Test 1
Regularities
- Every A has property B. - Whenever X happens Y also happens.
Necessities
- Something about the universe necessitates that every A has property B. - Something about the universe necessitates that whenever X happens Y also happens
Solutions to underdetermination of theory by data
1) The "No True Scotsman" ploy (a red herring) 2) Papineau's Inductive Justification of Induction 3) Probabilism 4) Carnap's Inductive Logic
Popperian science
1) We propose hypotheses (how we get them doesn't matter) 2) We test those hypotheses 3) Those falsified by observation or experiment are rejected 4) Those not yet falsified are corroborated
Falsificationism
A deductive approach to science that focuses on falsifying hypotheses as the key criterion of science Falsificationism is logically valid (unlike induction), so we are rational to use it Falsificationism avoids invalidity only if corroborating a theory is not the same as justifying it (i.e. showing it to be true or probably true).
The causal mechanical approach to explanation
A explains B only if A is a cause of B Salmon's solution to the problem of explanatory asymmetry was to base a new account of explanation on causal connection According to this account, the length of the shadow doesn't explain the height of the flagpole because it doesn't cause it But all events have innumerable causes. Which ones ought to be included in scientific explanations? Bas van Fraasen: the answers to such questions are entirely contextual
Lewis's Counterfactual account
A is a cause of B just if B would not have happened had A not happened This also helps us decide which causes ought to feature in particular explanations - Bas van Fraasen says we can pick out important causes once we recognise that explanations are questions of the form "Why X rather than Y?" - We are asking: "What are the factors without which we would end up in circumstance Y rather than circumstance X?" Counterfactual analysis - Counterfactual account suffers from difficult problem cases: - Smith is killed by a firing squad. Two of the members of the squad (call them A and B) actually hit him and both wounded him mortally. What caused his death on account of counterfactual claim?
Laws of Nature
A law of nature describes some regularity which holds at all times and in all places Can not be deduced from empirical evidence, can't have evidence about what is happening at every time and place Even if it were possible to inspect the entire universe we could not distinguish laws of nature from universal coincidences
Logical positivism
A philosophy that sees meaning in only those beliefs that can be empirically proven, and that therefore rejects most of the concerns of traditional philosophy (the existence of God / the meaning of happiness) as nonsense
No True Scotsman Fallacy
A way of reinterpreting evidence in order to prevent the refutation of one's position (If he's not mean, he's not a Scot) Ornithologists can not keep theory that "all swans are white" by declaring that Southern Hemisphere swans are not true swans? Science supposedly settles questions that are: 1. Empirical: could be justified by observation Contingent: not true in all possible worlds Falsifiable: one that could be shown to be false But if we make Scots mean or swans white by definition, apparently empirical questions become Necessarily true: true in all possible worlds Analytic: true by definition This makes science a deductive discipline that tells us nothing about the way the world is
Naive Inductivism
Analysis of scientific method was proposed in the 17th century: Francis Bacon (1) Collection of data from observation and experiment (2) Inductive inference to a general hypothesis (3) Further verification of the hypothesis by observation and experiment, eventually turning hypothesis into established law
Problems with Inductivism
Bacon's inductivism says ... (1) Collection of data from observation and experiment (2) Inductive inference* to a general hypothesis. (3) Further verification by observation and experiment, turning the hypothesis into an established law In assessing theories of science, we consider both plausibility and accuracy Scientists really infer from data to theory via induction?
Making Observations Less Fallible
But how is the falsificationist to understand the logic of checking and rechecking results? • How much checking should the falsificationist recommend that we do? If we say after a certain amount of checking that some experiment will always yield some result (that the result is a 'reproducible effect'), aren't we using induction? In short, while science is a distinct epistemological enterprise, it is not immune to the problems that are central to general epistemology
D-N explanations & probabilistic conclusions
Can D-N explanations have probabilistic conclusions? A implies that there is a 0.7 probability of B A is true ------------------------------ There is a 0.7 probability of B But what about... Smokers have 30% probability of cancer Fred is a smoker ---------------------------------------- Fred has cancer
Ampliative argument
Conclusion gives information not contained in the premises
Hypothetico-Deductive Method
Counteracts biased data collection. Proposing hypotheses and then looking for evidence that might prove them false. Bacon: "The human understanding when it has once adopted an opinion ... draws all things else to support and agree with it. And though there be a greater number and weight of instances to be found on the other side, yet these it either neglects and despises ... in order that by this great and pernicious predetermination the authority of its former conclusions may remain inviolate." If we propose a theory and use that as a guide for our collection of evidence, there is a danger that we will just collect evidence that supports our theory.
Induction and irrationality (Hume)
David Hume in Problem of Induction in A Treatise of Human Nature (1740): We must all reason inductively Inductive reasoning is invalid To reason invalidly is irrational ---------------------------- We must all be irrational
Causation
David Lewis argues that laws of nature (as appear in D-N explanations) need not refer to causation
Russell's inductivist Turkey
Day 1 the farmer brought food Day 2 the farmer brought food ... Yesterday n the farmer brought food -------------------------------- Today the farmer will bring food Problem: Inductive reasoning is ampliative, so logically invalid All ampliative arguments are defeasible
Ways humans learn about world
Deduction, intuition, guesswork, consulting authorities
Deducing universal statements
Deductive arguments rarely go from singular statements to universal ones I've got two cats One is black So is the other --------------- All my cats are black
Salmon's counter-example to I-S Explanation
Doctor puts you on a drug which has a known side effect In a small number of cases it gives rise to psychotic episodes Suppose you take the drug and have psychotic episodes This is not an inductive statistical explanation because nothing "makes highly probably" the fact that you have psychotic episodes One possible answer is that the prior probability of having a psychotic episode was tremendously low The drug explains the episodes because it has raised that very low probability
Solution to the Duhem-Quine Problem
Duhem said "scientific good sense" will tell us which hypothesis is at fault Popper — We may blame a falsification on some auxiliary assumption provided we do not do it in an ad hoc way
Occam's razor
Entities should not be multiplied beyond necessity "the simplest explanation is probably the right one" Does "grue" imply that the world contains extra entities / processes?
Hume's worry: Laws of Nature
Even if induction is good enough for ordinary purposes, it can't tell us about laws of nature
Statistical Relevance (S-R) Model of Explanation
Evidence that alters the probability of X. An explanans explains the explanandum if it is statistically relevant to that explanandum, i.e. the probability of the explanandum given the explanans differs from the prior probability of the explanandum
Inductive Statistical (I-S) model of Explanation
Evidence that makes X highly probable. The probability of i being an O, given that it is an R is high i is an R ------------------------------------- [mhp*] i is an O * MHP = Makes Highly Probable The probability of you getting cancer, given that you are a smoker is high You are a smoker ------------------------------------------[mhp] You have got cancer
Deductive nomological (D-N) model of Explanation
Explanation as a valid deductive argument - Valid deductive argument from the explanans to the explanandum - Explains phenomena by the action of laws Laws / Theories Initial Conditions ---------------- Observations Equal to (Explanans) ------------------ (Explanandum) In such explanations phenomena are said to be subsumed under "covering laws" The explanandum may be either a singular or a general statement
Explanation as Unification
Explanations reduce puzzlement by reducing the number of principles we need to explain the world. Phillip Kitcher: explanations work by unifying many disparate phenomena under a single, simple piece of theory Explanations dispel mystery by reducing the number of principles that we have to accept as "brute facts" - e.g. The theory of natural selection It also addresses relevance and symmetry problems as both inflate the number of "argument patterns" required - We don't need to explain the height of the flagpole using the length of the shadow as we already have social and economic explanations for that But what about "God explains everything"? - Kitcher avoids the problem by excluding any argument pattern that gives an explanation of absolutely anything
Inductivist science
Facts through observation ---(induction)---> Laws & theories ---(deduction)---> Predictions & explanations
Carnap's Inductive Logic
Facts through observation ---(induction)---> Predictions & explanations Probability is a logical relation between two types of statements: the hypothesis (conclusion) and the evidence (premises) Accordingly, a theory of induction should explain how, by pure logical analysis, we can ascertain that certain evidence establishes a degree of confirmation strong enough to confirm a given hypothesis
Quine on Natural Kinds
How do infants learn about the world if induction is so difficult? Natural selection causes infants to automatically attend to features which are reliable for induction in ordinary circumstances These abilities are refined as we learn but they are only ever detectors of the sort of things that creatures like us find useful Quine's characterisation of Natural Kinds isn't designed to solve the problem of induction. He's more interested in explaining ordinary language Natural selection is far from perfect at truth-tracking
Hume's worry: Natural Kinds
How do we know that there are any and why do we think science can detect them?
Modus tollens (valid)
If H then P Not P ----------- Not H
Affirming the consequent (invalid)
If H then P P ----------- H
The Principle of Induction
If a large number of As have been observed under a wide variety of conditions, and if all those observed As without exception possessed the property B, then all As have the property B. Naive inductivists believe that scientific knowledge is built by induction from the secure basis provided by observation As we get better at making observations, so we construct more and more laws and theories of ever increasing scope and so attain continuous progress in science
Problem with Naive Inductivism
If it is so good why are we calling it "naive"?
Confirmational holism
Individual hypotheses are not capable of being tested in isolation. Only groups of hypotheses are capable of being tested
Under-determination of theory by data
Induction as foundation of science has to bridge 3 gaps: 1) Generality gap 2) Precision gap 3) Observability gap
Hume on induction
Induction is a habit rather than a rational decision procedure So we can explain it psychologically but not logically
Nelson Goodman's New Riddle of Induction
Induction tells us that groups of entities have particular properties. Goodman doubts such inferences. Consider "All emeralds are green". Why does our evidence of emeralds not cause us to conclude that "All emeralds are grue*". * Grue = green if observed before 2050, otherwise blue. Goodman's point — There are some properties we would never think of inferring by induction. These properties are "non-projectible". Goodman's problem — What makes a property projectible? This is important because we expect science to detect projectible properties.
John Stuart Mill's "Kinds"
Induction works because sometimes we manage to hit on categories that correspond to real underlying Kinds in nature These real Kinds have many properties in common. Compare samples of Sulphur with the class of 'white things'... We can make an indefinite number of inferences about the uses to which sulphur might be put, the compounds it might form etc. But there are almost no useful inferences that you can perform about the class of white things. Does not solve the new riddle of induction By hypothesis, all the green things have the same number of properties in common as all the grue things
Regress of explanations
Instrumentalists agree that science doesn't provide explanations. For them, covering laws are useful ways to summarize data Realists argue that non-ultimate explanations are real explanations This is the Russian Doll Model of explanations - Popper argues that, since we cannot be sure that any laws are true, it makes sense to ask for an explanation of even our deepest laws Other realists argue that fundamental physical laws will provide ultimate explanations. - But what would such laws look like?
Boldness
Is falsifiability or testability sufficient as a demarcation criterion for science? Testability is a matter of degree Popper claims that the best/boldest theories have high empirical content The empirical content of a theory T is the class of propositions that would falsify T, if true But what about theories that are demonstrable a priori (without recourse to testing), e.g. Natural Selection? For Popper, such theories cannot be scientific, though "non-scientific" does not prevent such theories from being true
Appeals of Naive Inductivism
It gives an explanation, in terms of a methodology, of the success of science. • It explains the objectivity of science. • Having a formal theory of how science works allows scientists to evaluate their own performance and to justify their conclusions to others.
Duhem-Quine Problem
Just as tests employ instruments, they also often use theories. Hence we cannot test scientific theories in isolation from other "auxilliary hypotheses" Duhem— Auxilliary theories are "a group of theories accepted by the experimenter as beyond dispute" So falsificationism really looks like: (Theory + Auxiliary Hypotheses) implies P Not P -------------------------------------- Not (Theory + Auxiliary Hypotheses) Even if conclusive disproof is possible, which theory will we take to have been conclusive disproved?
Papineau's Inductive Justification of Induction
Justifying induction via induction appears circular but Papineau claims circularity isn't always a bad thing When person 1 induced from n observations of A going with B, that all As are Bs, this conclusion was true When person 2 induced from n observations of C going with D, that all Cs are Ds, this conclusion was true ... When person n induced from n observations of L going with M, that all Ls are Ms, this conclusion was true -----------------------------------------------Whenever someone induces, their conclusion is true This argument is "rule circular" but not "premise circular" Papineau asks, if in fact induction works, why shouldn't we be able to use it in arguments like his?
Scientific Laws
Laws make universal claims about large and open-ended numbers of cases 1. Regularities 2. Necessities "Metals expand when heated" How can you infer universal statements from singular statements?
Natural Kinds
Many philosophers have thought that there are some real "joints in nature" (Plato's idea) If so, then perhaps there are some real groups in nature and it is not arbitrary that we think of some properties as projectible and others are not So, while it might not in general be rational to perform inductions, it might be rational to perform them for metals or halides, or species or diabetics. So how can we best detect natural kinds? This is a difficult claim to assess primarily because there are many ways in which philosophers have understood the term "natural kind"
Aristotle's taxonomy of causes
Material causes we think of as constituents Formal causes we think of as essences or perhaps defining properties Efficient causes are what today we mostly call causes Final causes are today thought of as purposes or functions
Scientific explanation
Much of science concerns explanations that can be framed as why-questions Why do some people get cancer, and others not? Why is it that that almost nothing has an odd number of eyes? Why is it that people's perception is influenced by their expectations?
Nelson Goodman entrenchment
Nelson Goodman thinks that over time some theories about which kinds are "projectible" become "entrenched in" science But for Goodman, those theories are fictions that we use to construct theories with which we predict phenomena Entrenchment does not mark such theories as true or accurate, merely reflects their continued usefulness
Logical asymmetry between verification and falsification
No number of observations can confirm the conclusion of an inductive argument
The Fallibility of Observation
Observations are often made using instruments, galvanometers, thermometers, mass spectrometers etc. In so doing we assume that: The instrument is reliable It has been properly constructed It has been properly calibrated It is being properly used It is being properly read If one or more of these assumptions is false we may falsify a true theory
Generality gap
Observations are singular, but theories are general
explanatory asymmetry
One can derive the length of the shadow cast by a flagpole from the height of the pole and the angle of the sun above the horizon and laws about the rectilinear propagation of light. This derivation meets the DN criteria and seems explanatory "The reason the shadow is x metres long is that the pole is y metres tall and the sun is at z" On the other hand, a derivation of height from shadow and angle and the same laws also meets the DN criteria but does not seem explanatory "The reason the pole is y metres tall is that the shadow is x metres long and the sun is at z."
Hume's worry: Rationality
People act as if nature was regular, but we can't know that it is, so people are irrational
problems for the DN model
People rarely set out explanations as complete deductive arguments Good scientific explanations could be "sketches" of such arguments Many genuine explanations are not Deductive Nomological explanations Others are: Narrative explanations in history Explanations of people's behaviour Explanations in evolutionary biology Not all deductions are good explanations (explanatory asymmetry)
Popper's Rejection of Induction
Popper argues that Hume is wrong in thinking that "we must all reason inductively" In the context of discovery, induction is OK because it might lead to novel hypotheses. Here we care only about getting many bold hypotheses for later testing. In the context of justification we need not use induction at all. Instead we can rely on the asymmetry between verification and falsification If science is purely about falsification then it can be purely deductive
Convergent realism
Popper: if verisimilitude tells us how to replace falsified theories, science will gradually get closer to the truth
Defeasibility
Property of something which can be annulled, invalidated, or defeated
How pseudosciences deflect criticism
Pseudo-sciences are designed so as to deflect criticism Make claims that have no observable consequences Lie about empirical findings Make up a technical vocabulary of jargon words that have no clear definitions Base claims on anecdotal evidence Make up hypotheses about why particular tests don't work - ghosts don't like scientists or the 'shyness effect' A common variant of this is the very powerful deceiver
Deduction
Reasoning from general to specific The premises of deductive arguments guarantee their conclusions All people are mortal James is a person -------------------- James is mortal Impossible for the premises of a deductively valid argument to be true and its conclusion false
Induction
Reasoning from specific to general All the As observed up to now have had property B ---------------------------------------------- All As have property B Draws universal conclusions from singular premises (explains Bacon's choice of induction in his theory of scientific method)
Common explanation deficiencies
Relevance, Brevity, Clarity, Circularity, Validity, Falsity, Mystery
Problem with naive falsificationism
Rests on the idea that hypotheses can be subject to conclusive disproofs Valid, but is it sound? Testing is analogous to viewing an iceberg. It appears we are testing something small just as the visible part of an iceberg is only small part of the whole
Wesley Salmon on Causation
Salmon argues for causation as energy transference - A causes B just if B is a result of energy transferred from A. But many causes are not (fully) understood in terms of energy transference So in practice, this account of causation often rests on statistical relevance However we can usually infer the appropriate direction of causation But what about this case... - "I meant to tell Heather and Charles about the meeting, but I forgot so they missed it. " - There is no energy transference here
Popperian Picture
Science begins with bold imaginative solutions to problems These must be clear, precise so as to be maximally testable If falsified they are rejected Theories are never shown to be true Science consists of "Conjectures and Refutations" Despite learning no truths, "conjectural knowledge grows"
Problem with deducing universal statements
Science has to make universal statements about unobserved cases so deduction can't, on its own, explain how scientific inference works.
Observability gap
Scientific theories often describe processes and entities that are unobservable, but scientific data always concerns what we can observe.
Probabilism
Scientists make probabilistic claims, can we solve Hume's problem by acknowledging that science does not provide us with certainty? Hume — We still require a principle of uniformity in nature to make induction valid and rational, such as... "unobserved cases will probably resemble observed cases" Probabilists would need to show that for a hypothesis (h) evidence (e) and probability (p): p (h, e) > 0.5 But the evidence is finite and our theories make claims about an infinite number of situations... so p (h, e) = 0
Problem with Laws of Nature
Support counterfactuals (The Problem of Induction) - the thought "If I had not eaten so many potato chips, I wouldn't feel ill right now" implies eating too many potato chips caused the person to feel sick
Example of Tichy Millar Objection
T = "People and porpoises cannot fly because they are both mammals and mammals are all descendants of a non-flying ancestral species." (partly true) T' = T + "Bats are mammals" (i.e. an extra truth) So we ought to be able to replace T with T' BUT T' also has a new false consequence namely that bats can't fly because they're mammals. SO T and T' do not differ in their verisimilitude after all (even though we designed them so that they would). Therefore, by Popper's definition of verisimilitude, the only time that a new theory will count as more truth-like that an old one is if the new theory has no false consequences at all.
Good deductive-nomological explanations
The argument must be: - Valid - Non-circular The premises of the argument must be: - Testable - True/sound Should a good scientific explanation be objective? Should a good scientific explanation relieve puzzlement?
Logical positivists: Two paths to scientific hypotheses
The context of discovery The context of justification
The context of justification
The inferential process by which we establish a hypothesis as part of accepted science
Precision gap
Theories are precise, but observations are imprecise
A universal coincidence
There is no solid body of pure gold that weighs more than 3000 tonnes vs There is no solid body of pure plutonium that weighs more than 3000 tonnes (Deducible from a law of nature (because such a body would explode)
The context of discovery
This is the creative process involving many forms of inference (deduction, induction, intuition, guesswork...) by which we invent hypotheses in the first place
Locke on Nominal and Real Essences
Two ways to think about what there is in the world: 1) Real essences: the fundamental properties that distinguish some types of matter from other kinds. 2. Nominal essences: the properties that we use to make sense of the world Locke thought that we would not be able to detect real essences/natural kinds because we lack "microscopical eyes" Now science seems to have detected real essences Problem: begs the question (Infers that the problem of induction is solved)
Tichy Millar Objection
When you add a truth to a theory with one pre-existing falsehood, the new theory thereby gains a new falsehood. T is a theory, T' is theory that is supposed to replace it Suppose that T' has at least one T' entails p false consequence, f. This will also be a false consequence that T had (otherwise T' couldn't be more truth- like than T) • It also makes at least one prediction, p, that T does not make (as a good successor theory is supposed to do) Then we have: T' entails p T' entails f Together these imply that T' entails (p and f) But (p and f) is a false consequence that T does not have, so, by Popper's definition, T' is not more truth-like than T
Hume's worry: Projectible predicates / properties
Why are we prepared to infer that emeralds are green but not that they are grue? Science is supposed to be able to tell us which properties are "projectible" (in the way that green is but grue is not)
Deducing Explanations
Why do whale skeletons appear to have hind limbs? Taxonomy tells us that whales are mammals All mammals are evolved from a common ancestor (Natural Selection) The earliest mammal fossils have four limbs used for locomotion --------------------------------------------- Whales have evolved from four legged creatures Prediction from same premises: Taxonomy tells us that whales are mammals All mammals are evolved from a common ancestor (Natural Selection) The earliest mammal fossils have four limbs used for locomotion ------------------------------------------------- If new species of Whale is discovered, it will either be four legged or it will have vestigial hind-limbs
Problems with Popperian Science
Why use hypotheses that have not been shown to be true or even probably true? So we are not irrational for using induction. Now we are irrational for using theories that we have no reason to think true Some think Popper is smuggling in induction. Corroboration must be a sort of justification (i.e. a reason for thinking hypotheses are true) Popperians think that we use hypotheses without having reason to believe them Alan Musgrave argues that we have reason to believe unfalsified hypotheses, but not reason to believe that they are true - This is a form of belief voluntarism. - But if Musgrave is right, what is this reason to believe in merely corroborated hypotheses?
Sound argument
a valid argument with true premises
Hume objection to mechanical philosophy
accounts of nature in terms of mechanistic causation were in fact impossible At best science can tell us about constant conjunction which doesn't demonstrate necessary connection between causes and their effects
ad hoc adjustment
addition to a theory in order to save it from being falsified. Often, ad hoc hypothesizing is employed to compensate for anomalies not anticipated by the theory in its unmodified form
Enthymemes
an argument in which a premise or conclusion is unstated
crucial experiment
an experiment that has the power to decide between two competing theories We invent two theories which can be tested using the same set of auxiliary hypotheses. We derive different predictions from our two theories T + A predicts P T* + A predicts not If P does not happen do we know that T is false? T - dogs have two legs T* - dogs have four legs A - I am a dog So T* is false and dogs have two legs. Oops!
Mechanical Philosophy
believed in by Kepler and Descartes; the entire world is a machine that follows mechanical laws (Descartes extended this to human bodies); through this he introduced the concept of conservation of motion
Epistemology
study what knowledge is and how we get it
Verisimilitude
the appearance of being true or real Suppose that T and T' are theories. Then T' is more truth-like than T if and only if one of two conditions holds: T' has all the true consequences of T and but has fewer false consequences than T. or T' has all the same true consequences as T does and adds new ones and has no extra false consequences.
Explanandum
the phenomenon to be explained
Explanans
the theory which does the explaining
Naive Falsificationism
the thought that the failure of the prediction conclusively falsifies the hypothesis in question rests on the idea that hypotheses can be subject to conclusive disproofs Valid, but is it sound?