• No results found

The Semantics Of 'Knows' Again

§4.6 Gerken on epistemic focal bias

V. The Semantics Of 'Knows' Again

As discussed in §2, Williamson, Hawthorne, Nagel, and Gerken, all take their proposals to lend some support for invariantism about the semantics of 'knows'. In each case, the argument takes the same form: the existence of the error-salience phenomenon appears to pose a challenge to invariantism, since if invariantism were correct, it would show that people commit systematic errors in their knowledge attribution. However, if it can be shown that we should expect there to be such errors even if 'knows' has an invariantist semantics, because they are the result of well known psychological heuristics or biases, then the case against invariantism is at least partially undermined. And of course this is what the invariantists try to show - Williamson and Hawthorne with the availability heuristic explanation, Nagel with the epistemic egocentricity explanation, and Gerken with the epistemic focal bias account.

One might also think that the simulation heuristic explanation similarly blocks the argument against invariantism from the error-salience phenomenon. The thought would be that people rely on mental simulation to form judgements about whether the subject's belief is sufficiently modally robust to count as an item of knowledge, but because mental simulation is influenced by a heuristic which leads people to mutate salient aspects of a scenario more than non-salient

aspects, people are systematically led astray in their knowledge attributions when presented with salient, but epistemically irrelevant, error possibilities.

I have reservations about appealing to the psychological processes producing the error- salience phenomenon to defend invariantism. Though I certainly don't have a knock-down argument against the prospects of doing so, I will in closing, briefly sketch a line of thought according to which we should expect the error-salience phenomenon producing psychological processes to be constitutive of the concept of knowledge. We begin with the observation that according to traditional Bayesian decision theory, rational decision makers maximize expected utility, where expected utility is a function of the value of various possible outcomes, and the probability of their occurring. However, as Ross and Schroeder (2014) point out, for almost every proposition one should, and typically will, have a non-zero credence that it is true, since there is nothing that we can be completely certain is false. If we were to factor all of these non-zero probabilities into our decision making, then coming to a decision about how to act would very often be an unmanageably complex task for creatures, like ourselves, who have limited cognitive capacities and often need to engage in decision making in time-pressured circumstances. For this reason it has long been recognised by decision theorists - Savage (1972) and Joyce (1999) - and more recently by epistemologists - Weatherson (2005) and Ross and Schroeder (2014) - that when we are engaged in practical reasoning we must take some uncertain propositions to be certain for the purposes of decision making in order for decision making tasks to be manageable. One recent proposal is that our system of practical reasoning is governed by a norm stating that one should treat p as certain for the purposes of practical reasoning only if one knows that p. This is, of course, controversial.45 Suppose, however, that it is correct. In that case, it seems that one central role

for the concept of knowledge is to simplify decision making tasks by allowing us to take as certain in our practical reasoning uncertain propositions that are known to be true.46 This is

where the concern arises for invariantism. Given that we face a huge number of decision making tasks every day, we need a way of making good decisions in a cognitively cheap manner. If the function of the concept of knowledge is to help us to do this, by enabling us to simplify decision making tasks, then we will need a concept of knowledge that allows us to make accurate knowledge attributions in a cognitively cheap manner, if we are to be rational

45 Brown (2008), Neta (2009), Reed (2010), Schiffer (2007), and Gerken (2011), amongst others, dispute the

claim. This issue is discussed in greater detail in the essays 'Excuses and Epistemic Norms', and 'From Moore's Paradox To The Knowledge Norm Of Belief And Beyond' in this dissertation.

decision makers. And there is some tension between this need, and the invariantist suggestion that reliance on heuristics leads to systematic error in our knowledge attribution. If heuristic psychological processes have a tendency to lead us astray in our knowledge attributions in the way that the discussed proposals suggest when coupled with invariantism, then accurately judging whether a subject - either oneself of another - knows or not, will be a fairly hard cognitive task, requiring correction from system 2 psychological processes, rather than sole reliance on system 1 processes. Thus, making accurate knowledge attributions will be a relatively laborious and cognitively expensive task if invariantism is correct as compared to if contextualism is correct, for on contextualism, relying on such heuristics will tend not to lead one astray in one's knowledge attributions. For this reason I think it is unclear that an invariantist concept of knowledge would be fit to play one of the central normative roles that it is designed to play. The tension is between the fact that we employ the concept of knowledge to ease the cognitive burden of decision making and the fact that on invariantism accurate knowledge attribution will be a cognitively burdensome task. The tension appears to be lessened if our concept of knowledge is contextualist, and this, one might think, is a prima facie reason to prefer contextualism if, as the proposals discussed here suggest, knowledge attribution is influenced by heuristic processes.

As I said, this is hardly a conclusive argument. It is possible that even if invariantism is correct, sufficiently accurate knowledge attribution is cognitively cheap enough for the concept of knowledge to play the required role. Nevertheless, there does appear to be some cause for concern. This issue should be explored more, but this task is beyond the scope of this essay.

VI. Conclusion

Questions about what psychological processes produce the error-salience phenomenon are complex and relatively under-explored. Much more work will need to be done before we can

confidently draw any conclusions. Nevertheless, there is, I have argued, some reason to think that the phenomenon might be at least in part explained as the result of the influence of the simulation heuristic on knowledge attribution. The goal of this essay has been to sketch how such an explanation would go. This lays a foundation for further investigation.

§I. Norms, Justifications, and Excuses