Aristotle or Bacon? Two philosophers of science on scientific explanation. Klee uses Aristotle and Bacon to raise the difficulty of clearly defining the function and goal(s) of scientific explanation; this difficulty, moreover, is not resolved by some conception of "laws of nature." Rather, this conception is rejected by contemporary philosophers of science as "utopian," in part because of what Klee describes as its theological origin. These various failures to clearly define just what counts as scientific explanation then leads to a number of contemporary efforts to provide better accounts, beginning with...
6.1 The Deductive-Nomological (D-N) Model of Scientific Explanation. This model, proposed by Hempel and Oppenheim, seems to work best in physics - but it encounters at least two difficulties:
"Laws of Nature," as presumed in this model, are again highly problematic conceptions
The model can be easily trivialized, leading to powerful counterexamples which call it into question.
These difficulties lead to still other efforts to define scientific explanation:
6.2 The Aleatory Model of Scientific Explanation (Paul Humphreys)
6.3 The Unification Model of Scientific Explanation (Philip Kitcher). The problems with the previous two models leads Kitcher to take a more holistic approach. Kitcher argues that his approach is supported by earlier philosophers such as Kant and Charles Sanders Pierce - but this takes Kitcher in what for Klee is too much an "anti-realist" and/or social constructivist position.
[Ess: this is too either/or - Klee seems to assume that scientific theories can come in only two flavors, the realist and antirealist. This misses a middle ground of instrumentalism - one that does not force us into the extremes of social constructivism and antirealism which he sees as the only alternatives to his own version of realism.]
6.4 Natural Kinds, Homeostatic Property Clusters, and the Structural Basis of Reality (Richard Boyd). Klee prefers this approach because of its reliance on a (presumed) structural reality. Ironically, he must acknowledge the Aristotelian roots of this approach, despite his initial condemnation of Aristotle (!).
Aristotle or Bacon? Two philosophers of science on scientific explanation
-- The first to write with "systematic deliberateness" about scientific explanation was Aristotle; he said, "explanation in science unveils the essence or being (the ousia in Attic Greek) of the phenomenon explained"; explanation requires the correct "definition" of the phenomenon explained
--> problems:
1. entangled science too extensively with speculative metaphysics: scientist had to expose "essential being" of phenomenon, not just give account of "how" we experience it.
2. mixed explanation with conventions of language, i.e., categories and definitions of words
i.e., NK cells: to explain how or why NK cells kill cancer cells, it is not necessary to define "NK cell" or to expose the essence or being (ousia) of an NK cell
--Klee says this view was a great hindrance to the progress of science - left little room for experimentation - cut off listening to or responding to nature pushing back against our ideas because we were concerned with getting definitions and metaphysics right - concept of "a priori" mode of knowledge
[Ess's comment: Klee's judgment here appears to be limited in two ways. One, this appears to be a baldly anachronistic judgment - i.e., judging Aristotle's account of science in terms of 20th ct. criteria, rather than from a more historically-informed and contextualized standpoint. Two, this judgment fails precisely to recognize how Aristotle, the biologist (!) contributed precisely to a more empirically-oriented understanding of "science". Cf. especially Singer's defense of Aristotle on this point, as well as related comments in our initial outline.]
--Then, the view was that the structure of nature was a function of our familiar categories of thought; today, it's the other way around, and is called naturalized epistemology. Our familiar categories of thought are a function of the structure of nature.
-->Francis Bacon - first philosopher of science to begin dealing with and changing these ideas - he thought that the scientist was to observe and catalogue "trivial" differences in members of different classes of phenomena - main objection to Aristotelian view was that scientific questions are answered by observation or experimentation, not a priori methods - a "doctrinaire inductivist", but this led to the problem of lacking predictive power because it is difficult to predict anything about a system without general rules or ideas to work from.
--we know that Baconian methods cannot always be applied to science in reality because there is a certain structure that exists in our world, some things will never be true; Klee suggests these - Pine trees do not speak Latin, pure water does not solidify at 300 degrees Celsius, heavy exercise does not lower fever, and falling bodies do not accelerate proportional to their masses.
-->led to understanding these "regularities" as laws of nature - influenced by religious domination of culture during Renaissance; Klee says, "I say that this was unfortunate because the notion of a law of nature derives ultimately from purely theological doctrines about the natural world - doctrines that have no proper place in a rigorously empirical view of nature. A law requires a lawmaker, of course, and so it was easy to slip into the view that the laws of nature were conceived and ordained by a supreme deity." - made ideas acceptable to powerful church, but extremely inappropriate according to Klee (see discussion page 106)
--laws of nature popular today due to Carl Hempel, who along with Paul Oppenheim, published a paper presenting their model of scientific explanation, using laws of nature; in general now considered useless - "utopian"
6.1 The Deductive-Nomological (D-N) Model of Scientific Explanation
i.e., explanation of why mercury in thermometer drops briefly before rising when therm. placed in hot water
--H and O noticed use of laws of nature - i.e., laws of thermodynamics and coefficients of expansion - the event being explained (initial drop of mercury) is due to action of these laws of nature and some "contingent facts" about the thermometer
-they said the event is "subsumed" under these laws - "brought under" those laws in that it is a necessary consequence of the action of the laws, and that we have a "fairly precise" way of expressing the laws in a semantic form
-->Deductive-Nomological model basically says that one should be able to logically deduce a certain event-to-be-explained from a set of premises (laws of nature) and certain contingent facts (some starting conditions)
-H and O identified 4 "adequacy conditions" for this model to be effective
1. premise set of laws of nature and initial conditions must logically imply the description of the event to be explained. - not just that there is no contradiction between the premise set and the description of the event, but that the event is a "deductive consequence" of the premise set
2. premise set must contain at least one law of nature, it cannot be a set of statements, none of which deal identify an applicable regularity in the universe - rules out narrative explanations - places and times do not have predictive powers (according to Hempel) laws of nature do
3. statements of premise set must be empirically testable - independently of use in explanation, and NOT simply through a thought-experiment - goal was to rule out pseudo sciences like astrology and phrenology (idea that the shape of the skull determines mental abilities or capabilities - level of development), these pseudo sciences use nonempirical principles and hypotheses supported only by tradition and superstition
4. initially - all laws and initial-condition statements must be true laws and initial-condition statements, later changed to- must be well confirmed
-->example of mice:
a. two equal-sized groups of genetically identical mice
b. inject all of one group with drug to inactivate interferons (things that kill viruses), other group drug-free
**interferons' ability to kill viruses is lawlike feature of chemical structures
c. next, expose both groups to increasing concentrations of infectious pieces of viruses - start with small, non-infectious amounts up to concentration high enough to kill 50% of drug-free (normal) mice
d. we see that all of mice with drug to inactivate interferons (virus-killers) are dead at concentrations of pieces of viruses much less than the highest amount (see above part c)
--explanation: see structural outline pages 109-110; premise 1 is law of nature and premises 2-3 are initial-condition statements, 4-6 are deduced from these first three premises
--meets four conditions of adequacy
--another example on page 111 - physics ("the one domain of science in which the model works best")
-->appeal of D-N model: removes mystery of event and subjectivity of scientist
-->problems:
1. are there such things as laws of nature? what are they? how do we identify them?
2. model can be easily trivialized - easy to produce counterexamples
--remember Klee's view that the idea of laws of nature are based on theological ideas of divine plan - question arise, "why do philosophers of science continue to use the notion of laws of nature when the lawgiver has dropped out of the picture? What does one gain by adopting the language of lawlikeness without any sovereign in the background whose very existence would otherwise seem to be a precondition of lawlikeness?"
--Hempel says that laws of nature are the way they are to keep from attributing determinative powers to simple times and places
--Hume says laws of nature are "brute" - the most basic, cannot be simplified, even though they explain things more fundamental than themselves
--philosophers like their (laws of nature) "logical properties: their universal scope of application and their subjunctive import"
--universal scope: not system-specific, i.e., just for the planet Jupiter, but for planets in general or only for oxygen molecules within the city limits of Denver, CO
--subjunctive import: (positivist model) laws are universally quantified material conditionals
-->problem # 1: there are universally quantified material conditionals that are true but nevertheless not laws of nature, i.e., moon of earth has less than 1% oxygen in its atmosphere (see page 113), could be false under certain conditions
**a law of nature should not be false under any conditions
--Hempel again suggests staying away from using particular objects, times, places, or events (predicates in a law must be purely qualitative - no reference to these)
--many philosophers questioned the benefit of this idea - how is a meaning better than a mention of a particular object, etc.
-->problem #2: there may by laws of nature that are species specific (therefore they would violate the rule about mentioning specific times, places, etc.)
i.e., human pain maybe not physiologically same as canine pain
6.1.2 Standard Counterexamples to the D-N Model
-->Klee says:
...there is no explanation without determination. If we have an explanatory relation between a premise-set and a description of the event to be explained, then that can be only because something in the premise-set adequately captures the objective determinative connection between the events in question. The failure of the D-N model of scientific explanation has to do with the inability of mathematical logic alone to represent objective determinative connections; for all logical connectives represent are syntactic and semantic connections, not causal or supervenient ones. (114)
--all of the necessary conditions can be met, but there still be no explanatory relationship between the premise set and the description of the event being explained.
i.e., flagpole height
i.e., what you ate for dinner
6.2 The Aleatory Model of Scientific Explanation
--Paul Humphreys, basic idea: "an explanation cites a cause of the event being explained, and a cause is something that contributes invariantly to an alteration in the probability that its effect will occur."
--> key is "invariance" - phenomenon is invariant if, no matter what else you change in the situation, the phenomenon occurs nonetheless.
--counteractive causes - lower probability of effect
--contributive causes - raise probability of effect
--Humphreys says, "the reputed causal factor must modify the probability of the effect in the same way no matter how we alter all other circumstances consistently with the causal factor's occurring at all."
**exact value of the probability not important, only whether is increase or decrease
i.e., presence of NK cells and tumor remission
-->problem: requires that probability of effect is invariant across all possible circumstances ; factor can only be said to be a cause of a certain effect, if every possible condition has been explored - this is way too idealistic - can only discover "partial causes" and therefore can only ever have partial explanations
--this goes against the "common sense truth" that mature sciences do explain many events in their domains of inquiry, and do so relatively completely
--> more ontologically honest than the D-N model - Humphreys seeks a structural account of why and how causes modify probabilities - goes back to microstructure and microstructural changes - this is the real determinative force, however, Humphreys does not believe these microstructural changes to be beneficial or overly relevant; "real action" takes place at microscale level, but this model does not address that action
6.3 The Unification Model of Scientific Explanation
--Philip Kitcher believes that a main problem with the first two models, is that they evaluate scientific explanations one at a time. It is better to evaluate explanations as compared to one another.
--i.e., take all the beliefs of a community of scientists at one time and call them set K. Assume that they can be represented linguistically as one large set of statements accepted as true by a community of scientists. If we believe that metaphysically, the point of science is to remove the "mystery" from science and to minimize the number of "brute" (remember Hume?) states of affairs and to maximize the number of states of affairs derived from these brute (more basic) concepts, then all that matters is which beliefs are brute and which are derived from these more basic beliefs.
i.e., out of 100 "true" beliefs, perhaps 3 are brute and the other 97 are derived using logical inference patters , or 7 and 93 using 6 logical inference patterns
--> Kitcher would say that the first way (3 and 97) is the better way because there are fewer basic beliefs and more logically derived beliefs, and fewer logical inference patterns were used.
According to Kitcher, "To explain is to unify diverse phenomena by deriving their descriptions using the same small number of argument patterns over and over."
--similarities between Kitcher and D-N model:
1. both assume explanation to be a logical argument of some kind, they require logical derivation of some description of the event.
--Kitcher - "deductive chauvinism" - everything is deductive - no induction
2. both reject direct appeal to causation (opposite of aleatory model)
-->Basic idea of model:
"A derivation of t he description of an event is an explanation of that event provided that the pattern of argument represented by that derivation is a member of what Kitcher dubs the explanatory store of the community of practitioners in question. the explanatory store is that collection of argument patterns (not arguments themselves) that best unifies the body of beliefs currently held by the community of practitioners in question." - A set of argument patterns P is best unifying under 3 conditions:
1. there is no other set of argument patterns that implies the same set of beliefs K with fewer patterns
2. there is no other set of argument patterns the same size or with fewer patterns that implies a set of beliefs larger than K
3. if there is a set of argument patterns with the same number of patterns as set P, that implies a set of beliefs the same size as K, then those argument patterns are less stringent than set P.
--these conditions are rather confusing and difficult to really understand, especially to someone not trained in formal logic (i.e., the idea of stringency)
--this model is inherently quantitative because of all of the counting of beliefs, brute and derived, and argument patterns. This is why Kitcher uses logic, especially condition #3, so that logic symbols can be counted.
**there are some technical problems with this quantitative aspect of this model of explanation
-->specificity of antibodies to antigens - two theories to explain:
1. Antigen-template theory: the antigen ("bad guy" that gets bound-to) acts as a mold (Klee suggests like a cookie cutter), and the antibody ("good guy" that does the binding) conforms its shape to that of the antigen.
-this accounted for so much specificity, but did not explain why antibodies persist even when the antigen (mold or cookie cutter) is not there. Nor did it explain why a second, small dose of antigen ("bad guy") produced a huge antibody response (this is called anamnestic boosting). The final problem was that this molding idea did not fit with anything else biologically in any other system of the body
-some ideas suggested to account for these problems: enzyme modification, tertiary folding of preformed amino acid chains, indirect templates - none succeeded
**new evidence showed that repeated exposure to the same antigen produced qualitative changes in antibodies, not just quantitative (remember, this model is quantitatively oriented) - new antibodies might be more or less specific (likely to bind) to antigens
2. Clonal selection theory: (now the reigning model) this theory says that there exist immune cells that are each genetically programmed for the production of a certain antibody if necessary. The antigen ("bad guy") works to select which cells will produce antibodies that can bind to the antigens. Once these cells are selected, they start producing lots of copies of their own antibody - which can now bind to the antigen
-this explains why there are still antibodies even when the antigen is gone - the selected cells stay around for awhile and still produce some antibodies
-the anamnestic response happens because after an initial exposure to an antigen, the antibodies are floating around in a "ready" state, but not necessarily producing antibodies. When even a small dose of antigens are introduced to the blood again, these "ready" antibodies go crazy.
-qualitative changes in selection can be explained by the selection of pre-existing cells whose antibodies are more or less specific.
-->antigen-template theory is less unifying because the clonal selection theory (derived belief) is just a micro version of natural selection (a brute belief) - see bottom half of page 121
**critical question: Why should we prefer explanations that unify to those that do not unify? Why is it preferable to explain lots with little?
-if the universe is "messy" (as Kitcher says it is) and not ordered or explained by a few brute beliefs, then trying to unify everything to these few brute beliefs could lead us to inaccurate assumptions or explanations.
-To defend this, Kitcher assumes that the underlying causal order of events is dependent on what our theories say it is - he doesn't believe that there could exist a cause of a phenomenon that is not part of the explanatory store of the practitioners of the time.
--points to Immanuel Kant and Charles Sanders Peirce as supporters, especially Peirce - the notion that man's knowledge will progress in the future, but will never reach perfection, like a limit, or an asymptotic graph.
**Therefore, an explanation cannot consist of anything that is not represented by current scientific knowledge or understanding - the current scientific theories - this is dangerously close to antirealism, which says that facts are dependent of our being around to discover them - "The causal facts of the matter are 'constituted' by our practices of inquiry - they are not 'discovered' by our practices of inquiry." The realist would say that there are facts, such as NK cells killing their targets by punching holes in the membrane of the cells, that have existed ever since their presence in mammalian immune systems, and will probably exist from now on, unless they become extinct in the future, and these facts would have been true regardless of whether or not man developed the ability to observe and study them.
-->this started out looking good, but we have identified a serious problem. Klee says, "If one has to adopt the view that causal relations are socially constructed rather than discovered as they are in themselves in order to advocate the unification model of scientific explanation, then so much the worse for the unification model."
6.4 Natural Kinds, Homeostatic Property Clusters, and the Structural Basis of Reality
--We have just seen the
question of whether or not scientific facts are discovered as
they are in themselves or discovered as they are based on our
methods of inquiry and our discovering them.
[Ess: again, this is too either/or - the sorts of Kantian/Piercian
middle grounds of instrumentalism recognize both an external
reality independent of the observer and his/her culture, and
the role of culturally-relative categories, etc. in shaping one's
observation/understanding of that external reality.]
--It is usually thought that Aristotle worked out the first account of natural kinds - abstract essences whose defining attributes were eternally fixed, unchanging, and absolutely precise. Now, we generally don't believe that they must be eternally fixed, unchanging, or absolutely precise. They can be, according to Klee, "as sloppy as you'd like and still be useful in doing science." An example of a natural kind is an electron or an NK cell.
--Richard Boyd says that kinds in nature are not abstract essences or logical sets of words, but are homeostatic property clusters (h.p.c.) - clusters of properties that are homeostatically stable over a period of time - called a cluster and not a set because "cluster" allows for slight fluctuations in the properties whereas a "set" is clearly defined. Homeostatic signifies that changes in some property in the group is compensated for by other properties (i.e., homeostasis in the body) - Though individual properties may change or fluctuate, the overall effectiveness, or causal powers, remain basically constant over time. Klee suggests that the "underlying structures" in Humphreys' aleatory model that are responsible for variance in probabilities are h.p.c.'s.
-for example, NK cells may vary somewhat in their method of killing cells, but their overall effect is essentially the same; therefore, they may be a natural kind, but only as an h.p.c.
** in the case of antibodies as an h.p.c., I don't know that I agree because if one is dysfunctional, then the others cannot necessarily take up the slack, the person can have very serious health problems.
** if you consider the heart, and the cells in it an h.p.c., and a section quits working, you have what's called a heart attack!1 (technically "myocardial infarction" which means heart-muscle death)
-->Klee is a big fan of structure and its appeal - therefore, the D-N model and the unification model have a problem because they rely on linguistic activity of some kind, not the kind of structure Klee looks for. With this idea, then these models can only be true if the linguistic order of deriving beliefs from more basic beliefs and of reducing everything to laws of nature matches that of the true structural order of nature. Klee feels that they are not matched up, and therefore are weak models.
--Klee says, "such appeals [to structure] end explanation like nothing else ends explanation. Nothing grounds determination like structure. Once you know the structural determination, you know the determination. Once you know the determination, you know the explanation. Accordingly, I am inclined to the view that any model of scientific explanation that does not cash out the explanatory relation ultimately in terms of structural aspects of the phenomenon explained - be it h.p.c.'s or something else - is traipsing down the wrong path."