20. PHIL 160 Need auxiliary hypotheses to generate testable predictions: H1: Cadaveric matter causes childbed fever. H2: Cadaveric matter is transferred from cadavers to patients via hands, instruments. H3: Washing hands, instruments in chlorinated lime will remove or destroy cadaveric matter.
26. PHIL 160 1 st law of motion : If there is no force acting on a body, the momentum of that body will remain constant. Newton’s group of hypotheses
27. PHIL 160 2 nd law of motion : If there is a force acting on a body, that body will accelerate by an amount directly proportional to the strength of the force and inversely proportional to its mass. Newton’s group of hypotheses
28. PHIL 160 3 rd law of motion : If one body exerts a force on a second body, then the second body exerts a force on the first body that is equal in strength and opposite in direction. Newton’s group of hypotheses
29. PHIL 160 Law of universal gravitation : Any two bodies exert attractive forces on each other where the forces are in the direction of the line connecting the bodies and are proportional to the product of their masses divided by the square of the distance between them. Newton’s group of hypotheses
30. PHIL 160 3 laws of motion + law of universal gravitation No testable predictions! Newton’s group of hypotheses
31. PHIL 160 3 laws of motion + law of universal gravitation + auxiliary hypotheses about masses, positions of planets and sun Predictions of planets’ orbits Newton’s group of hypotheses
32. PHIL 160 But where? Negative test - there is a problem SOMEWHERE in the group of hypotheses Test group of hypotheses
33. PHIL 160 Bad prediction of Uranus’s orbit 3 laws of motion + law of universal gravitation + auxiliary hypotheses about masses, positions of planets and sun Newton’s group of hypotheses
34. PHIL 160 Bad prediction of Uranus’s orbit 3 laws of motion + law of universal gravitation + auxiliary hypotheses about masses, positions of planets and sun Missing a planet? Newton’s group of hypotheses
36. PHIL 160 Astronomers changed auxiliary hypothesis about how many planets, left Newton’s laws & UG unchanged. This fixed bad prediction of Uranus orbit. New planet (Neptune) observed by telescope.
38. 3 laws of motion + law of universal gravitation + auxiliary hypotheses about masses, positions of planets and sun PHIL 160 Bad prediction of Mercury’s orbit Newton’s group of hypotheses
39. PHIL 160 First attempt: stick another planet in (“Vulcan”) between Mercury and Sun.
40. PHIL 160 No repeatable observations of this new planet. First attempt: stick another planet in (“Vulcan”) between Mercury and Sun.
41. PHIL 160 No repeatable observations of this new planet. (Possible aux. hypothesis: the new planet is invisible!) First attempt: stick another planet in (“Vulcan”) between Mercury and Sun.
42. 3 laws of motion + law of universal gravitation + auxiliary hypotheses about masses, positions of planets and sun PHIL 160 Bad prediction of Mercury’s orbit Bad theory? Newton’s group of hypotheses
43. PHIL 160 Duhem says scientists must rely on “good sense” No set rule for how to fix group of hypotheses that makes a bad prediction.
44. PHIL 160 Holism in testing important in observations made with scientific instruments. Group includes hypotheses about the measuring device.
48. PHIL 160 In the group of hypotheses: Theory about salmonella growth. Theory of the microscope. Bad prediction could mean a problem with either theory (or both)
49. PHIL 160 Quine : Hypotheses are tested in groups and the group includes all human knowledge claims (including meanings of terms). “ Meaning holism” Can’t test hypotheses in isolation, and changes in one hypothesis can prompt changes throughout whole logical structure of the theory.
50. PHIL 160 What does this have to do with the Twin Earth thought experiment?
57. PHIL 160 Is the stuff from Twin Earth water? New test: elemental analysis Water from Earth: H 2 O Stuff from Twin Earth: XYZ But is Twin Earth stuff water?
58. PHIL 160 Is the stuff from Twin Earth water? Because of meaning holism, this depends on the choices you make in evaluating the group of hypotheses.
59. PHIL 160 1. Water is the stuff found in rivers, lakes, streams, and aqueducts that humans drink, cook with, bathe in, swim in, etc.
60.
61. PHIL 160 3. Although water from the ocean, water from a stream, water from a well, rainwater, etc., may differ in their impurities (salt, mud, etc.), they all count as water; that is, the substance containing these different impurities is the same.
62. PHIL 160 4. Water’s behavior is a result of what it is made of.
63. PHIL 160 5. If two samples display different microstructure (e.g., different elemental composition), they must be samples of different substances.
64. PHIL 160 6. Substances with different microstructures will display different macroscopic behavior.
69. Baptism: "That stuff is water" PHIL 160 Courtesy of the Frances Loeb Library, Graduate School of Design, Harvard University
70. Description: "Water has the following properties ..." PHIL 160 Freezing point Taste, color ... Boiling point Microscopic structure
71. PHIL 160 Courtesy of the Frances Loeb Library, Graduate School of Design, Harvard University "That stuff (water) has these properties"
72. Quine: PHIL 160 “ Any statement can be held to be true come what may, if we make drastic enough adjustments elsewhere in the system.”
Notes de l'éditeur
Note that we’re shifting the discussion a bit from “hypothesis testing” to “theory testing”. This is important – one of the issues Duhem raises is that the testing scientists do is of theories rather than of individual hypotheses. We’ll see why in just a moment.
Note that you can think of a theory as a group of hypotheses, some of which focus on ontology (what kind of stuff there is), others of which focus on behavior, maybe by positing laws of one sort or another.
This was the picture of science the logical empiricists found intuitively appealing. Hempel said that science couldn’t really work this way. There is just too much to observe. We’d never have all the facts, so any kind of generalizations we’d want to draw based on what we had observed would be vulnerable to the problem of induction . There would be no guarantee that what we had observed was an accurate indication of what the stuff we hadn’t observed was like.
Hempel says that there are two separate processes here, hypothesis formation and hypothesis testing. Pause a moment to notice the connection between hypothesis and theory here. First, an hypothesis is a claim – maybe about the existence or properties of a certain entity, maybe about how that entity evolves over time, maybe about how that entity interacts with other entities. When we call such a claim an hypothesis, it usually suggests that it’s a guess rather than something we’re certain about, but there can be hypotheses about which we’re pretty confident. A theory, then, will just be a group of hypotheses; each hypothesis is one of the claims that make up the theory’s account of the world. So, a strategy you might use to test a theory would be to test each of the hypotheses in the theory. First, we need to identify what Hempel called the “test implications” of the hypothesis – the observable consequences of the hypothesis that will help us figure out whether the hypothesis is true or false. Next, we set up the appropriate test conditions. And then, we see what actually happens.
If my hypothesis H is true, I expect to see observable outcome I under the test conditions. I set up the test conditions and don’t observe I. This test gives a negative result; the hypothesis must be false.
If my hypothesis H is true, I expect to see outcome I under the test conditions. I set up the test conditions and this time observe I. A positive test! Have I demonstrated that H is true? No. This is where the problem of induction comes in. Even though outcome I is consistent with H being true, it might also be consistent with a number of other hypotheses.
My observation of 10,000 black ravens and no ravens that aren’t black is consistent with the hypothesis “All ravens are black,” but this observation is also consistent with the hypothesis, “Some ravens are pink.”
My observation of 10,000 black ravens and no ravens that aren’t black is consistent with the hypothesis “All ravens are black,” but this observation is also consistent with the hypothesis, “Some ravens are pink.”
All the daily sunrises observed over the course of human history count as data to support the claim “The sun will rise tomorrow,” but they are also consistent with the claim, “The sun won’t rise tomorrow.”
Our scientific theories are underdetermined by our experience . There are lots of ways the world could be that fit with the data we’ve collected so far. This could be a world in which the “laws of nature” never change and sunrise will happen everyday, or it could be a world in which the “laws of nature” will change dramatically next Tuesday. The observations we’ve got can’t tell us which of these worlds we are in (at least not before next Tuesday). Since there are different ways the world could be that fit with our experiences of it, it is possible to generate equally many good theories to account for the world. From the point of view of testing, what underdetermination means is that a positive test doesn’t establish an hypothesis as certainly true.
Our scientific theories are underdetermined by our experience . There are lots of ways the world could be that fit with the data we’ve collected so far. This could be a world in which the “laws of nature” never change and sunrise will happen everyday, or it could be a world in which the “laws of nature” will change dramatically next Tuesday. The observations we’ve got can’t tell us which of these worlds we are in (at least not before next Tuesday). Since there are different ways the world could be that fit with our experiences of it, it is possible to generate equally many good theories to account for the world. From the point of view of testing, what underdetermination means is that a positive test doesn’t establish an hypothesis as certainly true.
It would be great to take our hypothesis one by one, subject them to tests, and get rid of the ones that fail the tests. However, for many hypotheses in scientific theories, we just can’t do this.
Pierre Duhem, a French philosopher and physicist, was the first to point out in the early 1900s that for the most part hypotheses are tested in groups because individual hypotheses are usually not testable .
Why not? Because individual hypotheses usually don’t have any observational consequences. In other words, they don’t make any predictions you could test with an observation.
Hypotheses like this are much more common than you might think. Recall the hypothesis Semmelweis had: “Cadaveric matter causes childbed fever.” Given that cadaveric matter is not something he could observe, how could he test this hypothesis? He could only test it if he included additional hypotheses about what kinds of activities spread cadaveric matter and what types of measures stopped its spread. In other words, Semmelweis’s test was really the test of a group of hypotheses.
If cadaveric matter causes childbed fever, and if cadaveric matter is spread to pregnant women by hands and instruments of physicians involved in autopsies, and if chlorinated lime deactivates cadaveric matter, then washing the hands and instruments in chlorinated lime after autopsies will lower the rate of childbed fever.
Semmelweis got a positive result to his test – washing with chlorinated lime resulted in a drop in the rate of childbed fever. But if the results had been negative, it might be very difficult to figure out just what the results could mean. Would a negative test result mean that chlorinated lime didn’t deactivate cadaveric matter? Or that the cadaveric matter was not transmitted from autopsies but through some other means? Or that cadaveric matter didn’t cause childbed fever?
Duhem points out that if your test yields a result that differs from the one you’d expect if your group of hypotheses was right, you know there must be a problem somewhere in the group of hypotheses. The difficulty is that the failed test doesn’t tell you where in the group the problem lies. The group of hypotheses stands or falls together; you can’t test them one by one to see where the problem is.
Duhem points out that if your test yields a result that differs from the one you’d expect if your group of hypotheses was right, you know there must be a problem somewhere in the group of hypotheses. The difficulty is that the failed test doesn’t tell you where in the group the problem lies. The group of hypotheses stands or falls together; you can’t test them one by one to see where the problem is.
It’s worth pointing out that there are some hypotheses that can be tested all by themselves. For example, “Mars has a triangular orbit” is testable. Observe the path of the orbit of Mars and see what shape it has. However, a great many important scientific hypotheses can only be tested when combined with other hypotheses.
It’s worth pointing out that there are some hypotheses that can be tested all by themselves. For example, “Mars has a triangular orbit” is testable. Observe the path of the orbit of Mars and see what shape it has. However, a great many important scientific hypotheses can only be tested when combined with other hypotheses.
In fact, here’s a theory that isn’t testable without additional hypotheses: Newtonian mechanics. The theory describes how bodies behave and interact.
What we need to add to Newtonian mechanics to get any testable predictions are auxiliary hypotheses about what bodies there are, their masses, positions, etc. But when you add auxiliary hypotheses about the masses and positions of the sun, Mercury, Venus, Earth, Mars, Jupiter, Saturn, and Uranus, [20] all of a sudden Newtonian mechanics gives lots of testable predictions about things like planetary orbits. Comparing observed planetary orbits to predicted planetary orbits is the kind of thing you might do if you were trying to test the theory of Newtonian mechanics. As long as the observations match up well with what the theory predicts, you can conclude that Newtonian mechanics is at least consistent with the observed planetary motions.
But what do we do if the observations don’t match up with the predictions? This was just the situation in which physicists and astronomers found themselves in the early 1800s. Although the hypotheses of Newtonian mechanics (plus those auxiliary hypotheses about the positions and masses of the celestial bodies) gave very good predictions of the orbits of Mars and Venus and Jupiter and Saturn, the prediction for the orbit or Uranus wasn’t even close to what was observed. [21] Clearly, something was wrong with this group of hypotheses, but what?
But what do we do if the observations don’t match up with the predictions? This was just the situation in which physicists and astronomers found themselves in the early 1800s. Although the hypotheses of Newtonian mechanics (plus those auxiliary hypotheses about the positions and masses of the celestial bodies) gave very good predictions of the orbits of Mars and Venus and Jupiter and Saturn, the prediction for the orbit or Uranus wasn’t even close to what was observed. [21] Clearly, something was wrong with this group of hypotheses, but what? One possibility was that the problem was in the law of universal gravitation, or in one of the three laws of motion, or even that two or more of these four hypotheses were false. Another possibility was that the problem was with the auxiliary hypotheses. Perhaps the predictions were made using a bad mass or position for one of the planets. Or maybe there were other planets in the system that were omitted from the auxiliary hypotheses!
This last hunch was the one that two separate astronomers pursued. They reasoned that if there were an additional planet beyond Uranus, the gravitational pull it exerted on Uranus could explain the observed orbit of Uranus. Indeed, they were able to use Newtonian mechanics to calculate the approximate mass and position of this additional planet, and only a few years later, this planet, Neptune, was observed by telescope.
This last hunch was the one that two separate astronomers pursued. They reasoned that if there were an additional planet beyond Uranus, the gravitational pull it exerted on Uranus could explain the observed orbit of Uranus. Indeed, they were able to use Newtonian mechanics to calculate the approximate mass and position of this additional planet, and only a few years later, this planet, Neptune, was observed by telescope.
This last hunch was the one that two separate astronomers pursued. They reasoned that if there were an additional planet beyond Uranus, the gravitational pull it exerted on Uranus could explain the observed orbit of Uranus. Indeed, they were able to use Newtonian mechanics to calculate the approximate mass and position of this additional planet, and only a few years later, this planet, Neptune, was observed by telescope. This case is a nice example of how scientists adjust a group of hypotheses to deal with a failed test. Here, the scientists found a way to add an auxiliary hypothesis that fixed the one bad prediction while leaving the good predictions of the other planets’ orbits untouched.
This last hunch was the one that two separate astronomers pursued. They reasoned that if there were an additional planet beyond Uranus, the gravitational pull it exerted on Uranus could explain the observed orbit of Uranus. Indeed, they were able to use Newtonian mechanics to calculate the approximate mass and position of this additional planet, and only a few years later, this planet, Neptune, was observed by telescope. This case is a nice example of how scientists adjust a group of hypotheses to deal with a failed test. Here, the scientists found a way to add an auxiliary hypothesis that fixed the one bad prediction while leaving the good predictions of the other planets’ orbits untouched.
It looks like a sensible strategy, but does it always do the job? As it turns out, Uranus wasn’t the only planet for which Newtonian mechanics predicted an orbit that was significantly different from the observed orbit. This was a problem for the planet Mercury as well (although it didn’t become apparent how big a problem it was until telescopes got fairly powerful, allowing for more precise observational data). Well, still aglow with their success with Uranus, astronomers tried to use the same approach to deal with this problem. They decided not to touch Newton’s laws. Rather, they hypothesized that there was an additional planet between Mercury and the sun. The gravitational pull from this extra planet would end up explaining why Mercury had the orbit that was actually observed. So sure were they that this strategy would work again that they actually named this additional planet Vulcan before the folks with telescopes even had a chance to locate it.
As it turns out, they needn’t have bothered. No matter how hard they looked, no one ever observed an extra planet in the vicinity of Mercury. [23] Ultimately, the astronomers did fix the predicted orbit of Mercury, but to do so they had to replace Newtonian mechanics with Einstein’s theory of relativity. Once again, the moral of the story about holism in testing is that it’s a group of hypotheses that generates the prediction. If what actually happens differs from what was predicted, you know there must be a problem somewhere in that group of hypotheses. But, you don’t know where in the group of hypotheses the problem actually is.
As it turns out, they needn’t have bothered. No matter how hard they looked, no one ever observed an extra planet in the vicinity of Mercury. [23] Ultimately, the astronomers did fix the predicted orbit of Mercury, but to do so they had to replace Newtonian mechanics with Einstein’s theory of relativity. Once again, the moral of the story about holism in testing is that it’s a group of hypotheses that generates the prediction. If what actually happens differs from what was predicted, you know there must be a problem somewhere in that group of hypotheses. But, you don’t know where in the group of hypotheses the problem actually is.
As it turns out, they needn’t have bothered. No matter how hard they looked, no one ever observed an extra planet in the vicinity of Mercury. [23] Ultimately, the astronomers did fix the predicted orbit of Mercury, but to do so they had to replace Newtonian mechanics with Einstein’s theory of relativity. Once again, the moral of the story about holism in testing is that it’s a group of hypotheses that generates the prediction. If what actually happens differs from what was predicted, you know there must be a problem somewhere in that group of hypotheses. But, you don’t know where in the group of hypotheses the problem actually is.
As it turns out, they needn’t have bothered. No matter how hard they looked, no one ever observed an extra planet in the vicinity of Mercury. [23] Ultimately, the astronomers did fix the predicted orbit of Mercury, but to do so they had to replace Newtonian mechanics with Einstein’s theory of relativity. Once again, the moral of the story about holism in testing is that it’s a group of hypotheses that generates the prediction. If what actually happens differs from what was predicted, you know there must be a problem somewhere in that group of hypotheses. But, you don’t know where in the group of hypotheses the problem actually is.
What should we do when we encounter failed predictions, then? Duhem says the scientist must rely on “good sense”. Perhaps the scientist can change one hypothesis in the group and see what the new group predicts. If that doesn’t bring the prediction in line with reality, maybe change a different hypothesis and see what happens then. It may seem prudent to start making the changes in the auxiliary hypotheses before moving on to the central hypotheses of the theory. But, there’s no sure way to know ahead of time how – if at all – you’ll be able to turn bad predictions into good ones.
Here’s one more consequence of this holism about testing Duhem points out: there are even hypotheses built into the observations we make in order to test our hypotheses in the first place, at least when we make those observations using measuring devices.
. If you test a theory that predicts your room-temperature chicken leg is crawling with salmonella bacteria by looking at that chicken leg under the microscope to see the bacteria, your group of hypotheses now includes a set of hypotheses about how the microscope works and what kinds of entities it can let you observe under what conditions. If the predicted detection of the salmonella fails, the problem could be with your hypotheses about room-temperature chicken, or it could be with your hypotheses about the microscope.
. If you test a theory that predicts your room-temperature chicken leg is crawling with salmonella bacteria by looking at that chicken leg under the microscope to see the bacteria, your group of hypotheses now includes a set of hypotheses about how the microscope works and what kinds of entities it can let you observe under what conditions. If the predicted detection of the salmonella fails, the problem could be with your hypotheses about room-temperature chicken, or it could be with your hypotheses about the microscope.
. If you test a theory that predicts your room-temperature chicken leg is crawling with salmonella bacteria by looking at that chicken leg under the microscope to see the bacteria, your group of hypotheses now includes a set of hypotheses about how the microscope works and what kinds of entities it can let you observe under what conditions. If the predicted detection of the salmonella fails, the problem could be with your hypotheses about room-temperature chicken, or it could be with your hypotheses about the microscope.
Duhem’s holism about testing makes evaluating the fit between theory and world pretty strenuous work. Quine, however, tells us that the holism that comes into play in our attempts to test a theory is even more far-reaching. Indeed, Quine asserts that all human knowledge claims, whether scientific hypotheses or claims about the relation between the New York Yankees and the source of all evil, are tied up together in one big group. In other words, when we make a prediction and observe a different outcome, the group of hypotheses we must figure out how to adjust includes everything , even meanings of words.
If science is holistic, then scientists make choices – rather than being forced by the data to a single conclusion, they have to use the data to decide how to update their hypotheses and concepts.
You considered a thought experiment starting in the year 1600. You are a scientist who has that essential piece of laboratory equipment, an interplanetary transporter. With this device you travel to Twin Earth, a planet very much like our own Earth.
While there, you discover that the rivers, lakes, streams, and aqueducts of Twin Earth are filled with a liquid that Twin Earthlings drink, cook with, bathe in, swim in, etc. With their permission, you collect several buckets of this stuff and teleport back to Earth.
The obvious question: is this stuff water? Certainly, to the average person in the 1600s, it would seem to be. But you are a scientist . You are not interested in mere appearances. You want to establish by experimental means whether this stuff really is water. In the framework of 1600s science, this will not be a matter of fancy equipment but rather of empirical determination of the behavior of this stuff in different conditions.
Can this stuff be drunk like water? You summon the two village idiots, give one a glass of water and the other a glass of the stuff from Twin Earth. Neither dies, both affirm that their thirst is quenched. So far, the Twin Earth stuff acts like water. You heat some of the Twin Earth stuff in a cauldron; after a while it boils, just like water. You use some of it to irrigate a plot in your garden; the plants do just as well as in the part of your garden irrigated with water. You climb the high peak with a bucket of each; the water and the stuff from Twin Earth freeze at the same altitude. For every test you perform, water and the stuff from Twin Earth behave identically. Thus, you decide, it is reasonable to conclude that the stuff from Twin Earth is water.
By the next stage of the thought experiment, in 1785, you might want to reconsider this conclusion. For now, you are a scientist in the laboratory of Lavoisier, the discoverer of oxygen. As a scientist, you no longer see the world as made up of earth, air, fire, and water. Instead, you recognize that the world is made up of many different elements, including oxygen and hydrogen.
One of your experiments involves getting a bucket of water from the Seine River, running electricity through it, and collecting the gasses the bubble off. You determine that twice as much hydrogen gas as oxygen gas is produced in this experiment and conclude from this that water is two parts hydrogen to one part oxygen, or H 2 O. What about the buckets of stuff still sitting in the storeroom from that trip to Twin Earth back in 1600? Just for fun, you have a young assistant perform the same electrolysis experiment on the water from Twin Earth. But surprise: this stuff has a different elemental composition (call it XYZ). You repeat the experiment yourself to be sure and you get the same results. Whatever this stuff from Twin Earth is, it’s not H 2 O.
What, then, should we say about whether the stuff from Twin Earth is water? In 1600, there was no observation that allowed us to distinguish samples of water from Earth from the liquid from Twin Earth. Now, elemental analysis provides a way to distinguish them. However, this is the only detectable difference between these to substances. Is this difference enough for us to tell whether the stuff from Twin Earth is water? One natural response might be to say, if these two substances have different elemental compositions – if they’re made of different stuff – then they must be different substances. So no matter how similarly the stuff from Twin Earth behaves to water, the fact that it is not H 2 O means it is not water. We know from the elemental analysis that water is H 2 O, and this is enough information for us to figure out that the stuff from Twin Earth is not water. But do we want to say “Water is H 2 O” is some sort of definition of water? It’s certainly not a definition that would have made any sense at all in 1600, and people then spoke of water without any confusion. So, at least in 1600, H 2 O was no part of the meaning of the word “water”. The central question may be whether the elemental analysis in 1785 necessarily changed the meaning of the word “water”. Because, I would argue, it might be equally reasonable to respond to the news that the stuff from the Seine is H 2 O and the stuff from Twin Earth is XYZ by saying, “Look, water has at least two different elemental compositions!” In other words, I’m claiming that we’re not forced into a certain conclusion by these experimental results. Rather, we have to make a choice .
On its own, each of these hypotheses seems reasonable. But, in light of the elemental analyses of the stuff from the Seine and the stuff from Twin Earth, we can’t commit to all of these hypotheses. [30] As a group, the hypotheses get us into trouble. The stuff from Twin Earth qualifies as water according to hypotheses 1 and 2, but not according to hypothesis 5. It seems like it would be pretty hard to keep hypothesis 6, no matter what we decide to call the stuff from Twin Earth. But if we get rid of hypothesis 6, then hypothesis 5 seems pretty arbitrary. Something’s got to give.
Here’s one way to update the group of hypotheses: [31] drop hypothesis 6 (which just doesn’t seem to fit with the experimental evidence) and hypothesis 5. The hypotheses that are left define water in terms of its macroscopic behavior (freezing, boiling, dissovling substances, etc.). While hypothesis 4 links water’s macroscopic behavior with what water is made of, it doesn’t specify that only a single sort of microscopic composition can produce the macroscopic behavior. Revising the group of hypotheses this way lets us say that the Twin Earth stuff is water, even though it has a different microscopic structure than Earth water.
Alternatively, [32] we could hold onto hypothesis 5 and replace hypotheses 1 through 3 with the definition “Water is H 2 O.” (Since hypothesis 6 is still at odds with the experimental results, it’s still out of the group.) This sort of updating reflects a decision that the microscopic structure matters more than the macroscopic behavior in identifying what is water and what is not. But it is crucial to notice here that [33] this way of updating the group of hypotheses is changing the meaning of “water”. So, Quine argues, maybe there are no analytic claims – no claims that are true or false based on meanings alone. Why? Because meanings get updated, and they get updated based on choices we make.
This is an illustration of meaning holism . What we mean by a term like “water” is tied up with all sorts of experiences we have with water and with other kinds of things. It depends on how narrowly or broadly we choose to draw our categories, what kinds of variations we can tolerate between two samples while still considering them “the same” stuff. And because meanings are connected to other like this, an experimental test could just as easily change what we mean by “water” as disprove our hypothesis that the stuff from Twin Earth is water. If you’re nervous about this — if it just seems like a bad idea to consider changing the meaning of a term that’s as basic as water — we need to consider the question of what exactly fixes the meaning of “water” in the first place. Here are two possibilities.
Let’s call the first possibility “baptism”. [34] We point at stuff in the world and say, “This stuff here is water.” I’m not committing to knowing much about what I’m naming. I’m just picking something out of the world – whatever it is – and giving it a name. But, it’s worth noting that when I point to a bucket of stuff from the Seine River and name it “water”, I probably mean to include more than just the stuff in this particular bucket. I probably want to include the other stuff that is relevantly similar – the rest of the stuff in the Seine plus the stuff in other lakes, rivers and streams.
But there’s another way I could fix the meaning of water: “description”. [35] Here, I’d say, “Water is whatever has these properties,” and then I’d list the properties. The properties I specify here will determine what counts as water and what doesn’t. (And, it’s entirely possible to list a set of properties that nothing has. For example, a descriptive definition of “unicorn” is a list of properties that no actual beast has).
Is the stuff from Twin Earth water? If the meaning of “water” is fixed by baptism, it will depend on what we point at as “water” and whether the Twin Earth stuff is similar enough. If the meaning of “water” is fixed by description, it will depend on what properties are part of the defining description. And, complicating matters, remember that science is trying to describe the stuff in our world. [36] In other words, baptism and description get intertwined with each other. So, scientists could go either way in classifying the stuff from Twin Earth as water or something else. Where does this get us? Quine wants us to notice the flexibility of our definitions and abandon the analytic/synthetic distinction. He says all our meanings are tied up together, and all of them are connected with our experience. So, there may not be any claims that are true based on meanings alone. The other big consequence of meaning holism is that an unexpected experimental result signals a problem in a group of “hypotheses” that includes definitions, too. In other words, I can deal with the bad prediction by changing a hypothesis. But, I might equally well deal with the bad prediction by changing a definition as well.
But here’s where Quine’s radical holism seems to throw the project of scientific testing into serious confusion. [37] For Quine says, “Any statement can be held to be true come what may, if we make drastic enough adjustments elsewhere in the system.” Regardless of the experimental outcome, we can hold a particular hypothesis true, as long as we change other hypotheses or definitions. So, if the hypothesis can be kept regardless of what happens, how is this a test?