Against Intellectual Autonomy: Social Animals Need Social Virtues

ABSTRACT

We are constantly called upon to evaluate the evidential weight of testimony, and to balance its deliverances against our own independent thinking. ‘Intellectual autonomy’ is the virtue that is supposed to be displayed by those who engage in cognition in this domain well. I argue that this is at best a misleading label for the virtue, because virtuous cognition in this domain consists in thinking with others, and intelligently responding to testimony. I argue that the existing label supports an excessively individualistic conception of good thinking, both within and outside philosophy. I propose replacing ‘intellectual autonomy’ with ‘intellectual interdependence’, which properly emphasises the depth of our reliance on one another, without suggesting we ought ever to be epistemically servile.

KEYWORDS:

• Epistemic autonomy
• virtue epistemology
• interdependence

On Twin Earth, just like the actual world, philosophers engage in debates about ethics and metaphysics and epistemology, developing analyses and constructing counterexamples, or looking to the sciences for evidence. The lines of debate are familiar too: compatibilists argue with incompatibilists about free will, deontologists debate consequentialists and physicalists attempt to deflate accounts of phenomenal consciousness. There too, philosophers advance and debate the value of virtue theoretical accounts of morality and of epistemology. There’s one glaring difference: among more familiar dispositions that feature in their canonical versions of the epistemic virtues, there’s an unfamiliar one. In addition to open-mindedness, perseverance, and humility, Twin Earth virtue epistemologists praise agents for exhibiting a virtue for which we lack a name, but that I will call shouting down. Encyclopedias and textbooks on Twin Earth give first-pass definitions of ‘shouting down’ roughly as follows: the virtue of shouting down consists in refusing to listen to others and may include attempts to prevent them being heard by third parties.

As the many similarities between them and us suggest, Twin Earth people are psychologically identical to us. So why does their list of virtues feature shouting down? I can only speculate that their history is rather different to our own: perhaps in the distant past malign actors were so common that this disposition was essential. Today, however, the intellectual environment is very much like our own and a disposition to refuse to listen seems an odd inclusion in a list of virtues. Put that objection to a philosopher on Twin Earth and they’ll soon set you right. The virtue of shouting down, they’ll tell you, consists in a disposition to refuse to listen to others in the right way, at the right time and to the right degree. Of course we should often listen respectfully, they’ll agree. But we shouldn’t do that all the time. Sometimes speakers are in bad faith and known to be so. Sometimes they’re trolls or liars, sometimes their speech is racist or misogynist. When the occasion calls for it, the virtue of shouting down requires respectful listening; on other occasions, it requires turning from, perhaps mocking the speaker or refusing them a platform.

The debate between us and them over the ‘virtue’ of shouting down might seem to be entirely verbal. Every case in which it’s clear to us that we ought to listen respectfully to others is a case on which they’d agree, and such cases are no less common there than here. Unclear cases are equally common at both worlds – and we tend to have the same sorts of debates over them here and there. We just cast them in different terms. Given how closely the debates mirror one another, we might shrug off the difference as an interesting quirk or a historical curiosity and move on.

I think that would be a mistake. As Confucius once insisted, the ‘rectification of names’ is important: ‘If names are not rectified, speech will not accord with reality; when speech does not accord with reality, things will not be successfully accomplished’ (2003, 13.3). Words are sometimes mere labels – it doesn’t matter whether we call the animal dog, chien or Hund – but they can matter a great deal when they are normatively inflected. They can bias debates and pump intuitions in one way, rather than another. If we recognize a canonical virtue of shouting down, we will be more sensitive to opportunities to close down debate than to those that call for listening respectfully, even if we explicitly recognize that the virtue can be manifested in that way. We’ll find ourselves criticizing others for failing to shout down when they interrupt and heckle, and the linguistic oddity of doing so will shape our responses, leaving us more tolerant of such behavior than we should be. We need a better name for the virtue; one that better matches the kind of behavior that it calls for.

The philosophers of Twin Earth must work things out for themselves. But we can rectify the names in our debates. I’m going to argue that the virtue we call intellectual autonomy¹ is almost as badly named as shouting down and that the misnaming isn’t harmless. In fact, the problems closely parallel the problems with shouting down. ‘Shouting down’ is a misleading label for a virtue most typically manifested in respectful listening; it does violence to language, and distorts thought, to make that the focal point of the virtue. Similarly, ‘intellectual autonomy’ is a misleading and distorting label for a virtue that manifests primarily in thinking with or deferring to others.

‘Autonomy’ features in a number of debates within, and beyond, philosophy. Given the variety of uses, it may well be that there’s no single disposition or property at issue across all these debates. It is more than possible, indeed, that this variety leads contributors to a single debate sometimes talking past one another, perhaps because they first encounter the concept in a different philosophical context and carry the conceptions prevalent in that context to the new debate. In the light of this apparent polysemy, it’s important to clarify that I am concerned with autonomy here solely as it is taken to manifest in the virtue that governs our intellectual relations to one another; in particular, in the regulation of our inescapable epistemic dependence on others. It may well be that ‘autonomy’, in this context, is entirely distinct from the cognate notion in debates about informed consent, moral responsibility or political thought. This diversity of connotations of the term gives us all the more reason to rectify the names.²

Just as every philosopher on Twin Earth is quick to concede that shouting down can’t be what it sounds like, every philosopher here agrees that intellectual autonomy (IA) can be a misleading label. It suggests an individualism and a self-sufficiency completely at odds with our pervasive epistemic dependence. Virtuous thinking is so heavily dependent on others that there’s at least a prima facie case for thinking that it’s misleading to think there’s a virtue that deserves to be called IA. Most of what we know we know on the basis of testimony. We rely on testimony for mundane facts about places far away and times long past. We also rely on it for current events, including the politics of our own country. We rely on expert judgment pervasively: there are many, many topics on which others know much more than I do, and some of them matter deeply to me. Is climate change a pressing concern? How should we respond to it? Is this lump a concern for my health? Why is my car making a funny noise? On these questions and many, many more, we routinely defer to the judgment of those who are better placed than us. Moreover, we recognize that new knowledge is rarely (or never) generated by lone individuals: it is the product of distributed intellectual labor. A naïve conception of IA as independent thinking cannot survive contact with the epistemic facts.

In recognition of our deep epistemic dependence, philosophers stress that IA does not consist in striking out on one’s own. But that doesn’t worry most of us: after all, we also recognize that virtues always require judgment – phronesis – in their application. The virtuous agent exhibits their character traits ‘at the right time, about the right things, towards the right people, for the right end, and in the right way’ (Aristotle 2014, 1106b). Insufficient reliance on others is fully compatible with a sophisticated understanding of IA. Accordingly, accounts of IA stress the need for good judgment in its application. For King (2021), the virtue ‘requires thinking for ourselves while relying on others appropriately – neither too much nor too little’ (2020, 59). For Roberts and Wood (2007), it consists in the virtue of being regulated by others in the right kinds of way. For Jonathan Matheson, it consists (inter alia) in the disposition ‘to make good judgments about how, and when, to rely on your own thinking, as well as how, and when, to rely on the thinking of others’ (Matheson 2021, 183.) For Battaly (2021) it’s a virtue that must be balanced by another, the virtue of ‘intellectual interdependence’, and must be applied judiciously.³

So defenders of IA have a ready and (prima facie) plausible reply to the concern that the virtue places too great an emphasis on independence from others. They recognize the danger, but argue that this just shows the need for good judgment. They may go on to argue that any virtue term carries with it analogous risks. Consider intellectual humility. The intellectually humble person must avoid the error of servility (Church 2016; Whitcomb et al. 2017). She must show proper pride in her own intellect (Tanesini 2021b). To exhibit an excess of humility is, alongside a deficiency of the trait, a vice. We can’t avoid this kind of risk by relabeling the virtue. Call it ‘intellectual pride’ and we risk the opposite sort of error. Any virtue can become a vice if the disposition it names is exhibited in excess.

Of course, advocates of the virtue of shouting down can defend their categorization on precisely the same sort of grounds. We must shout down the right people, on the right occasions, in the right way, they’ll say. They’ll concede that there’s some linguistic oddity in criticizing someone who is, say, attempting to silence fair and well-motivated criticism on the grounds they’re not manifesting the virtue of shouting down, but that’s a linguistic oddity we just have to put up with. Both Battaly (2021) and Matheson (2021) note the oddity of urging someone to exhibit the virtue of IA by deferring more, but that just comes with the territory.

To make our case that the philosophers of Twin Earth should abandon talk of shouting down, we must show that their categorization causes avoidable problems. Similarly, calls for a different label for the dispositions that philosophers now take to be characteristic of IA must show that there’s something at stake beyond terminology. It might be odd to call respectful listening ‘shouting down’, but perhaps that’s an oddity we should live with. Respect for tradition, and the utility of shared language, might justify it. I’m going to argue that there’s a better label – intellectual interdependence (II) – for the relevant traits, and that adopting it will reorient our debates productively. It will be clearer to us how we ought to manifest the virtues in this domain: the domain of balancing the tasks of learning from, with, and occasionally against others. I take the label, and an initial definition, of intellectual interdependence from Battaly (2021).⁴ She argues that between them, II and IA can do all the work we want from an account of the virtues in this domain. I’m going to argue that II can all the work by itself.

It might be best to begin by clarifying the value of IA. What work do we call upon it to do? There are a number of different accounts of the value of IA. I will note some of the more promising and prominent without attempting to adjudicate between them – my aim is not to identify the best account, but rather to claim that the values that IA is taken to protect or instantiate can equally well or better be protected or instantiated by II.

On some accounts, IA is intrinsically valuable: IA is a constituent of the good life or of flourishing (Roberts and Wood 2007). Vega-Encabo (2021) adumbrates several accounts that ground its value in this sort of way. One possibility is that IA owes its value to its apparent necessity for the dignity or self-respect of persons. On this sort of view, to lack IA would be to be a creature molded by others, and that seems incompatible with the respect due to rational beings. On another view, IA is valuable due to its connection to agency; specifically, by making agents answerable for their beliefs. Vega-Encabo develops this thought by reference to a credit account of knowledge, on which ‘knowledge’ is attributable to an agent only if her true belief is the product of the virtuous deployment of cognitive capacities.⁵ Since on this sort of account, knowledge must be due to the agent herself, and not to others, she must believe autonomously.

Tanesini (2021a) grounds the value of IA in agency via a different route. On her view, IA is valuable, at least in part, because only an autonomous agent qualifies as an informant, rather than a mere source of information. Autonomy is required for being an informant, because only autonomous agents are answerable for their beliefs, and therefore their testimony. Vega-Encabo’s own view is rooted in how IA manifests in the judgments the virtuous agent is called upon to make. IA manifests in how we balance skepticism versus deference in interactions with others. On his view, IA should be identified not only with exploring ideas for oneself and making up one’s mind, but especially with these higher-order decisions concerning when it’s appropriate to engage in such activities.

Another family of accounts emphasizes the instrumental value of IA: it’s valuable (at least in part) because it conduces to or is required for true beliefs, knowledge or understanding. Matheson and Lougheed (2021) suggest that IA might be valuable because groups of independent thinkers are epistemically more reliable than those in which people are disposed to defer to one another (one might think here of the Condorcet jury theorem, which states that under some conditions the larger a group the more likely it is to be accurate: one of these conditions is epistemic independence between group members). Other views hold that IA is necessary for understanding, because understanding can’t be conveyed by testimony (Zagzebski 2007; Pritchard 2016; see Boyd 2017 for dissent). Matheson (2022) gives perhaps the most thorough defence of the instrumental value of IA: IA enables understanding why p, in what he calls the good case, and understanding of the shape of the debate whether p in the bad case. In addition, he argues, the person who engages in first-order deliberation whether p is more likely to be able to deal with apparent defeaters than the person who accepts that p on the basis of reliable testimony, since the latter agent has no way of knowing whether the apparent defeater was taken into consideration by the testifier.⁶ Further, intellectually autonomous thinking conduces to the inculcation of other epistemic virtues, such as the virtue of intellectual humility, Matheson suggests.

If my claim, that IA should be replaced by II, is to be made plausible, it’s incumbent on me to show that II can secure the goods that IA secures, while minimizing any costs that IA carries with it (and without adding significant costs of its own). I will focus, first, on the costs of IA, before showing that II can do the work it’s called upon to do. I will then return to the value of IA, to show that II can do as good a job of securing these goods.

The very fact that theorists begin almost every account of IA with a disclaimer – IA is not what it sounds like – suggests that the terminology is misleading, and that we need to work to avoid these misleading connotations. More disturbingly, there’s evidence that our cultural infatuation with IA distorts public debate. Think of the prevalence of calls to ‘do your own research’ and other such invocations of IA on social media. While philosophers will be quick to point out that IA, as they conceive it, doesn’t justify such calls (and several philosophers have specifically addressed the usage that conspiracy theorists and contrarians make of them; see Buzzell and Rini 2022; Levy 2022; Ballantyne and Dunning 2022), it’s harder to rebut such calls when one is oneself in the business of defending a virtue that seems to support it.

The strength of our cultural infatuation with IA is revealed by examining how the educated public responds to calls to ‘do your own research’. Such responses never (to my knowledge) consist in putting forward a more nuanced conception of IA. Instead, the most common response is to mock those who make them as hypocrites. We think they haven’t done their own research. In fact, we take them to be especially gullible, accepting the misinformation regurgitated on social media or talk radio, rather than the free thinkers they claim to be.⁷ As this response shows, ordinary people seem to accept the propriety of such calls to IA: they see no conflict between the heroic independence called for and what the virtue demands. That suggests that intuitions are shaped to our epistemic detriment by our advocacy of the virtue of intellectual autonomy.

Even the most sophisticated philosophers, who are acutely aware of the need for a nuanced and modest account of IA, have their intuitions and their argumentative strategies distorted by the canonical name for the virtue: they underestimate the work that II does in their own discussions, or put forward accounts distorted by a lack of attention to the virtue. To show this, I’m going to work through discussion of the virtue by Heather Battaly (from whom I’ve borrowed the label ‘II’ and whose account of the virtue I’m building on). Battaly can’t be accused of ignoring II, but (even) she underestimates just how much work it does even in her own discussion.

Battaly argues that we ought to recognize two distinct virtues governing how we weigh our own contribution against that of others: IA and II. The two virtues are typically correlated, but they can dissociate.⁸ That is, agents may possess IA without II, or II without IA; they may also possess both or neither. IA is, roughly, the behavioral and motivational dispositions to think independently. II is the behavioral and motivational dispositions to think interdependently – ‘to think with other agents (and sources) and consult other agents (and sources)’ (160). To show they can dissociate, as well as to clarify each, Battaly briefly describes several different sorts of cases. In what follows, I work through some of these cases, both to show that she’s invoking II more than she takes herself to be, and that (correlatively) II can do all the work that’s needed in this domain by itself.

Battaly adopts an explicitly Aristotelian framework: for her, virtues are means between deficiency and excess. The agent who manifests an excess of IA is disposed to think for themselves when it’s inappropriate to do so, or in the wrong ways (for example, when they’re unreliable in the domain) or with respect to some of the wrong questions (e.g. questions that can’t be answered). This latter sort of case plays an especially important role in Battaly’s argument: such cases illustrate how IA and II can dissociate. An agent can display excessive IA by forming beliefs concerning matters on which no one is (currently) reliable. ‘She, for example, believes – rather than merely hypothesizes or predicts – that there is/isn’t sophisticated intelligent life elsewhere in the universe’ (157). In this case, the agent manifests an excess of IA without an insufficiency of II. Since (by hypothesis) this is a question on which other agents aren’t reliable either, it would be as inappropriate for the agent to defer to others for beliefs as it is for her to form them off her own bat. Equally, a deficiency of IA need not correlate with an excess of II, Battaly argues. The agent who is excessively deferential – in domains where she is reliable, for example, or when she defers to the unreliable – is deficient in IA. But the agent who simply gives up attempting to answer a question that is important for her is deficient in both IA and II.

Battaly also argues for the dissociation of IA and II beginning from cases focused on the latter. The agent with the trait, or the virtue, of II is disposed to think and consult with others at the right time, in the right manner, for the right reasons and in the right respects. Someone may display an excess of II if they consult the unreliable – members of a cult, for example. They might even display excessive II in relying on the reliable, if they do so in contexts in which it is more important to develop their own thinking than get at the truth. Again, though, an excess of II need not correlate with a deficiency of IA. The agent who defers to others on questions in which no one is reliable is excessive in II but not deficient in IA (since these are questions on which it would be equally inappropriate for them to form their own beliefs autonomously). Conversely, deficiencies of II need not be correlated with an excess of IA, though they often are. The agent who just gives up when it would be appropriate to think with others is deficient in II, but not excessive in IA.

These dissociation cases are central to Battaly’s case for thinking that we need to identify not one but two virtues in this domain: IA can’t do all the work in identifying what goes wrong in these cases. When agents independently form beliefs on questions on which others are much more reliable than they are, they go wrong epistemically, but not by failing to manifest IA (‘if excessive thinking for oneself isn’t a kind of autonomous thinking, then what is it?’ (168)). We need to invoke another disposition or virtue to explain where they go wrong. The fault lies not in IA, but in a deficiency in II. Similarly, to understand what goes wrong in some cases in which an agent manifests excessive II, we need to invoke IA. Finally, we need both to understand how agents should alter their dispositions depending on context. There are hostile environments in which II is inappropriate, and virtuous agents must rely exclusively on IA; perhaps there are friendly environments in which II is the only virtue.⁹

Tradition identifies the central virtue in the domain of managing our epistemic relations with others as IA. Because Battaly bucks that tradition, she defends her approach by arguing that IA can’t do all the work all by itself. She doesn’t consider a still less traditional approach: calling on II to do all the work. It’s that approach I aim to defend here. As I’m sure she’d agree, the question isn’t ‘can we identify a virtue such that that virtue can fit all the cases?’ Tradition has more or less succeeded at that task. The question is instead ‘what is the most perspicuous way of dividing up virtues such that they can do the epistemic work we require of them?’ Tradition has succeeded in explaining all the cases, Battaly implicitly suggests, by loading up IA with all sorts of dispositions that fit uneasily with its core. It has asked a virtue that is focused around thinking independently to contort itself to count apt deference as virtuous. It would be clearer and provide a better map for the epistemic landscape, she suggests, if we identified distinct virtues to do this work. That said, we risk going too far in this endeavor. We mustn’t make the dispositions too narrow (we’d face an unruly explosion of dispositions if we went that way). I’m going to suggest that II can do all the work traditionally thought to fall to IA, and that Battaly thinks is best done by a combination of virtues, without loading the term with dispositions that sit uneasily with it.

The first thing to note is that II is already doing much more work in Battaly’s discussion than she gives it credit for. Consider a case in which an agent independently comes to hold a belief about a question on which no one is reliable; Battaly gives the example of forming a confident belief whether there’s sophisticated intelligent life elsewhere in the universe. Now ask: why is it the case that agents should refrain from forming such beliefs? In all such cases, reliable testimony from other agents is central to the answer: her independent belief-formation is wrong, at least in part, because it manifests insufficient deference to others. Suppose she’s not an expert on the domain. In that case, it’s only on the basis of testimony from experts that she could know that this is a question on which she should refrain from belief formation (after all, this is a question that turns on specialist matters in mathematics, physics, evolutionary theory, astronomy, and so on – it’s consistent with common sense that intelligent life elsewhere is a mathematical certainty or that it’s vanishingly unlikely). Even an expert must defer in order to be in a position to know that she ought to refrain from confident belief formation: after all, multiple disciplines bear on the question, not one (actually, experts must often defer to other experts within their own domain: it’s a reality check that keeps them from spiraling into confirmation bias).

Even the agent who forms beliefs about such matters by deference displays a problem with II: she fails to defer to the right agents. Again, a lack of appropriate II is explanatorily central to such cases. Either way, the person who forms a confident belief on this kind of question, whether independently or by deference to the unreliable, is at fault because they’re not deferring as they should. Better deference is what’s called for here: she should defer to the reliable (about the question on which she defers, or about the reliability of those she defers to).¹⁰

Battaly’s cases require more work from II than she thinks. I now want to argue that they require much less work from IA than she thinks. In fact, II does all the work in these cases – and in every other case, too. It’s important to recall that II is not simply a disposition to defer to others, though it entails such a disposition on many occasions. Rather, it is the disposition ‘to think interdependently – to think with other agents (and sources) and consult other agents (and sources)’ (160). Thinking with and consulting is very different from mere deference: even when it calls for deference, it’s never mere deference (which Matheson 2022 rightly identifies as antithetical to intellectual virtue).

The agent who thinks with others is intelligently active from the get-go. She’s highly sensitive to the domain in which a question arises. When she’s not herself an expert on the question, she relies on others to identify whether this is a question on which there’s expertise to rely on, and on whether this is the kind of occasion on which one should defer. If she then defers, she does so intelligently. She does not simply adopt the beliefs of those around her. She filters testimony on the basis of cues to competence, consensus, and with reference to the values of those providing it (perhaps because shared values provide reassurance that the person respects our epistemic interests and is not seeking to exploit us (Levy 2019)). She is also attentive to the plausibility of what is said: what Goldman (2001) calls a ‘non-discriminating reflector’ who does not filter testimony for plausibility is in fact very rare, because deference is never mere deference (Coady 2006). Rather, it’s always flexible and intelligent (Buskell 2016). Even the members of cults assess the plausibility of cult doctrine, though they are more motivated to leave the group by conflicts between their moral values and those of the cult than by the implausibility of cult beliefs (Sauvayre 2015).

Virtuous cognition is never merely deferential, nor is it ever truly independent. Rather, it intertwines deference and independence. The person who thinks with others thinks: she is active in evidence gathering and assessment, in conversation and debate with others. Such agents are ‘contributing members in the intellectual division of labor’ (as Matheson (2021) demands of the autonomous agent). When a question arises that is (partly) within her domain of expertise, she’s ready to provide testimony about it, or to work with others to answer it. Even in these cases, she’s attentive to testimony: no expert is able to strike out on their own, without guidance from other experts within and beyond their domain of expertise (peer review, or some other way of crowdsourcing quality checks, is important because even expert-level thinking must be scaffolded, checked and supported by other experts, and every expert contribution draws on multiple sources). No thinker, good or bad, ever strikes out truly on her own; the virtuous thinker least of all.

All the work of virtuous cognition seems to be easily shouldered by II; it’s hard to see what’s left for IA to do. Perhaps, though, my discussion has been distorted by its base: by building on Battaly’s work (with its explicit invocation of II). Perhaps if I started elsewhere, it’d be more apparent what IA is needed for. Consider, then, Roberts and Wood’s (2007) canonical account of IA. On their view, IA consists in the disposition to resist ‘alien hetero-regulators’ (282), while embracing appropriate ‘proper hetero-regulators’ (285). A hetero-regulator is a source of information, authority, sanction and criticism. An agent embraces a hetero-regulator by ‘appropriating’ it. To appropriate a hetero-regulator is to make it one’s own, such that by being regulated by it, one is governing oneself. Being governed by appropriated hetero-regulators is governing oneself: it is auto-nomos (self-rule).

Roberts and Wood seem to provide substance to the claim that virtuous social cognition is autonomous. They recognize the extent and the need for intellectual dependence, but clarify how and why such dependence can count as self-rule. This, I will suggest, is an illusion: their condition is far too demanding. It’s too demanding in two different ways. First, we can satisfy it only in very narrow domains, entailing that we’re all non-autonomous in much of our mental lives. Second, in those domains in which we succeed in satisfying the condition, in satisfying it we eliminate the need for hetero-regulation altogether.

Roberts and Wood’s exemplars are supposed to have appropriated alien hetero-regulators by coming to understand them. A person who has ‘mere propositional knowledge of the regulator … cannot be said to have appropriated the regulator in the way required for autonomy’ (278). Understanding must be deep: the person who has truly appropriated the hetero-regulator can improvise and invent on its basis. To this they add a second condition: not only must she understand the tradition she appropriates, she must also understand the pressures she resists. ‘Otherwise his resistance can hardly be intelligent, and without intelligence it cannot be autonomous’ (278).¹¹

The first problem with these conditions is that they are very demanding; too demanding for anyone to meet them with regard to more than a narrow domain of knowledge. Genuine understanding is hard to come by. Roberts and Wood give as their example the Catholic layperson who understands the tradition she belongs to so deeply that she spontaneously thinks in its terms and make creative and intelligent use of it. The empirical evidence indicates, however, that very few believers meet this condition: most have shallow and inconsistent understandings of their own traditions. They do not spontaneously think in its terms: religious believers who know what their theology officially commits them to spontaneously think in intuitive terms inconsistent with doctrine (Barrett 1999; Slone 2004). For example, they know that doctrine requires them to accept that God is omniscient and omnipresent, but they spontaneously attribute limited knowledge and the need to complete one task before undertaking another to the deity. That’s unsurprising: theology is hard and doctrine the product of many centuries of work by impressive thinkers. It’s hard to wrap one’s head around the divine attributes like omnipresence – not to mention the Trinity.

There’s nothing distinctive about religious believers in this respect: non-believers (and believers too, of course) govern much of their thinking by science, but fail to meet the demanding (individual) understanding condition Roberts and Wood set down. We may think we do, but we’re subject to the so-called illusion of explanatory depth: we can’t explain the claims we take ourselves to understand (Rozenblit and Keil 2002; Mills and Keil 2004). Even those with some college-level education in the sciences manifest spontaneous thought about life at odds with the theory they’ve learned (Shtulman 2015). Believers and non-believers alike, we outsource much of our cognition to sources we trust. We genuinely understand only domains we’ve immersed ourselves in for very extended periods of time.

We might respond to this concern in the same sort of way in which some virtue ethicists and epistemologists have responded to the situationist challenge to virtue theory. The situationist challenge leverages empirical research that appears to show that behavior is best predicted by features of the situation in which agents find themselves, and not by their characters. The most common response this worry has been to argue that the data is consistent with a virtue framework, because the virtues are rare (see Alfano and Fairweather 2017 for the situationist challenge to virtue epistemology; Baehr 2017, in the same volume, discusses the rarity response). Similarly, we might take the evidence that few of us satisfy Roberts and Wood’s condition for appropriation of an intellectual tradition to show that few of us possess the virtue of IA.

The problem with this account is that it does not make IA (merely) rare; it ensures that no one possesses the virtue in the very domains in which it is called for. The person who fully understands an intellectual tradition is not in fact governed by a proper hetero-regulator: there’s no real sense in which the regulator is ‘hetero’ to the agent. It doesn’t give Roberts and Wood what they want: an account of proper ‘other regulation’. It’s too successful as an account of autonomous thought: it gives conditions under which the self governs the self, with the ‘other’ out of the picture. But it’s also not successful enough: it ensures that none of us can be autonomous with regard to bodies of thought we don’t fully understand. It entails that there can be no virtuous acceptance of expert testimony, unless we’re ourselves experts in that very domain. It entails an account of IA that can’t do the work of properly regulating our intellectual relations to those outside these domains.

We need, therefore, to replace this account with something much less demanding. Unsurprisingly in this light, immediately after having advanced understanding as a seemingly necessary condition of autonomy (‘Mother Teresa could not be an autonomous adherent of the Christian moral tradition without understanding it’ [Roberts and Wood 2007, 279]), they weaken it considerably (‘autonomy must be compatible with accepting particular things the rationale for which one does not grasp’ [279]). It’s enough, Roberts and Wood seem to say, if the agent understands why the source is authoritative and willingly submits to it for that reason. That’s surely a more plausible condition, albeit one their official account seems to rule it out. More pointedly, it’s a condition that can be satisfied by II: the agent who submits to the authority of a tradition because she sees its authority as earned is better described as thinking with it than independent of it.

II is a far less demanding condition: one we can satisfy with regard to the broad range of domains across which a virtuous agent might reasonably defer and contribute. It requires not understanding or appropriation, but intelligent and flexible deference, utilizing cues to expertise, guided by testimony. We don’t appropriate the knowledge we defer to, but we are intelligently receptive to it. II reflects the kind of epistemically social animal we are and should be. Indeed, the illusion of explanatory depth itself illustrates our flexible and intelligent deference. It arises not from our vices, but from the intelligent outsourcing of cognition to other agents (Wilson and Keil 1998; Rabb, Fernbach, and Sloman 2019): we take ourselves to understand because we have epistemic access to understanding (distributed across other agents and across the world). This is our thinking together (see Levy 2021 for further discussion).

It should immediately be added that Roberts and Wood have many insightful things to say about virtue in this domain. Their discussion of how the virtuous individual should respond to criticism, to sanction, to models and to testimony, when and how they should resist pressures extraneous to truth-seeking and so on is valuable. However, none of this is inconsistent with II – rather, it is better explained by II than by their official account. II does a better job than their official account of capturing the ‘dependent independence’ (285) of the virtuous agent. Even the regulation that Roberts and Wood most prize, self-governance by an intellectual tradition one has made one’s own, is fully compatible with II. The agent who manifests II plays her part in the division of cognitive labor, and that requires her to contribute her expertise when it’s appropriate to do so. She will identify alien hetero-regulators as alien in the light of testimony from others, or by its conflict with authoritative information, and be herself a source of authoritative testimony when the occasion calls for it.¹²

Let’s conclude, finally, by returning to the values supposedly underwritten by IA, to assess whether we’d lose anything were we to jettison it in favor of II. IA is supposed to ground dignity or the self-respect of persons. These properties may be inconsistent with mere deference, but they’re fully compatible with II: the person who manifests it testifies authoritatively when she may, and accepts testimony flexibly and intelligently when she must. IA is supposed to ground knowledge because autonomous belief is credit-worthy. But there’s no reason to think the agent who manifests II is less credit-worthy: certainly not in their authoritative testimony, but also in their intelligent and flexible deference. IA is supposed to be necessary for an agent to count as an informant, because only the autonomous agent is answerable for her beliefs. Again, there’s no reason to think that the agent who manifests II is any less answerable: their testimony is either authoritative, or has been appropriately filtered for credibility. IA is supposed to be instrumental to knowledge and to understanding. There’s no reason to think II inferior on these grounds; quite the opposite. IA is supposed to be manifested in the intelligent assessment of when to defer and when to hold firm. That’s the very business of II, not something it needs extraneous help with.

The only good that IA might conceivably secure that II might not is the good that arises from a network of independent thinkers. There’s empirical evidence that group deliberation is more reliable than individual deliberation, and that some degree of obstinacy contributes to this ‘assembly bonus’ (Sunstein 2006). It is conceivable that agents who see themselves as thinking with others might be too easily cowed. They might be less willing to table private information, when they take it to be incompatible with the group’s conclusion, or less willing to express conflicting opinions. Conversely, one-sided cognition, with agents subjecting conflicting opinions to searching scrutiny while being easily satisfied with arguments for views they find congenial, might be adaptive in as much as it ensures that groups explore the space of reasons much more effectively than any individual could (Mercier and Sperber 2011). On this view, the value of IA would, paradoxically, lie in how it conduces to group deliberation.

We might concede the point and argue that the benefits of such obduracy are outweighed by the costs: the lab-based studies that demonstrate how myside bias can lead to better results pales into insignificance in the face of the ill-effects of the bias in the world at large. In fact, I think a stronger response is available: the assembly bonus effect is found outside the laboratory, but only when dissent is appropriately channeled and shaped. Institutions like peer review shape dissent in this way: by making it anonymous, they free people from social pressures and by formalizing it, they channel it in productive directions. Frank but respectful criticism, backed by evidence and argument, is familiar from the seminar room. So, of course, is rudeness and sheer obstinacy. The first is distinguished from the second because it qualifies as a kind of thinking with, and therefore manifests II. It’s worth adding, moreover, that though reliable belief formation often requires dissenting voices, it can’t end with dissent: the aim is instead to converge on the truth. This, too, is a reason to see dissent, channeled by institutions and relations of respect, as a way of thinking with others.

We’ve seen that II can secure the goods supposedly underwritten by IA. But the concept of IA also has a role in explaining darker phenomena. Matheson (2022) calls on it to explain certain kinds of epistemic injustices; testimonial injustice might arise from a failure to accept testimony, in the name of thinking for oneself. Tanesini (2021a) appeals to deficits and excesses in autonomy (and the social conditions that give rise to such deficits and excesses) to explain a range of further epistemic vices: arrogance, silencing, servility. These phenomena can be as easily and as perspicuously explained as arising from pathologies of II as IA. The agent who refuses to countenance the testimony of others due to identity prejudice (Fricker 2007), is deficient or inappropriate in their II: they fail to think with or to consult properly with others. The agent who manifests excessive autonomy, as Tanesini conceives of it, is clearly deficient in II; they’re refusing to think with others. The agent who is not answerable for her beliefs because her confidence has been eroded due to a lack of recognition is prevented from thinking with others; through no fault of her own, she’s unable to be a contributor to the distributed process of cognition.

Finally, it’s worth mentioning one more role IA is called upon to play by Battaly. She argues that in an epistemically hostile environment, in which deceptive informants vastly outnumber reliable sources of information, being an ‘epistemic loner’ might be virtuous. I have deep worries about the cogency of this thought experiment. Given the deep social dependence of human cognition, a hostile environment like the one she envisages is hostile to all virtues. There’s no way to flourish epistemically in it: being an epistemic loner protects one from certain falsehoods, perhaps, but does not replace them with truths (indeed, it probably randomizes the falsehoods). Further, identifying that one is an epistemically hostile environment is not in any realistic situation itself something that one can do on one’s own. Even to discover that one should not rely on others, it is necessary to rely on others.

Conclusion

IA has its costs. We should be suspicious of any concept that must be introduced with the caveat ‘it’s not what it sounds like’. We should expect the cultural infatuation with a concept like that to have ill-effects – and there’s reason to suspect that IA does indeed have such effects. I’ve pointed to those effects in the public domain, but I think they’re there in philosophical debate too. We tend to give social cognition too individualistic an interpretation – a tendency that arises from and reinforces a virtue centred around thinking for oneself. Good thinking is never isolated thinking. Philosophers ask ‘when should we think for ourselves?’ That’s a bad question, that leads us in the wrong direction. We should ask ‘how should we think together’ and ‘how do I best think with others?’¹³

Let me finish with a brief discussion of three objections. First, it might be objected that there’s no need to be so revisionary as to replace IA with II. The notion of relational autonomy (Mackenzie and Stoljar 2000) already has a firm place in the lexicon, and it does all the work I’ve called upon II to do. In particular, it lacks the individualistic and heroic overtones of IA. Whatever we call the virtue, the notion of II I have in mind is quite distinct from relational autonomy as it currently features in the virtue epistemological literature. Elzinga’s (2019) development of the concept centres on how the dispositions – in particular the conceptual know-how and the epistemic confidence – required for virtuous thinking are causally and constitutively dependent on relations to others. That’s a point well worth making, but it’s somewhat orthogonal to the question I’m concerned with. I’m concerned with how we ought to conduct ourselves in the social-epistemic domain, not with how the dispositions in question are caused or constituted. Given that relational autonomy already has an assigned philosophical meaning, we do better to adopt new terminology for the virtue I have in mind. It’s important to add that the proposal to replace IA with II is not intended to have ramifications for other uses of autonomy in other debates. There are of course cases in which independence is valuable, and perhaps there’s a corresponding virtue that deserves the name ‘autonomy’ (I take no stand on this question). The proposal is simply that epistemic dependence is so pervasive and genuine independence so corrosive that ‘autonomy’ is misleading when it comes to the regulation of our intellectual relations with one another.

Second, there is (I concede) a role IA is sometimes called upon to play in philosophical argument that II is ill-suited to. I have in mind Carter’s (2022) historical account of autonomy. On his view, an agent counts as knowing that p only if they satisfy a historical condition: roughly, their belief must not have been formed in a way that pre-empted or bypassed her cognitive competencies. II doesn’t seem a good label for the problems that arise when agents fail to satisfy this condition. Well, perhaps that’s a reason to keep IA within our terminological repertoire. But the role Carter calls upon IA to play is very different from the role that most virtue epistemologists call upon it to play; perhaps we should restrict IA to playing this former role, and replace it with II for the latter. It’s important to note that using IA to refer to this kind of condition doesn’t carry the risk of giving apparent approval to calls to ‘do our own research’, since the condition is backward-looking and far easier to satisfy than the tacit call for intellectual independence that accompanies IA as it is currently understood.

Third, it might be objected that II is too undemanding and too blunt to do the work we require from the virtue. Agents who grow up in epistemically benign environments might display the virtue simply by accepting testimony undiscriminatingly. Conversely, we’d be unable to criticize agents who grow up in environments in which false claims circulate if they accept them: in doing so, they’d display appropriate deference. We’d also have no grounds for criticizing agents who develop in environments that are so hostile to the virtues that they could not develop the capacity to discriminate among sources of testimony.

We should accept that II will not provide us with the tools to diagnosis every epistemic deficit in regulating our intellectual relationships. In some cases, virtue theoretical concepts simply may not be applicable, because intellectual agency is not sufficiently developed for these demanding concepts to get a grip. In defence of II, it should be noted that when these sorts of cases are a problem, they’re a problem for every conception of a virtue that focuses on how we regulate our intellectual interdependence: when the environment does not allow for the development or deployment of the virtues, the solution cannot turn on how we regulate our mental lives. We need a variety of concepts to diagnose epistemic ills, rather than calling upon any one concept to do too broad a range of tasks. We risk confusion if we load up our concepts with too many, and too disparate, demands.

Virtue epistemologists – in the responsibilist tradition – invoke IA to provide a regulative ideal. It’s because it plays this role that it’s risky: calls to be more autonomous lead us away from thinking well with others and leave us without an easy way to identify the problem with those mavericks who strike out on their own. IA, as Carter invokes it, is much more fully something outside of the agent. It plays a small and only indirect role in guiding virtuous cognition: calling on someone to be more autonomous in this sense can only be calling on them to avoid certain situations (e.g. joining a cult, or having a very radical cognitive enhancement), not to govern their cognition in certain ways. Obviously, there’s some overlap between the cognitive competences we have in mind in the first case and the second, but they’re different enough to warrant considering separate labels for them. We need to rectify the names!

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The work was supported by the Arts and Humanities Research Council [AH/W005077/1].

Notes on contributors

Neil Levy

Neil Levy is a Senior Research Fellow at the Oxford Uehiro Centre for Practical Ethics, and a professor of philosophy at Macquarie University, Sydney.

Notes

1. Some philosophers talk about epistemic and not intellectual autonomy. To my ear, the second has a slightly broader scope than the first: knowledge is central to, but does not exhaust, cognition. Nevertheless, I think most philosophers take them to be synonymous. Any differences don’t matter here: the names are both misleading and for the same reasons.

2. I am grateful to an anonymous reviewer for stressing the need to distinguish the sense of autonomy at issue in this debate from other, albeit related, senses. The reviewer suggests, intriguingly, that even within virtue epistemology, people may talk past one another because they have different conceptions of autonomy in mind. The reviewer suggests, for example, that Battaly (2021) is concerned with a notion of intellectual independence that is quite distinct from Roberts and Wood’s (2007) more Kantian account of autonomy. While I’m sympathetic to the suggestion, I will follow Battaly (2021, 154) herself in treating the accounts as concerned with the same virtue, since this reflects how the accounts function in contemporary virtue epistemological debates. In the conclusion to this paper, I will note one usage of ‘intellectual autonomy’ within epistemology that seems sufficiently distinct from the regulative sense to deserve its own name.

3. Batally notes the whiff of ad hoccery about the appeal to good judgment. That, at any rate, is how I interpret her note 23: ‘One worry about good judgment is that it is designed to magically solve this and other problems!’ (171).

4. Elgin (2021) also emphasises the role of epistemic interdependence alongside autonomy. She argues that the two are mutually supporting. But the role she calls upon IA to play is, precisely, contributing to the conversation: there seems no need for any virtue other than II.

5. Carter’s (2022) is a book-length argument for a different – historical – autonomy condition on knowledge.

6. Matheson notes that this benefit is ‘not guaranteed’. I am sceptical that in general first-order deliberation improves a novice’s capacity to respond to defeaters, compared to accepting expert testimony. Compare two agents who have come to believe that anthropogenic climate change is a real and serious threat to human flourishing; the first by doing her own research and the second on the basis of testimony from an expert. They are now each told that human-generated CO₂ emissions are vastly outweighed by emissions from natural sources, and that therefore these emissions can’t be responsible for any climate change. It’s far from obvious to me that the first is better off than the second. The first may or may not have encountered this very claim before. If she hasn’t, she’s no better off for her research. If she has, it’s only if the claim is repeated in much the same form as before and she’s aware of a good response to it in that form, and she trusts that response (by hypothesis, she’s no expert!) that she’s in a position to see that the evidence is misleading. If the second agent is in the position of most of us who accept the consensus, though, she’s likely to shrug, secure in the knowledge that the consensus is extremely strong and that it is vanishingly unlikely that this sort of evidence hasn’t been taken into account.

7. Substantiating this claim is of course difficult: I’m not aware of any systematic research undertaken on the prevalence of such responses. I can at least provide evidence that my impression is widely shared. One of Urban Dictionary’s top two definitions of ‘do your own research’ is ‘a phrase meaning “Google until you find something that agrees with you”’ (BaronCoop 2021) (its other top definition points to a different function: to avoid having to do the work of providing references). Similarly, the Thinking is Power website, which aims to teach critical thinking, defines ‘do your own research’ as ‘using a search engine to find information that confirms what they already think is true’ (Trecek-King 2021). Buzzell and Rini (2022) offer a more nuanced account, on which those who laud doing their own research inevitably slide back into epistemic dependency.

8. It’s worth noting the structural similarity between Battaly’s account of these twin virtues and her (co-authored) account of intellectual humility. The latter, too, identifies two distinct dispositions that together do the work that others take to be done by one. In the case of intellectual humility, the second disposition is the vice of arrogance: on this account, an agent can manifest the virtue of intellectual humility while also being excessively arrogant (Whitcomb et al. 2017).

9. Battaly also invokes linguistic propriety to argue that we need more than one virtue. Suppose it’s epistemically appropriate for an agent almost always to defer to others, due to the epistemic properties of the world they inhabit. If we called upon IA to do all the explanatory work in this sort of case, we’d be forced to say that such an agent exhibits the virtue of IA, despite never thinking for themselves. Equally, we’d be forced to call upon the excessively arrogant individual to be more autonomous in their thinking.

10. It’s widely held that we can identify who to defer to for ourselves, on the basis of cues to expertise: track record, acceptance of the consensus, honesty, argumentative capacity, intellectual honesty and so on (Anderson 2011; Goldman 2001; Guerrero 2017). To a large extent, though, we’re reliant on testimony to utilize these very cues. How do I know that this prize or these publications are a marker of expertise? How do I know that the person has exhibited intellectual honesty? How do I know what the consensus is? Only argumentative capacity is something that I can, to some degree, assess for myself – but unless I’m an expert, it’s a heuristic that allows me to distinguish the practiced from the unpracticed, not the reliable from the unreliable (Levy 2021).

11. We’ve already noted that there are grounds for denying this: the flexibility and selectiveness of deferential mechanisms manifest genuine intelligence. See Levy and Alfano (2019) for discussion.

12. There are grounds for suspicion that thinking in terms of IA distorts Roberts and Wood’s intuitions, just as it distorts public debate in the direction of what Buzzell and Rini (2022) call heroic epistemology. Consider their discussion of ‘sanctions’, which they think of as extrinsic rewards and penalties. Sanctions range from money to prizes to acceptance and rejection of journal articles. These sanctions, they claim, can be useful and can train the person to orient themselves toward epistemic goods. But as merely extrinsic supports, they ought to fall away. Autonomy, they write, ‘is freedom from the push and pull of sanctions in the same way in which Wittgenstein’s Tractarian philosopher is free from the ladder of referential language’ (270). But sanctions are, inter alia, epistemic signals – signals that provide us with guidance as to when we’re on the right track. To think we can float above such signals is to embrace an individualistic epistemology, albeit one on which the person has first internalized the thoughts of others. Epistemic scaffolding is not merely developmentally necessary; it’s required in an ongoing way. The maverick, no matter how intelligent and educated, runs enormous epistemic risks and is badly wrong far, far more often than they’re right.

13. McKenna (2023) argues that the yearning for IA tends to frustrate other epistemic goals. His argument might support mine indirectly, by demonstrating we have less to lose and more to gain from abandoning it.