists’ testimony about their own research results is more reliable, because there are processes in place to continually fact check this testimony, whereas religious testimony concerns matters that are even in principle beyond empirical verification. This, among other factors, means that we have much more reason to trust scientific results, as compared to religious beliefs.²

I will make four replies to this argument. First (1), assuming that Nieminen and colleagues do not mean to argue against the reliability of testimony in general, their argument does not yet seem to undermine the parallel between scientific and religious trust previously argued. It seems that granting the general usefulness of testimony would be required, as scientific researchers must also rely on testimony to know about the widespread use of processes like peer review. They must also rely on testimony to know the extent to which science is about phenomena that can be empirically verified. Likewise, statistical methods simply represent the gathering of a large amount of testimonial evidence about people’s behavior, symptoms, test results, and so on. Yet if testimonial evidence has no epistemic value on its own (e.g., to inform us about what kind of symptoms patients and their doctors have observed), then it seems a collection of testimonial evidence will also have difficulty in creating reliable conclusions. This does not mean that the vetting of low-value epistemic sources cannot provide more epistemic value – but at least some value has to be there at the start (cf. Hietanen et al., 2020, pp. 532–533). Moreover, as our confidence that a proper statistical study has been conducted need not itself be based on a statistical study, but can be based on testimony, it is not clear why normal testimony should always be considered less certain than statistical studies.

Second (2), as long as the general usefulness of testimony is admitted, it seems difficult to make the empirical verifiability of claims a necessary condition of trusting others. It seems that we routinely believe others about many things that we ourselves are not in a position to verify. For example, we trust others to inform us about their mental states and past events that they have experienced. Lacking telepathic capabilities and time machines, such claims are beyond direct verification, although they can certainly be corroborated by other evidence. It is true that our inability to check claims can create a kind of epistemic risk (Koskinen, 2018) that a clever liar might utilize to get us to believe false things. Thus, we might think that knowledge that has been verified by multiple witnesses who were in a position to contradict false claims is more reliable. Moreover, setting in place processes that fact check claims can help mitigate the epistemic risk. Yet, this is not necessarily the only way to mitigate the risk, and there might be reason to trust testimony even if it goes beyond what we ourselves and other can check, or even if it is given by only one person. For example, I do not need to have a detective constantly shadow my wife to mitigate the epistemic risk that she might secretly be a secret agent; I can just trust her.

Third (3), building on the previous, it seems the kind of factors and evidential ranking cited by Nieminen and colleagues serve only to highlight how some factors might make some testimony more reliable in particular circumstances. For example, it seems that a collection of testimonial evidence corroborating each other will in many situations be of more value than just an individual testimony, and testimony from reliable sources who are in a position to know what they are talking about is better than rumors. However, the criteria appropriate for one case need not be universally necessary. As Wagenknecht notes:

[It is] important to be domain-specific. After all, what applies to interactions between children or friends in non-professional everyday life need not apply to the professional interaction between scientists. (Wagenknecht, 2016, p. 131)

This also works vice versa: what applies to science need not apply universally. For example, in medicine (the example utilized by Nieminen and colleagues), statistical studies are often required to generate reliable results, but this is because human biology is highly complex, which means that individual people are often not able to reliably infer the causes of their health. For example, one person’s experience of getting lung cancer after habitually smoking for many years, and another person’s experience of life-long smoking without getting cancer does not yet tell us much. Because so many things affect a person’s health, a much broader study collecting lots of evidence was required to enable the present firm scientific conclusion that smoking does cause cancer. Yet the complexity of many issues does not mean that testimony is therefore not able to provide us reliable knowledge about many other things.³ For example, it can reliably tell us about the existence of such in-depth studies and meta-analyses of the effects of smoking and inform us of the efforts of many tobacco companies to discredit public belief in this scientific research (Oreskes & Conway, 2010). If ordinary testimony about such matters could not provide us knowledge, then how could we have any certainty in our scientific beliefs either? It seems clear that reliable knowledge cannot be built out of just statistical studies all the way down. Likewise, completely ordinary testimony can plausibly establish, say, that someone seems to have had a religious experience resulting in drastic changes to their life.⁴

Fourth (4), the use of reliability-increasing factors like intersubjectivity is not limited to scientific practice. Many such factors were already known beforehand, and these topics have also been long discussed in religious epistemology. My point here is not that these examples necessarily suffice to establish the reliability of Christian testimony – just that the boundary between the use of testimony in “science” and “religion” does not seem as clear as the argument from the superiority of science seems to assume. For example, in Christian apologetics, much is often made of the multiple witnesses to Jesus’s resurrection, their capacity to really know that Jesus died and really rose from the dead, and the way the claim that many witnesses existed could have been falsified (e.g. Loke, 2020). There have even been attempts to utilize the best scientific knowledge of the day to help identify miracles (Duffy, 2008), even though divine action is outside what is usually understood as science (though cf. Donahue, 2024). A person’s observed character and motives have been thought to impact the reliability of the testimony of Christians to outsiders, as when John Chrysostom (Homily 12 on Matthew, 5; Chrysostom 1888, p. 5) complains that people will not believe Christians are preparing for the future life when they see Christians spending all their efforts on acquiring material possessions. Similar graded evaluations of testimony can also be seen within Christian theology itself. For instance, the testimony of ecumenical Church councils, and of what Christians have believed “everywhere, always, and by all” (the Vincentian canon) has traditionally been seen as especially weighty (though cf. Pelikan, 1971). Likewise, prominent theological figures have seen faith based only on testimony as inferior to faith that also includes a reflective, reasoned understanding (Baker-Hytch, 2018; Cross, 2018).

Thus, it seems that the argument from the superiority of science is not sufficient to undermine the case for the necessity and potential usefulness of testimony in religious matters. Scientific norms might help criticize particular features of some religious social belief forming processes, such as situations in which confirmation bias reigns. However, care should be taken in developing general-level criteria for the rationality of trusting testimony, as this plausibly varies in different contexts. Scientific objectivity plausibly involves the recognition and management of epistemic risks (Koskinen, 2018), and restricting all rationality by the criteria of one discipline involves plenty of such dangers. For example, an overly narrow view of what counts as evidence might lead to the dismissal of qualitative research, and difficult-to-measure features of reality in favor of relying only on quantitative methods. This would be a form of scientific imperialism (Mäki et al., 2018), with potentially undesirable consequences. For instance, probably every scientist and scholar working at a modern university can recognize how management strategies focusing purely on measurable metrics, like the quantity of publications, might be problematic if it means neglecting less measurable values like the quality of publications (see also Ioannides, 2005). As some other examples, Altanian (2024) has argued that in discussion of genocides, emphasis on official sources has led to the epistemic injustice of disvaluing victim testimony, while Dormandy and Grimley (2024) argue that scientific “gatekeeping”, while necessary, can itself exemplify motivated reasoning and lead to ignoring valuable nonmainstream perspectives.

As noted, sometimes tales spread socially for reasons quite independent from their truth value. As Russell (1991) notes in relation to the myth of the flat Earth, much of the force of the myth comes down to Washington Irving’s fictional tale A History of the Life and Voyages of Christopher Columbus (1828), which opposed Columbus and flat Earthers. Irving’s well-written prose made much more money and provided more entertainment than the comparatively dry work of historians, which contributed to the greater influence of Irving’s work. The popularity of fake news likewise owes much to persuasively crafted propaganda and cognitive biases, rather than truthfulness (McBrayer, 2020; Levy, 2021).

Normally, when we think we can explain widespread belief in some testimony as resulting from factors that are not sensitive to truthfulness, as in the case of the myth of the Flat Earth, we do not think people’s repetition of that testimony has epistemic force. In the study of cultural evolution, it has similarly been noted that it is not only the truthfulness of ideas, but also factors like social usefulness and cognitive biases that influence their cultural success. In other words, cultural evolution does not necessarily “select” beliefs for their truth value. This means that one might potentially undercut the reliability of some testimonially transmitted beliefs by studying the underlying selection effects, and discovering they are not sensitive to the truth.

Philosopher Taylor Davis (2020) has tried to build a debunking argument for religious beliefs based on just such considerations. According to Davis, the cultural transmission and survival of religious beliefs can be plausibly explained entirely as the result of selective effects that have no relation to truth. Prominently, the “big gods” account claims that beliefs about gods have helped enforce prosocial moral behaviors, and that this helps explain the cultural stability of such beliefs (Narenzayan et al., 2013). The plausibility of such explanations, Davis argues, distinguishes religion from science, because the cultural transmission of science is set up in a way that depends on truth value, meaning that the prevalence of scientific beliefs cannot be explained purely as the result of their prosocial effects. The knowledge that their testimonially based beliefs are not truth sensitive then acts as an undercutting defeater for the reliability of religious belief based on testimony.

Davis’s argument is a reworking of the much-discussed evolutionary debunking argument of John Wilkins and Paul Griffiths (2013). These have traditionally focused on purported features of evolved human cognition that make religious belief “natural” in the sense of being cognitively intuitive. It is then argued that insofar as religious beliefs can be explained as being generated by such factors, this undermines or debunks the rationality of religious belief. Many different theistic responses have been given to such arguments, with continuing discussion (e.g. Launonen, 2021; Kvandal, 2023). Davis develops this further by focusing on cultural evolution, which he uses to contrast the relative reliability of scientific and religious testimonial practices.

Beliefs having prosocial benefits is not in itself sufficient for a debunking argument, and Davis acknowledges that usefulness also influences the popularity of scientific beliefs. However, Davis argues that in science, but not in religion, the selective value of beliefs is linked to their truth value. This is because “scientific norms and methods ensure that false scientific theories are eventually rejected.” (Davis, 2020, p. 206) The effectiveness of scientific norms is revealed, Davis argues, by two factors: (1) the improvement of scientific accuracy over time, and (2) the convergence of scientific beliefs around the world, even in otherwise very different cultural contexts. In contrast, Davis’s central evidence against the truth sensitiveness of religious beliefs is their divergence across cultures. The assumption is that since people in different cultural have equivalent reasoning capacities, it is to be expected that rationally based beliefs should converge across cultures. In contrast, religious diversity might show that similar truth-seeking processes are not the primary driving force in religious belief. Launonen and Visala (2023, p. 194) formalize Davis’s argument as follows:

Cross-cultural divergence is good evidence that beliefs in a given domain have been favored by cultural selection in virtue of other factors except their truth sensitivity.

As the formulation by Launonen and Visala makes clear, Davis does not intend his argument to debunk all religious belief – he allows that some minority (theistic philosophers of religion, for example) might have reflective reasons for religious belief. He points out, however, that most religious beliefs are acquired through “cultural inheritance, or social learning” (Davis, 2020, p. 205) – in other words, testimony. Thus, if cultural evolution undercuts the reliability of religious testimony, then this would undermine most religious beliefs.

I will present four counterarguments to Davis. First (1), it seems there is quite a leap from prosocial benefits playing a large role in the emergence and sustainability of belief in “big gods,” to religious beliefs being fully explained by such factors. The veracity of big gods theory continues to be debated (Whitehouse et al., 2023), but the theory does not in any case even claim to provide a full explanation for monotheism and other more specific religious beliefs, practices, ideas, and texts. For example, it would not explain the specific content of the Bible or the Qur’an, or the specific way Christianity and Islam developed historically. Moreover, big gods theory has been developed within a methodological naturalist framework, bypassing evaluation of the possibility of conjoining it with theistic explanations (cf. Finnegan et al., 2023).⁵ Thus, it seems that a fair bit more work would be required to show that such theories really show that only non-truth sensitive factors are required to explain the emergence and sustainability of religious testimony. Instead, given the fact that many religious believers seek to be reflective and value the truth, it seems at least intuitively plausible that religious belief systems with some truth sensitivity and some capacity of forming supportive arguments might in fact have a selective advantage in comparison to other forms of religion. This may not be fatal to Davis’s argument, since it does not necessarily depend as much on big gods theory as on religious diversity. However, it does certainly affect the force of premises 3, 4, and 5.

Second, (2), given that Davis’s purpose is not to argue against the reliability of testimony in general, it seems cultural divergence and cultural convergence can hardly be the sole criteria for deciding whether some testimonial tradition is trustworthy or not. It even seems that Davis’s own argument could not get off the ground if these were the only relevant criteria. This is because the only way we can know about cultural divergence and cultural convergence is through testimony. And this knowledge will have to spread through social learning, rather than being discovered independently by separate cultures. Thus, it seems the defender of Davis’s debunking argument will themselves have to allow for a more permissive understanding of testimony-based belief, rather than requiring that the reliability of useful beliefs must always be confirmed by cross-cultural convergence. And such an account will also be needed for scientific practice and the public communication of science, as pointed out above. This means that, as Launonen and Visala (2022, section 6) argue against Davis, the criteria for expert scientific beliefs should not be generalized to all testimonial beliefs. Rather, one needs to consider the possibility that, just like scientific beliefs can be rationally acquired from scientific experts, religious beliefs can also be acquired rationally from religious experts. As Visala points out,

Religious beliefs are often acquired via social learning. Sometimes, they are not even clearly understood. In both these senses, they do not differ from everyday scientific beliefs. Furthermore, both can be justified because there are experts who, arguably, have access to relevant evidences and arguments. In the case of religion, these experts are often theologians and philosophers, sometimes scientists. (Visala, 2018, p. 218)

Visala admits that scientific beliefs typically have reflective reasons that can be queried from the experts, but he sees this as a convergence between science and religion: similarly, most people who hold religious beliefs have access to religious experts who can present evidence and arguments for the religious view. But in both cases, laypersons will not usually need to be able to provide independent evidence themselves before accepting beliefs (relevant for premise 5).

Social epistemologists (e.g. Zagzebski, 2012), and psychologists (e.g. Mercier, 2020) do not necessarily agree on what sort of factors influence (or should influence) trusting and doubting testimony. But it seems that cultural convergence will not typically be required for trust. The independent agreement of many witnesses does often make testimony more credible, and disagreement among experts may at least sometimes reduce confidence. When experts, or communities of experts disagree, this can be a difficult situation for laypersons, although it can be quite difficult to determine just who are “epistemic peers” (Bogardus, 2013). Nevertheless, many have argued forcefully that laypersons can in many cases have reason to trust one group of experts over the other (Goldman, 2002, chapter 7; Zagzebski, 2012; Ballantyne, 2022). For example, one might have some particular view of the significance of disagreement, even though this is disagreed on. All this means that premise 5 will have to be amended to allow for a more permissive understanding of when testimony is trustworthy.

Third (3), disregard for truth is only one explanation for cultural divergence (against premise 1). There are also many other, epistemically legitimate reasons why certain types of knowledge might be tied to a particular culture or social group. For instance, knowledge of the precise location of Westminster Abbey will be more common in London than elsewhere, and it is possible to understand the social divergence of religious beliefs as being truth sensitive in the same way (Baker-Hytch, 2018). Davis argues that “assuming that Christian beliefs are true, we should observe that people in Europe and people in the Philippines converged upon Christian beliefs in a manner that was independent of cultural contact” (Davis, 2020, p. 206). However, the prediction that Christian beliefs should spread in a way that is independent of geographical locations and social networks seems incongruent with Christian theology. The religion claims to be based on special, rather than just general revelation, God typically working through the Church to spread the message of Jesus (Visala, 2020). All major religions have their own ways of explaining religious diversity, and often in ways that do not require arguing that the other religion does not get anything right, or that its testimonial processes are fully unreliable. Likewise, cultural convergence might sometimes be explained as the result of common human biases, rather than as the result of independent reasoning.

Fourth (4), Davis seems to be exaggerating the convergence of scientific beliefs and downplaying the convergence of religious beliefs. As Launonen and Visala (2022, section 6) point out, the definition of “science” should include not only the “hard” sciences like physics and chemistry, but also “soft” sciences like psychology, sociology, and history. There is certainly merit in calling attention to features that might make the natural sciences special, but the criteria used should not exclude other sciences, as otherwise our knowledge will be much too limited. Although there is some disagreement even in the hard sciences, there is certainly more in the soft sciences. Yet this does not mean that the methods used in these fields are not truth sensitive. Likewise, the disagreement of philosophers and theologians does not mean that these disciplines (in the various ways they are practiced) are not truth sensitive. Nor are particular features of some natural sciences, like the ability to make scientifically testable predictions, necessary conditions for truth sensitiveness, given the validity of these other disciplines. Thus, one would need to take a broader look at fields beyond the natural sciences before formulating one’s general criteria for reliable testimony.

Launonen and Visala (2022, section 6) also point to examples of convergence on religious matters. They mention particularly the emergence of similar theistic beliefs and theistic arguments in many different cultures, which would be expected based on the doctrine of general revelation (cf. Brown, 2012; Clark & Winslett, 2023), as well as the way non-Christian scholars commonly agree with some historical facts that Christian apologists use to argue for Jesus’ resurrection (cf. Loke, 2020). This is not the place to assess the extent of agreement on these issues. However, while admitting that there is far more diversity in religious beliefs than in the natural sciences, such seeming convergence in religious beliefs would need to be analyzed to develop Davis’s argument.

Although this is not pointed out by Launonen and Visala, examples of convergence might also be found within a religious tradition. As Thomas Kelly (2011) points out, it is clearly true that the billions of religious believers in the world have not arrived at their beliefs completely independently, because these beliefs are indeed socially transmitted. Yet Kelly argues that some sort of convergence across generations might still be required to explain the persistence of such beliefs over time. If something about people’s traditional religious beliefs had not resonated with their own experience and reflection at least to some extent, then such beliefs would more likely be replaced by other belief systems over time, even if part of that stability might also be explained by the lack of religious freedom. Similarly, Kelly points out, the belief that 2+2=4 is typically acquired socially from authority figures (with social incentives for repeating the right answers), but the fact that it strikes people as true is important for explaining why this belief persists. Insofar as this plausibly applies to religious belief, then even socially connected, persistent religious beliefs might be considered somewhat “convergent” or partly “independent” testimony.

In summary, factors like disagreement and independent convergence can certainly impact the force of testimony. It is also plausible that religious diversity indeed calls for reflection of the reasons for trusting one religious tradition rather than another (Baker-Hytch, 2018). Nevertheless, it seems the cultural evolutionary debunking argument is yet incomplete, requiring more work on the nature of testimonial beliefs, and alternative ways of understanding the diversity and convergence of beliefs. Existing theories do not yet prove that religions always only make use of “existing cognitive by-products” (Talmont-Kaminski, 2013) rather than being truth sensitive regarding many testimonially transmitted religious beliefs.

Thus, it seems that both the argument from the superiority of scientific practice, and the argument from cultural evolution fail to debunk the rationality of trusting in religious testimony in general. However, this does not yet mean that scientific norms and practices might not help develop critique of particular religious social belief forming processes. Even though norms like “being able to accurately predict events” (Davis, 2020, p. 204) are not universal for truth-seeking practices, I will argue in the next section that examples from science might indeed be helpful in other ways.