For the past half century it has been largely agreed that one cannot observe without prior theory. This is rarely explicated, however, and there seems to be some ambiguity in the claims made. So I will do a rough taxonomy of the TDOT.
When N. R. Hanson introduced the claim it was this:
Weak TDOT: “There is a sense, then, in which seeing is a ‘theory-laden’ undertaking. Observation of x is shaped by prior knowledge of x.” (Hanson 1958:19)
This is rather vague, and it remained vague when Kuhn took it up and defended it:
What a man sees depends both upon what he looks at and also upon what his previous visual-conceptual experience has taught him to see. (Kuhn 1996:113)
The target of Hanson and Kuhn’s discussions was the positivist claim that observations were framed in the language of “sense-data”, which were pretheoretic as experiences and in terms of language (the usual example used was “red here now”, which is pretty far from actual scientific observations). And there is something to Hanson’s claim, but it remains rather unlikely that Kuhn’s claim that an Aristotelian and a Galilean will see a pendulum differently because they are “[p]racticing in different worlds”. While we can say that prior experience modulates observation, it seems overwrought to say that one’s “worldview” (a term I despise as meaningless) constructs our observations like this.
[I have a similar problem with WIttgenstein’s comment in the Philosophical Investigations (II, xi, p.190) that “If a lion could talk, we could not understand him.” If we share a form of life, as we clearly do, both being social and sexual mammals in an ecology, then we should be able to understand quite a lot of a lion’s discourse. It would, I suspect, boil down to “My mates! My food! My territory!” But we’d understand it well enough. Likewise, we understand the motivations of dogs, primates and to a lesser extent cats, either because our form of life is closely evolutionarily related to them (primates) or because we have evolved in parallel fashion (dogs). Lebensformen are not hermeneutically sealed from each other, contra Kuhn and WIttgenstein and many other German-Romantic-inspired thinkers. Had Wittgenstein not dismissed evolutionary biology so quickly, he may have understood this. Instead he reasoned from a sample species set of one.]
The strong version of TDOT that Kuhn asserted is this:
Strong TDOT: One can only perceive what one’s worldview [global theoretical commitments] permits and requires.
This is extremely circumscriptive. And I think it is (rarely for a philosophical claim) empirically false, unless we reduce the notion of “theory” to “dispositions”, in which case it is barely trivially true. Children have no worldviews (although they certainly have experience to guide them) and yet they learn to see, feel, and hear, etc. A recent study has finally resolved, empirically, the issue of whether we have the capacity to perceive objects when we were born blind but regain sight as adults (the Molyneux Question; answer: not unless they can correlate sight with touch and proprioception), which raises the question how children can possibly learn to recognise objects at all. It seems that infants have a disposition to treat motion and touching as a reward feedback, and so they move their limbs at random until they touch something. Is that “theory”? Only in the mind of a theory-obsessed philosopher could it be considered to be.
Harold I. Brown (1995) gives what I think is the best summary discussion of the TDOT. Brown distinguishes six different theses:
- The items we perceive are already infected with material from the theories we accept.
- Scientists ignore evidence that contradicts their favored theories.
- Observations that are undertaken to evaluate a comprehensive theory presuppose that very theory in a way that prevents an objective test of that theory.
- All scientifically significant observations assume some theories besides the theory being evaluated; it is always possible to protect a favored theory by challenging these auxiliary theories.
- Which observations scientists undertake is determined by accepted theory.
- Observation reports must be expressed in the language of the theory being tested if they are to be relevant to the evaluation of that theory.
Thesis 1 is undoubtedly true. Nobody begins an investigation tabula rasa, and having some theories, even if they are half-formed and vague, must influence at least some acts of observation. But it is unclear to me this is necessarily the case or that the “infection” is so significant as to determine the outcome of the observation.
Thesis 2 is also likely to be occasionally true. That doesn’t mean it is always true. This claim is rather trivial, if occasional, or false, if the claim is it that it always occurs.
Thesis 3 is the main subject of discussion in the period after Kuhn. If your theory makes you take observations in ways that beg the theoretical question, then the practice of science becomes an exercise in telling oneself stories for whatever reason. That way, postmodernism lies.
Thesis 4 is true, and forms the basis for Lakatos’ claim that a research program has a protected core (which view I think is not true, but is contingently a fact about some disciplines or programs, and only a few). The Duhem-Quine hypothesis makes this point against falsification os theories, but again I think it claims too much. A truly perverse scientist can always retain a favoured hypothesis (like creationism) by abandoning other theories (like, say, the whole of physics), but this won’t fly in scientific arenas, or at least it used not to.
Thesis 5 is the crucial sense I am concerned with in this series. The point is not that prior experience and theoretical commitments influence observation, but that they determine what is observed. This determination leaves no room for untheoretical phenomena.
Thesis 6 is a claim about the language of the reports of observation. It basically means that even if you can perceive a phenomenon, you can’t represent it except in the language of the theory. This thesis leaves no room for representations of observations except in terms of the meanings assigned by theory.
Of course two or more of these senses may be included in a particular claim for TDOT, but there is a general characterisation we may give of it that leads to a better grip on the notion:
General TDOT: An observation report O is dependent upon a theory T
We can focus on
- the report and the language used
- the dependency relation
- the theory or theories and their scope, and degree of influence upon the observation report O.
If we focus on the report (thesis 6), then the argument might be that using the terms of the theory in some way means that theories are incommensurable because they do not denote the same parts of the world and so we cannot prefer one over another on empirical grounds. This extends to technical terms and mathematical variables. Moreover, we can usefully consider ways in which neurobiology modifies and processes signals. Kuhn and others thought this was crucial here. I do not. If we perceive, it is not because an image falls on our retina, but because the entire apparatus is working, from lens to visual cortex. The older view is like saying that a camera does not record accurately because it uses silver halide chemistry to indicate where photons fell, or in a digital camera, a CCD chip and internal processing and storage in RAW format. [See this post by John Holbo for a discussion of “picture theories of observation”.]
If we consider the dependency relation, a weak dependency (theory biases but doesn’t determine observation) causes us only a little discomfort. For it to be problematic we need to have a strong claim, that is determines wholly (or mostly) what we can perceive and how. Only then will it exclude ordinary observation, naive observation, and phenomena that are not in the Bogen-Woodward definition: patterns in (theoretically determinate) data. If we allow that the dependency is partial, either because the theory is only slightly relevant or because it only biases our observations, then the way is clear for a kind of empiricism to both develop and establish theories more or less objectively.
Therefore we really need to consider what the theories are on which observations depend, and in what manner, which will be the subject of a future post.
Brown, Harold I. 1995. Empirical testing. Inquiry: An Interdisciplinary Journal of Philosophy 38 (4):353 – 399.
Hanson, N. R. 1958. Patterns of Discovery. New York: Cambridge University Press.
Kuhn, Thomas S. 1996. The structure of scientific revolutions. 3rd ed. Chicago, IL: University of Chicago Press.
Wittgenstein, Ludwig. 1968. Philosophical investigations. Translated by G. E. M. Anscombe. Repr. of [3rd ed.] English text, with index. ed. Oxford: Basil Blackwell.
One of the flat earther arguments is that the spherical earthers are so constrained by their theory that they cannot properly understand their observations. I recognized various populations of Austrofundulus as sorta different. Now that I know they are different species, it amuses me how much more different they are now than they were then.
Personally, I am a skeptic of “sense-data” theories.
In what follows, I will use the terms “syntactic” and “semantic”. No doubt, I will be misusing them, but they are the closest ideas that fit. If I measure something as 10 inches, you might measure the same thing as 25.4 cm. I’ll use “syntactic” to refer to “10 inches” and “25.4 cm” (thought of as strings of characters), and I’ll use “semantic” to refer to the actual length on which we both agree, albeit with different syntax.
Suppose that I take a picture with my digital camera, and you take a picture of the same scene with your camera. The syntax here is the numeric value assigned to the pixels, while the semantics is what is pictured. We probably have different cameras, different pixel resolutions, etc. There is no easy relation between my syntactic data and yours, though no doubt the semantics is “similar”. (I put quotes around “similar” because I don’t think we have a useful concept of similarity for comparing semantics).
If my camera has a telephoto lens, and yours has a closeup lens, there might be significant differences in the semantics. But they will be in the form of differences in the amount of detail that is available.
In ordinary seeing, we are mainly conscious of semantics. Presumably, neural firings are syntactic, but we are not aware of individual neural firings. In science, we mainly deal with the syntactic, and it is up to us individually to interpret that in semantic terms.
I see Kuhn as mainly pointing to syntactic differences. Thus he saw a lot of variation that is theory dependent. My personal view of a scientific theory, is that its main role is a defining how to give syntactic representations of reality. Thus a paradigm shift is a change in how to map reality to syntax. And of course, syntactic data is theory laden. But semantic data (how we interpret that syntactic data) is mostly dominated by reality rather than by our theories. Thus I can talk about the earth going around the sun (Copernican syntax), and in the next sentence talk of the sun rising in the East (Ptolemaic syntax) without even noticing the paradigm shift between those two ways of talking.
I’m looking now, at Brown’s thesis 3. Based on how I see theories, it makes no sense. Specifically, “objective test of that theory” makes no sense. A theory is not a description. Rather, it is a mapping from semantic reality to objective syntactic data. We can test it pragmatically – does it do the job well enough. But we cannot test whether it is true, which is what I take an objective test to entail. If truth is correspondence with reality, then a theory establishes a particular correspondence (or mapping) and is prerequisite to any talk about the truth of the syntactic representations.
As for that Wittgenstein comment on talking lions – I agree with Wittgenstein. But I think Wittgenstein was making a comment about the nature of language, rather than about our relation to reality. Specifically, I see language as intensively social/cultural, and as being part of what binds a society. Come to think of it, there is almost nothing in what I am typing that could make sense to a lion.
I too disbelieve in sense data. I take that to be established by now. Sense data language makes even less sense.
I am not quite sure if your syntax-semantics distinction is the same as the Syntactic versus Semantic Conception of Theories. The latter is similar tot he idea that a family of models needs an interpretation to be explanatory. So that might be the point being made (I prefer Suppe’s version to van Fraassen’s).
Of course Brown is not arguing for any of these theses.
Another one of those posts I just need to thank you for. Cheers. I have grasped the gist of theory-ladeness before, but this really is a useful disambiguation for me.
I dislike the notion of theory dependence because “theory” in these debates is even more slippery than “paradigm” in Kuhn. At any given time, however, I accept that some features/frameworks are functionally privileged or, as a left-over Kantian might put it, synthetic a priori for the time being. That’s how I understand the approach to intellectual history of people like Foucault and Ian Hacking. What I do object to is the notion that not all propositions are equal somehow makes it mysterious that new ideas emerge at all, It seems to me that the impression that science and other cognitive activities require inexplicable quantum leaps is mostly an artifact of the position of the observer. If you’re analyzing how Frenchmen classified plants in 1720, you aren’t entitled to go beyond the evidence of what these folks did and wrote just as a linguist studying ancient Greek can only claim that the second person middle aorist of luo is such and such if and only if the usage is attested. The aforementioned Frenchmen and Greeks, however, didn’t operate under any such methodological stricture since they had citizen’s rights in their own discipline/language. A static picture of a science or a language creates the illusion of magic barriers that don’t exist from a dynamic point of view. Can I find words for 47 kinds of snow? Sure, if you give me the time to cook them up, which is precisely what I and my cohorts will do if there’s some reason to elaborate terminology about snow.
Of course, I’m not claiming that there really aren’t any barriers to changing the rules, just that the actual as opposed to the imagined barriers are sociological in nature and perfectly amenable to study. It takes work to alter the perspectives that filter experience in actual human disciplines precisely because activities like sciences are social so that nothing meaningful occurs unless you can bring the others along—as the psychoanalysts used to say, the moment of insight occurs in the ear of the analyst. The persistence of older ways of thinking isn’t just a matter of inertia either. There are specific mechanisms that work to police discourse. In Galileo’s time, for example, hewing to the text of Aristotle was required by the bylaws of the University of Padua. And the crucial thing to remember is that this sort of censorship/discipline is absolutely necessary. It is like the air whose resistance slows the bird but is actually what makes it possible to fly. (I’m inclined to believe that a crucial feature of the Scientific Revolution was the development of a new system of thought control that turned out to be more flexible in some ways than the older censorship of church and academy but was actually much more rigorous. That’s a different story.)
Have to agree with Wittgenstein. I know people who seem quite fluent with language, but I haven’t the foggiest idea what they’re on about. If I can’t even understand another human’s interpretation of reality, I’d probably make even less sense of any lion utterances. Language is also used to express the subjective as well as the objective, and the common factors there may be even fewer.
My point is that Wittgenstein himself shows how we might understand a lion (and by extension speakers of other theoretical languages) because we share a form of life. He seems to have thought that use is restricted to within a language community; I think it is able to bridge them because forms of life are not coterminous with language communities.
For #4: Roentgen’s discovery of X-rays was the result of an unexpected observation. He was fiddling with a cathode ray tube (CRT) in a dark room, and saw something he did not expect. (At the time people did not know much about such matters, and were only beginning to understand them in terms of both earlier and new theories.) This was the period when the surprising discoveries of X-rays, other CRT phenomena, and radioactivity led finally to pretty much our current view of the structure of matter. (The discovery of radioactivity also was an accidents.)
For the usefulness of theory in guiding observation, I recall Walter Keller (a well-know geologist) remarking to me that he was flying along the Andes and recognized a great deal about the geology he was seeing that he would not have seen before the theory of plate tectonics. (This was about 20 years ago. Keller’s DOB was March 13, 1900, so he was around 80 when he saw this.)
Maybe you mentioned this in one of your posts a week or two back, but: Which arguments that appeal to TDOT are you interested in here? Different versions are going to seem more or less important depending on what implications you care about.
For instance, I work on science and values, so I’m interested in arguments that `values’ (whatever we mean by those) can/do/should influence theory choice. The first part of version 4, which you pass over quickly, is most important in contemporary such arguments, because (so these arguments go) values get involved by way of the auxiliary theories.
I think I would have huge difficulty grasping what an individual lion had to say.
But if I spent some time observing the culture, society, social structure and environment the lion has emerged from I would begin to have some idea of the context of what is being said.
The observations would be theory dependant to some extent so I would expect a degree of error.
Bit like trying to translate a 6th century Welsh poem. Seems utterly mad when you get the first literal transcript; but you reconstruct the context with reference to a range of subjects and theories from history, archaeology, anthropology and other social sciences.
You live with the fact that conclusions are theory dependant and will contain a degree of error.
But even with these problems you can run quite far with things.
I think you just have to learn to live with the fact that what you are constructing is fluid and subject to sudden change and try not to get to attached to what you are working with, as it is prone to transformation as new tools are discovered.
Comments are closed.