Science as maximal reliableness

I'm reading a book called Metascience and Politics by one A.J. Gregor (some PoliSci professor at Cal). I'm only in the second chapter, but came across this interesting passage:

Some of the critical issues that seem to divide political science as a discipline into factions turn on what science is understood to be. First, the suggestion that "unscientific" methods are essential to political inquiry requires that we have at least some understanding of what "science" might be--and secondly, the identification of "behavioralists" seems, generally, to turn on the disposition of some political scientists to attempt to employ the methods of science in their work. Science, and the methods of science, should be, therefore, the first objects of analysis if one wishes to assess the cognitive merits of the "behavioralist-non-behavioralist" dispute.

The effort to unpack the concept "science" and to offer a summary account of "the scientific method," is beset, of course, with innumerable difficulties. The term "science" obviously has persuasive force.... It has also enjoyed a long history. Its specific cognitive meaning is, as a result, difficult to isolate, not only because of the persuasive employments of the term, but because the term has had a long and confused philological history....

Once this qualification is made, it can be argued that the term "science" has taken on a relatively (but not undisputed) specific contemporary meaning: it refers to those procedures, which, as a matter of historic fact, have provided a systematically articulated and comprehensive body of maximally reliable knowledge claims that afford men survival and adaptive advantage by affording explanatory and predictive leverage. Such an explication thus includes both a consideration of "science" as process (that is to say, the term refers to methods and procedures invoked to provide for maximally reliable truth ascriptions) and "science" as product (that is to say, the term has as referent that corpus of linguistic entities to which truth can be ascribed with maximal domain variant reliability). The former is commonly spoken of as the "scientific method," and the latter "scientific knowledge." Both are historic products and both are historically relative.

What this implies is that neither specific observational nor experimental procedures, nor methods of generalization, nor specific logicodeductive strategies, nor any collection of procedural assumptions or presuppositions, nor any technique of measurement or instrumentation is absolutely essential to the method of science. Furthermore, no single existential assertion delivered by scientists, nor any conjunction of such assertions, is absolutely essential to the corpus of science. Science is neither a specific collection of procedures nor a specific body of essential truths. Every truth warranting procedure and every truth claim remains, in principle, subject to review--none are specific to science.

...

The one feature we have gradually come to realize as constitutive of science in whatever form it has taken, a feature pervasive of science both as process and product, is its necessary concern with reliability. Without reliability argument and deliberation cannot proceed. Knowing anything becomes impossible.

...

Our preoccupation with reliability is dictated by the generic human concern with adapting to, and controlling, our complex and changing environment. We find it necessary to anticipate futures--and we do that best by exercising the imagal trial-and-error techniques, and their subtending rationale, that we identify as rational behavior....

[pp. 21-23; all emphasis his]

What I think is interesting here is that science is given such a purely procedural definition: science is X, where X is whatever set of methods and procedures that have generated the most reliable set of knowledge claims that help us predict stuff and adapt to our world. There's no a priori constraint on what those methods or knowledge claims could be--indeed, Gregor says elsewhere that "[i]f intuitions, mystic insights, the dialectic, phantasy, or whispered intelligences from God proved to be maximally reliable in permitting men to empirically adapt to and effectively control their environment, they would become, ipso facto, constituent procedures of science." In other words: science is whatever gets results, dammit!

While the definition of science is in this view procedural, it is still the case that all science candidates must output empirically testable results--otherwise, you would have no common measuring stick by which to compare the reliableness of the different candidates. So a possible science in which you ask the High Priest for the correct answers is permissible, so long as the High Priest is giving predictions that can be verified in an uncontroversial way later on.

Now, Gregor is not coming up with all this in a vacuum--he has a specific purpose in mind, which is to show that--because science has no immutable, first-order claims, it therefore cannot be considered an "ideology". Gregor identifies as an essential feature of ideologies that they have one or more first-order claims that one cannot negate while still claiming to be an adherent of that ideology. For example, one cannot negate that some races are superior to others while still claiming to be an adherent to National Socialism--thus, National Socialism is an ideology. Since science only contains a second-order claim (that is, a claim that is not directly about objects in the world, but rather, one which is about claims that are about objects in the world--in this case, the reliability of such claims)--that science is X, where X is the most reliable set of methods etc.--it contains no individual substantive first-order claim that is truly immutable. It is in this sense forever self-correcting and irrigid--and so definitely not an ideology.

But I wonder if Gregor makes a hash of things in this defense of science. Indeed, it seems that by defining a science solely according to its empirical track record, it collapses the distinction between method and theory. Let's say that we have set of scientific methods M that taken together define a science, and that, adhering to M, scientists generate a set of theories T. In the normal course of events, the scientists would conduct empirical tests of all the various theories in T, and the ones that yielded reliable predictions would be kept as "(provisionally) true" and the ones that yielded bad predictions would be scrapped as "false"--and over time more theories (generated in accordance to M) would be added to T and so on. But now if we face the question as to the legitimacy of M--of the methods that generated all these theories--Gregor would have us apply the exact same criterion to M as we did to the theories in T, namely, that M yields reliable predictions.

But this seems confused. In the first place, it's odd to think of "reliability" as a property of M, when in fact M is not itself making a set of predictions about the world that are true or false--M generates theories that make these predictions. So when Gregor defines science as "those procedures, which, as a matter of historic fact, have provided a systematically articulated and comprehensive body of maximally reliable knowledge claims", he is moving too fast and skipping a step: the procedures that define a science do not themselves "provide" "knowledge claims", they provide testable theories; and it is these theories, once tested, that provide the knowledge claims. Perhaps we can infer that Gregor meant that science is the set of procedures that, as a matter of historic fact, has provided the most successful theories. However, this understanding presents a number of difficulties, chief among them being: what is the point of a scientific method, and what work is it doing for us? Certainly we could not, on the basis of the methods it utilized, dismiss a priori any questionable undertaking that called itself science, since any set of methods is fair game: we would have to wait for this set of methods to to churn out theories over a period of time sufficient for it to be deemed a "historic fact" that the set of methods did not yield a "comprehensive body maximally reliable knowledge claims". In other words, a scientific method, on Gregor's view, can only be evaluated a posteriori. But in this, a scientific method becomes functionally equivalent to what Gregor elsewhere describes as a "methodology"--a specific approach within a science that yields a specific kind of data and prediction, and whose value is determined by the quality of the predictions it makes (for example, in sociology there may be different methodologies--some qualitative, some quantitative, whatever--for determining the quality of life of a class of people, but these methodologies are all consistent with the same scientific method).

Indeed, I was under the impression that the whole point of defining a "scientific method" was so that we could impose some a priori constraints on just what kinds of theories were allowable, so that we could a) situate science in general in a broader philosophical/epistemological framework (or, put another way: as a philosopher, I should be able to talk about science such that it is abstracted from any particular claims it makes, in case I want to have a discussion about, say, the general difference between scientific and ethical claims), but more importantly , b) provide a procedural guarantee that the sort of knowledge claims that the science ultimately produces are of a kind consistent with science qua "human concern with adapting to, and controlling, our complex and changing environment". Now, about (a), don't get me wrong: it is not as if Gregor's definition leaves philosophers with nothing to say about science--he still allows it to be defined such that it must be about generating empirical claims, and science is necessarily tied to a general human activity (adapting to the environment, etc.). But it does seem weird that in a discussion about the rules of induction, you could not, by Gregor's lights, claim to necessarily be talking about something that is relevant to science (the "scientific method" could, after all, be nothing more than writing down the predictions of the High Priest). With regards to (b), it may be true that Gregor's conception of "science as maximal reliableness" is sufficient in most cases to convince us that its claims are of a common type--a type of claim that is characterized by its being in some essential way concerned with helping humans to adapt to their environment. However, the problem is that--while in lots of cases the notion of reliability is straightforward because theories are put to work making predictions--in some cases, a theory is put to work to explain something that happened in the past, and direct empirical testing of such theories is difficult or impossible. For example, cosmologists posit theories about how the universe first unfolded, or evolutionary biologists hypothesize about the sequence of events that led to life developing on earth. In these cases, a theory is created to retroactively explain historical evidence--and so any theory consistent with the evidence must be considered as "reliable" as any other, no matter what method that theory was generated in accordance with. So it is precisely in these cases where the idea a scientific method actually does some useful work: it brings unity to predictive knowledge claims and historical ones by bringing them under the same criterion, not of reliability--which isn't really applicable to historical claims--but of the procedural criterion of having been generated by the proper, "scientific" method. And with this in place, we are thus free to rule out historical (as opposed to predictive) theories a priori that otherwise would be impossible to rule out for lack of the ability to test their reliability.

(As a quick addendum, let me just add that, really, this idea of a One True A Priori Scientific Method is required if we are make any sense of science as something that is concerned with the human activity of controlling and adapting to our environment. The crucial distinction to keep in mind is between that of a science and that of a history. A history is just a straightforward record of stuff that happened, and so of course there's no a priori constraint on what a history might contain besides that it's logically consistent--so for example, there is nothing wrong with a history in which God parts an ocean or where the universe is created 4000 years ago. Its just data, which--true or false--is always admissible as data, as a history. But a scientific theory is not a record, and it is not simply a set of data--it is something like a schematized conditional claim of the form "if X occurs under conditions Y, then Z will occur". So when a cosmologist theorizes that the big bang occurred, properly speaking there is the scientific claim that goes something like "when you have a universe that is infinitely dense, then it will rapidly expand outwards", and, distinct from this, a historical claim that goes something like "15 billion years ago, the Big Bang happened". Of course, anyone who believes that it is acceptable for scientific claims to be used as the basis for the actual Historical Record of the World would have no trouble accepting all of what the cosmologist says, but someone who had, say, certain theological commitments could consistently accept the cosmologist's scientific claim while rejecting the cosmologist's historical claim. But, crucially, what our theologian cannot do is reject the cosmologist's scientific claim (except on scientific grounds)--nor can the theologian ever even submit a scientific theory in which a deity causes something to happen. Such a theory would not be admissible. Why? Because the introduction of divine causation breaks the schema in virtue of which scientific claims are tools of human adaption to the environment (you can use a conditional statement to help you successfully do stuff in the world; but you can't use raw data to help you do anything in the world...although you do need it to generate those helpful conditional statements). You cannot posit "if God comes down from the sky when the Israelites are on the run, then He will part the Red Sea", precisely because God doesn't necessarily act in a rule-like way--he acts according to His own divine will! And indeed, isn't it quite appropriate that a supreme deity who has total control over the universe and all of Creation would be ruled out a priori from a system of generating knowledge claims that purports to be an extension of humankind's dominion over its environment? How could God come under the dominion of humankind?)

The threat of nonlinear history

I've often wondered, in a lazy sort of way, in what way or to what extent political science is supposed to be a science. From doing some cursory looking around on the internet--and from casting my memory back to that one undergraduate PoliSci course that I took--it seems like wide swaths of political science look like your standard social science, doing things like finding correlations between levels of wealth and frequency of political revolutions and the like. But then there are the other parts of political science that get more play, the ones that boldly impose models on politics and history and try to make long-run predictions. So, for instance, you have a debate like Huntington's "Clash of Civilizations" (the world will fracture along cultural lines) versus Fukuyama's "Last Man" (liberal capitialism will win History), where two different theories offer competing outlooks for the next several decades (or more), and these are the kinds of views that inform the decision-making of the world's foremost power-wielders.

Now, I think it's fair to say that a whole lot of the heavy lifting that these theories do is done by the particular understanding of history and causation in history that they are founded on. For example, Fukuyama's contention that liberal capitalism is the final historical phase of humanity relies on the notion that history is a procession of battles between ideologies, and with this conception comes all of the theory-concepts (e.g., "ideology", "an ideology triumphing and continuing through history") and models he needs to identify historical patterns and then extrapolate those patterns into the future, forming (something like) a prediction (the "ideology" liberal capitalism has "triumphed" over Fascism and Communism, and so it "is the big winner of History"). And Huntington's idea about cultural conflict relies on notions about how human beings behave on a large scale, about how their cultural identity governs this behavior. And so Huntington will have a different set of theory-concepts going to work for him, not looking at "ideologies" so much as "civilizations" and their various "features", and perhaps having a different idea about what constitutes a "conflict" between them. But the point of all this is to show that though these theories impose different models on the world that make use of different sets of concepts, and that these concepts causally interact with each other in the model in different ways, and thus offer not only different predictions of the future but different sets of conceptual levers of history for power-wielders to pull, it is still the case that both theories have concepts, that they both have some causal scheme for how these concepts interact, and that they both make long-term predictions of some kind.

The thing I want to explore is: what kinds of assumptions are made by these theories such that they can have these formal features about them (theory-concepts or models, causal scheme, long-term predictions)? To me, it seems like they necessarily view history as something like a linear system that can be lassoed with broadly applicable concepts, that these concepts interact with each other in a regular mechanistic way, and that this mechanistic motion of broadly applicable concepts is what provides the basis for extrapolations of the system's behavior into the future (i.e., long-term predictions). When I say a "linear system", I mean a system that has this property: if you start the system in nearly the same state this time as you did last time, then the results will be nearly the same this time as they were last time (or at least, the difference in results between the two runs will be predictable/regular and proportional in some way to the difference in the starting states). A system's being linear is critical for the formation of broadly applicable concepts, because of the pragmatic limitations of being able to precisely repeat an experiment over and over again. For example, take Newtonian physics. At some point someone wanted to be able to predict how far a ball would travel given that it was thrown with a certain force at a certain angle. Of course, we know now that there is an elegant little algebraic formula that will yield the right answer. But for that first person doing the investigation, he had to start off by taking a ball and propelling with some force at some angle, and measuring how far it travelled. And he ran this same experiment over and over, and recorded all the measurements. Of course, the initial conditions for all of the different runs were not precisely the same. With each throw, there were arbitrary small variances in amount of force applied and the precise angle thrown brought about by shaky hands or uneven springs or whatever. But because of the linearity of the system, this didn't matter too much--the differences in starting state were small, and so the differences in end states was also (commensurately) small, and these small differences could be cleverly disappeared, either by deliberately reducing the granularity of the measurements so that the differences could not be detected ("significant figures"), or by leveraging the randomness of the differences by averaging the results together, thus cancelling out the "noise" and converging on the "true answer". What this means is that, at the end of the day, our experimenter has a long list of what can for all intents and purposes be considered the same exact test run over and over again. And he can take the regularities found here, and see if they hold with other objects--and if they still hold, he can try tweaking different variables in the initial state (halve the force, double the angle, etc.) and see if the results are altered in a predictable and proportional way. The broader the set of scenarios that he finds his regularities in, the more general and broadly applicable will be the concepts that eventually populate his finished theory. And in this case of Newtonian physics, the concepts are very general and broadly applicable indeed: he gets to say that "an object" weighing W will travel X distance when propelled with Y force at Z angle. But of course, none of this would be possible if Newtonian physics of this type were nonlinear--that is, if even imperceptible differences in starting conditions caused erratic and unpredictable differences in the result. If this were the case, then on the first test run, the ball might go 5 feet--and on the next run, might fly 18 feet, all due to a very slight difference in the number of air molecules that happen to strike the ball as it accelerated through the air. And then on the next test the ball might fly a measly .2 inches, because on the last run a few molecules were shaved off the ball when it hit the ground, and so the weight variable changed ever so slightly. Under these conditions, there are no tricks the experimenter could deploy to cancel out the "noise" of the results--because of the imperfect fidelity of each test to the previous one and the system's nonlinearity, his results would amount to nothing more than a list of random numbers (reflecting the randomness of the variances in the initial conditions of each test), and he could no more average the results than one can average the results of a television screen full of static and get an image. With no regularities to extract from his results, he cannot broaden his theory by generalizing it to other scenarios, and so cannot form an appropriately general theory with appropriately general theory-concepts, like "an object" and "X distance". Rather than a very useful and interesting generally applicable statement of a causal relation that can be used in the future to make predictions in all sorts of different situations involving objects being propelled through space, the experimenter has a very not-useful and uninteresting dead historical record that describes some meaningless events (ten separate occasions of some ball being thrown in some guy's backyard) that will have no bearing or impact on anything that comes after it. He will have not a theory, but a collection of arbitrary data. And of course, with no broadly applicable and causally interactive concepts with which to formulate a theory, there can be no mechanistic fast-forwarding of a model to make predictions about the future (indeed, if Newtonian physics were nonlinear, the world would be craaaaazy).

So you can see where I'm going with this. Could it be that the engine of history--the vast system of human beings that interact with each other, each with its own idiosyncratic motives and behaviors and tendencies--is essentially nonlinear? Could it be that if you were to somehow model all of humanity on a computer and run it in fast forward, that if you went and removed, say, some insignificant English peasant in 1132 AD, then the result in 2008 would be vastly different, Sliders-esque world with totally different leaders, countries, and dominant ideologies? Or we can go even smaller: if a butterfly gives its wings an extra flap in 540 BC somewhere in Greece, does the same kind of wild and erratic variation in world-makeup occur? You may think: well what does a butterfly have to do with human affairs? But as we all know by now, the weather is famously sensitive to changes of even this scale. If the butterfly did this, it would unleash a whole alternate history of weather patterns, weather patterns that would effect human society in countless different ways over the centuries.

How can we tell if human history is, indeed, massively nonlinear? This I do not know. Maybe the surest way we would ever be able to tell is by somehow coming up with a computer model of humanity that everyone could agree really does model humanity accurately. Then we could simply experiment by running the model (a computer model has the benefit that it can be run in precisely the same way as many times as you like. If the model is deterministic, then even a nonlinear system will spit out the same result over and over). But it seems to me that you could also make a fairly educated guess at humanity's nonlinearity by applying some armchair common sense. Just by reflecting on your own life, you can easily see how incredibly and delicately contingent are so many basic features of your life. For example, when my mother was a little girl and her family was moving out West, they were originally planning on settling down in Seattle. However, things changed when they decided that they would take a detour to Los Angeles to take the kids to Disneyland--they ended up taking a liking to the then-undeveloped San Fernando Valley, and lived there ever since. So if it wasn't for the decision to visit Disneyland--and if it wasn't for Disneyland being located in LA--and if it wasn't for Walt Disney having the idea for Disneyland--and if it wasn't for Walt Disney's mother missing a train and having to share a taxi with a charming fellow by the name of John Disney--and on and on and on, branching through history in a million parallel branches going as far back as you want to go--if it wasn't for all those meaningless contingencies, I wouldn't have been born. And I think when you take this and multiply it by every little event and fact and all the people in all of history, and factor in on top of this other nonlinear systems like the weather that affect everyone all throughout the world through history, I think there can only be one sobering conclusion: that humanity, that history, is massively, terrifyingly, unendingly, unsolvably nonlinear.

But hold on--all is not lost. Because some theory-concepts just so happen to be durable in the face of nonlinearity. For example, it may be impossible to predict a year from now exactly which days will be rainy in May in San Francisco do to the nonlinearity of weather, but it is possible to make a prediction like, next year there will be more rainy days in May than in September. And certainly we can safely predict that it will be colder here in the summer than in the fall. So it is not as if nonlinearity automatically equals total unpredictabliity. There could be some local parts of the system that show regularities, perhaps because some outside thing is significantly constraining the possible outputs of the nonlinear system (in our case, it would be the shape of the land and the affects of ocean temperature on barometric pressure and the fog and whatnot). And so the question for political scientists is: are the concepts that populate your theory durable in the face of nonlinear history? Is a "winning ideology" subject to the frail infinitude of contingencies, like my birth? Or will that "winning ideology" stubbornly remain no matter what the contingencies are, like cold summers in San Francicso?

Perhaps any theory in political science is, ultimately, a statement about what holds steady in the face of the mass contingencies of history. Still, though, one can worry about the potential damage nonlinear history can wreak on our finely tuned theories, and wonder if there might be some way of verifying that we are indeed engaging in some kind of prediction-yielding science, and not an obscure form of historical interpretation that comments on past events without shedding light on future ones.