Over at “Brian Leiter’s blog”:http://leiterreports.typepad.com/blog/2008/07/a-proposal-abou.html there was a long thread recently about journal refereeing and reviewing practices. I thought I’d make a few points here that are getting lost in the crush.

1) In my experience, most absolute disasters with delays about refereeing concern (a) potential referees who simply don’t answer requests to referee, and (b) cases where the editors run out of people they know/trust on the relevant topic. If everyone who received a request to referee a paper could answer it, even in the negative, that day, and if answering negatively suggest 1-3 names of people with some expertise in the field, that would make things flow much more smoothly.

2) Relatedly, I think a lot of people, when refereeing, don’t take into account how time sensitive it is. Imagine you’ve got a paper that you’ve promised to referee within the month. And you’ve got a project of your own that is due at the end of that month. And you’ve got enough time in the month to do both. What should you do? I think the answer is that you should referee the paper straight away. Usually getting your paper done earlier won’t make a difference to anyone. Getting the report done earlier will make a difference. I think the system would work a lot more smoothly if every referee, upon getting a paper, seriously considered the question “Can I do this today?” Obviously if you have to present a lecture that day, or the next day, and it isn’t done, then the answer is no. But often times the answer is yes. It’s not like you’ll often spend more than a few hours on the paper, or that doing the paper that day will take more time, but it will make a difference to editors and writers.

3) If we want to keep the model of some journals being run through departments, rather than through publishers, then some amount of delay is going to be inevitable. If nothing else, most journals run by departments have a support staff of 1. If that one person is sick, or on annual leave for a time, the whole system basically creaks to a halt. If that person is spending literally all their time for a two or three week period getting an issue readyopt print, nothing happens with submissions. I’ve never had to deal with this, but I imagine if you don’t have good staff (or, more likely, don’t have good staff management) things are worse.

Probably the single biggest thing that could be done to improve journal response times would be to find a way to keep the system running when less than fully staffed. But it’s hard to do that in a small operation, when you can’t simply move staff from elsewhere onto the project.

4) The journal management software systems that are currently being rolled out make a huge difference. There’s nothing as good as keeping a paper from dropping off the face of the earth as reminders every few days that your report on it is overdue. (Since I sign off on every paper on Compass, I get a lot of these, but I’m not that late on too many.) Potentially these systems can, by automating processes now done by staff, help a lot with point (1). And that’s important, because otherwise point (1) seems to me to be intractable short of handing over all the journals to commercial presses.

Having said that, everyone hates the software when it is being rolled out. But it really makes all the difference in the world.

5) There’s been some discussion of cutting back on referee reports. I think this is basically a good idea. It’s true that referees need to say something to editors about what’s good or bad about a paper. But from experience I’ve learned that it’s *much* easier to find something informative to say about a paper to an editor than it is to say something informative and polite to an author. And anything that speeds up the process is probably good.

6) But I really don’t think the comments thread at Leiter is taking seriously how much of the problem is caused by there being too many papers being submitted. If every paper being submitted was a real philosophical advance, that wouldn’t be a problem – it would be paradise. But I don’t really think this is so.

Lots of papers I see to referee are basically glorified blog points that don’t attempt to make more than a very small point. Some of them would be quite good blog posts. But most journals aim a little higher than that. (Note this is different to the length point. Lots of good papers, even papers in top anthologies, are short. But they are all ambitious.)

Disturbingly, many papers seem to be largely unaware of the relevant literature, especially with the most recent developments. I see too many papers that simply don’t pay attention to relevant work from the last 10 years.

Now I don’t want to pretend that I’ve never written (or published) papers that fall in one or other of these categories. But I do think that many papers get sent out when the author could profitably have either rolled the paper into a larger paper, or spent time talking to colleagues/friends/blog readers about relevant literature that should be consulted.

I used to think this was a tragedy of the commons problem. (Mark van Roojen makes this suggestion in the Leiter thread.) The pressures to publish meant not quite cooked papers were being frequently released. And that’s too bad, but an inevitable consequence of everyone acting in enlightened self-interest. But really I don’t think that’s true.

That’s because I don’t think most people appreciate how important very very good papers are to one’s philosophical career. If you’re Tim Williamson or David Lewis you can write several papers a year that are important and groundbreaking. But most of us aren’t like that. Most of us will be such that most papers we write will sink without much trace. The vast bulk of attention will be paid to just a few papers. This can be seen in public through looking at citation rates. (Here are “mine”:http://scholar.google.com/scholar?q=weatherson&hl=en&lr=&btnG=Search on Google Scholar for example.) The most cited papers have an order of magnitude more citations than the bulk of papers, especially when self-citations are removed.

And if we care about professional advancement as much as contribution to philosophical thought, the same story really holds. People tend to get hired based on their best papers. (And they tend to get passed over based on their worst papers.) This shouldn’t be too surprising. People are busy. They don’t have time to read a job candidates full dissertation, let alone their full output if they’re more senior. They read what is (reputed to be) the best work. And that’s what goes into hiring decisions. As we see every year when looking at junior hires, it doesn’t really matter if that best paper was published in _Philosophical Review_, the Proceedings of the Philosistan grad conference, or (more likely) the candidate’s own website. What matters is how good it is, or appears. As a rule, spending more time improving your best paper will do more for your professional prospects than sending it off and moving on to another paper.

Indeed, even if one just cares about publication, I imagine a lot of people (probably me included) could do with being slower on the “submit” button. Most, though not all, bad papers get rejected. And that takes time. Spending time making a good paper very good, rather than submitting the (seemingly) good paper may well mean one fewer rejection, and hence quicker publication.

So, simple solution to the problem of journals being so slow – don’t submit so much!

One of the things that’s been a running thread through my recent thoughts about the epistemology of philosophy is that it is importantly a group activity. This is largely for prudential reasons. For those of us who aren’t Aristotle or Kant, by far the best way to regiment our philosophical thinking is subjecting it to the criticisms of others. That’s a substantial constraint; it means giving up points that can’t convince our peers. And sometimes that will have costs; we’ll be right and our peers wrong. Sometimes we might even know we’re right and they’re wrong. But as a rule one does better philosophy if one subjects oneself to this kind of constraint from the group.

Or so it seems to me. A thorough empirical investigation would be useful here, especially in terms of trying to figure our just what exceptions, if any, exist to this general principle. But given the relatively low quality of philosophy produced by most people who don’t regard themselves as being regulated by criticisms of their peers, I think it’s pretty clear the rule as a whole is a good one.

That all suggests that the metaphor of “armchair theorising” or “armchair philosophy” is very much mistaken. For armchairs are really places where one engages in solitary activities. And contemporary philosophy is a group activity par excellence.

So we need a new metaphor. “Conference room philosophy” sounds dreary even to me. “Coffeeshop philosophy” is better. But it might be better still to keep the idea of a seat. After all, most philosophy is done sitting down. I suggest “barstool philosophy”. I’m not convinced the best philosophy is done during/after drinking, but the image is pleasingly social at least!

Plurals and Deferred Ostension

I was trying to use some other examples of deferred ostension in order to put some constraints on what might be happening with the ‘we’ in “We won 4-2 last night”. The canonical example is (1)

(1) The ham sandwich is getting impatient.

This manages to communicate that the person who ordered the ham sandwich is getting impatient. That is, “the ham sandwich” somehow manages to pick out the person who ordered the ham sandwich.

Both the explicit term “the ham sandwich” and the intended referent, its orderer, are singular. I was wondering what happened when we made either plural. First, imagine that the person ordering hadn’t ordered a ham sandwich, but had instead ordered the olives. Then I think (2a) would be more or less appropriate, but (2b) would be infelicitous.

(2a) ?The olives are getting impatient.
(2b) #The olives is getting impatient.

Second, imagine that the intended referent is plural, but the phrase used is singular. So a table of people ordered the paella, and they are getting impatient. I think (3a) is a little better than (3b).

(3a) ?The paella are getting impatient.
(3b) ??The paella is getting impatient.

Do others agree with those judgments? If they’re right, they suggest that plurality ‘trumps’. That is, if either the noun phrase used, or the intended referent, is plural, then the verb should be plural as well.

Ishani and I have been talking about an odd usage of “we” that seems to raise interesting philosophical issues. I’ll just set up the puzzle today, and hopefully over the week there will be some attempts to solve the issue.

It’s common to say that “we” is a first-person plural pronoun. It’s also common to use “we” when referring to the activities of a group that, strictly speaking, you’re not part of. So, when asked about Geelong’s latest game, I might say something like “We were three goals down at half time, but we played well in the second half and won by ten points.” Now there’s a group of 22 guys who, in the example, played well in the second half. But I’m not one of them. I’m too old, too unfit, too useless and, crucially, not a registered player for the club. What’s going on in cases like this?

The easiest thing to say is that this is simply a mistaken use of language. But I don’t think that will do. For one thing, it’s simply too widespread a mistake to be written off so easily. In some sense, a usage that widespread can’t be simply mistaken. For another, the usage shows some degree of systematicity, the kind of systematicity that we as philosophers/semanticists should be in the business of explaining. We’ll see some of the respects of systematicity as we go along, but for now let me note just two of them. The first is that it’s very hard to have this kind of usage for first-person pronouns. (There are exceptions, but this is the rule.) So (1) is fine, but (2) is marked.

(1) We played well in the second half.
(2) *I played well in the second half.

The other is that there aren’t that many cases where we can say _We did X_ to mean that some group of which you’re particularly fond did X. So it is possible to say it about (most) liked sporting teams, but not about, say, your favourite restaurant. No matter how much you like _Le Rat_, if you’re simply a fan (rather than an employee) you can’t say

(3) *We got three stars from Bruni in the _Times_.

Similarly, it is possible to say _We did X_ to mean that a political group you affiliate with did X, but not a rock band you are a fan of. So if you’re a fan and supporter of Peter Garrett both as a rock star and a politician, and Garrett has a number 1 single and an 8 point lead in the polls, then (4) could be permissible, but (5) seems considerably more marked.

(4) We have an 8 point lead in the polls.
(5) *We have a number 1 single.

So it looks like there is something interesting to explain about the pattern of usage here. In fact, there seem to be two distinct questions to ask.

The first of these we might call the *truthmaker* question. That is, what relation must hold between the speaker and the group whose actions constituted X happening for _We did X_ to be true? (Or, if you don’t think these utterances are generally true, for it to be appropriate.)

The second of these we might call the *semantic* question. Say that we settle the truthmaker question by saying that the speaker S has to stand in some distinctive relation R to the group G that did X for _We did X_ to be true. There remains a question about how _We did X_ comes to have those truth conditions.

It could be that _we_ picks out the group G. That would be an odd way for _we_ to behave, since the speaker isn’t among the G. Call this result a kind of _deferred ostension_.

Or it could be that _did X_ picks out a property that can be applied to a larger group than those that directly did X. So even if 22 guys on a field in Geelong won the game, _won_ in _We won_ could pick out a property that’s instantiated by a larger group, perhaps the group of all Geelong’s supporters. Call this result a kind of _deferred predication_.

The semantic question then is whether examples like (1) and (4) involve deferred predication or deferred ostension.

The truthmaker and semantic questions are related, we think, and hopefully by the end of the week we’ll have answers to them.

Asssume we have a radioactive particle with a half-life of 1. Then there is a countably additive probability function, whose domain includes all open intervals (x, y) and is closed under union and complementation, such that Pr(S) is the probability that the particle’s decay time is in S.

In cases where Pr(T) is non-zero, we can define Pr(S|T) in the usual way as Pr(S&T)/Pr(T). But even in cases where Pr(T) is zero, we might like to be able to have Pr(S|T) as defined.

Let T then be the set of rational numbers. (Note that if the domain of Pr is closed under countable union and complementation, then T will be in the domain.) Now we might wonder what Pr( |T) looks like. That is, we might wonder what Pr looks like when we conditionalise on T.

I think, and if I’m wrong here I’d welcome having this pointed out, that these conditional probabilities are not defined. And not because Pr(T)=0. In lots of cases probability conditional on a zero-probability event can be sensibly defined. But in this case, if there were such a thing as Pr( |T), then for any rational number _x_, Pr({x)|T) would be 0. And that would lead to a failure of countable additivity.

I imagine all of this is well known, but I hadn’t realised the consequences of this. Let D be the smallest set of sets of positive reals that includes all open intervals (x, y) and is closed under countable union and complementation with respect to the reals. Then there is no _conditional_ probability function from D x D\{} -> [0, 1] such that for any open interval (x, y), Pr((x, y)|R) is the chance that the particle will decay in (x, y). (By R here I mean the set of all reals.) If there is any function that has this last property, it must be defined over a narrower domain than D x D\{} -> [0, 1].

An interesting technical question came up in my probability lectures at St Andrews the other day, and it took me until now to realise the correct answer to it.

The question was whether there’s any good reason to think that credences can be irrational numbers. Why, went the question, couldn’t we hold the structure of credences to have the topology of the rationals rather than the reals?

Now one possible answer is that we want to preserve the Principal Principle and since physical theory gives us irrational chances, we might allow irrational credences. But I think this puts the cart before the horse. If we didn’t think that credences and chances had the right kind of topology to support the Principal Principle, I don’t think the Principal Principle would look that plausible.

A better answer involves countable additivity. The rationals are closed under finite addition, multiplication and non-zero division. But they’re not closed under countable additivity. (For examples, think of the expansions of _e_ or _pi_ as infinite series of rationals.) Since, I hold, we should think countable additivity is a coherence constraint on credences, we should think that credences have a structure that is closed under countable addition. And that means they should be (or at least include) the reals, not that they should be confined to the rationals.

For a long time I thought it was established that (given a standard axiomatisation of the probability calculus) countable additivity and countable conglomerability were equivalent. But I’ve lost confidence in my belief. So I’m wondering if anyone can tell me exactly what the answers are to a few questions below.

Just to make sure we’re clear, I’m taking countable additivity to be the principle that if each of the E_i in {E₁, …, E_n, …} are disjoint, then Pr(E₁ v … v E_n v …) = Pr(E₁) + … + Pr(E_n) + ….

And I’m taking countable conglomerability to be the following. Again, if each of the E_i in {E₁, …, E_n, …} are disjoint, then there is some E_i such that Pr(E | E_i) <= Pr(E).

_Question One_: Does a failure of countable additivity entail a failure of countable conglomerability?

I'm pretty sure that, as stated, the answer to that is *no*. Consider a standard finitely additive probability function. So there's some random variable X, and for all natural x, Pr(X=x)=0, while the Pr(X is a natural number)=1. Now insist that Pr is only defined over propositions of the form _X is in S_, where S is a finite or cofinite set of natural numbers. (By a cofinite set, I mean a set whose complement, relative to the naturals, is finite.) I'm reasonably sure that there's no way to generate a failure of countable conglomerability.

_Question Two_: Assume there is a random variable X such that Pr(X is in S1 | X is in S2) is defined for every S1, S2 that are non-empty subsets of the naturals. And assume that whenever S2 is infinite, and the intersection of S1 with S2 is finite, then Pr(X is in S1 | X is in S2) is 0. (So Pr violates countable additivity.) Does Pr fail to respect countable conglomerability?

I'm even more confident that the answer to this is *yes*. Here's the proof. Any positive integer _x_ can be uniquely represented in the form 2ⁿ(2m+1), with _n_ and _m_ non-negative integers. For short, let a statement of the form _n=x_ mean that X is one of the numbers such that when represented this way, _n=x_, and similarly for _m_. Then for any non-negative integer, Pr(X is odd | _m=x_) = 0, since for any given _m_ there is one way to be odd, and infinitely many ways to be even. By conglomerability, that implies Pr(X is odd) = 0. But an exactly parallel argument can be used to argue that Pr(X+1 is odd) = 0. And this leads to a contradiction.

_Question Three_: Assume there is a random variable X such that for any x, Pr(X=x)=0, while Pr(X is a natural number)=1, and that Pr(X is in S1 | X is in S2) is defined for every S1, S2 that are non-empty subsets of the naturals. Does Pr fail to respect countable conglomerability?

This is what I don’t know the answer to. I think the answer is *yes*, but I can’t see any obvious proof. Nor can I come up with a counterexample. Does anyone know (a) what the answer to this question is, and (b) where I might find a nice proof of the answer?

Much thanks in advance for helpful replies!

There is one other argument that Williamson deploys against Evidence Neutrality: it is unattainable. EN requires that the community be able to decide what its evidence is. But an individual can’t, in all cases, even decide what her own evidence is. In hard cases, EN doesn’t just fail as a theory of group evidence, it fails as a theory of individual evidence.

This isn’t something special about evidence. Williamson thinks there is almost nothing that we can, in all cases, tell whether it obtains. Evidence is undecidable because, he argues, practically everything is undecidable in hard cases. The latter conclusion has constraints for norms. If there are norms, then they can’t be things that we know to obtain. Williamson gives a nice example. When one is speaking to a group, the rule _Adjust the volume of your voice to the size of the room_ is a good rule, an ideal to aim for, even if we don’t know, and can’t in principle know, the exact size of the room. Such a norm is a regulative ideal; we aim for it, even if we can’t always tell how close we are to hitting it.

So there can be norms that we can’t always obtain, or perhaps can at best obtain by luck. EN might, for all Williamson has said, have such a position. We should use evidence that all the members of our community recognise as evidence. The benefits of such a rule can be seen by looking at the relative success, over the course of human history, of individual and group research projects. The great majority of our knowledge of the world is the outcome of research by large, and often widely dispersed, communities of researchers. Even in cases where a great individual advances knowledge, such as Darwin in his theorising about evolution, the individual’s work is typically improved by holding themselves to EN as a norm. In Darwin’s case, the reason for this is relatively clear, and I think instructive. Darwin collected so much evidence over such a long period of time, that the only way his younger self could convince his later self that it was all part of his evidence was by the same methods that his younger self could convince the community of biologists that it was part of his evidence. It was holding to EN that allowed him to engage in a fruitful long-term research project.

In many ways, EN is quite a weak norm. In earlier posts I discussed what amount to two major exceptions to it. First, EN doesn’t require rule neutrality. So the maverick scientist can hold EN while coming to quite bizarre conclusions by adopting various odd rules. As we saw above, we can put some constraints on what makes a good rule, but those constraints won’t individuate the good rules. Second, EN, as I’m interpreting it, allows one to choose one’s own community. One of the ways we uphold EN in science is by excluding from the community those who doubt the relevant evidence collecting methods. That means we exclude the odd crank and sceptic, but it also means we exclude, from this particular community for the while, those scientists who carefully study the evidence collection methods that we use. In the latter case at least, there is a very real risk that our community’s work will be wasted because we are using bad methods. But the alternative, waiting until there is a rigorous defence of a method before we start using it, threatens a collapse into Cartesian scepticism.

Even if EN is a norm of evidence, a regulative ideal, rather than a constitutive principle of evidence, we might still be pushed hard towards taking intuitions to be evidence. Or at least we might be so pushed some of the time. It doesn’t violate EN to take what nutritionists tell us about a healthy diet at face value; the reports of nutrition science are common ground among the community of ethicists. But we can hardly take facts about disputed examples, for instance, as given, even if they are quite intuitive to some of us. And even if, as it turns out, we know the answer. If there are people who are, by any decent standard, part of our community of philosophers, who disagree about the cases, we should be able to give our grounds for disagreement. Not because this is necessary for knowledge, but because the policy of subjecting our evidence to the community’s judgment is a better policy than any known alternative.

To be sure, some work needs to be done to show that that taking intuitions as basic does conform to this idea. As Williamson notes, one thing that might (even in somewhat realistic cases) be in dispute is the strength of an intuition. So taking EN as normative might require some modification to intuition-driven philosophical practice. But I don’t think it will require as big a diversion as Williamson’s preferred anti-psychologistic approach.

If Evidence Neutrality (EN) is true, it is presumably true everywhere. One way to argue against it then is to argue that it doesn’t hold in other subjects. And that’s what Williamson does. He argues that it doesn’t hold in particular in science.

bq. If Evidence Neutrality psychologises evidence in philosophy, it psychologises evidence in the natural sciences too. But it is fanciful to regard evidence in the natural sciences as consisting of psychological facts rather than, for example, facts about the results of experiments and measurements. When scientists state their evidence in their publications, they state mainly non-psychological facts (unless they are psychologists); are they not best placed to know what their evidence is?

If this were a true description of the position of evidence in science, it would be a problem for EN. But it isn’t. EN doesn’t psychologise evidence in science, it _institutionalises_ it. Let’s recall the original statement of EN.

bq. [W]hether a proposition constitutes evidence is in principle uncontentiously decidable, in the sense that a community of inquirers can always in principle achieve common knowledge as to whether any given proposition constitutes evidence for the inquiry. (Emphasis added)

Here’s one way to preserve EN in a field. Adopt some standards for something being evidence in that field, standards that are in practice (if not always in theory) decidable. Then take questions about whether those standards are good standards to belong to another field. That is, take it that people who are questioning the standards, questioning whether these standards genuinely generate evidence, to be outside the community in the sense relevant to EN. They might of course be part of another intellectual community, but they aren’t part of this community. That way we can preserve EN within every given community.

Compare a principle we might call Foul Neutrality (FN) governing a sport. It’s pretty important for playing football that we have a quick method for deciding what’s a foul and what isn’t. And this must be decidable independent of one’s interest in the game. We don’t get FN by psychologising fouls; we get it by having referees. The referees could be wrong, and indeed we could have interesting projects about improving the quality of referees. But when we engage in that project we’ve stopped playing football. The community of footballers (as such) satisfies FN because it’s part of being in that community that we take the referee’s word as final.

Science isn’t like football in that it requires absolute respect of the referees judgment. But it is frequently true that the project of using methods or devices to produce evidence is quite distinct from the project of evaluating whether those methods or devices are good. And we can sensibly individuate communities by looking at which methods they take as given. The short version of my response to this argument is that that’s really how science works; i.e. that science consists of communities so individuated. Each community has a refereeing institution. Or, at least, it is how it works in the vast majority of cases. In cases where the refereeing institutions break down, where there isn’t some other community to serve in effect as referee for your community, then we might have to fall back on psychological states. But EN doesn’t systematically psychologise evidence in science.

We might think that evidence must consist of facts measured rather than something about their measurement, because those are the kinds of things we can submit to statistical testing. But that argument, if it works, proves too much. Williamson’s initial description of scientific evidence was that it consisted of “the results of experiments and measurements”. But that’s ambiguous between two readings. On the first, scientists just state the outcomes of their measurements. That is the kind of thing that you can do statistical analysis on. On the second, they state the results of the measurement, and describe what kind of measurement it is. And that’s, I think, the true reading. At least for results of any interest, you have to describe how you got them, as well as what you got. But you can’t do statistical analysis on a description of a kind of measurement. So it isn’t true that all scientific evidence consists of things you can plug into mathematical equations.

On the other hand, this picture of scientific practice does seem to support the institutional picture of evidence. Why is it that we report the methods as well as the result? One simple answer is that it is settled (relative to the kind of science we’re engaged with) that using that method produces scientific evidence. That’s not to say that the method is beyond dispute. It might be that some other science studies the workings of the very machines that a particular science takes for granted in their operations. It’s merely to say that this science has approved the method in question.

We can see this even more clearly if we look at engineering settings rather than science settings. Imagine we’re working on a bridge construction project, and we need to know the density of some concrete. We’ve got a machine that measures concrete density, so we use it and, assuming the answers are plausible, we’ll take those answers as given. Evidence Neutrality will be ssatisfied because we’ll agree to use the machine. Of course, the only reason we trust the machine is that there is someone, typically someone else, whose job it is to test the machine on a regular basis, and service it, or have it serviced, if it isn’t, and although we might not know the details of how this process works, we’ll have a nice certificate saying the machine is in good condition to use. Now the folks who calibrate machines like this aren’t perfect, so there are other people whose job it is to audit them on a regular basis. And auditors aren’t perfect either, so there will be some body, perhaps a certification body, that oversees them. A positive mark from an auditor only licences a calibrator to approve a machine if the auditor is in turn certified. The board itself may need to be checked, so maybe it will have a board, perhaps including representatives of people like bridge builders who use the machines that we’re all interested in.

The crucial point about this story is that at every stage in the process, EN is satisfied. It is similar, I think, in sciences, though the structure is more fluid. Just which sciences will validate the use of the measurement techniques in other sciences is not as straightforward as in engineering. And the precise boundary between questions that are internal to a given science and questions external to it will change over time. When many questions central to the science start to turn on a particular kind of question about measurement, then those measurement questions may become part of the science. (For instance, if experimental philosophy really takes off, perhaps questions about survey design will be regarded as philosophical questions in the future. More prominently, in recent years questions about the behaviour of satellites have become part of climate science because of the importance of satellites to climate measurement.) But still the broad structure is fairly similar.

The big difference between science and engineering is what happens at the end of the process. The way I described the bridge building case was that eventually, the people responsible for checking the activities of others were the very people (or at least the representatives of them) who were being watched over to start with. That obviously isn’t what happens in science. We don’t check the activities of (say) particle physicists by putting together a board of psychiatrists, nutritionists, economists etc. How might we satisfy EN in basic physics?

Two obvious answers spring to mind. One, either common sense or philosophy tells us that we can take perceptual evidence as given. So even in fundamental physics we can individuate the community in such a way that those who are raising sceptical doubts are doing something else, namely philosophy.

The other answer is that we might take scientific evidence, at the most fundamental level, to be psychological states. Certainly it isn’t uncommon for _philosophers_ of physics to take the role of physical theory to explain our observings. That’s part of why we’ve ended up with such psychologically flavoured interpretations of quantum mechanics, from the Copenhagen interpretation to the many minds interpretation. Perhaps that’s just philosophers bringing in bad philosophical prejudices, but it seems like we _can_ do science respecting EN. That’s because EN mostly is satisfied by the institutional structure of science, and when it isn’t, it doesn’t seem to destroy science to take some evidence to be psychological. So there isn’t an argument from science against EN.

In the previous post we argued that as well as evidence, we need a notion of an unjustified rule that takes us from evidence to conclusion. Some may think that this notion is too obscure, or at least philosophically disreputable to do the work it’s put to. This section is then a discussion of rules that aims to increase their respectability. The idea is obviously not new; it traces back as least as far as Wittgenstein on rule-following, if not back to Carroll himself. And it has links to contemporary epistemology, wherever someone says that there are things we are entitled to assume without argument. But since it is doing so much work here, and since in conversation it has often been the most puzzling aspect of the argument to others, it may be worth saying just a little about what I’m taking rules to be here.

The argument in that section relied on Lewis Carroll’s example involving modus ponens, and that might suggest that modus ponens is the kind of rule that we need. Indeed, it often seems that some philosophers think that it’s the _only_ rule we need. (Many philosophers have said that conditionalisation, which is really just a probabilistic form of modus ponens, is the only rule we need.) I think this example is misleading for four related reasons.

First, modus ponens is a rule of *implication* and what we’re really looking for here are rules of *inference*. We’re looking for rules that tell you what to do with evidence. And what you do with evidence is draw inferences from it. These may not, in any interesting sense, be implications of the evidence.

Second, modus ponens is necessarily truth preserving, and this might be thought to be related to its acceptability. This seems like a mistake twice over to me. It’s arguable (indeed I’ll briefly argue for it presently) that we need some ampliative rules to explain the rationality of induction. And there are necessarily truth preserving rules that we cannot employ without justification. The rule that lets us infer _p and oxygen has atomic number 8_ from _p_ is necessarily truth preserving, but not a rule we could freely employ without independent justification.

Third, modus ponens can be used in suppositional reasoning just as well as in regular reasoning. It would take us way too far afield to investigate this properly, but I suspect some of the rules we’ll look at can’t be properly be used in all suppositions. (Of course some rules of _implication_, at least in some formulations, also have restrictions on when they can be used; think of the restrictions on necessessitation or universal-introduction.)

Fourth, modus ponens might (although this is controversial) be constitutively related to the meaning of the conditional. Perhaps, as some inferentialists believe, the meaning is determined by the acceptability of the rule. Perhaps the meaning directly implies that it is an acceptable rule. If either of those things are true, they aren’t I think things we’d want to generalise to, say, rules for rational inductive inference. (There is obviously a lot to be said here, such as considering what might justify inferences related to logic if not meaning. Many recent papers by Boghossian, Williamson, Wright and others are relevant here. But I’m just going to avoid that issue for today, in large part because I do find it mysterious how to generalise much of that debate from issues about modus ponens to issues about, say, enumerative induction.)

Now all that is just to say what rules are not. Can we say what they are? As noted above, in some ways rules are very familiar. Any time any philosopher claims that we are warranted, or entitled, to, without justification, rely on the deliverances of some source, it seems they are proposing that the inference from _The source says p_ to _p_ is a good rule. So we can take dogmatists about perception (e.g. Pryor’s “The Sceptic and the Dogmatist”) as endorsing a rule that lets us move from _Appears that p_ to _p_ without antecedent justification. And some anti-reductionists about testimony seem to hold that the rule which licences the inference from _I’m told that p_ to _p_ is a good one. So in some ways this isn’t a new idea, it’s just a way of framing an old idea. There are, however, two reasons we might think that this is a good framing. First, it lets us ask some relatively precise questions about the statement of the rules. Second, by thinking about rules as a class, we can formulate restrictions on what could be a rule.

It is harder than one might like to actually state rules that we can or do use. Clearly it isn’t a rule that we can, in every case, infer from _Appears that p_ to _p_; there are illusions, some of which we know about. Nor is it a rule that we can, in every case, infer from _I have observed many Fs in widespread conditions, and they have all been Gs_ to _The next F I observe will be G_; there are gruesome predicates. We could try to incorporate the exceptions into the rule, but a quick glance at how one might do this reveals that it isn’t too attractive, at least as long as we want to have non-trivial rules. Probably we need something different.

In the case of appearance, there is a natural move to make. Instead of a single rule with all sorts of qualifications, we might try to defend the following two unqualified rules.

(A) If it appears to you that _p_, then you have a reason to believe that _p_.

(ND) If you have a reason to believe that _p_, and this reason isn’t defeated, believe that _p_

The conjunction of (A) and (ND) entails Pryor’s dogmatism, although that conjunction is considerably stronger than dogmatism. For example, it entails that appearances that you have reason to believe are deceptive provide some reasons; dogmatism as such is silent on such appearances. But both rules seem at least defensible, even given the existence of illusions. (Of course, if they are good rules, they don’t need a defence. But we can reasonably expect that a good rule won’t be such that we can have good reason to believe it is bad, and (A) and (ND) pass that rule.)

The case of induction is a little trickier, because as soon as we think about the rule a hard question arises. We can only project non-gruesome predicates. Does that mean (a) that the non-gruesomeness of the projected predicates should be an input to inferences licenced by the rule, or (b) that the rule requires no such input, but it only licences inferences when the predicates are in fact non-grue? The latter sounds more plausible; otherwise no one who lacked the concept of a gruesome predicate could rationally make inductive inferences. But it raises tricky questions about using that rule in suppositional reasoning. What if (assuming this is possible) different predicates would be gruesome if _p_ were true, and we are inferring under the supposition that _p_? It isn’t obvious just what restrictions should be put on the use of this rule in a suppositional context. Perhaps those restrictions are quite tight. This has important consquences for arguments for the contingent a priori that assume that we can make inductive inferences in suppositional contexts, such as John Hawthorne’s “Deeply Contingent A Priori Knowledge” and my “Scepticism, Rationalism and Externalism”.

That was all about the nature of rules; we might wonder whether there is anything that we can say about which rules there are. Different ways of thinking about Carroll’s example suggest two different constraints on rules, one more liberal and the other more conservative.

The more liberal constraint is a kind of transcendental consideration. In many cases it seems, at least prima facie, that we can get knowledge from a certain source, but we couldn’t antecedently justify the use of that source. A classic example of this kind of reasoning is C.A.J. Coady’s arguments for against reductionism in testimony. The thought there is that so much of the time the only way we have of checking one person’s testimony is through the testimony of another, that if we weren’t able to take some testimony as basic knowledge, we’d be led to a debilitating scepticism. That seems unacceptable, so we might take such a rule as given.

The more conservative constraint takes more seriously the particular way in which we need a rule to sidestep Carroll’s tortoise. The core problem isn’t just that when we add another premise, one that justifies a particular use of the rule, we need yet more to get to the conclusion. Rather, the core problem is that when we add the kind of premise that could justify the rule, we need another step of the very same rule. Justifying this particular use of the rule doesn’t seem to get us any closer to where we need to be. Perhaps those cases, where justifying a particular use of the rule still requires the rule, are the only cases where there are unjustified rules.

This seems to be a more conservative principle because whenever it obtains, we’ll be able to give a transcendental argument for the existence of a rule. But the converse doesn’t seem to hold. We might justify taking someone’s testimony about being true because we believe them to be generally reliable. We still need a rule saying that it’s good to believe those who are generally reliable, but that doesn’t seem like the same rule. Similarly, we might accept (A) as a derived, and hence justified, rule because the best explanation of our experiences is that they are generally reliable. (Jonathan Vogel has argued for this at length over the years.) We still need a rule saying that we should believe the best explanation of a phenomena, but that doesn’t look like rule (A) again. On the other hand, Hume’s arguments about induction arguably do show that a justification of induction will need to use induction. And a justification of (ND) will, I imagine, still use something equivalent to it. So even on the more conservative conception of rules, they may still be rules.

These are enormous questions, to say the least, so I don’t think this goes close to settling anything. But I hope it’s enough to suggest that we haven’t given up on systematic epistemology once we admit the notion of unjustified rules that justify inferences. And admitting that, which Carroll’s example suggests we must, is enough to sidestep Williamson’s argument against EN.