Presentation on theme: "Basic Scientific Reasoning 11/19/2004. Outline What is Science? 2-4-6 Problem H-D / D-N Method The Quine-Duhem Thesis: Christine Ladd-Franklin’s falsification."— Presentation transcript:
Basic Scientific Reasoning 11/19/2004
Outline What is Science? 2-4-6 Problem H-D / D-N Method The Quine-Duhem Thesis: Christine Ladd-Franklin’s falsification of Helmholtz’s theory of color perception.
What is Science? Astrology Psychoanalysis Physics Psychophysics Mathematics Phrenology Biology Psychology Scientology Economics Cognitive Science Chemistry Sociology History Cognitive Ethology
Defining Science Is science defined by its method of reasoning? –Its results? –The phenomena it studies? –An attitude towards evidence? –A certain methodology? –Scientists? –The power relationships in which it is embedded? –Its value to the larger culture? –Use of (grounding on) mathematics?
Why should I care? ‘Science says that…’ ‘The top scientists think that…’ States that have (or are) debated teaching creationism in elementary science education (2002-present): –Georgia –Kentucky –Kansas –Dover, Pennsylvania –Maryland –The President
A Demonstration So how does ‘real’ science work? 2-4-6 problem
Preliminary Lessons from 2-4- 6 1.Science is an activity of real people who work in a social environment 2.There is nothing ‘special’ about the methods of reasoning used in science 3.When scientists ‘converge’ on the answer is often determined by social factors.
Structures of Reasoning on display in 2-4-6 Verification - Affirming the consequent: IF P THEN Q Q THEREFORE, P Example: IF aliens killed JFK, THEN there would be questions regarding the ‘single-shooter’ theory. There are questions regarding the ‘single-shooter’ theory. THEREFORE, aliens killed JFK.
Verification IF the rule is ‘Any three numbers such that x < y < z’, THEN 2-4-6 will fit. 2-4-6 does fit THEREFORE, the rule is ‘Any three numbers such that x < y < z’ IF the rule is ‘Any three numbers (x, y, z > 0) such that y = x+2 and z = y+2’, THEN 2-4-6 will fit. 2-4-6 does fit. THEREFORE, the rule is ‘Any three numbers (x, y, z > 0) such that y = x+2 and z = y+2’
Verification’s Problems Can’t ensure the truth of the theory that entails the verified prediction Can’t distinguish between two theories that both entail the verified prediction
Structures of Reasoning on display in 2-4-6 Falsification - Modus Tollens: IF P THEN Q NOT-Q THEREFORE, NOT-P Example: IF aliens killed JFK, THEN Jack Ruby is an alien. Jack Ruby is NOT an alien. THEREFORE, aliens did NOT kill JFK.
Falsification Application to 2-4-6: IF the rule is ‘any three numbers (x,y,z) such that x > y < z’, THEN the sequence ‘4-2-8’ will fit. The sequence 4-2-8 does NOT fit. THEREFORE, the rule is NOT ‘any three numbers (x,y,z) such that x > y < z’ IF the rule is ‘any three symbols (x,y,z) such that x < y < z in their standard order’, THEN the sequence ‘a-b-c’ will fit. The sequence a-b-c does NOT fit. THEREFORE, the rule is NOT ‘any three symbols (x,y,z) such that x < y < z in their standard order’
Falsification’s Virtues & Vices Still can’t ensure the truth of the theory that entails the verified prediction CAN distinguish between two theories, so long as there is at least one prediction that is entailed by one & NOT by the other.
The DN Model of Explanation Explanations are sound arguments (valid w/ true premises) that entail the event to be explained. L 1, L 2, … L n F 1, F 2, … F n E Explanans Explanandum Laws (usually conditionals) Facts Phenomenon Explained
Example Explanations are sound arguments (valid w/ true premises) that entail the event to be explained. For all cases, if you stimulate the L cone, the subject experiences a sensation of red. My L cone is stimulated Therefore, I have a sensation of red. Explanans Explanandum
H-D Notice the relation to falsification’s H-D method: For all cases, if you stimulate the L cone, the subject will experience a sensation of red. I am not experience red Therefore, my cone is not stimulated Explanans Explanandum Notice also that all explanandi are potential predictions!
Real scientists don’t work that way.
Quine-Duhem Thesis Any hypothesis can be defended in light of any evidence. (if you’re willing to make “drastic enough adjustments elsewhere” in your system of beliefs.) IF Hypothesis is true, THEN Prediction. Prediction is NOT true THEREFORE, the hypothesis is NOT true VALID
Dogmatism Any proposition can be made consistent with any scientific theory, if one tries hard enough –Gender, Intelligence, Variability and Pre- and Post-Darwin Biology –Myths of Gender by Anne Fausto-Sterling
Here, the Devil lurks IF my Hypothesis is true, Prediction is NOT true THEREFORE, at least one of (1) – (N) is NOT true (1) AND My apparatus works, (2) AND My experiment is correctly designed, (3) AND My subject is of interest to others, … THEN, prediction. (N) AND ….
Delineating Phenomena w/in ‘Memory’ Damn that Microsoft! Have to do it online.
Raise your hand if you had: BAG DOG FAN GAS HAT KID LOG PAD SOD VEX WIN ZIP
Definitions of ‘Memory’ Maybe, my memory for IP addresses isn’t really ‘memory’? http://inquiry.wustl.edu/newFrames/modules.p hp?mod_id=437 For all normal humans, we only have the ability to remember 7 digits +/- 2 I’m a normal human Therefore, I can only remember 7 digits +/- 2
Philosophers: (Plato / Aristotle?), Bergson, Russell, Ryle all distinguish between knowing how and knowing that. This distinction is transposed into memory: –I remember how to do arithmetic. –I remember that 2+2=4. –I remember how to get to Sears –I remember that Sears is on the corner of …
Further Distinctions: I just remembered that I am supposed to be in class! Remember when Janet Jackson had her ‘wardrobe malfunction’? Remember where you were when the Towers collapsed? Pavlov’s dog. Priming
Motor Skills Memory LTMSTM Non-Declarative (Implicit) Declarative (Explicit) Episodic (Events) Semantic (Facts) PrimingClassical Conditioning 1. Splitting Dissociable Kinds of Memory 1. Splitting Dissociable Kinds of Memory
H.M. from the Perspective of Cognitive Neuropsychology Develop Taxonomies of Memory Characterize Different Types of Memory Understand Neuro- cognitive Mechanisms Memory LTMSTM Non-Declarative (Implicit) Declarative (Explicit) Episodic (Events) Semantic (Facts) PrimingClassical Conditioning Motor Skills
The Hippocampus Identify set of tasks on which H.M. succeeds (T S ). Identify set of tasks on which he fails (T F ). Conjecture some cognitive faculty required for each task in T S and for no task in T F.
On which tasks does H.M. succeed? Perceptual and motor. I.Q. tests. Mirror Drawing. Gollins Partial Pictures. Priming. Classical and Operant Conditioning. Language.
On which tasks does H.M. fail? Conscious recognition of facts and events (Squire) Regardless of kind of test (free recall, cued recall, recognition) Regardless of material (e.g., words, digits, faces, mazes, life events) Regardless of sensory modality Conclusion: Loss of “Declarative Memory”
Characteristics of Observational Research Make some sort of record and analyse data obtained from it Does NOT manipulate or ‘intervene’ in the scenario.
Complexity of Observation Expectations & Perception –Anomalous playing cards –Multi-modal feedback (Data from lyric study) –Underdetermined Perception Influence of early hypotheses –Ratman study data Extending Perception w/ Instruments –Galileo / Hale-Bopp
Anomalous Playing Cards Link
Naturalistic Observation Observations made in the ‘natural’ setting of the organism. Researchers must immerse themselves in the setting. Task: to describe the setting, events, individuals observed w/out influencing the situation Often Qualitative, not Quantitative Often NOT a matter of testing a hypothesis, but rather gathering data to develop a testable hypothesis.
Data Field notes, journal entries, interviews, recording ‘artifacts’
Famous Naturalistic Observations Jane Goodall Charles Darwin Survivor?
Goodall ‘Termite fishing’ and tool-use Click Here
Darwin Sexual dimorphism is caused by three possible mechanisms: 1.mechanisms of sexual selection, 2.fecundity selection 3.ecological causation, e.g., resource- partitioning
Naturalistic Study 1 Marmots Cows Horses Hare Squirrel Deer
Problems for Systematic Observation Equipment (Nielson ratings) Reactivity of subjects Reliability (Mate selection in Blue Tit) Sampling Confirmation Bias
Systematic Observation Careful, often quantitative, observations of one or more specific behaviors. Observations made in ‘quasi-natural’ setting Researchers often do NOT immerse themselves in the setting Quantitative, not qualitative (using coding systems) Often a matter of testing a hypothesis
Wash U Fa06 (n=3 / 1m / 2f)
UCSD Wi05 (N=84 / 54f / 30m)
McDaniel Fa04 (n=4 / 3m / 1f)
Wash U Sp04 (n=11 / 4f / 7m)
UCSD Fa03 (n=10 / 5f / 5m)
Wash U Fa05 (n=3 / 1m / 2f)
UCSD Wi05 (N=84 / 54f / 30m)
McDaniel Fa04 (n=4 / 3m / 1f)
Wash U Sp04 (n=11 / 4f / 7m)
UCSD Fa03 (n=10 / 5f / 5m)
Coding Systems Marmots 2
Sampling Continuous Time Sampling Event Sampling
Dissociation and the Taxonomy of Memory A major theme in current studies of both humans and experimental animals is that memory is not a single entity but is composed of separate systems (Weiskrantz, 1990; Squire, 1992; Schacter and Tulving, 1994). The dissociation between declarative (explicit) and nondeclarative (implicit) memory is based on studies of experimental animals as well as amnesic patients and normal subjects showing that fact- and-event memory is distinct from other kinds of memory (skills, habits, and priming). –Squire and Knowlton
Memory Dissociation Argument in Philosophy Some X is a ‘natural kind’ iff in no possible world do parts of X exist without the whole and X still exists in that world (I.e. H2 w/out the O is not water). Hume’s claim: Two events are causally connected if and only if one might occur without the other. Why? –Because nothing can cause itself. –And if two events are necessarily conjoined, they are the same event (with the exception of two events that are both caused by a third event like ‘Socrates dying’ and ‘Xantippe becoming a widow’).
Biases in the characterization of ambiguous phenomena
American 19 th century ‘Polygeny’ The hypothesis: the ranking of races according to intelligence can be established objectively by a physical measurement, namely brain size Samuel George Morton 1830s – 1850s studied the cranial capacity of a library of skulls categorized by race.
Morton: Data? –Obtained a collection of over 600 skulls, mostly of native Americans & published a study Crania Americana in 1839 –Then obtained a collection of mummy skulls from Egypt and published Crania Aegyptiaca in 1844. Measurement device? –= mustard seed, that is, until it started producing unfavorable results, then switch to BBs (1/8 inch diameter steel ball).
Crania America Categories Malay American Ethiopian Caucasian Mongolian
The Data 65947829Ethiopian 6010082144American 64898118Malay 69938310Mongolian 751098752Caucasian SmallestLargestMeanNRace
1 st : Crania America Morton: mean 82 inches. –Morton divided the ‘American’ skulls into ‘Toltecans’ and ‘Barbarous tribes’. 82 inches is the average of the ‘Barbarous tribes’. The real average is 80.2 BUT, Morton’s failed to distinguish other groups = such as the Incan Peruvians who have an mean of 74.36, BUT make up 25% of the sample. Iroquois, on the other hand, contribute only 3 skulls that have a mean near 87. Gould corrected the biases and came up with an mean of 83.79
2 nd : Over-count Caucasians The 17 ‘Hindu’ skulls, whose mean is 75, were eliminated from the Caucasian sample BUT 3, whose mean was near 87 were admitted. Why? Once these are restored, and the samples weighted, the Caucasian mean is 84.45 (And Eskimos, if pulled out from the ‘Mongol’ group, get a mean of 86.8)
Creeping Bais 1: Categorization The skulls were from Mummies – so on what grounds is he categorizing & sub- categorizing race? “Negroid” is someone he believed was black, but had some ‘caucasian’ blood. His subdivision of the Caucasian race is based on, guess what? The bulbous- ness of the forehead. The mean of the entire group is 82.15
Creeping Bias 2: Gender Male heads tend to be bigger than female heads (because male bodies tend to be bigger than female bodies). Since this data is based on mummified remains, we can adjust for gender. 87.5 (2)75.5 (4)Negroid 73 (1)Negro 86.5 (24)77.2 (22)Caucasian MaleFemaleRace
Incidentally… There is a great variation in the body size of native Americans. If we rank Morton’s Crania America categories according to typical body size (Seminole largest, Peruvians smallest), we match his cranial capacity ranking exactly.
Creeping Bias 3: Subconscious mis- measurement Morton published his entire data tables, including a couple of the tables measured with both seed AND lead shot. The averages were adjusted thus: –111 Indian skulls: +2.2 inches –19 Caucasians: +1.8 inches –18 Africans: +5.4 inches –The measurement tool most likely to exemplify a priori bias did.
Here are the full categories: Modern Caucasian Group –Teutonic Family Germans English Anglo-Americans –Pelasgic Family –Celtic Family –Indostanic Family –Semitic Family –Nilotic Family –Malay Group –Malayan Family –Polynesian Family Mongolian Group – Chinese Family Ancient Caucasian Group – Pelasgic Family – Nilotic Family Negro Group – Native African Family – American-born Negros – Hottentot Family – Australians American Group – Toltecan Family Peruvians Mexicans – Barbarous Tribes
And here’s his data: Excel File
Summary: Biases creep: 1.Shifting categories 2.Ambiguous measurements will reveal prejudices (artifacts of instruments) 3.Failure to consider alternative hypotheses (I.e. body size / gender) 4.Miscalculations (confirmation bais)
(Brief) History of Color Science Basic Schema:
Hermann von Helmholtz (1821-1894) Short = Purple Middle = GreenLong = Red
Historical Note: In 1877, Ladd-Franklin became the first woman to attend (albeit unofficially) Johns Hopkins where she studied mathematics. She wrote a dissertation under the supervision of C.S. Pierce. It was published in 1883, but her Ph.D. was not awarded until 1926! Even though she had studied under Helmholtz and had published a great deal in psychological journals, she was never admitted to the American Psychological Association meetings to present her papers. While she lectured at John Hopkins, Columbia, Clark, Harvard and Chicago, she never held an official academic post, and she was rarely paid. Her book Color and Color Theories was finally published in 1929, one year before her death.
Ladd-Franklin (1847-1930) IF stimulating the long-wavelength cone yields a red experience, and stimulating the middle-wavelength cone yields a green experience, THEN stimulating both the long and middle- wavelength cone would…. yield an experience of reddish-green
L-F’s argument (≈1892) IF stimulating the long-wavelength cone yields a red experience, and stimulating the middle-wavelength cone yields a green experience, THEN stimulating both the long and middle- wavelength cone would yield an experience of reddish-green Stimulating L and M yields an experience of yellow. THEREFORE, Helmholtz’s theory is NOT true Yellow does NOT look like reddish-green. THEREFORE, yellow is NOT reddish-green. Good Argument Right?
Why not? “Helmholtz deemed it illegitimate or at least untrustworthy to draw conclusions as to physiological processes from the direct psychological character of the sensations” -Von Kries
Q-D for Helmholtz IF stimulating the Long-wavelength cone yeilds a red experience, and stimulating the middle-wavelength cone yields a green experience, THEN stimulating both the Long and Middle- wavelength cone would yield an experience of reddish-green Stimulating L and M yields an experience of yellow. THEREFORE, Helmholtz’s theory is NOT true THEREFORE, Yellow is NOT reddish-green. Yellow does NOT look like reddish-green. BUT: One cannot draw conclusions about the physiology of color from this fact, so it does not follow that: yellow is not reddish-green or greenish-red.
Note: The Gestalt Psychologist David Katz made the phenomenology of color appearance the starting point for a theory of color (1908).
Vindication: In 1957, Dorothea Jameson & Dale Hurvich proposed what is now called the ‘opponent-processing’ model of color perception. According to it, colors are given by mixing 4 primary colors which are arranged in opponent pairs: RedGreen BlueYellow This is now the dominant theory in color science
Vindication Performed by Jameson and Hurvich in 1957. A test light is shown to a subject. If the light appears greenish, a red- appearing light is added until the test light no longer appears at all greenish.
Jameson and Hurvich Results
Cone Sensitivity Curves
Mathematical Transformation of Cone Sensitivity Functions We decorrelate the responses of the L, M and S cones by weighting each signal with a constant, and combining those results: C 1 () = 1.0L() + 0.0M() + 0.0S() C 2 () = -0.59L() + 0.80M() + -0.12S() C 3 () = -0.34L() + -0.11M() + 0.93S()
Opponent Processing Model
Falsification (finally) In 1977, Edwin Land produced a falsification: The particular relationship between the stimulus of the L, M and S cones is both NECESSARY and SUFFICIENT for determining a particular color sensation. (Given trichromatic perceivers and normal conditions) Helmholtz (sufficiency) Produce a case where two objects stimulate the L, M and S cones in exactly the same way, but produce two different color sensations (necessity) Produce a case where two objects stimulate the L, M and S cones in different ways, yet produce the same color sensation. Two falsifications required:
Mondrians 5.8 (L) 3.2 (M) 1.6 (S) These values held for red, green and blue patches, yet the patches still appeared red, green and blue! 5.8 (L) 3.2 (M) 1.6 (S)
Tristimulus values sufficient for color appearance According to the Helmholtz theory, IF objects reflect the same tristimulus values, THEN they will appear to be the same color. A patch in the first Mondrian reflects the tristimulus values of 5.8(L), 3.2 (M) and 1.6 (S), yet looks red. A patch in the second Mondrian reflects the tristimulus values of 5.8(L), 3.2 (M) and 1.6 (S), yet looks blue. (from 1) IF the Helmholtz theory is correct, THEN the patch in the first Mondrian will appear to be the same color as the patch in the second Mondrian. They do NOT appear to be the same color. THEREFORE, the Helmholtz theory is NOT correct.
Tristimulus values necessary for color appearance According to the Helmholtz theory, IF two objects appear to be the same color, THEN they will reflect the same tristimulus values. Before Land turned the projectors on, the red patch looks red. After Land turned the projectors on, the red patch looks red. (from 1) IF the Helmholtz theory is correct, THEN the patch that continues to look red must reflect the same tristimulus values when the projectors are on & off. The patch that continues to look red does NOT reflect the same tristimulus values when the projectors are on as it does when they are off. THEREFORE, the Helmholtz theory is NOT correct.
Helmholtz’ response (from 1) IF the Helmholtz theory is correct, THEN the patch that continues to look red must reflect the same tristimulus values when the projectors are on & off. The patch looks red w/ the projector OFF and it reflects tristimulus values x, y, z THEREFORE, the Helmholtz theory is NOT correct. The patch looks red w/ the projector ON and it reflects tristimulus values a, b, c BUT: One cannot draw conclusions about the physiology of color from this fact, so… The patch looks red w/ the projectors OFF, but it is really isn’t. (or vice versa)
Timeline 1856: Helmholtz proposes his theory 1892: Christine Ladd-Franklin formulates her argument. 1957: Theory change following Jameson & Hurvich 1977: Land’s Falsification 64 years!20 years 1908: Gestalts.
Why Theory Change? New Technology (Galileo's Telescope) Socio-cultural factors (Ladd-Franklin) Mathematics (Jameson & Hurvich)
Alternatives for Demarcation 1.Verification 2.Falsification 3.Lakatos’ ‘Progressive v. Degenerative’ 4.Sociological factors 5.Explanation in terms of mechanism.