Against manifestationalism

I recently read Asay and Bordner's (A&B) interesting paper "A modest defense of manifestationalism". Last week I discussed a stricter form of constructive empiricism which holds that we should believe that our best theories correctly describe not observable phenomena but only observed phenomena. A&B explore the same view under the name "manifestationalism". More precisely, they given the following taxonomy:

Scientific realism (SR): Science aims to give us theories that are literally true; acceptance of a theory involves the belief that it is literally true.

Constructive empiricism (CE): Science aims to give us theories that are empirically adequate; acceptance of a theory involves only the belief that it is empirically adequate.

Manifestationalism (M): Science aims to give us theories that are comprehensively manifestly adequate; acceptance of a theory involves only the belief that it is comprehensively manifestly adequate. (A theory is manifestly adequate just in case it correctly describes all observed phenomena.)

I have already mentioned that I don't think it's particularly helpful to frame the realism debate in terms of the aim of science. The interesting position is that of the manifestationalist who believes that scientific theories correctly describe only observed phenomena, and who is agnostic about the unobservable and the observabled-but-unobserved.

A&B clarify that manifest adequacy is not simply truth to the phenomena observed so far, but truth to all phenomena that have ever been or will ever be observed, including those observations that take place outside a scientific context (this is what they mean by comprehensive manifest adequacy). This clarification is important. Belief in nothing more than what has been observed so far entails radical scepticism about the future, a position which is not just implausible but probably psychologically impossible to adopt. M avoids this radical scepticism. M endorses the belief, for example, that I will continue observing my computer screen for as long as I sit here typing; that when I go downstairs to get some food I will observe some bananas in the fruit bowl; etc.

The case for manifestationalism

A&B do not attempt to present a conclusive case for M. They merely try to show that it is not absurd or irrational, but a serious contender in the philosophy of science. I will argue that they fail. Much of A&B's paper is devoted to answering a criticism of M based on the aim of science, which I discussed briefly in my previous post; I will focus on their arguments for M, of which there are three:

1. One possible argument for M appeals to van Fraassen's "vulnerability criterion" of belief, expressed in this passage:

If I believe the theory to be true and not just empirically adequate, my risk of being shown wrong is exactly the risk that the weaker, entailed belief will conflict with actual experience. Meanwhile, by avowing the stronger belief, I place myself in the position of being able to answer more questions, of having a richer, fuller picture of the world, a wealth of opinion so to say, that I can dole out to those who wonder. But, since the extra opinion is not additionally vulnerable, the risk is — in human terms — illusory, and therefore so is the wealth. ... What can I do except express disdain for this appearance of greater courage in embracing additional beliefs which will ex hypothesi never brave a more serious test?

Given two theories that are equally vulnerable, we should believe the logically weaker theory. Van Fraassen uses this to argue for CE, but A&B claim that this argument is really an argument for M, not for CE. We will only ever be able to test our theories against those phenomena that are actually observed; we will never obtain any evidence that could decide between two rival manifestly adequate theories. If I believe a theory to be empirically adequate and not just manifestly adequate, my risk of being shown wrong is exactly the risk that the weaker, entailed belief will conflict with actual experience... M and CE are equally vulnerable and M is logically weaker than CE. So we should accept M.

Of course, the force of this argument for M depends on whether we accept van Fraassen's vulnerability criterion of belief. I won't discuss the case for or against this here. Another question is, are CE and M equally vulnerable? Once the end of science occurs, and all the observations we will ever make have been made, the answer is clearly "yes". However the situation is somewhat trickier from our current position. We can see the problem clearly by asking, exactly which aspects of our best theories should we believe? It's easy for the constructive empiricist to answer this. CE tells us to believe what our theories tell us about the observable world. But M tells us to believe what our theories tell us about the observed world - observed in the past, the present, and the future.

The obvious difficulty here is that we don't know what observations will be made in the future. Consider the following case. My friend Frank has an apparently normal chicken egg. Nobody has ever observed the interior of this egg. Frank decides to put the egg in a box and throw it into a volcano, so that nobody will ever observe its interior (suppose this is part of some bizarre religious ritual). Well, if I accept M, then I must be agnostic about what it in the interior of the egg. Its interior will, in all likelihood, never be observed. But then suppose that while carrying the egg to the volcano, Frank drops it and it breaks, revealing the yolk and egg white.

The point, in the context of the "vulnerability criterion" argument for M, is that since we don't know what will be observed in the future, it's not at all clear that, at the present time, CE and M are equally vulnerable. The manifestationalist believes only that theories are true for what will be observed - if we don't know whether X will be observed, or we have good reason to believe that X won't be observed, then we must be agnostic about X. For this reason the manifestionalist, but not the constructive empiricist, must be agnostic about many things that will ultimately end up being observed. CE is therefore riskier than M, and this pushes us towards CE.

2. A&B suggest that M better captures the spirit of empiricism, which counsels us to limit belief to what in principle to tested against experience. The observations that we can make are constrained by our spatiotemporal location. CE involves belief in entities and processes billions of light-years distant or billions of years in the past, that we could not possibly observe. M offers a more modest position based on the specific position of humans in the universe and the epistemic limits entailed by this.

With this argument, I think that A&B have identified a genuine problem with van Fraassen's formulation of CE, but other versions of CE are less vulnerable to their criticism. We simply need to reconceive what is meant by "observable". I offer one suggestion in this post. The constructive empiricist hold the observable is what could in practice be detected by the unaided senses, given the technology of the time; and no feasible technology currently available would allow us to travel to distant quasars. This conception of observable is of course not precise - there is no absolutely clear line between the observable and unobservable - but this is equally true of van Fraassen's way of drawing the distinction.

3. Finally, A&B argue that the aim of an enterprise should be related to "what an ideal community of practitioners of that enterprise would accomplish under ideal circumstances." The aim of an enterprise should be attainable at least in ideal circumstances. A&B then suggest that the ideal scientific community is a community of people just like us, with the same sensory capacities and occupying the same spatiotemporal location in the universe, but who are devoted exclusively to the development of science. They meticulously and flawless record every observation. They generate theories that perfectly fit the observations. What will the final product of this community be? They will have a theories that are manifestly adequate - but not necessarily true or even empirically adequate. Rival empirically adequate theories will fit the data equally well. Even in ideal circumstances, there is no guarantee that we would produce empirically adequate theories, but we would produce manifestly adequate theories. This favours M.

I have two objections to this argument. First, why should the aim of an enterprise be decided by what an ideal community would achieve? Consider games with impossible aims. The aim of Klondike solitaire is to move all cards to their foundation piles; however, this is not possible even in principle for about 20% of deals. An ideal community of Klondike solitaire players could not move all the cards to their foundation piles in those deals - nevertheless, the aim remains the same.

Second, why suppose that there is a definable "ideal community" of scientists? A&B suppose that the ideal community is limited to our place in space and time; but why? Surely an even better scientific community would be one that existed as soon as complex life was possible, and had been meticulously collecting observations ever since. Would all members of the ideal community have perfectly functioning sensory capacities, e.g. 20/20 vision? Perhaps they would have superhuman capacities, 20/10 or even 20/1 vision perhaps. It isn't at all clear what counts as ideal in this respect.

It's clear that the ideal community will need some superhuman capacities. A&B suggest that ideal community would devote all their time to science, and they record every observation they make. Neither of these things is possible for humans. We need to devote time to other enterprises such as eating food, and we wouldn't be able to record every observation even if we tried. There are simply too many observations, and in any case, a person could not record observations while at the same time generating theories to account for those observations.

The problem for A&B is that in order to ensure that the ideal community generates manifestly adequate theories, they must suppose that this ideal community is capable of feats that go far beyond human capacities - indeed, feats that go far beyond the capacities of any conceivable organism. But then why not say that the ideal community is one that observes every observable? This community would produce empirically adequate theories - and this would give us an argument for CE.

What's wrong with manifestationalism?

There are many objections we might raise against M. I want to outline what I regard as a knock-down argument against it. Consider again Frank and his egg. We saw that Frank broke the first egg. So now he goes and out and purchases a second one, which again he intends to put in a box and throw into a volcane. This time, he takes many precautions to avoid breaking the egg, such as wrapping it thickly with bubble wrap. If I accept M, I must be agnostic about the interior of the egg, since I have good reason to believe that this will never be observed.

But now suppose that on the trip up the volcano, Frank gets hungry, and decides he wants to eat the egg. Well, now I have good reason to believe that Frank will observe the interior of the egg, and since our best theories tell us that he will observe a yolk and egg white, I should believe that he will observe a yolk and egg white. Of course, this example can be run in reverse as well, where Frank initially intends to eat the egg, and then intends to throw it into the volcano: in this case, I first believe that he will observe a yolk and egg white, and then I must give up that belief and become agnostic.

This is of course an absurd conclusion, absurd enough in my view that it's enough in itself to remove M from serious consideration. One response to this is to hold that we should remain agnostic about anything that has not yet been observed - even if we have good reason to think that Frank will open the egg, we should remain agnostic about the interior of the egg until he actually does it. But this entails radical scepticism about the future, which is no less absurd; and as I noted earlier, M was supposed to avoid radical scepticism.

Ultimately, it seems to me that A&B fail to show that M is a serious contender. There are very serious, very obvious problems with M, and no plausible arguments in its favour.

Constructive empiricism and the past

Derek Turner argues, in chapter 7 of Making Prehistory, that van Fraassen's constructive empiricism entails radical scepticism of the past. Here is Turner's argument:

(P1) All of our knowledge is limited to that which we can observe.
(P2) We cannot observe things which no longer exist or events which occurred in the past.
(C) Therefore, we cannot know anything about past things and events.

The constructive empiricist must suspend belief in dinosaurs, the Black Death, WWII, even what she had for breakfast this morning. Turner takes this conclusion to be far too radical, hence a good reason to reject constructive empiricism. I agree with him that such radical scepticism should be avoided, but I'm not convinced that the constructive empiricist is committed to it.

It seems clear that the constructive empiricist is committed to (P1), since this merely expresses the constructive empiricist's scepticism about unobservables. Van Fraassen would presumably want to reject (P2), but as Turner argues, it's not clear how he could do so. The natural response for van Fraassen is to say that some past entities such as dinosaurs were such that, if they were around today, we would see them. So dinosaurs are observable. The problem with this move, as I discuss in this earlier post, is that we may as well also say that if we stepped into a miniaturization machine, we would see bacteria.

In any case, the interesting question is not whether past things and events are observable per some definition, but whether there is good justification for believing in past things and events. My point is, we can surely draw the observable/unobservable distinction so that some past things and events count as observable - but similarly, we can draw that distinction so that electrons and neutrinos count as observable (we might say that to observe something is simply to detect it either with the senses or with instruments). In practice, dinosaurs are just as inaccessible to us as electrons and neutrinos - indeed, as Turner argues, dinosaurs are if anything even less accessible, since if electrons and neutrinos do exist we might interact with them and manipulate them instrumentally.

The constructive empiricist must accept (P1) and (P2), and the conclusion straightforwardly follows. However, the significance of the conclusion depends on how we interpret the word "we". Turner seems to assume that "we" refers to presently existing humans. I think we should resist this interpretation. To explain why, consider this very similar argument, which suggests that constructive empiricist to an even more extreme form of scepticism:

(P3) All of my knowledge is limited to that which I can observe.
(P4) I cannot observe anything other than my immediate surroundings.
(C) Therefore, I cannot know anything abuot that which is not in my immediate surroundings.

The problem with this argument is that science is fundamentally a human and communal activity. Perhaps there could in principle be one lone scientist, who correctly applies the scientific method (whatever that is) but who never communicates with any other people - clearly, however, her investigations wouldn't get very far. Every scientist relies on the testimony of others, and although testimony is often unreliable, there are circumstances where we can trust it. Science depends on this because a single scientist can't test everything. The vast majority of beliefs that a scientist holds even about the observable world will never have been tested by her. Speaking as a layperson, I've never observed the planet Neptune. I rely on the testimony of other people, who claim to have seen it through their telescopes. We work as a community to generate scientific knowledge; if you reject that community aspect, you undermine the very conditions of modern science.

Observability is relativized to the epistemic community. Who counts as part of that community? Not just contemporary people, but people in general - past, present, and future - and all their observations count. The Black Death was not simply observable, it was actually observed by hundreds of thousands of people, and we have numerous reports of it, many of which are reliable. So we should believe that the Black Death occurred. Of course, the people who observed the Black Death no longer exist. But why would that matter? Science is something that develops across many times and places. Indeed, building and testing theories can only be done over time, sometimes only over literally decades.

My point is that the existence of human communities over time, and certain conditions under which we can reasonably rely on the testimony of others, is simply taken for granted in any philosophy of science, because if this assumption rejected there is no reason to think that anything resembling science actually takes place. Having made this assumption, there is then an additional question about what else we should believe: for example, does science require us to extend belief beyond human observation?

Of course, a question remains about what justifies this assumption. Why should I believe that other people exist, that I can sometimes rely on what they say, etc? This questions arises when we "step back" from philosophy of science and it is, of course, a familiar question of philosophical scepticism. Perhaps it will turn out that whatever justifies belief in the existence of other people will also justify belief in some unobservable entities such as electrons. This would refute constructive empiricism, or at least it would make the constructive empiricist's selective scepticism of unobservables arbitrary. But in principle, the constructive empiricist need not be committed to scepticism of human history.

What about prehistory? Can the constructive empiricist believe in dinosaurs? On this point, I'm inclined to agree with Turner: constructive empiricism implies total scepticism about the distant past. I don't, however, share Turner's feeling that this is an especially extreme position. Indeed, it's curious that Turner regards such scepticism as so repugnant, because he devotes a great deal of his book to defending an epistemic asymmetry thesis that we have better access to the microscopic and microphysical than to the distant past; as he puts it, "there is a rough sense in which we can know more about the tiny than about the past" (pg 25). However, Turner is rather sanguine about the constructive empiricist's scepticism of microphysical unobservables; he sees this as a fairly modest feature of the theory. If we have better access to the tiny than to the past, and if scepticism of the tiny is at least reasonable (though perhaps not rationally compelled), why would scepticism of the past be repugnant?

Observable or observed?

In my previous post I discussed some difficulties with the observable/unobservable distinction in constructive empiricism (CE). Here I'll focus on a related problem, the distinction between the observable and the observed.

Extreme constructive empiricism

In accepting that our best scientific theories correctly describe all observable phenomena, CE goes far beyond the empirical evidence. Most of what is observable will never be observed. Consider this position, which I will call "extreme constructive empiricism" or "ECE" for short: our best theories are true for the observed phenomena, but we should be agnostic about everything else. Why not adopt this more sceptical position? There are some obvious benefits to ECE. In holding that science delivers truth about the observable, two common arguments for antirealism are unavailable to CE:

(1) The pessimistic induction. Put briefly, this argument goes as follows: (P1) Most scientific theories accepted in the past have been rejected; (C) so, most currently accepted theories will be rejected in the future.

However, every case in which claims about unobservables have been rejected has also involved rejection of claims about observables. Those theories that we rejected in the past were rejected because they conflicted with observations. If our current theories are going to be rejected in the future, this will presumably be because they confront new observations that contradict them. So no antirealist who takes current accepted theories to be true for observable phenomena could endorse the pessimistic induction. Obviously this difficulty doesn't arise for the ECEist who only takes those theories to be true of the observed phenomena.

(2) The underdetermination argument. This argument rests on the claim that for any theory that correctly describes observable phenomena, there are numerous other theories that also correctly describe observable phenomena but that make different claims about unobservables. Again, it is easy to see that this can be run against CE, because for any theory that correctly describes what has been observed, there are numerous other theories that also correctly describe what has been observed but that make different claims about observable-but-as-yet-unobserved phenomena.

So what is the justification for belief in the observable over the merely observed? I will discuss two arguments.

The aim of science

First, Van Fraassen and Monton argue that CE is an attempt to describe the aim of science, with an assumption that the behaviour of scientists is rational. If the aim of science were truth about the merely observed, rather than the observable, then:

there would be no scientific reason for someone to do an experiment which would generate a phenomenon that had never been observed before. But one of the hallmarks of good scientists is that they perform experiments pushing beyond the limits of what has been observed so far.

Suppose I develop a theory of galaxy evolution that accommodates all that has been observed so far. I could derive further predictions from this theory, and then probe the sky to check if those predictions are accurate. But if the aim of science is simply truth about what has been observed, why bother? My theory already achieves this aim; it is therefore as good as any theory could possibly be. It would be unreasonable to risk turning an perfect theory into an imperfect one by making further observations that could contradict it. By contrast, in holding that the aim of science is truth about the observable, CE accommodates the fact that scientists spend so much time generating new observations.

There are a couple of things to say about this. First, assuming that it even makes sense to talk of the aim of science as a whole (as opposed to the aims of particular scientists or particular research groups), why would we suppose that science has only one aim? Two others goals that science plausibly aims at are the discovery of new phenomena and the manipulation and control of natural phenomena. If science aims at discovery, that immediately makes it rational for scientists to make new observations. If science aims at control and manipulation, new observations are always worthwhile because the more we know about the world, the better we are at controlling and manipulating it.

Second, we must distinguish the aim of science from the epistemology of science. Just because science aims at empirical adequacy, it doesn't follow that we should believe that our best theories are empirically adequate. Perhaps science sometimes falls short of its aims. After all, we all know that science has developed over time, and we hope that it will continue to develop in the future, that in the future it will be closer to achieving its aims than it is today. So the question remains, why suppose that our scientific theories save the observable phenomena, rather than merely save the observed phenomena?

Epistemic risk

A second argument for belief in the observable is hinted at by van Fraassen in his contribution to Churchland & Hooker's Images of Science. Although this is stated as an argument against realism, we can use the ideas here to support CE over ECE:

If I believe the theory to be true and not just empirically adequate, my risk of being shown wrong is exactly the risk that the weaker, entailed belief will conflict with actual experience. Meanwhile, by avowing the stronger belief, I place myself in the position of being able to answer more questions, of having a richer, fuller picture of the world, a wealth of opinion so to say, that I can dole out to those who wonder. But, since the extra opinion is not additionally vulnerable, the risk is — in human terms — illusory, and therefore so is the wealth. ... What can I do except express disdain for this appearance of greater courage in embracing additional beliefs which will ex hypothesi never brave a more serious test?

The realist believes everything the CEist believes, but the realist also holds additional beliefs that cannot face any additional tests. As Kukla (pg 98) puts it, van Fraassen seems to adopt a "vulnerability criterion of belief". Two theories are equivalently vulnerable when they are disconfirmed by precisely the same observations. Van Fraassen holds that if two theories are equivalently vulnerable, we should believe the theory that is logically weaker.

CE is obviously riskier than ECE, since the CEist is committed to beliefs about what has not yet been observed whereas the ECEist is agnostic about this. The two positions are not equally vulnerable. Of course, this is not yet an argument for CE. Indeed, if anything this looks like an argument against CE: intuitively, if a belief is riskier, if it is more likely to be wrong, then that is a reason not to hold that belief.

But notice that if our goal is simply to minimize the risk of holding false beliefs, then we should become radical sceptics and believe nothing at all (or perhaps we should believe a few propositions that we think are beyond doubt, but certainly we should believe nothing substantial about the world). Clearly we have other goals. There are, broadly speaking, two types of epistemic error: type I errors consist in believing what is false; type II errors consist in failing to believe what is true. Attempting to avoid one type of error pulls us in a different direction than attempting to avoid the other type, and an important question in epistemology and philosophy of science is how to balance these two goals.

Here is a simple suggestion: We avoid type I errors through the vulnerability criterion of belief; don't hold beliefs that go beyond what can be tested empirically. We avoid type II errors by holding the strongest possible beliefs given the vulnerability constraint. This is exactly what CE does. ECE makes very severe type II errors: to be agnostic about everything beyond what has been observed entails total agnosticism about the future, and this would be disastrous even assuming that it's psychologically possible. ECE undoubtedly commits fewer type I errors than CE, but the price in type II errors is simply too high.