On the Possibility of Group Knowledge without Belief

Abstract

Endorsing the idea of group knowledge seems to entail the possibility of group belief as well, because it is usually held that knowledge entails belief. It is here studied whether it would be possible to grant that groups can have knowledge without being committed to the controversial view that groups can have beliefs. The answer is positive on the assumption that knowledge can be based on acceptance as well as belief. The distinction between belief and acceptance can be seen as a refinement of the ordinary language concept of belief, and it may be useful in understanding the nature of epistemic justification and classifying various types of epistemic subjects.

Keywords:

• Group Knowledge
• Group Belief
• Rejectionism
• Acceptance

Acknowledgements

The author wishes to thank Giedre Vasiliauskaite for her valuable comments, Pekka Mäkelä for his good remarks and Raimo Tuomela for his encouragement and advice. The author is grateful to Kay Mathiesen who suggested writing about group knowledge and who made several important points during our long discussion in Helsinki. Thanks are due also to Nike Parland who pointed out some matters that still require thinking on the author’s part. Also, the author wishes to thank the audience of the presentation at the Conference on Collective Intentionality V for their critical questions and discussions afterwards. This work has been partially supported by the Academy of Finland, grant numbers 214244 and 210406.

Notes

[1] For a different approach to group beliefs see Miller (2006), who takes group beliefs to be mutual beliefs with the content that the group has accepted a view.

[2] To avoid possible misunderstandings, it should be stressed that claim (REJ) concerns non‐summative beliefs. Its intended meaning is that there is no such thing as a proper belief whose carrier is a group. In particular, accepted group views are not proper beliefs. The claim does not deny the existence of summative group beliefs whose carriers are the individual group members.

[3] I am not sure whether these should be taken as defining features or whether there could be more sophisticated ways of spelling out their differences in terms of their respective functional roles (see, for example, Lehrer 2000), but at least voluntariness seems to work better as a heuristic method of identifying cases of acceptance compared with the other features typically proposed in the literature.

[4] There seems to be a problem in Clarke’s account, however, as Cohen (2000) has noticed. Clarke (1994, 146) acknowledges that beliefs are involuntary and acceptances are voluntary, but this is not compatible with his acceptance–belief entailment thesis because that would make beliefs voluntary too. Perhaps Clarke could settle for acceptances’ being only half‐voluntary and say that we can only choose, of those propositions that we believe, whether we also accept them or not. It is doubtful whether this would solve the problem: Suppose we believe p. According to Clarke, we can decide not to accept p; that is, we can reject p. A slightly odd result of half‐voluntarity would be that we cannot then accept not‐p, because otherwise our beliefs would be contradictory.

[5] I have considered some other arguments in an earlier paper (see Hakli 2006).

[6] Actually, Goldman calls summative all states that depend on all the group members’ corresponding individual states, but the more usual use originating from Quinton (1976) takes summative to mean any notion that functionally depends only on the states of (at least some) of the individuals, or stands as a shorthand expression for certain combinations of corresponding individual mental states.

[7] Common knowledge of p means that everybody knows p and everybody knows that everybody knows p and everybody knows that everybody knows that everybody knows p, and so on. Distributed knowledge (sometimes called aggregative knowledge) of p means that p is derivable from the totality of what the individual group members know. See, for example, Fagin et al. (1995).

[8] Cases where we might say that a group knows something even if none of the individuals know, except maybe as derived from the group’s knowledge, can be constructed by considering Goldman’s (1999, 81) example of a weather forecast agency in which the official weather forecast is based on the reports of several individual experts, giving more weight to those experts who are known to be most accurate. The resulting forecast may not only differ from all the expert reports (so the individuals would not personally believe the group view, but rather their own prediction concerning, say, tomorrow’s maximum temperature), but it may even be more accurate than any of them. Supposing that the same procedure has been found successful and reliable based on past experience, it could be argued that the group is justified in its view and, supposing further that it actually will be accurate, that the group knows what the weather will probably be like tomorrow. (The word “probably” seems accurate here because the topic concerns future and the weather, but this does not cast doubt on there being knowledge because it occurs only in the content of the knowledge.)

[9] Although it seems that, for Kusch, the knowing agent is still an individual. It is just that an individual cannot know alone but knowledge always requires that someone else acknowledges it as knowledge, according to Kusch.

[10] Suppose a group like a panel of judges (A, B and C) must decide on a verdict. The panel must find the defendant liable if and only if it finds that the defendant’s negligence caused injury to the plaintiff and that the defendant had a duty to care for the plaintiff. The votes of the judges are as follows:

If the panel votes simply on the issue of liability, each judge considers the issue independently of the two others and the panel’s decision is made using majority principle on the basis of the last column in the above table. Thus the panel comes to the conclusion that the defendant is not liable; this method Pettit (2003) calls the conclusion‐driven approach. On the other hand, the panel may use the premise‐driven approach and vote on the two premises separately, and then aggregate the results of the majority votes on each premise by deciding that the defendant is liable if and only if both premises are endorsed by a majority. As can be easily seen from the above table, this gives a contrary result, finding the defendant liable. The latter method is said to collectivize reason, because although it may lead to decisions that most group members would individually reject, it guarantees that the decisions of the group are mutually consistent. This is crucial especially in temporally extended situations where the group’s new decisions should be consistent with its decisions in the past.

[11] However, Mathiesen (2006, 163) thinks that performative group views are not appropriate as the basis for knowledge, because they have a world‐to‐mind direction of fit and thus cannot be evaluated as true or false, unlike ordinary beliefs that have a mind‐to‐world direction of fit. Tuomela (2007, chapter 8), on the other hand, holds that these kinds of collective acceptances have both directions of fit.

[12] With respect to objection (i), I agree—but Tollefsen’s argument does not rest on such an inference. As we have acknowledged, the attributions of knowledge only give initial plausibility to the possibility of group knowledge. With respect to (ii), I would like to point out that group mind theories have been discussed in the literature concerning collective intentionality and it seems that, although talk of intentional states, deliberation and decision‐making does presuppose minds, talk of groups’ intentional states does not require a “group mind” but the intentional states of the group supervene on the intentional states of the individuals. For example, Pettit (2003) takes it sufficient for a collectivity to count as an intentional subject if (a) it can be described as having intentional states and (b) the collectivity is rationally unified so that we can expect these intentional states to satisfy rational constraints.

[13] I can give here only a brief version; for more details, see Hakli (2006): Involuntariness of an attitude B for an agent A entails that it is possible for A to will to have B but to not succeed. For groups it is impossible to have this kind of mismatch between will and belief because there is no “group will” that could will to take a view distinct from the actually adopted view. The will of the group gets constituted in the decision‐making process and must thus accord with the resulting group view, the “group belief” that is. This is different from the case of individuals who may want to believe p but their perceptions and experiences force them to believe not‐p.

Admittedly, in some cases the decision‐making process is highly automated and regulated by the rules that the group has adopted in the past, as in the example given by Mathiesen (2006) in which a group’s secretary records the views of the individual group members, combines them into a “group view”, and then informs the group members about the adopted view. It seems that the control is limited here, so perhaps voluntariness is to some extent a matter of degree. Still, even in this case there are choice points for the agents; namely, they can decide what to report as their own views to the secretary. Thus, the inputs to the system are acceptances that are voluntary. So supposing that the group members know how the procedure works and want it to produce a certain view as a result, they can still choose the group view, not alone of course but jointly, as a group. So this seems to be an instance of a group choosing its view. If we wanted to have a totally automatic procedure without any control left for the agents, we would have to assume the possibility of sophisticated equipment with sensors monitoring the brain states (or whatever) of the group members, capable of calculating an aggregated view automatically. Now the resulting “group views” are completely involuntary, but it seems that they have also lost the properties that made them non‐summative group views in the first place. We have just imagined a device for reporting summative group beliefs.

[14] Also others, like Lehrer (1987) and Tuomela (2004), have considered knowledge that is based on acceptance.

[15] For discussion on the epistemological distinction between internalist and externalist accounts of justification, see for example, BonJour (2005). According to internalism, whatever justifies an agent’s belief must be internal and cognitively accessible to the agent; for example, the agent’s beliefs or other mental states that the agent can be aware of, or perhaps a priori necessary truths that the agent can grasp. One intuition behind this requirement is that for someone to have a justified belief, he or she is supposed to be able to provide reasons for his or her belief. Externalism, on the other hand, denies these requirements and allows the justification to be external to the agent’s cognitive perspective and perhaps something that the agent is not at all aware of. What is important for externalists is that whatever justifies beliefs is something truth‐conducive, something that makes it likely that the belief is true. For instance, reliabilists require that the processes that produce beliefs must be reliable in the sense that they produce mostly true beliefs. One intuition here is that what matters for an agent to be a good source of information is whether his or her beliefs can be relied on; that is, that most of his or her beliefs are actually true, not whether they have access to the grounds of his or her beliefs. Here I would like to slightly depart from the established terminology and employ a tripartite division into purely externalist, purely internalist and mixed views depending on whether the justification is based only on external elements like reliability, only on internalist elements like access to reasons, or on both.

[16] Schmitt (1994), however, when considering what he calls substantive theories of individual justification and applying them to the collective case, ends up rejecting internalist accounts and endorsing an externalist account, namely reliabilism. His analysis is as follows: “A group G is justified in believing p just in case G exercises a reliable belief‐forming process that yields G’s belief that p.” This seems peculiar to me. Taken together with his view that “G is justified in believing p just in case all members of G would properly express openly a willingness to accept r jointly as the group’s reason to believe p”, it would seem that the group may have accepted r as the group’s reason to believe p but r does not play any role whatsoever in the actual (reliabilist) justification of p. It may be the case that r is accepted as the group’s reason to believe p although it is not, in fact, a reason to believe p. On this account, the belief p is still justified if the group has, maybe without even realizing it, exercised a reliable belief‐forming process.

[17] However, as Cohen (1992) notes, although beliefs are involuntary, they (and knowledge resting on beliefs) may still be subject to some normative evaluation because we can indirectly affect our beliefs (e.g. by voluntarily choosing which books to read) and this voluntary decision may result in adopting false beliefs instead of true ones, so we do have some responsibility for what we believe too, not just for what we accept.

[18] However, some would prefer not to talk about justification here but, perhaps, rather some sort of knowledge that does not require justification.

[19] In addition to Cohen’s (1992), related views have been proposed by Lehrer (1987, 1999) and Corlett (1996, 91–105). Lehrer and Corlett take knowledge to require not only belief but also acceptance, which in their accounts amounts to higher‐order evaluation of belief. The distinction between belief‐based knowledge and acceptance‐based knowledge can also be related to Sosa’s (1991) distinction between animal knowledge and reflective knowledge. In Sosa’s account, if an agent’s true belief p is a product of reliable faculties, it is justified in a weak sense that does not require that the agent is aware of its reliability, and may thus count as knowledge in this weak sense. Reflective knowledge, on the other hand, additionally requires that the agent has an epistemic perspective and is aware of the belief’s origin in a reliable faculty. It seems that Sosa’s animal knowledge is closely related to what we have here called belief‐based knowledge, and Sosa’s reflective knowledge is somewhat similar to our acceptance‐based knowledge, except that in Sosa’s account reflective knowledge presupposes animal knowledge, which is not required in the present account because acceptance does not entail belief.

[20] Just before the article was going to print I found an important book, the name of which had not indicated to me that it was about belief and acceptance (Frankish 2004). I can only make a brief remark now. Frankish too distinguishes between belief and acceptance, but for him these classes are not mutually exclusive: certain beliefs (that he calls strand 2 beliefs) are actually acceptances of a certain type. These beliefs are then voluntary in his account whereas in my classification of doxastic states (Hakli 2006), beliefs are taken to be involuntary. In spite of the apparent clash of views, my impression is that our accounts are to a large extent compatible and the differences are terminological rather than substantial. Roughly put, what Frankish calls strand 1 beliefs correspond to what I have here called beliefs, strand 2 beliefs correspond to what I have (following Tuomela 2000) called acceptances as true, and those acceptances that are not beliefs correspond to pragmatic acceptances. As said, making the distinctions is more important than selecting the labels, but one benefit of adopting the terminology suggested by Frankish (or Tuomela’s term “acceptance belief” for that matter) would be that the tension between acceptance‐based knowledge and the knowledge‐belief entailment thesis (KEB) would vanish. On the other hand, this choice would require endorsing both belief voluntariness and the possibility of group beliefs. Both of these views are controversial and I have tried to avoid committing to them in this paper, but I also hold that these disputes are to some extent merely terminological.