My colleague Maria Lasonen-Aarnio’s great paper Higher-Order Evidence and the Limits of Defeat is just out in PPR. I agree with her conclusions about higher-order evidence. Indeed, on a number of points I agree with her because I’ve been convinced by her arguments. But I did want to register one quibble, one that I don’t think undermines the position she ultimately adopts. In fact, it offers a way to respond to an objection.
Lasonen-Aarnio is interested in cases with the following structure:
1. S has evidence E, and let’s assume S knows she has E, and E is in fact excellent evidence for p.
2. S has strong but misleading evidence, call it H, that E is not any kind of evidence for p.
3. S has no other evidence that tells in favour of p.
There is a widespread intuition in these cases that S should not believe p, because H undermines the support that E provides for p. Lasonen-Aarnio wants to argue against this intuition. Or, at least, she wants to argue against this intuition given a ‘rule-based’ conception of epistemic rationality. The difference between the two possible conclusions will turn out to matter for what I say.
By a ‘rule-based’ conception of epistemic rationality, I mean (and I think Lasonen-Aarnio means too), a theory with the following two principles:
1. All epistemic norms are to be explained by the existence of epistemic rules.
2. For any one of these rules that explain epistemic norms, there is a distinction between following the rule, and merely complying with it, and full rationality requires following not merely complying.
Lasonen-Aarnio’s first step is to argue that even if you accept the intuition here, you should still think that rules like “Believe what the evidence supports” are good rules. It’s true that the intuition in question implies that the rule is somehow trumped. But that doesn’t mean we should have a more restrictive rule saying “Believe what the evidence supports, unless one of these conditions obtain”, where the conditions are the conditions where the rule is trumped. The reason for this is that the intuition that started us down this track is completely general. Any time an agent gets evidence that a rule is unreliable or untrustworthy, the intuition says that rule is trumped. So no finite rule can accommodate all the possible trumping conditions.
One possibility would be to have an infinite rule. It isn’t hard to describe what such a rule would look like. Consider the function from possible situations to belief states that are permissible in that situation. (Ignore the possibly serious problem that such there are too many situations for this to even be a function.) Call this an über-rule of rationality. Such a rule could cover all cases, including the one S faces in our example at the start.
But there is a pressing problem that Lasonen-Aarnio raises here. The reasons for thinking that no finite rule will capture the intuition are reasons for thinking that the über-rule will be infinitely complex. And that in turn means it will be hard for agents to genuinely follow the rule. Put another way, once we start considering über-rules, the distinction between following a rule and merely complying with it will be obliterated, since the most a non-Godlike agent could do is merely comply with it. But once this distinction is obliterated, we have abandoned the rule-based conception of epistemic rationality. After all, any theory whatsoever can be rephrased as a rule-based theory, provided we let rules be nothing more than functions from situations to evaluations of actions in that situation, and the most we expect of agents is complaince with that ‘rule’. The rule-based conception wasn’t meant to be this trivial!
One natural move here, one taken by David Christensen, is to say that agents like S face a dilemma. S should believe p, in virtue of 1, and should not believe p, in virtue of 2. But Christensen, like others moved by the intuition we started with, doesn’t think this is a ‘pure’ dilemma. Rather, he thinks that although S will do something irrational whatever she does, it is worse for her to hold on to her belief in p than to scrap it. Lasonen-Aarnio argues against this possibility, and it is here I want to quibble with her.
As Lasonen-Aarnio sets things up, S faces two rules.
- Rule 1 – With evidence E, believe p!
- Rule 2 – With evidence that E is not good evidence for p, don’t believe p!
But these rules don’t imply that it is better to comply with Rule 2 than with Rule 1. That needs further explanation.
Lasonen-Aarnio assumes that the further explanation will consist of a third rule, one that instructs agents to comply with Rule 2 rather than Rule 1 when given a choice. And on the strong conception of a rule-based conception of epistemology that we started with, she has to be right. Moreover, I think the intuition she’s contesting, the one that says S should give up the belief in p, really doesn’t look that plausible without this very strong conception of epistemology as rule-based.
If we drop this way of thinking about the rule-based conception of epistemology, there is a way out. Let’s say there is an extra normative fact that it is better to comply with Rule 2 rather than Rule 1. (Or, for that matter, that it is better to comply with Rule 1 than Rule 2. I’m really most interested here in the general issue of whether there can be impure dilemmas, not which way this particular one should be resolved.) That normative fact is not explained by, or grounded in, the existence of an epistemic rule. It’s true that the normative fact can be converted into a rule, in much the way that we generated the über-rule above. But that conversion would be misleading, for the distinction between following and complying is not intuitively relevant here.
This is a very general point about impure dilemmas. It is very plausible that the moral worth of an action depends on the action being done for the right reasons. If we’re sympathetic to moral rules, we’ll say that this means that the action is done by a person following, and not merely complying with, the moral rule. Following Nomy Arpaly and Julia Markovits, I don’t think it is important that the agent recognise the rule as a moral rule, or perhaps even as any kind of rule. But that doesn’t undermine the importance of the following/complying distinction, since an agent can genuinely follow a rule without thinking of it as a moral rule. An agent might refuse to lie out of respect for her interlocutors, even if she mistakenly believes (perhaps on general consequentialist grounds) that there is nothing particularly wrong with lying. In that case I think she is genuinely following, not merely complying with, a rule against lying.
In cases of moral dilemmas, there will be some undefeated moral rule that the agent breaks no matter what she does. In some of these, there will be a less bad choice to make. An agent faced with a choice between lying and breaking a promise might do wrong either way, but it may be worse to break the promise. (This isn’t because promise-breaking is always worse than lying; I’m just making an assumption about the particular case.) Now we could say that in such a situation there is a rule saying it is worse to break the rule against promise-breaking than the rule against lying. But this would be a misleading way of speaking, since intuitively there isn’t a difference between following and merely complying with this meta-rule. Or, if there is a difference, it is that ‘mere’ compliance is better. Someone who keeps the promise out of respect for the promisee is merely complying with the meta-rule. But that’s better than either of the obvious ways of following the meta-rule. Someone who lies rather than breaks a promise because it maximises their moral value is excessively self-concerned. And someone who thinks that they should minimise how disrespectful they are to others is treating the person they are lying to too much as a means. It’s better to simply follow the rule against promise-breaking, and comply with the meta-rule. And that’s to say the meta-rule isn’t really a rule in the relevant sense; it’s simply a normative fact that in this circumstance, one action is worse than another.
The same story holds, without the attendant moralising, in the epistemic case. It could be perfectly rational to merely comply with a meta-rule, while following the underlying rules. That position allows for impure epistemic dilemmas, at the cost of giving up a fully rule-based conception of epistemology. I don’t think that’s in conflict with what Lasonen-Aarnio says. Indeed, it helps her overall project I think.
On first reading her paper, one might worry that Lasonen-Aarnio’s arguments overgeneralised; that if they worked they would show that there was no such thing as an impure dilemma. And that seems like an implausible result, at least to me. But in fact she doesn’t ‘prove’ any such thing. Rather, her conclusion is that the thoroughly rule-based conception of epistemology is incompatible with impure dilemmas. Since the best versions of normative internalism in epistemology seem to end up committed to a thoroughly rule-based conception, and to impure dilemmas, her argument is (as she says) a strong argument for normative externalism in epistemology.