Nothing at Stake in Knowledge*
*This paper is dedicated to Richard Nisbett, who helped create the field of experimental philosophy, on the occasion of his retirement.
Many philosophers hold that stakes affect ordinary knowledge ascriptions. Here's a version of a pair of cases aimed at supporting this: Bob and his wife are driving home on Friday and considering whether to stop at the bank to deposit a check. The lines at the bank are very long and so Bob considers coming back on Saturday. In the low stakes version, nothing of importance hinges on whether the check is deposited; in the high stakes version, it is very important that the check be deposited. Bob's wife asks whether the bank will be open on Saturday. Bob says he drove past the bank last Saturday, and it was open. However, his wife points out that banks sometimes change their hours. Bob says “I know the bank will be open tomorrow”. In the low stakes case, many philosophers maintain that Bob does indeed know that the bank will be open; in the high stakes case, these philosophers maintain that Bob is ignorant – his statement that he knows the bank will be open tomorrow is false. These philosophers also maintain that this pattern of judgments is what we would expect from competent speakers confronted with this and similar cases (e.g., Cohen, 1999, 2013; DeRose, 1992, 2009; Fantl and McGrath, 2002; Nagel, 2008; Rysiew, 2001; Stanley, 2005).
Though many philosophers agree that stakes play a role in ordinary knowledge ascriptions, there is disagreement about what explains this. One view, epistemic contextualism, holds that “to know” is a context sensitive verb and that the truth conditions for knowledge ascriptions can vary across conversational contexts (e.g., DeRose, 2009). For instance, Bob's statement “I know the bank will be open tomorrow” can be true in low stakes contexts and false in high stakes contexts. Another view, interest‐relative invariantism, denies that “to know” is a context sensitive verb and that the truth conditions for knowledge ascriptions vary according to conversational contexts. Instead, cases like the Bank cases show that practical factors—i.e., stakes—play a distinctive role in determining whether the knowledge relation obtains (e.g., Stanley, 2005). Yet another alternative, which we'll call classical invariantism, denies that “to know” is a context sensitive verb and that practical factors, such as stakes, play a direct role in determining whether the knowledge relation obtains. Instead, stakes affect knowledge ascriptions only by affecting our assessment of factors that have traditionally been taken to constitute or be necessary for knowledge, such as e.g., belief, quality of evidence, etc. (e.g., Bach, 2005; Weatherson, 2005; Ganson, 2007; Nagel, 2008). If this is right, then the role of stakes in knowledge ascriptions fails to motivate such surprising views as epistemic contextualism or interest‐relative invariantism. Naturally, epistemic contextualists and interest‐relative invariantists deny this, claiming that even when the factors that have traditionally been taken to constitute or be necessary for knowledge are held fixed, stakes continue to play a role in ordinary knowledge ascriptions (e.g., DeRose, 2009; Lawlor, 2013).
So we see a dispute over what best explains the role of stakes in ordinary knowledge ascriptions. It is thus extremely surprising that a wide range of empirical evidence suggests that ordinary knowledge ascriptions fail to display any sensitivity to stakes (e.g., Buckwalter, 2010; Buckwalter and Schaffer, 2015; Feltz and Zarpentine, 2010; May, Sinnott‐Armstrong, Hull, and Zimmerman, 2010; Turri, 2017; though see e.g., Pinillos, 2012; Pinillos and Simpson, 2014; Sripada and Stanley, 2012). If stakes really do not play any role in ordinary knowledge ascriptions, one of the main motivations for epistemic contextualism and interest‐relative invariantism would be undermined. Perhaps these different explanations of the role of stakes in ordinary knowledge ascription are born out of nothing more than a myth (Schaffer and Knobe, 2012). If so, classical invariantism about knowledge might be best supported—not because it provides the best explanation of the role of stakes in ordinary knowledge ascriptions, but rather because the failure of stakes to play a role in ordinary knowledge ascription would undercut an important motivation for its two competitors, epistemic contextualism and interest‐relative invariantism. These radical alternatives to classical invariantism, lacking evidence in support of one of their important motivations, should perhaps then fall. Classical invariantism would stand.
In the remainder of this article, we will disarm an important motivation for epistemic contextualism and interest‐relative invariantism. We will accomplish this by presenting a stringent test of whether there is a stakes effect on ordinary knowledge ascription. Having shown that, even on a stringent way of testing, stakes fail to impact ordinary knowledge ascription, we will conclude that we should take another look at classical invariantism.
Here is how we will proceed. Section 1. lays out some limitations of previous research on stakes. Section 2. presents our study and concludes that there is little evidence for a substantial stakes effect. Section 3. responds to objections. The conclusion clears the way for classical invariantism.
1. Strengthening the Case for Stakes
The role of stakes in ordinary knowledge ascriptions is taken to be illustrated by patterns of judgments allegedly made by competent speakers. While a number of philosophers have taken for granted the sensitivity to stakes of knowledge ascription among competent speakers, empirical evidence has suggested otherwise. A wide range of empirical research has failed to uncover evidence that stakes play a role in ordinary knowledge ascriptions (see below for discussion of evidence seemingly supporting such a role). In light of this evidence, it would be tempting to conclude that perhaps stakes do not, after all, play a role in ordinary knowledge ascription. As tempting as this may be, however, it seems to us that there are a number of issues that have yet to be addressed.
First, virtually every study exploring the role of stakes in knowledge ascription has failed to ask participants whether they view the situation as a high or low stakes situation. This is especially surprising given that the main finding is a null result i.e., that knowledge ascriptions do not differ between high and low stakes cases. It may well be that the manipulation—i.e., high vs. low stakes—was ineffective, perhaps because participants failed to pay sufficient attention to key details varying between the cases. For instance, in the Bank cases, it may be that participants fail to appreciate that in one case it is “very important” that a check be deposited, while in the other it is “not very important.” If so, then the fact that no stakes effect was found wouldn't show that competent speakers fail to display a sensitivity to stakes. So, in the study reported below, the first question examined whether participants have understood what was at stake.
Another, perhaps more serious issue is that some of the results suggesting that stakes fail to play a role in knowledge ascriptions might be due to protagonist projection. Protagonist projection occurs when a subject takes up a protagonist's perspective and imagines what seems true from the protagonist's point of view (Holton, 1997). Importantly, protagonist projection looks to be at least partly responsible in producing otherwise surprising findings. For instance, Weinberg, Nichols, and Stich (2001) present evidence that Westerners and East Asians view Gettier cases differently: Westerners judge that protagonists in Gettier cases do not know the relevant proposition while East Asians judge that protagonists in Gettier cases do know the relevant proposition. But recent research suggests that the differences uncovered by Weinberg, Nichols, and Stich disappear when taking into account participants’ tendency to engage in protagonist projection. Indeed, recent work by Machery et al. (2015) found that, across four cultures—USA, India, Japan, and Brazil—rates of knowledge denial when confronted with Gettier cases were both high and similar across each of the four cultures sampled when a question targeting protagonist projection was introduced. While some participants attributed knowledge to a Gettierized protagonist when asked whether that protagonist “knows” or “does not know,” when given a question aimed at probing for protagonist projection—i.e., being asked whether the Gettierized protagonist “really knew” or “didn't really know but only thought she knew”—rates of knowledge denial increased with the vast majority of participants indicating that the Gettierized subject “didn't really know” but “only thought that she knew” (see also Machery et al. forthcoming).
Similarly, protagonist projection also looks to be behind apparently non‐factive knowledge ascriptions such as “Everyone knew that stress caused ulcers, before two Australian doctors in the early 1980s proved that ulcers are actually caused by bacterial infection” (Buckwalter, 2014, p. 396). It is not the case that people accept statements like this because they think that knowledge is non‐factive. Instead, they accept statements like this because they are engaging in protagonist projection by imagining what seems true from the perspective of individuals prior to the discovery that ulcers are caused by bacterial infection.
Recent work by Turri (2017) suggests that something like protagonist projection—what he calls “deferral”—may play a crucial role in some of the cases aimed at showing that stakes play a role in knowledge ascriptions. Instead of people's knowledge ascriptions shifting along with variations in stakes, people may simply defer to others’ mental‐state reports. For instance, in one version of the Bank cases (but not in the version we used), the protagonist in the high stakes case says, “I don't know it will be open tomorrow,” while the protagonist in the low stakes case says, “I know the bank will be open tomorrow” (see, e.g., DeRose, 2009, 2011). In this version of the Bank cases, agreement with both statements might have nothing to do with a shift in stakes. Instead, people may naturally defer to the mental state reports of others.
In addition to producing the misleading appearance of a stakes effect, protagonist projection can also mask a genuine stakes effect. This can happen when the protagonist says in both conditions, “I know the bank will be open tomorrow,” as is the case in the vignettes we used. To ensure that protagonist projection does not mask any genuine effect of stakes, we introduced a probe aimed at capturing whether participants are making genuine knowledge ascriptions or merely projecting: It contrasts “knows” and “thinks he knows, but doesn't actually know.” By comparing participants’ answer to this probe to their answers to a “know/does not know” probe, we will also be able to examine the role of protagonist projection in knowledge ascription in a context where stakes are manipulated.
Finally, the failure of stakes to play a role in knowledge ascription may only reflect something peculiar about the practice of knowledge ascription within a narrow linguistic community. Indeed, all of the empirical work done thus far has been conducted with participants drawn from the USA. Perhaps a wide range of other linguistic communities display a sensitivity to stakes. So we would like to know whether the effect of stakes (or lack thereof) in knowledge ascription is cross‐culturally robust. Putting all of this together, our questions are:
- Is the difference in stakes appreciated by participants?
- Is the lack of a stakes effect due to protagonist projection?
- Is the effect of stakes (or lack thereof) in knowledge ascriptions cross‐culturally robust?
In taking these up, our strategy was to undertake a cross‐cultural study, introducing a number of measures aimed at addressing the questions under consideration in order to determine whether stakes sensitivity (or the lack thereof) reflects a core aspect of folk epistemology.
2. A Study in Folk Epistemology
2.1. Method
We collected data from 4504 people across nineteen sites, spanning sixteen countries. Each participant was randomly assigned to one of two conditions, a low or high stakes version of a Bank case. Here is the low stakes version:
Bob and his wife are driving home on a Friday afternoon. They both received some money earlier in the day, and so they plan to stop at the bank on the way home to deposit it. But as they drive past the bank, they notice that the lines inside are very long, as they often are on Friday afternoons. Although they generally like to deposit any money they receive at the bank as soon as possible, it is not especially important in this case that it be deposited right away, and so Bob suggests that they drive straight home and deposit their money on Saturday morning. His wife says, “Maybe the bank won't be open tomorrow. Lots of banks are closed on Saturdays.” Bob replies, “No, I know the bank will be open. I was just there two weeks ago on Saturday. It was open until noon.” As a matter of fact, the bank will be open on Saturday morning.
And here is the high stakes version:
Bob and his wife are driving home on a Friday afternoon. They both received some money earlier in the day and so they plan to stop at the bank on the way home to deposit it. But as they drive past the bank, they notice that the lines inside are very long, as they often are on Friday afternoons. They have recently written a very large and very important check. If the money is not deposited into their bank account before Monday morning, the important check they wrote will not be accepted by the bank, leaving them in a very bad situation. Bob suggests that they drive straight home and deposit their money on Saturday morning. His wife says, “Maybe the bank won't be open tomorrow. Lots of banks are closed on Saturdays.” Bob replies, “No, I know it'll be open. I was just there two weeks ago on Saturday. It was open until noon.” As a matter of fact, the bank will be open on Saturday morning.
To see whether stakes are appreciated, participants were first asked:
Comprehension: According to the story, which of the following statements is correct? [It is not very important that Bob and his wife deposit their money/It is very important that Bob and his wife deposit their money.]
They were then asked:
Knowledge Attribution: In your personal opinion, when Bob says “I know the bank will be open” is his statement true? [Yes, Bob's statement is true./No, Bob's statement is not true.]
And finally to see if responses to Knowledge Attribution are due to protagonist projection, we asked:
Strict Knowledge Attribution: In your personal opinion, which of the following sentences better describes Bob's situation? [Bob knows the bank will be open on Saturday./Bob thinks he knows the bank will be open on Saturday, but he doesn't actually know it will be open.]1
1 There was a slight difference in the translation of this question in the low and high stakes cases for the Japanese version. The low stakes version read as reported above but the high stakes version used, “Of the following two sentences, which do you think better describes Takeshi's situation?”.
The cases were translated into fourteen languages by competent native speakers and presented in the respective native language for each group.
2.2. Results
First, do participants appreciate the difference in stakes? It turns out that they do. Overall, 78% of participants passed Comprehension (see Table 1 for Demographics). So we take our first concern—that participants may not be appreciating a difference in stakes—to be resolved.
| Sample | Students | Method | Payment | Language | N |
|---|---|---|---|---|---|
| Europe | |||||
| Bulgaria | N | Web‐based | Volunteers | Bulgarian | 327 |
| France | N | Web‐based | Compensation & volunteers | French | 367 |
| Germany | N | Web‐based | Compensation | German | 153 |
| Italy | Y | Paper‐pencil | Volunteers | Italian | 139 |
| Portugal | Y | Paper‐pencil | Volunteers | Portuguese | 139 |
| Spain | N | Web‐based | Compensation | Spanish | 239 |
| Switzerland | N | Paper‐pencil & web‐based | Volunteers | French | 54 |
| Switzerland | Y | Paper‐pencil & web‐based | Compensation & volunteers | French | 30 |
| UK | N | Web‐based | Compensation | English | 255 |
| Middle East | |||||
| Iran | N | Paper‐pencil | Volunteers | Persian | 164 |
| Central & North America | |||||
| Mexico | N | Paper‐pencil | Volunteers | Spanish | 133 |
| USA | N | Web‐based | Compensation | English | 225 |
| South America | |||||
| Brazil | Y | Paper‐pencil | Volunteers | Portuguese | 135 |
| East Asia | |||||
| China | Y | Paper‐pencil | Volunteers | Chinese, Traditional | 128 |
| Guangzhou China | Y | Paper‐pencil | Volunteers | Chinese, Simplified | 109 |
| Mainland China | N | Web‐based | Compensation | Chinese, Simplified | 180 |
| Hong Kong | Y | Web‐based | Compensation | Chinese, Traditional | 146 |
| Japan | N | Web‐based | Compensation | Japanese | 151 |
| Japan | Y | Paper‐pencil | Volunteers | Japanese | 165 |
| Mongolia | N | Paper‐pencil | Volunteers | Mongolian | 115 |
| South Asia | |||||
| India | Y | Paper‐pencil | Volunteers | Bengali | 162 |
Next, do stakes affect Knowledge Attribution? Removing participants who failed Comprehension, we analyzed responses from the remaining 3530 participants. Overall, we found a significant, but negligible effect of stakes on Knowledge Attribution (χ2(1, 3530) = 9.040, p < .01). Ascription of knowledge in the low stakes condition (85%) is only 3% larger than in the high stakes condition (82%), and its conventional effect size is very small (Cramer's V = .051): On a standard interpretation of conventional effect sizes (see Ellis, 2010) it does not even count as small.2
More importantly, across sites we find virtually no evidence that stakes affect knowledge attribution (see Table 2 and Figure 1).
| Sample | N | χ2 | p‐value | Cramer's V |
|---|---|---|---|---|
| Central and North America | ||||
| Mexico | 133 | .000 | .990 | .001 |
| USA | 225 | 3.661 | .056 | .128 |
| South America | ||||
| Brazil | 135 | .318 | .573 | .049 |
| Europe | ||||
| Bulgaria | 327 | .084 | .773 | .016 |
| France | 367 | 2.990 | .084 | .090 |
| Germany | 153 | .555 | .456 | .060 |
| Italy | 139 | .117 | .732 | .029 |
| Portugal | 139 | 1.229 | .268 | .094 |
| Spain | 239 | 6.219 | * | .161 |
| Switzerland | 84 | .841 | .359 | .100 |
| UK | 255 | 4.470 | * | .132 |
| Middle East | ||||
| Iran | 164 | .020 | .889 | .011 |
| East Asia | ||||
| China | 128 | .522 | .470 | .064 |
| Hong Kong | 146 | .272 | .602 | .043 |
| Guangzhou China | 109 | .690 | .406 | .080 |
| Mainland China | 180 | 1.345 | .246 | .086 |
| Mongolia | 115 | .003 | .959 | .005 |
| Japan | 316 | 5.728 | * | .135 |
| South Asia | ||||
| India | 162 | 1.747 | .186 | .104 |
Out of the nineteen sites sampled, only three (16%)—Spain, UK, and Japan—displayed a significant, small‐sized effect of stakes on knowledge ascriptions (the data from the USA is also near significant; see Figures 2 and 3). Despite this, a logistic regression model revealed that there was no interaction between stakes and site on Knowledge Attribution, which suggests that there is surprising stability in the lack of a stakes effect across sites (see Table 3).3
| Sample | B(SE) | Wald χ2 | p‐value | Exp(B) |
|---|---|---|---|---|
| Central and North America | ||||
| USA | .231(.641) | .130 | .718 | 1.260 |
| Mexico | −.686(.579) | 1.405 | .236 | .503 |
| South America | ||||
| Brazil | −.942(.602) | 2.445 | .118 | .390 |
| Europe | ||||
| Bulgaria | −.614(.486) | 1,601 | .206 | .541 |
| France (Contrast) | — | — | — | — |
| Germany | −4.12(.553) | .556 | .456 | .662 |
| Italy | −.872(.666) | 1.716 | .190 | .418 |
| Portugal | −1.237(.641) | 3.723 | .054 | .290 |
| Spain | .162(.535) | .091 | .762 | 1.176 |
| Switzerland | −1.314(.798) | 2.713 | .100 | .269 |
| UK | .142(.571) | .061 | .804 | 1.152 |
| Middle East | ||||
| Iran | −.740(.535) | 1.914 | .166 | .477 |
| East Asia | ||||
| China | −1.041(.633) | 2.706 | .100 | .353 |
| Hong Kong | −.360(.756) | .227 | .633 | .697 |
| Guangzhou China | −1.110(.649) | 2.924 | .087 | .329 |
| Mainland China | .564(.331) | .000 | .996 | .000 |
| Mongolia | −.669(.589) | 1.289 | .256 | .512 |
| Japan | .027(.507) | .003 | .957 | 1.027 |
| South Asia | ||||
| India | −.077(.621) | .015 | .901 | .926 |
Moreover, these three linguistic communities, along with every other linguistic community sampled, displayed overall high rates of knowledge attribution regardless of whether the case was low or high stakes (Figure 1). These results fit with a range of similar findings4
Given that we find virtually no evidence of a stakes effect on Knowledge Attribution, we want to know whether this may be due in part to protagonist projection. We look at this in two ways, first, using our second measure (“Strict Knowledge Attribution”). Overall, we again find a significant, but negligible effect of stakes on Strict Knowledge Attribution (χ2(1, 3522) = 10.451, p < .01). Ascription of knowledge in the low stakes condition (63%) is only 5% larger than in the high stakes condition (58%), and its conventional effect size is tiny (Cramer's V = .054): Again, on a standard interpretation of conventional effect sizes (see Ellis, 2010) it does not even count as small.
Moreover, across sites, we continue to fail to find evidence of a stakes effect on Strict Knowledge Attribution (see Table 4 and Figure 4).
| Sample | N | χ2 | p‐value | Cramer's V |
|---|---|---|---|---|
| Central and North America | ||||
| Mexico | 131 | 1.740 | .187 | .115 |
| USA | 225 | 5.453 | * | .156 |
| South America | ||||
| Brazil | 135 | 2.603 | .107 | .139 |
| Europe | ||||
| Bulgaria | 327 | .248 | .618 | .028 |
| France | 365 | .505 | .477 | .037 |
| Germany | 151 | 3.860 | * | .160 |
| Italy | 139 | .024 | .878 | .013 |
| Portugal | 139 | .033 | .856 | .015 |
| Spain | 239 | 1.930 | .165 | .090 |
| Switzerland | 84 | .350 | .554 | .065 |
| UK | 252 | 2.640 | .104 | .102 |
| Middle East | ||||
| Iran | 164 | 2.159 | .142 | .115 |
| East Asia | ||||
| China | 128 | 2.092 | .148 | .128 |
| Hong Kong | 146 | 2.776 | .096 | .138 |
| Guangzhou China | 109 | .002 | .969 | .004 |
| Mainland China | 180 | .479 | .489 | .052 |
| Mongolia | 116 | .556 | .456 | .069 |
| Japan | 316 | 2.988 | .084 | .097 |
| South Asia | ||||
| India | 162 | .144 | .704 | .030 |
Indeed, only two linguistic communities (11%)—Germany and USA—out of the nineteen sampled displayed evidence of a significant, small‐sized effect of stakes on Strict Knowledge Attribution (Figures 5 and 6). But despite this, there was no interaction between stakes and site on Strict Knowledge Attribution (see Table 5).5
| Sample | B(SE) | Wald χ2 | p‐value | Exp(B) |
|---|---|---|---|---|
| Central and North America | ||||
| USA | .477(.375) | 1.621 | .203 | 1.611 |
| Mexico | .330(.458) | .519 | .471 | 1.390 |
| South America | ||||
| Brazil | −.769(.445) | 2.990 | .084 | .464 |
| Europe | ||||
| Bulgaria | −.061(.339) | .032 | .857 | .941 |
| France (Contrast) | — | — | — | — |
| Germany | .539(.443) | 1.483 | .223 | 1.715 |
| Italy | −.122(.429) | .081 | .776 | .885 |
| Portugal | −.111(.437) | .064 | .800 | .895 |
| Spain | .187(.360) | .269 | .604 | 1.206 |
| Switzerland | .090(.514) | .031 | .861 | 1.094 |
| UK | .248(.360) | .473 | .492 | 1.281 |
| Middle East | ||||
| Iran | −.640(.402) | 2.535 | .111 | .528 |
| East Asia | ||||
| China | .364(.449) | .656 | .418 | 1.438 |
| Hong Kong | .450(.452) | .991 | .320 | 1.568 |
| Guangzhou China | −.191(.459) | .173 | .677 | .826 |
| Mainland China | −.504(.536) | .884 | .347 | .604 |
| Mongolia | −.481(.480) | 1.009 | .315 | .618 |
| Japan | .334(.387) | .747 | .387 | 1.397 |
| South Asia | ||||
| India | −.050(.415) | .014 | .905 | .952 |
On the second way of looking at whether the lack of a stakes effect might be due to protagonist projection, we looked at whether there was an effect of stakes on Strict Knowledge Attribution among those who attributed knowledge on Knowledge Attribution. Here we find a marginally significant effect (χ2(1, 2923) = 3.567, p = .059, Cramer's V = .035): Of those who attributed knowledge on Knowledge Attribution in the high‐stakes case, 68% selected “really knows” on Strict Knowledge Attribution; of those who attributed knowledge on Knowledge Attribution in the low‐stakes case, 71% selected “really knows” on Strict Knowledge Attribution. This amounts to a mere 3% difference.
Yet again, across sites, we continue to find virtually no evidence of a stakes effect on Strict Knowledge Attribution among those who attributed knowledge on Knowledge Attribution (Table 6 and Figure 7).
| Sample | N | χ2 | p‐value | Cramer's V |
|---|---|---|---|---|
| Central and North America | ||||
| Mexico | 92 | 1.448 | .229 | .125 |
| USA | 202 | 3.182 | .074 | .126 |
| South America | ||||
| Brazil | 110 | 2.303 | .129 | .145 |
| Europe | ||||
| Bulgaria | 245 | .056 | .812 | .015 |
| France | 333 | .003 | .995 | .003 |
| Germany | 109 | 1.351 | .245 | .111 |
| Italy | 122 | .095 | .758 | .028 |
| Portugal | 120 | .650 | .420 | .074 |
| Spain | 190 | .297 | .586 | .040 |
| Switzerland | 74 | .724 | .395 | .099 |
| UK | 218 | 1.327 | .249 | .078 |
| Middle East | ||||
| Iran | 118 | 2.013 | .156 | .131 |
| East Asia | ||||
| China | 107 | .613 | .434 | .076 |
| Hong Kong | 134 | 3.581 | .058 | .163 |
| Guangzhou China | 90 | .356 | .551 | .063 |
| Mainland China | 179 | .210 | .617 | .034 |
| Mongolia | 80 | 1.074 | .300 | .116 |
| Japan | 249 | .034 | .854 | .012 |
| South Asia | ||||
| India | 139 | .416 | .519 | .055 |
On this way of looking at whether a stakes effect might be masked by projection, we find that stakes had no significant effect in any site, and approached significance in only two sites: Hong Kong and USA (Figures 8 and 9). Yet again, a logistic regression model revealed that there was no interaction between stakes and site (see Table 7).6
| Sample | B(SE) | Wald χ2 | p‐value | Exp(B) |
|---|---|---|---|---|
| Central and North America | ||||
| USA | .523(.417) | 1.572 | .210 | 1.688 |
| Mexico | .570(.568) | 1.009 | .315 | 1.769 |
| South America | ||||
| Brazil | −.613(.488) | 1.577 | .209 | .542 |
| Europe | ||||
| Bulgaria | .048(.393) | .015 | .903 | 1.049 |
| France (Contrast) | — | — | — | — |
| Germany | .641(.640) | 1.004 | .316 | 1.899 |
| Italy | .102(.477) | .046 | .831 | 1.107 |
| Portugal | .329(.516) | .407 | .524 | 1.389 |
| Spain | .147(.413) | .126 | .723 | 1.158 |
| Switzerland | .415(.583) | .506 | .477 | 1.514 |
| UK | .317(.407) | .606 | .436 | 1.373 |
| Middle East | ||||
| Iran | −.544(.470) | 1.337 | .248 | .580 |
| East Asia | ||||
| China | .309(.504) | .374 | .541 | 1.361 |
| Hong Kong | .776(.551) | 2.301 | .129 | 2.173 |
| Guangzhou China | .248(.528) | .221 | .638 | 1.282 |
| Mainland China | −.239(.565) | .178 | .673 | .788 |
| Mongolia | −.500(.549) | .829 | .363 | .607 |
| Japan | .058(.493) | .014 | .906 | 1.060 |
| South Asia | ||||
| India | −.274(.492) | .310 | .578 | .760 |
Even on our second way of looking at whether a stakes effect might be masked by projection, we continue to find virtually no evidence that stakes affect knowledge attribution. We conclude that the lack of a stakes effect on knowledge ascriptions cannot be dismissed by appealing to protagonist projection.
2.3. Discussion
Given these results and the wide swath of research indicating that stakes do not play a role in ordinary knowledge ascription, the scales tilt against epistemic contextualism and interest‐relative invariantism, at least to the extent that they attempt to account for everyday knowledge ascription. One of the important motivations for these views—that stakes play a role in ordinary knowledge attributions—is undermined. These views are “idle hypotheses” (Turri, 2017).
Interestingly, stakes had a marginally significant effect on both Knowledge Attribution and Strict Knowledge Attribution among those who attributed knowledge on Knowledge Attribution for the USA sample. Stakes also had a significant, but small effect for the Strict Knowledge Attribution question. Perhaps this explains why some epistemologists who have alleged there is a stakes effect, the most influential of whom are Americans, thought there was a stakes effect. In any case, the small effect of stakes is a far cry from what we would expect if stakes played an important role in knowledge ascription. On the whole, we find that there is surprising stability in the lack of a stakes effect on knowledge ascriptions across cultures.
3. Objections and Responses
3.1. Evidence for a Stakes Effect
Some experimental studies claim to have found some evidence for a stakes effect (Pinillos, 2012; Sripada & Stanley, 2012). These results clearly conflict with our findings as well as a range of other research. So perhaps there really is a genuine stakes effect on ordinary knowledge ascription.
We are skeptical that these studies provide genuine evidence that there is a stakes effect on ordinary knowledge ascription. Pinillos (2012) gave people a case about an individual, Peter, writing a paper for an English class. In the low stakes version, it is not very important whether the paper has typos; in the high stakes version, it is very important that the paper not have typos. Pinillos found that when participants were asked, “How many times do you think Peter has to proofread his paper before he knows that there are no typos?”, the median response was 2 in the low stakes version while the median response was 5 in the high stakes version. Pinillos interprets this as evidence that stakes do indeed affect ordinary knowledge ascriptions.
But Buckwalter (2014) and Buckwalter and Schaffer (2015) present compelling evidence that the effect Pinillos found has nothing to do with knowledge. The effect persists when “knows” in the probe Pinillos used is replaced with “believes,” “guesses,” and “hopes.” Rather than being an effect on knowledge ascription, the stakes effect Pinillos has uncovered is instead an effect on the modal expression “has to” (Buckwalter and Schaffer, 2015).
In Sripada and Stanley (2012), participants read about an individual, Hannah, who is allergic to Mongolian pine nuts. She is having dinner at a Mongolian restaurant, and the possibility is raised that there may be Mongolian pine nuts in her dish. They found an effect of stakes on knowledge ascription in two separate pairs of cases. However, the effect looks to be quite small in both cases (less than a 1 point scale difference on a 7 point scale).7
3.2. The Role of Linguistic Evidence
Proponents of contextualism or interest‐relative invariantism may object that their theories are not meant to account for everyday knowledge ascriptions. They are, after all, theories of knowledge, rather than linguistic theories about “to know.”
We have two brief responses to this concern. To the extent that contextualism and interest‐relative invariantism are not meant to account for knowledge ascriptions—perhaps they are theories of knowledge to be developed largely independently of how people talk about knowledge and what they think about it—then, we acknowledge, our findings have little to say about contextualism and interest‐relative invariantism. They are only relevant for those philosophical views that aim at accounting for everyday knowledge ascription.
We add that as a matter of fact, many contextualists and interest‐relative invariantists propose to account for everyday knowledge ascription. DeRose (1992) formulates contextualism as a semantic theory and he engages with the early experimental philosophy literature on stakes effects in bank cases (2011). Moreover DeRose (2009) is explicit that “[t]he best grounds for accepting contextualism comes from how knowledge‐attributing (and knowledge‐denying) sentences are used in ordinary, non‐philosophical talk: What ordinary speakers will count as ‘knowledge’ in some non‐philosophical contexts they will deny is such in others” (p. 47). Ludlow (2005, 11) too explicitly formulates contextualism as a semantic thesis: “According to the thesis of contextualism in epistemology, many of our knowledge attributions (including self‐attributions) are context sensitive.”
3.3. A Negative Result Limited to the Bank Cases
One might argue that our results merely show that the bank cases are inappropriate to elicit a stakes effect, not that there is no stakes effect. Other cases would elicit a stakes effect. However, the bank cases were put forward by some advocates of the view that stakes affect ordinary knowledge ascriptions as being the best cases for eliciting a stakes effect (Buckwalter and Schaffer, 2015, 222). Furthermore, even Sripada and Stanley, who think that their pine nut cases are the best cases for eliciting a stakes effect on ordinary knowledge ascription, find at best a very small effect of stakes, which shifts people from being slightly more to slightly less neutral about whether to ascribe knowledge. Finally, other studies have used a range of cases with each failing to uncover an effect of stakes on ordinary knowledge ascription (see e.g., Feltz and Zarpentine, 2010; Turri, 2017; Turri and Buckwalter, 2017). Indeed, in light of our findings and a range of research failing to uncover a stakes effect in ordinary knowledge ascriptions, we think that stakes fail to reflect a core aspect of folk epistemology.
4. Conclusion
Epistemic contextualists hold that knowledge ascription shifts across conversational contexts; interest‐relative invariantists recognize a distinctive epistemic role for practical factors such as stakes in knowledge ascription. If either of these is correct, serious pressure is put on classical invariantism since the classical invariantist rejects both claims. However, a wide range of empirical evidence now suggests that stakes do not play any role in ordinary knowledge ascription.
We set out to provide what we take to be a stringent test of whether stakes play a role in ordinary knowledge ascriptions. In doing so we pursued three main questions:
- Is the difference in stakes appreciated by participants?
- Is the lack of a stakes effect due to protagonist projection?
- Is the effect of stakes (or lack thereof) in knowledge ascriptions cross culturally robust?
Our results indicate that subjects do indeed appreciate the difference in stakes, that a stakes effect is not being masked by protagonist projection, and that the lack of a stakes effect in ordinary knowledge ascription is cross‐culturally robust. In light of our evidence and a wide range of previous empirical findings on the role of stakes in knowledge ascription, one of the core motivations for epistemic contextualism and interest relative invariantism is undercut. Although we won't defend this claim in detail here, we conclude that classical invariantism should be taken seriously, now that its challengers have been undermined.8
References
Number of times cited according to CrossRef: 6
- Jumbly Grindrod, James Andow and Nat Hansen, Third‐person knowledge ascriptions: A crucial experiment for contextualism, Mind & Language, 10.1111/mila.12196, 34, 2, (158-182), (2018).
- Masashi Kasaki, Problems of Translation for Cross-Cultural Experimental Philosophy, Journal of Indian Council of Philosophical Research, 10.1007/s40961-017-0119-5, 34, 3, (481-500), (2017).
- Edouard Machery, Stephen Stich, David Rose, Mario Alai, Adriano Angelucci, Renatas Berniūnas, Emma E. Buchtel, Amita Chatterjee, Hyundeuk Cheon, In-Rae Cho, Daniel Cohnitz, Florian Cova, Vilius Dranseika, Ángeles Eraña Lagos, Laleh Ghadakpour, Maurice Grinberg, Ivar Hannikainen, Takaaki Hashimoto, Amir Horowitz, Evgeniya Hristova, Yasmina Jraissati, Veselina Kadreva, Kaori Karasawa, Hackjin Kim, Yeonjeong Kim, Minwoo Lee, Carlos Mauro, Masaharu Mizumoto, Sebastiano Moruzzi, Christopher Y. Olivola, Jorge Ornelas, Barbara Osimani, Carlos Romero, Alejandro Rosas Lopez, Massimo Sangoi, Andrea Sereni, Sarah Songhorian, Paulo Sousa, Noel Struchiner, Vera Tripodi, Naoki Usui, Alejandro Vázquez del Mercado, Giorgio Volpe, Hrag Abraham Vosgerichian, Xueyi Zhang and Jing Zhu, The Gettier Intuition from South America to Asia, Journal of Indian Council of Philosophical Research, 10.1007/s40961-017-0113-y, 34, 3, (517-541), (2017).
- Blake Roeber, How to argue for pragmatic encroachment, Synthese, 10.1007/s11229-018-1850-4, (2018).
- Wesley Buckwalter, Epistemic Injustice in Social Cognition, Australasian Journal of Philosophy, 10.1080/00048402.2018.1459754, (1-15), (2018).
- Simon Langford, Knowledge judgements and cognitive psychology, Synthese, 10.1007/s11229-018-1880-y, (2018).
