11 Casting in Bold Relief
Il faut reculer pour mieux sauter.
Earlier I said that felicitous falsehoods figure in understanding by contributing to an account’s overall tenability. The question is how. It might seem that considerations that are acknowledged to be false cannot be initially tenable. That is not so obvious, since initial tenability turns on what people actually accept. That Amherst is ninety miles from Cambridge, although strictly false, is probably accepted and acceptable when considering driving from one city to the other. If so, it is initially tenable. People might well accept considerations they consider false, particularly if they consider them true enough. In any case, falsehoods that lack initial tenability can still in principle contribute to the tenability of the account they belong to. As we saw earlier, not all elements of an acceptable account need be initially tenable. Some initially nontenable or even untenable considerations earn their tenability by strengthening the tenability of a suitably tethered account. Thus, for example, early twentieth-century physics rightly accepted the positron because the alternatives—physical theories committed to the existence of negatively charged electrons with no positively charged counterparts—were less tenable than theories that postulated an as yet undetected positively charged particle. The effectiveness of such a strategy does not yet vindicate felicitous falsehoods, though. ‘There exist positrons’ was held to be true, although not yet independently confirmed.
How exactly can acknowledged falsehoods enhance tenability? In chapter 2, I characterized felicitous falsehoods as false in ways that do not matter. In this chapter I discuss laboratory experiments, thought experiments, and literary fictions. I argue that they all distance themselves from the facts in order to screen out irrelevant, potentially confounding complications. This enables them to exemplify important features and patterns that would otherwise be obscured. Once we recognize and appreciate the importance of those features and patterns in artificial settings, we can test whether they afford insight into what obtains in fact. Novels and thought experiments are clearly fictions. Some are felicitous. Whether laboratory experiments are fictions is not so clear. Whatever the verdict, they are like fictions in distancing themselves from the world in order to exemplify factors that obtain but are hard to discern or whose significance is hard to appreciate in their natural settings. Although laboratory experiments, thought experiments, and fictions are unfaithful to the facts, their infidelity does not discredit them, for their contribution to understanding lies in what they exemplify, not in what they describe.
Laboratory Experiments
Every discernible item has indefinitely many discernible properties and stands in indefinitely many discernible relations. Our only hope of understanding and coping with the chaos that confronts us is to ignore most of what is there to be seen. To a considerable extent, selective disregard is automatic. In order to see anything, we overlook a lot. Where automaticity fails, we purposely block things out. We scan the forest by purposely ignoring individual trees, or focus on a tree by screening off the rest of the forest. In some cases, though, more than an act of will is required to selectively disregard what we need to. Then we may resort to experimentation. If it is unfeasible to simply pay no attention to the man behind the curtain, we may contrive a situation that excludes him.
A laboratory experiment is no mere matter of bringing nature indoors. It is a controlled manipulation of events, designed and executed to make some particular phenomenon salient. Natural entities are multifaceted. Important properties and relations are often masked by the welter of complexities that embed them. In experimenting, a scientist isolates a phenomenon from many of the forces that ordinarily impinge on it. To the extent possible, she eliminates confounding factors. She holds most ineliminable factors fixed, effectively consigning them to the cognitive background of things to be taken for granted. This enables the effect of the experimental intervention on the remaining variable to stand out. Through such a strategy, she casts into bold relief factors that might typically be hidden from view.
Suppose a population of wild mice who were accidentally exposed to bisphenol-A later exhibited a high rate of liver cancer. To conclude that exposure to bisphenol-A caused their disease would be premature. For all we know, those mice might have been peculiarly susceptible to liver cancer or have been exposed to an unnoticed carcinogen. To glean direct, nonanecdotal evidence of a connection between exposure to bisphenol-A and liver cancer, investigators place genetically identical mice in otherwise identical environments, exposing half of them to massive doses of the chemical while leaving the rest unexposed. Because the assignment of a given mouse to one group or the other is random, the mice exposed to bisphenol-A and the mice in the control group are initially equal in expectation of developing liver cancer. Their common genetic endowment and otherwise identical environments neutralize a multitude of genetic and environmental factors believed to standardly influence the incidence of cancer. This blocks rival explanations that might be proposed for the elevated rate of cancer in the wild population.
The result of the experiment exemplifies the difference (if any) in the incidence of liver cancer between the two groups of mice. It not only instantiates the difference, it also makes that difference manifest. If the difference is statistically significant, then the result exemplifies a correlation between exposure to bisphenol-A and the incidence of liver cancer. (For simplicity, I will assume that the correlation, if there is one, indicates that exposed mice are more likely to develop liver cancer than unexposed mice. It could go the other way. Perhaps exposure to bisphenol-A protects against liver cancer.) Although correlation does not imply causation, a robust correlation often provides strong evidence of causation. In this case, the background assumption that moves us from a mere correlation to a causal judgment is the well-founded conviction that the experiment was so rigorously designed and executed that nothing but the exposure to bisphenol-A could have caused the difference. That being so, the result may also exemplify a causal relation.
So far we are just talking about the particular mice in the experiment. But the goal of the investigation is not primarily to discover their medical fates. It is to use their medical fates to learn something more general. Since the mice in the experiment were chosen arbitrarily from the class of mice with a particular genome, it is straightforward to project to other mice of the same strain. The experiment then also exemplifies the increased propensity of mice of that strain to develop liver cancer when exposed to bisphenol-A. Moreover, the mice are model organisms, so there is reason to think that what holds for them also holds for the organisms they serve as models for—in this case, mammals, including humans. The experiment’s result instantiates propensities to develop cancer in a variety of groups: the mice in the experiment, other mice of the same strain, mice in general, rodents, mammals. It also instantiates a vast and variegated host of other properties. But given the background assumptions that motivate and shape the inquiry, it exemplifies the propensity of mammals exposed to bisphenol-A to develop liver cancer. The inference from mice to mammals is a big jump. But if the background assumptions legitimating treating the mice as model organisms are cogent, it is reasonable to treat the experiment as exemplifying a causal connection between exposure to bisphenol-A and cancer in mammals in general.
The interpretive path I just sketched is reminiscent of Capdevila’s discussion of Rosso Galera, Rosso Guzzi. Here the connecting links derive from medical science rather than Italian motorcycle culture. The interpretation I’ve given is mediated by background assumptions drawn from our current best understanding of carcinogens, of mammals, and of methods for investigating carcinogenicity. That understanding may be wrong in ways that would vitiate my interpretation. So it is fallible. But if the understanding is sufficiently accurate and adequate (even if not true in every respect), the interpretation is too. If the exposed mice show a sufficiently elevated rate of liver cancer, we are right to think that bisphenol-A is carcinogenic.
What if we are wrong? Exemplification requires instantiation. Although we have no reason to think so, suppose there is a sharp threshold. Exposure below a certain level is causally inert. Above that level, cells go wild. Then the consequence of a small animal’s abrupt exposure to massive doses of bisphenol-A over a short period of time does not exemplify what happens to a large animal exposed to small doses over a long period of time. Given that humans are never exposed to levels or spikes in level comparable to those of the mice, the experiment does not exemplify a danger to humans. This could be so. An important mediating assumption might be false. In that case, the result does not exemplify anything about human vulnerability to cancer.
Still, the result exemplifies a connection that justifies our thinking that exposure to bisphenol-A increases the likelihood of developing cancer. Even if the result is misleading, it affords insight into the structure of our current understanding of the subject. The attribution of the difference in incidence of cancer to exposure to the chemical is reasonable to the extent that the scientists contrived a situation where rival explanations of the difference between the exposed mice and the control group have been blocked. The projection to other mammals is reasonable to the extent that the grounds for taking the mice to be model organisms are sound. Because the experiment takes place against a cluster of fallible background assumptions, it does not afford conclusive evidence. But because of its rigorous controls, it affords stronger and more direct evidence than a mere correlation between exposure and cancer in a wild population would.
Designing an experiment is setting a stage where events can play out. Conducting an experiment involves initiating and perhaps intervening in a course of events. Experiments are dynamic (Nersessian, 2007). They unfold as they do in part because of the scientist’s actions. She instigates and perhaps interrupts, deflects, impedes, or amplifies a natural sequence. She may isolate phenomena from their normal concomitants and introduce unusual provocations. The mice in the experiment are exposed to massive doses of bisphenol-A. This allows the scientist to obtain a pronounced effect in a relatively short time. Her working assumption is that a small mammal’s exposure to a large dose over a short period is equivalent to a large mammal’s exposure to a lower dose over a longer period. If that assumption is true enough, she can safely extrapolate from the laboratory situation to mammals in general, including humans.
Experiments often involve creating and using items that are nowhere to be found in nature (Cartwright, 1983; Hacking, 1982). Genetically identical mice are artifacts; their genetic makeup is designed to suit the sorts of experiments in which they will be used. Pure forms of chemicals are artifacts as well, being synthesized under carefully controlled conditions to avoid contamination. This can be crucial.
The Miller—Urey experiment begins with chemicals believed to be present on Earth in prebiotic times. The experiment consists of a sequence of chemical reactions whose ultimate output consists of new organic chemicals and amino acids. It thus shows how life could have emerged from nonliving matter (Ball, 2005). For the experiment to work, the chemicals—methane, ammonia, hydrogen, and water—had to be pure. Any hint of contamination would discredit the result. Moreover, to ensure that no organic material was accidentally introduced during the course of the experiment, the chemical processes had to be completely isolated from the environment. Although—indeed because—the experimental components and conditions were unnatural, the experiment revealed something important about the natural world. It did so not by saying that something is the case, but by showing it to be the case. It exemplified a path from four initial chemicals to amino acids.
The sequence of chemical reactions that took place in the Miller—Urey experiment may have been instantiated not only at the dawn of life, but innumerable times since. According to Goodman, any of those instantiations could exemplify the emergence of organic compounds from the four original chemicals. Just ignore all the other chemicals in the neighborhood; ignore the prevalence of oxygen and nitrogen in the atmosphere; concentrate on the reactions due entirely to hydrogen, methane, ammonia, water, and electricity. But such a natural instantiation of the sequence of reactions would be untrustworthy. The possibility of contamination would inevitably loom large. What distinguishes the experiment from other instances of the sequence is that it constitutes a context where it is manifest that nothing but the four chemicals and a bit of electricity was necessary to initiate the process. The experiment thus affords not just an instance, but a telling instance. By exemplifying that those chemicals and electricity suffice to generate new organic compounds, the experiment affords genuine insight into an aspect of nature.
An experiment unfolds over time. It has a narrative structure (Nersessian, 1993), with a well-defined beginning, middle, and end. Like a Greek tragedy, the experiment begins in medias res. The scientist conducts his investigation against a background of established findings and shared assumptions that frame the events and circumscribe their interpretation. The Miller—Urey experiment takes place against a cluster of background assumptions about chemical reactions. Its narrative arc starts with the enclosure of the pure chemicals in a carefully crafted, tightly sealed apparatus; it develops through the heating of the water and the emission of an occasional spark that mimics the effect of lightning; and it climaxes when the reactions are complete. The denouement consists in extracting the resulting solution and subjecting it to chromatography to determine its chemical composition.
Experimental results do not speak for themselves. They require interpretation. Their interpretation draws on background assumptions, beliefs about instrumentation, experimental design, the course of events that constitute the experiment, and the outcome. A change in any one of these factors can prompt revisions in others. Against the background of the theory of relativity, we interpret the Michelson—Morley experiment as affording evidence of the nonexistence of luminiferous ether; prior to the advent of the theory of relativity, it afforded evidence that the ether might not exist; long after most physicists accepted the theory of relativity, Morley took the experiment to demonstrate that his interferometer was not sensitive enough to measure ether drift, which he still believed was there to be measured.
These features of experiments are well known. I mention them to highlight how distant many scientific experiments and their results are from the natural phenomena they illuminate. The items experimented upon are often artifacts constructed expressly for experimentation (see Cartwright, 1983). The circumstances in which they are placed are artificial; they are carefully contrived situations, often ones that do not naturally occur but that are designed expressly to exemplify telling features of the phenomena. For an experiment to disclose something about a range of phenomena, it must exemplify features it shares with those phenomena. But it may, and in some cases must, diverge from the phenomena in other important respects.
‘Natural experiments’ take advantage of natural, or anyway unarranged, distributions that mimic the laboratory scientist’s random assignments to treatment and control groups. The subjects are populations that are suitably alike except for exposure to the factor under investigation. John Snow’s study of the London cholera epidemic can be seen as a natural experiment in that, as far as anyone can tell, but for their source of water, the people who used the Broad Street pump were suitably similar to the denizens of the workhouse, who used a different pump, and the workers in the brewery, who did not drink water. Those two groups served as controls, enabling Snow to identify the pump as the source of the epidemic and contaminated water as the cause of cholera. Natural experiments are welcome where it is immoral or unfeasible to conduct controlled experiments. But they are less trustworthy than laboratory experiments, for they rest on the optimistic conviction that the natural arrangement controls for all the plausible sources of variation between the two subpopulations.
For scientific results to be acceptable, they must be replicable. Clearly not every aspect of the original experiment must be reproduced. That would be impossible. Never again can an experiment instantiate being the first time a particular phenomenon was observed. Failing to replicate that feature is no flaw. Other features could be replicated, but there would be no epistemic value in doing so. Although the original experiment was conducted by a left-handed, squash-playing postdoc, it would be ludicrous to insist that only a left-handed, squash-playing postdoc could replicate it. Evidently, some features of an experiment are irrelevant and need not be replicated. But features that are irrelevant to one experiment may be relevant to another. Ordinarily, for example, the altitude at which an experiment is conducted is irrelevant. So many experiments originally performed in Mexico City can be replicated in New Orleans. But where altitude matters, an experiment performed in New Orleans would not qualify as a replication of one done in Mexico City.
Replication requires that subsequent experiments match the first in every respect that is pertinent to the hypothesis being tested. The second experiment must exemplify the same features as the original when both are interpreted as tests of the same hypotheses. This allows for considerable variation in the other features the various experiments instantiate. Indeed, such variation is desirable. If subsequent experiments instantiate the same unexemplified features as well as the same exemplified ones, their agreement might be misleading. Factors currently deemed irrelevant might in fact account for their common result. Instead of seeking to reproduce the original experiment as closely as possible, it is methodologically preferable to perform a variety of experiments that reproduce the features that the original exemplifies but differ from the original and from one another in those that are not exemplified (Elgin, 1988, 89).
Experiments are conducted; they do not just happen. They have a narrative structure. They are subject to interpretation and to reinterpretation if background assumptions change. They are repeatable. In short, they are close kin to dramatic enactments. This is not quite to say that experiments are works of fiction; but it is to suggest that the gulf between what we consider fact and what we consider fiction may be narrower than is typically supposed (see Cartwright 1983).
Nevertheless, one might think, there is a crucial difference. Unlike the ‘events’ in a novel or thought experiment, the events constituting an experiment actually occur. They are processes that real things really undergo. The real mice exposed to bisphenol-A really exhibited an elevated rate of liver cancer. The Miller–Urey experiment produced real amino acids from real chemicals. The worry is that, unlike a standard experiment, a mere simulation—be it a computer simulation, a thought experiment, or a work of literary fiction—might omit something significant. In that case, what occurs in the simulation is not representative of what occurs in reality. But the same holds when scientists move from the field to the lab. The controls introduced in the lab might inadvertently omit something significant. If so, the result does not directly project onto the phenomena it concerns. The trustworthiness of a laboratory experiment, a natural experiment, or a thought experiment depends on the adequacy of the background assumptions that frame it and the scrupulousness with which the experiment is designed and executed.
Thought Experiments
In laboratory experiments, scientists simplify, streamline, manipulate, and omit, so that the effects of potentially confounding factors are minimized, marginalized, or canceled out. An experiment deliberately departs from nature in order to advance an understanding of nature. Rather than invalidating the experiment, this departure is what enables it to disclose barely detectable or standardly overshadowed aspects of nature. Thought experiments involve further distancing. They are not actual, and often not even possible, experiments. They are imaginative exercises designed to disclose what would happen if certain, perhaps unrealizable, conditions were met.
Their reliance on imagination may give hard-nosed epistemologists pause. The imagination is, after all, free to entertain any ideas it likes. It is not bound to respect conceptual connections, evidence, laws of nature, or the dictates of common sense. Familiar scientific thought experiments violate all of these. But as Kant (1981) emphasizes, freedom is not lawlessness. Freedom consists in being bound by laws we set for ourselves—laws we reflectively endorse as reasonable and rational. As the locus of the free play of ideas, the imagination is not a realm in which ideas are utterly unconstrained, bouncing off one another like gas molecules in random motion. It is a realm in which the play of ideas is bound by constraints the imaginer sets. Although the constraints are self-imposed and vary from one imaginative setting to the next, they are real. The power of thought experiments to illuminate the facts lies in no small measure in the flexible, variable, but nonetheless binding character of the constraints that the imagination imposes on them. By setting such constraints and drawing out their consequences, the imagination serves as a laboratory of the mind, a venue in which hypotheses can be contrived, elaborated, and tested. Moreover, in scientific thought experiments, the constraints, even if tacit, are recognized, shared, and reflectively endorsed by a scientific community. Even so, how can a thought experiment claim to yield any insight into the facts? Why isn’t it simply an exercise in fantasy?
Unlike fantasies, thought experiments are not essentially private; nor are they particularly mental. Although they are imaginative exercises, they are publicly articulated, discussed, illustrated and disputed. They consist of verbal or pictorial representations. Their claim to be imaginative stems from the fact that, like works of fiction, they are typically not, and in any case need not be, representations of anything real. But the unreality of the objects that ostensibly figure in them does not undermine their function.
In designing a laboratory experiment, a scientist may begin by performing something like a thought experiment. She runs through the expected course of events in her head and perhaps describes it to her research team before attempting to implement it in the lab. This suggests that the difference between real experiments and thought experiments lies in the fact that thought experiments sharply truncate the experimental process. They omit the implementation step. But scientists can’t do this whenever they please. The question is when is stopping short legitimate?
Sometimes an actual experiment of the sort envisioned cannot be carried out. It is impossible or impracticable. As we saw, Einstein demonstrated the equivalence of gravitational and inertial mass via a thought experiment that could not be carried out. It would require sending an unconscious subject in a windowless enclosure into a region of deep space distant from any significant source of gravity, rousing him, and querying him about his experiences. This is morally, practically, and physically unfeasible. Still, the recognition that we cannot do a real experiment does not by itself legitimate the results of stopping short. Sometimes, the infeasibility of an experiment translates into the infeasibility of finding out a particular fact. The reason Einstein’s thought experiment is effective is that it takes the form of a challenge: Suppose the specified conditions were met. How could a subject tell whether he was in one situation or the other? If our best efforts to identify a way to tell the difference fail, and fail for scientifically principled reasons, we have evidence of the equivalence. Our failure indicates that, if our theories are close to correct, there is no difference to detect.
Sometimes the imaginative rehearsal reveals that an actual experiment need not be carried out. The mental run-through itself discloses the relevant information. Without physical implementation, Galileo’s thought experiment discredits the Aristotelian contention that the rate at which bodies fall is proportional to their weight. According to Aristotle, a boulder should fall more quickly than a pebble, assuming they are composed of the same material. What, Galileo wonders, are the implications for a composite object? He imagines an object consisting of a boulder tethered to a pebble. Being composed of two rocks and some rope, the composite object is heavier than either rock alone. If Aristotle is right, it should fall more quickly than the boulder. But since, according to Aristotle, the pebble falls more slowly than the boulder, once the two are roped together, the pebble should retard the boulder’s fall. It should serve as a brake. Hence the rate at which the composite object falls should be between that of the boulder and that of the pebble. The composite object cannot fall both more quickly and more slowly than the boulder, so the Aristotelian commitments are inconsistent. By exemplifying the inconsistency, Galileo’s thought experiment demonstrates that the Aristotelian account cannot be correct.
One might argue that Galileo’s thought experiment discredits my analysis.1 Exemplification, I said, requires instantiation. Real chemical reactions occur in the Miller–Urey experiment. So it is plausible that by exemplifying those reactions, the experiment affords epistemic access to them, enabling us to recognize them and appreciate their significance, not only in the experimental setting but also outside of it. In a mere thought experiment, however, no physical changes actually occur. A thought experiment, not being material, cannot exemplify material properties. This is so. The sequence of ideas that constitutes Galileo’s thought experiment does not instantiate material properties of falling bodies. There is, for example, no danger of being hit on the head by the falling rocks. But the rate at which bodies fall and the independence of that rate from the weight of those bodies are abstract, mathematical properties. They can be instantiated by material and immaterial sequences alike. There is then no bar to saying that via exemplification thought experiments afford epistemic access to abstract properties that are in some cases instantiated in material objects. A thought experiment is a representation—a re-presentation—of abstract features, an imaginative re-embodiment of them. We are to imagine—that is mentally, verbally or pictorially present—a situation where the abstract features are realized. In effect, we are to investigate what would happen in a virtual reality where certain constraints are supposed to hold.2
Philosophers sometimes think that we resort to thought experiments only when, for one reason or another, a real experiment cannot be carried out. Perhaps Galileo could not have conducted a real experiment to conclusively demonstrate his point. Maybe he did not have sufficiently accurate timers or a high enough tower from which to run the test. Maybe he did not have the resources to eliminate the effects of air resistance, and so on. Now, however, we could conduct the experiment. Shouldn’t we? Probably not. Rather than concluding that the thought experiment was a second-best strategy resorted to because of circumstances beyond the scientist’s control, we should recognize that a real experiment would not have made Galileo’s case any more forcefully than his thought experiment did. It would simply have muddied the waters. Once we start dropping objects from towers, we face the problem that cancer-ridden wild mice pose for biologists and the emergence of amino acids in nonisolated situations pose for chemists. How do we know that unrecognized confounding factors do not explain our finding? By deploying an austere thought experiment where the distance and duration of the fall, the presence or absence of air resistance, and a host of other possible sources of interference are simply omitted, Galileo blocks such challenges. The thought experiment demonstrates an inconsistency in the Aristotelian position—an inconsistency that would obtain regardless of the conditions under which the experiment was conducted. Just as a laboratory experiment is preferable to a ‘natural experiment’, the thought experiment is preferable to an actual experiment, because it is invulnerable to a host of potentially misleading challenges that an actual experiment would face.
Even in the empirical sciences, not every question can or need be answered by direct appeal to observational evidence. Thought experiments are often appropriate where observation is not apt. This leads some to conclude that thought experiments are effective only where the issues concern conceptual or theoretical commitments. If so, thought experiments disclose something about our concepts and our theories, not something about the world. Since we can tease out commitments in the armchair, it is no surprise that mere thought experiments are effective for revealing them. But to conclude that thought experiments do not reveal anything about the way the world is would be too hasty. Galileo’s thought experiment did not just exemplify an inconsistency in Aristotle’s theory. It also showed that any theory that took the rate at which bodies fall to depend on their weight would be inconsistent. From this it follows that the rate at which bodies fall is independent of their weight. The thought experiment thus exemplifies a feature of reality, not just of theoretical or conceptual commitments. That feature is, or is a consequence of, a modal fact. The rate at which objects fall is independent of their weight because they could not fall any other way. Thought experiments, it seems, afford epistemic access at least to theoretical, conceptual, and modal matters.
A thought experiment fixes certain parameters (e.g., about the relevant laws of nature and the supposed initial conditions), provides a description of the experimental situation that sets out all and only the features considered relevant, and works out the consequences. Galileo’s two rocks are assumed to be made of the same material and to be the same shape, thereby obviating the effects of material constitution and shape on the outcome. They are assumed to be falling through the same medium. Weight is assumed to be additive, and a tethered material object is assumed to be subject to the same laws of nature as untethered ones (Gendler, 2010). In effect, the thought experiment invites us to consider what would happen if certain conditions, some expressly specified and some tacitly assumed, obtain.
Like literary fictions and ordinary experiments, thought experiments have a narrative structure. We perform thought experiments by imagining a scenario in which something happens—a sequence of events with a beginning, middle, and end. Thought experiments can be construed as tightly constrained, highly focused, minimalist fictions, like some of the works of Borges. If Borges’s minimalist stories are genuine fictions, there seems no reason to deny that thought experiments are too.
To understand a thought experiment requires a suspension of disbelief. We grant its (tacit and explicit) assumptions even though we know that they do not—and in some cases cannot—obtain. Maxwell’s demon provides a telling instance. According to the second law of thermodynamics, when ensembles of gases at different temperatures are brought into contact with one another and isolated from the wider environment, they will evolve to thermodynamic equilibrium. That is, the resulting ensemble will reach a stable, uniform temperature between the two original temperatures. Maxwell’s question is whether this law is inviolable. To answer it, he mounts this thought experiment. Consider a vessel of air at uniform temperature. Its individual air molecules move at different velocities, even though the mean velocity of any sufficiently large arbitrary collection of those molecules is almost exactly uniform. Divide the vessel into two sections, A and B. Now imagine a demon capable of seeing and tracking individual molecules. He monitors a door between A and B, allowing only fast molecules to move from A to B and only slow molecules to move from B to A. Eventually most of the fast molecules will be in B and most of the slow ones will be in A. Simply by controlling access across the divide, the demon will raise the temperature of B and lower the temperature of A, thereby violating the second law of thermodynamics. Although we know full well that no one is, or arguably could be, endowed with the abilities ascribed to Maxwell’s demon, we bracket that knowledge and see whether such a being could defy entropy. Considerable ingenuity may be required to figure out which beliefs should be suspended and which ones should be retained. This is why scientific thought experiments are embedded in theoretical discussions that fix their parameters. And it is why their implications are subject to dispute. Still, we are inclined to agree with Maxwell that the demon thought experiment shows that the second law of thermodynamics is in principle violable. It is a merely statistical law.
Like literary works and ordinary experiments, thought experiments require interpretation. Sometimes interpretations diverge. The Einstein–Podolsky–Rosen experiment is a case in point. Very roughly, the scenario is this: Two particles interact, then fly off in opposite directions. Once they are separated, the measurement of one should have no effect on the state of the other. But if we measure, say, the position of one and apply the Schrödinger equation, we can determine that the other also has a definite position. This seems to violate the uncertainty principle: being unexamined, the second particle should have no definite position. What does the thought experiment show? The answer is not at all clear. Apparently, even the authors disagreed. Podolsky thought it demonstrated that quantum mechanics is incomplete, while Einstein thought it showed either that quantum mechanics is incomplete or that states of spatially separated objects are not independent of each other (Bokulich, 2001).
To recap: a thought experiment is an imaginative exercise designed to investigate what would happen if certain conditions were satisfied. Conducting it requires a suspension of belief, in that the conditions imagined are not realized in fact, and may be inconsistent with conditions we know to obtain in fact. It requires a suspension of disbelief, in that it asks us to entertain scenarios that we know do not and often could not obtain. It depends on background assumptions about what commitments are to be retained, what commitments are to be relaxed, and what commitments are to be abandoned in entertaining the imaginative scenario. Whether the constraints are tacit or explicit, in conducting the thought experiment the epistemic agent is bound by them. A thought experiment has a narrative structure, with a beginning, middle, and end. It is subject to interpretation, and to reinterpretation if the background assumptions change. Schrödinger’s cat, originally introduced to criticize the Copenhagen interpretation, now appears in every interpretation of quantum mechanics, each offering its own account of the poor beast’s state. Finally, we saw that thought experiments are valuable in investigating what is not open to direct empirical inspection—conceptual or theoretical commitments and their consequences, as well as modal properties, and, I would add, relations of cotenability on noncotenability.
Fictions
The thought experiments I have mentioned so far have been drawn from the physical sciences. But thought experiments are ubiquitous in philosophy as well. Like scientific thought experiments, those in philosophy illuminate factors that are not readily accessible to direct inspection. Because of a difference in subject matter, the range of factors illuminated by philosophical thought experiments is broader. Familiar philosophical thought experiments afford insights into normative properties (trolley problems, the experience machine), introspectively available properties (Mary, brains in a vat), and metaphysical properties (fission and fusion in personal identity, the ship of Theseus). No more than conceptual, theoretical, and modal properties, are these open to direct empirical inspection.
Like scientific thought experiments, many philosophical thought experiments are fairly austere. But philosophical thought experiments often are not so tightly bound by theoretical constraints as scientific thought experiments tend to be. Thus there is more controversy over what we should conclude from them. (Confession: I have no idea what the Chinese Room shows.) Many are relatively autonomous. We can fruitfully entertain them against the background of multiple sets of philosophical assumptions—indeed without even being aware that we are making philosophical assumptions. The trolley problem was introduced to disclose a consequence of the doctrine of double effect. It now has a life of its own (and seemingly innumerable children and grandchildren).
If an austere thought experiment can afford epistemic access to a range of properties, and can do so in a context that is not tightly beholden to a particular theory, there seems to be no reason to deny that a more extensive thought experiment can do the same. Indeed, philosophy provides examples. Plato’s Republic and Rousseau’s Emile are cases in point. This opens the way to construing works of literary fiction as extended, elaborate thought experiments. They afford epistemic access to aspects of the world that are normally inaccessible—in particular, to the normative, psychological, and metaphysical aspects that philosophical thought experiments concern.
Again one might worry about the capacity of exemplification to account for this. A work of fiction, not being alive, cannot instantiate psychological or moral properties. If it cannot instantiate them, it cannot exemplify them. But it can instantiate and exemplify abstract properties that are concretized in human agents. Suppose Meg instantiates a pattern of psychological features—a network of beliefs, desires, aversions, and preferences, for example. Although the specific elements of her network are psychological, the pattern is abstract. In principle, it can be instantiated by something other than psychological elements. A fiction writer might create a scenario where that pattern is instantiated and exemplified via a sequence of descriptions. In effect, she takes a pattern that is embodied in fact, abstracts it, and re-embodies it in fiction. (Or, more likely, she abstracts individual elements instantiated in fact, finds or devises an appropriate pattern, and embodies that pattern in fiction.) Strictly, in the fictional setting it is not a pattern of psychological features. But it is a pattern that is, or that may be, instantiated by psychological features. So it affords epistemic access to a pattern that we may find ourselves or our fellows instantiating.
Like an experiment, a work of fiction selects and isolates, manipulating circumstances so that particular properties, patterns, and connections, as well as disparities and irregularities, are brought to the fore. It may localize and isolate factors that underlie or are interwoven into everyday life or natural events, but that are apt to pass unnoticed because other factors typically overshadow them. This is why Jane Austen maintained that “three or four families in a country village is the very thing to work on” (1814). The relations among the three or four families are sufficiently complicated and the demands of village life sufficiently mundane that the story can exemplify something worth noting about ordinary life and the development of moral character. By restricting her attention to three or four families, Austen in effect devises a tightly controlled thought experiment. Drastically limiting the factors that affect her protagonists enables her to elaborate the consequences of the relatively few that remain.
If our interests are cognitive, though, it might seem that this detour through fiction is both unnecessary and unwise. Instead of resorting to fiction, wouldn’t it be cognitively preferable to study three or four real families in a real country village? Probably not, if we want to glean the insights that Austen’s novels afford. Even three or four families in a relatively isolated country village are affected by far too many factors for the social and moral trajectories that Austen’s novels exemplify to be salient in their interactions. Too many forces impinge on them and too many descriptions are available for characterizing their interactions. Any such sociological study would be vulnerable to the charge that unexamined factors played a nonnegligible role in the interactions studied, that other forces were significant. Austen evades that worry. She omits such factors from her account and in effect asks: suppose we leave them out, then what would we see? Similarly, the model pendulum omits friction and air resistance, allowing the scientist in effect to ask: suppose we leave them out, then what would we see?
The question is how what we see in the fiction is supposed to inform our understanding of reality. That Elizabeth Bennet and Mr. Darcy, who do not exist, are said to behave thus and so does not demonstrate anything about how real people really behave. That an idealized pendulum, which also does not exist, is said to behave thus and so does not demonstrate anything about how actual pendulums behave.
Recall the paint company’s sample cards discussed earlier. Most people speak of them, and probably think of them, as samples of paint—the sort of stuff you use to paint the porch. They are not. The cards are infused with inks or dyes that are the same color as the paints whose colors they exemplify. It is a fiction that they are samples of paint. But since the sole function of such a card is to convey the paint’s color, the fiction is no lie. All that is needed is something that is the same color as the paint. A fiction thus conveys the property we are interested in because, in the respect that matters, it is no different from an actual instance. The exemplars on the paint cards need not themselves be paint. Similarly in literary or scientific cases. If the sole epistemic objective is to exemplify particular properties, then in a suitable context, any symbol that exemplifies those properties will do. If a fiction exemplifies the properties more clearly, simply, or effectively than a strictly factual representation, it is to be preferred to the factual representation.
In Nicomachean Ethics, Aristotle (1985) suggests that we should call no man happy until he is dead. Initially this seems implausible. As is well known, part of the problem is that what Aristotle means by ‘happy’ is not what we mean. ‘Flourishing’ would be a better term. But ‘call no man flourishing until he is dead’, even if not quite so implausible, still seems a bit extreme. Surely, one wants to object, we can easily discern that some of our fellows are currently flourishing. Aristotle defends his idea by contending that severe enough reversals of fortune late in life would justify the conclusion that a man’s life had not been a happy (or flourishing) one. Maybe so. But even if someone suffered serious misfortunes late in life, it is tempting to object, ‘Well he was happy (or flourishing) up until then’. Aristotle’s own example of Priam, the elderly king of Troy, invites this response. Until the Trojan War, Priam prospered. His life ended in misfortune; but throughout most of it, he apparently flourished. This is the objection students typically raise, and if one just reads Aristotle, it does not seem unreasonable.
Oedipus Rex can be read as a thought experiment that vindicates Aristotle’s claim. For most of his life, Oedipus, like Priam, seemed blessed with the gifts of fortune; and as far as anyone could tell, he lived a life of Aristotelian virtue. Evidently, he flourished and deserved to flourish. When Thebes suffered a plague, which the oracle blamed on him, Oedipus discovered that he had unwittingly killed his father and married his mother. This discovery did not just doom his future happiness, as the sack of Troy doomed Priam’s. It discredited his past happiness. He had, through no fault of his own, been living a lie. We might exonerate him for the wrongs he had done, since he acted out of nonculpable ignorance. But even if he was blameless, his relations to himself, to his wife/mother, to his children/siblings, to the citizens of Thebes who suffered for his iniquities, and to his own past, were forever changed, and would henceforth be tinged with revulsion. It turns out that he had not been flourishing during the early years, even though he and everyone else thought that he was.
Oedipus Rex is a work of fiction that advances our understanding not only of Aristotle’s ethics, but also of the human predicament. It underscores the limits on human knowledge and the vulnerabilities that stem from those limits; the value of knowing oneself and one’s situation, and the limits on the human capacity to do so. It does not constitute a proof that Aristotle is right, but it poses a challenge: you should be wary of calling a man flourishing during his lifetime unless you are sure his situation is not like Oedipus’s. It is hard to see how the challenge can be met.
Kant (1981) maintains that not only is it impossible to know whether someone else has acted morally, it is impossible to know whether you yourself have done so. For any action that accords with the categorical imperative, there is always an available self-interested maxim that might have been the real motive. So it is impossible to glean unequivocal empirical evidence for moral action. But even if we can never discern that someone has behaved morally (that is, acted on account of respect for the moral law), an author can effectively stipulate it. In A Tale of Two Cities Dickens portrays Sidney Carton’s self-sacrifice as stemming from purely moral motives. An author can portray a situation, develop a character, and convey his thoughts, feelings, and reasons for acting, thereby blocking the explanation from self-interest. He can, that is, exemplify a pattern and demonstrate that were that pattern to be instantiated, it would be an instance of acting morally.
If Kant is right about the impossibility of finding unequivocal instances of acting morally, how do we learn what a moral action is? Even if the categorical imperative is a deliverance of reason, we need to know how to apply it. Non-Kantians face a similar problem. Everyone recognizes that it is difficult in practice to distinguish acting morally from acting out of self-interest. If I am right, fiction can play a major role in moral education. By exemplifying genuinely moral patterns of deliberation and action, it can teach us something that is at least hard, and might be impossible, to reliably discern in real-life interactions.
People have inner lives replete with motivations, perceptions, emotions, and thoughts. Because a variety of combinations of psychological elements might yield the same outward behavior, it is impossible to uniquely determine the underlying psychological states from observations of overt behavior alone. Moreover, people often misunderstand themselves, and often have reason to mask what they think and feel. So even if we ask them and they tell us, we ought not be confident that we know how things are with them. This raises the question: how do we learn (indeed, what even leads us to suspect) that other people have inner lives that are quite unlike ours? Their feelings may not only be unlike what we actually feel about things, but may also be unlike what we would feel if we were in their place. It is one thing to be able to imagine oneself having had someone else’s experiences. That is hard enough. But we can do things that require yet more imaginative dexterity. Even if Jim recognizes that had he been treated as Jane was, he would be bitter, he may also understand why she is willing to let bygones be bygones. This is an amazing cognitive achievement. I suggest that fiction plays a major role in equipping us for this kind of feat. In reading a work of fiction we take up a point of view and try it on for size. In effect, we experiment with the perspective and see how things look from there. Many works portray the world through a protagonist’s eyes, conveying her experiences, feelings, and thoughts. They disclose the limitations of her perspective. Some do more. A work may afford multiple perspectives on the same series of events, disclosing the resources and limitations of each. In effect, each filters events through a different sieve.
Philosophers as well as nonphilosophers have a tendency to take the fruits of introspection at face value, or at least to grant them a higher epistemic status than outsiders’ opinions about a subject’s state of mind. A work like Lolita, written from the perspective of an utterly unreliable narrator, affords insight into the limits of introspection. A character’s perspective can be so skewed or benighted that he is simply wrong about the central events of his life. Such a work can be construed as a thought experiment that undermines the conviction that a person’s access to his own motives, beliefs, and other attitudes always affords better evidence than the evidence that one’s words and actions afford to others. Perhaps there is privileged access in the sense that each of us knows herself in a way that she knows no one else. But self-deceptive fictional characters undermine the conviction that we always know ourselves better than others know us.
Cavell maintains that the problem of other minds is not, or not only, the problem of ascertaining whether an entity has a mind, but the problem of figuring out what is on someone’s mind. Even if I am confident that an individual has beliefs, desires, preferences, and feelings, I am woefully underequipped to identify those fine-grained mental states. The problem is not just a problem about other minds, though. I may be equally under-equipped to know what my mental states are. Do I really expect I’ll get the paper done, or have I managed to deceive myself into taking a hope for an expectation? Do I really want the promotion, or do I just think I want it because I know that it is the sort of thing that people in my position are supposed to want? As Cavell (1987) reads Shakespeare’s tragedies, they afford evidence of the uncertainty of mental state ascriptions. For the same sorts of reasons that Lear cannot recognize Cordelia’s devotion, that Othello cannot recognize Iago’s malevolence or Desdemona’s fidelity, that Hamlet cannot trust his judgment, we cannot be sure of the mental states that we ascribe.
A work of literature can also function as something akin to an impossibility proof—a thought experiment that exemplifies the inadequacy of its own grounding assumptions. Davenport (1983) reads Middlemarch as a thought experiment about marriage. Both Dorothea Brooke and Dr. Lydgate are in deeply unhappy marriages. Because in the world of the novel divorce is unthinkable (and unthought of), they are doomed to serve life sentences for their unwise choices of mates. By exemplifying the intractability of the problem they face, the novel provides reason to think that divorce, or something like it, should be an option. Davenport considers Middlemarch flawed because it does not allow for the possibility of divorce. I disagree. I consider it a powerful thought experiment that reveals the unacceptable consequences of institutional arrangements that do not allow for exit visas.
Metaphysical thought experiments are often science fictional. Some are so austere that in their philosophical settings we do not know what to think. Literary and cinematic fictions help us out. What should we make of Putnam’s brains in a vat? The Matrix supplies an answer. What would a computer that passed the Turing test be like? His name is Hal. Could beings without inner lives actually be indistinguishable from us? The way to settle such matters (even tentatively and revisably) is to design a scenario in which the consequences of such hypotheses play out. Write a story about the love lives of zombies, or about the lives of zombies incapable of love. We may find that our off-the-cuff intuitions do not stand up under elaboration or that the consequences of our assumptions are quite different from what the austere philosophical thought experiments led us to suppose.
Despite what I have said, the idea that fictions function as thought experiments that afford insight into actual human experience may seem a stretch. I do not think that it is. An example shows why. One of the mysteries of the Penn State pedophilia scandal is how Joe Paterno, the longtime football coach whom many considered a bastion of integrity, could have turned a blind eye to the actions of his assistant, Jerry Sandusky. Sportswriter Thomas Boswell ventures the following answer:
Everybody has weak spots in their character, fault lines where the right earthquake at the wrong time can lead to personal catastrophe. Most of us are fortunate that our worst experience doesn’t hit us with its biggest jolt in exactly the areas where our flaws or poor judgment or vanity is most dangerously in play. It’s part good luck if we don’t disgrace ourselves. But when it does happen, as appears to be the case with Joe Paterno, that’s when we witness personal disasters that seem so painful and, in the context of a well-lived life, so unfair that we feel a deep sadness even as we simultaneously realize that the person at the center of the storm can never avoid full accountability.… Forces collide, conspire, confuse, and an icon of integrity fails to act, fails to see. (2011)
If this sounds familiar to those who do not read the sports pages, it is because the passage is a precis of Oedipus Rex as filtered through Aristotle’s Poetics (1973).3 The great man, beset by hubris, does terrible things and is brought down by his tragic flaw. One can quarrel with Aristotle’s reading of Sophocles or with Boswell’s implicit endorsement of that reading. One can doubt that Paterno was the man of integrity he was alleged to be. One can even think that ignoring rampant pedophilia is vastly worse than inadvertent patricide and incest among consenting adults. Still, Oedipus Rex provides a template for understanding Paterno. Having seen the pattern in fiction, we are in a position to entertain the possibility that it affords an understanding of what happened in fact.
One objection to the thesis that fictions function epistemically is that they provide no evidence about the way the world is (Carroll, 2002).4 Davies (2010) contends that my account is vulnerable to this objection. Whether he is correct depends on what it is to provide evidence. According to David Lewis,
We who have lived in the world for a while have plenty of evidence, but we may not have learned as much from it as we could have done. This evidence bears on a certain proposition. If only that proposition is formulated, straightway it will be apparent that we have very good evidence for it.… If we are given a fiction such that the proposition is obviously true in it, we are led to ask: and is it also true simpliciter? And sometimes, when we have plenty of unappreciated evidence, to ask the question is to know the answer. (1983, 279)
Like Lewis and Davies, I maintain that fictions equip us to recognize and marshal information we already have. This enables us to treat that information as evidence. Whether this amounts to providing evidence or merely to providing access to evidence we already have makes no difference. Either way, they advance understanding of the actual world.
Fictions also exemplify abstract features and patterns, affording direct evidence of them, and indirect evidence that their real-world instantiation is possible (Elgin, 1996). Davies maintains that this feature makes fiction a source of hypotheses, rather than evidence of a truth. Perhaps so. But fictions do not just generate hypotheses. By exemplifying features and patterns that, if instantiated, are significant, they also provide reasons to think that the hypotheses are worth taking seriously. Some scientific simulations do the same. I see no reason to exclude symbols that play this role from the epistemic realm. If a work of fiction can afford reason to suspect that a pattern obtains in fact, or that discovering whether it obtains would be epistemically worthwhile, the fiction functions epistemically, whether or not we cash out ‘reason to believe’ in terms of evidence. A fiction can then render a hypothesis initially tenable.
Not every falsehood is felicitous. So the fact that we can write a fiction that exemplifies a pattern does not ensure that the pattern can be projected onto the world. “The Elephant’s Child” is a fictional tale purporting to explain how the elephant got its trunk (Kipling, 1978). Once, the story goes, elephants had ordinary noses. But an insatiably curious elephant’s child wanted to know what crocodiles eat for breakfast. He bent down to ask a crocodile. The crocodile, smacking his lips and thinking that young elephant would be a delicious answer to that very question, bit the youthful elephant’s nose and tugged on it. As the elephant (with the help of a friendly python) pulled back, its nose elongated until it took the shape of a trunk. That basically is the story. What is wrong with it? As a children’s story, perhaps nothing. But there is surely something wrong with it as a contribution to the understanding (even a child’s understanding) of biology. Since I contend that science can incorporate fictions and falsehoods, I am in no position to say, as others would, that what is wrong with the story is that it is false. So is the ideal gas law and the thought experiment about Maxwell’s demon. The difference is that while the ideal gas law and Maxwell’s demon afford some understanding of the behavior of actual gases, Kipling’s story affords no understanding of the phylogenesis of elephant trunks. If the story, with its tacit commitment to the heritability of acquired traits, were true, it might yield some understanding of how the elephant got its trunk. But once we acknowledge that both it and its Lamarckian presuppositions are false, the story seems to have nothing to do with the phylogenesis of elephant trunks. We need to appeal to natural selection for an explanation, and that explanation will be entirely unsympathetic to any tugging-on-noses hypothesis. Kipling’s story, I suggest, is neither directly nor indirectly suitably tethered to the phylogenic facts it purportedly bears on.
Experiments yield evidence, not proof. And evidence is sometimes misleading. The status of an item as a laboratory experiment or thought experiment does not hinge on its being successful in advancing understanding. I do not claim that every work of fiction succeeds. Nor do I claim that every laboratory experiment or thought experiment does. Some are muddled or confused. Some overlook real possibilities or fail to control for important variables. Some replicate what is already known or widely accepted. Some are trivial. Some thought experiments and literary works may even be the equivalent of high school science experiments, exemplifying what is already understood in an effort to show how such symbols advance understanding. Any finding must be tested by its fit with what we already have reason to believe. Galileo’s thought experiment revealed an inconsistency in Aristotle’s theory. It is, or is very close to, a crucial experiment. We need nothing further to show that Aristotle’s theory is false. But most experiments and thought experiments, and most works of literature, work within a context of background assumptions. If the assumptions are incorrect or incomplete, an experiment or thought experiment may inherit and reinforce their inadequacies. If they are (close enough to) correct and complete, the experimental result is prima facie informative.
Still, we might resist identifying works of fiction with thought experiments. The argument from banality is the contention that the knowledge imparted by fictions amounts to little more than truisms (Carroll, 2002).5 The guiding idea seems to be that if fiction has an epistemic function, it is to impart truths like the morals of Aesop’s fables. These truths are inferred inductively or deductively from stories. Since such truths are banal, they are, for the most part at least, epistemically inert. But the patterns and features that works of fiction exemplify are far from truisms. Because exemplars display rather than merely state, they can be exceedingly fine-grained. Hitchcock’s North by Northwest exemplifies delicate nuances in the texture of fear—a virtual continuum from trepidation to terror. There is nothing banal about learning to recognize subtle differences and project them properly onto members of the classes the exemplars typify. Nor is it likely that we will be able to capture in a pithy proposition just what such a fiction discloses.
Davies worries that I ignore a fundamental difference between literary fictions and thought experiments. Thought experiments are part of science; so to project their results onto the phenomena they pertain to is to remain within the realm of science. But to project insights drawn from fiction onto actual human relations is to move beyond literature to an extraliterary realm. This, Davies charges, is illicit, for such projection is not “intrinsic to the proper engagement of literary works as literature” (2010, 65). Such policing of disciplinary boundaries strikes me as ill advised. If, as I have argued, works of fiction advance understanding in much the way that thought experiments do, it makes no difference whether that advance counts as literary or extraliterary.
That a work is a rich source of insights is not a reason to doubt that it advances understanding. But it may be a reason to doubt that the work is a thought experiment. Stereotypical thought experiments tend to be austere. They symbolize along relatively few dimensions. Although they require interpretation, the interpretation of any given thought experiment is supposed to be univocal, at least until the relevant background assumptions change. But univocality is not a virtue in literary or dramatic fictions. That Henry V can be interpreted as pro-war and antiwar is not a defect in the play. Literary works are semantically dense and replete (Goodman, 1968). A work is semantically dense in that the characters, events, perspectives, and their alternatives are selected from a dense field of alternatives. There is no limit to the factors, modes of description, or level of generality it could have deployed. Nor are there general rules, like the rules about statistical significance, that readily carry over from one novel to the next. In principle, every semantic difference can make a difference to what the work conveys. And a feature that makes a difference to the interpretation of one work may be irrelevant to the interpretation of another. A work is replete to the extent that it symbolizes along multiple dimensions. Syntax, vocabulary, perspective, tone, and timbre may all function symbolically. Even punctuation can matter. Any number of aspects of a fictional work can in principle contribute to its epistemic functioning. Moreover, works of fiction are apt to bear multiple correct interpretations. In several respects, then, they differ from austere, univocal thought experiments. Perhaps this is a reason to deny that works of fiction are thought experiments; perhaps it is a reason to say that only under an interpretation is a work of fiction a thought experiment; perhaps it is a reason to think that some thought experiments are more austere than others.
I favor the last alternative. Although stereotypical thought experiments are austere, there is a continuum of cases from Maxwell’s demon and trolley problems through the myth of the cave and Emile to ‘didactic fictions’ like Animal Farm and Uncle Tom’s Cabin, to Middlemarch and Oedipus Rex. I doubt that there is a sharp boundary between thought experiments, strictly so called, and works of fiction. But demarcating the boundary is not so important. Whether or not we call works of fiction thought experiments, I have urged that fictions, thought experiments, and laboratory experiments function in much the same way. By distancing themselves from the facts, by resorting to artifices, by bracketing a variety of things known to be true, all three exemplify features they share with the facts. Since these features may be difficult or impossible to discern in our everyday encounters with things, fictions, thought experiments, and laboratory experiments advance our understanding of the world and of ourselves.
In chapter 2 I said that despite or even because of their inaccuracy, felicitous falsehoods embody and advance understanding. Pretty clearly, thought experiments are felicitous falsehoods if any symbols are. In this chapter I have argued that fictions are, or are closely akin to, thought experiments. If so, they too can be felicitous falsehoods. I do not contend that every fiction or every thought experiment is felicitous. Some are banal. They exemplify features that are obvious and uncontroversial anyway. Some are muddled. They exemplify nothing more than confusions that went into their design. They defy consistent interpretation and thus disclose nothing beyond their own incoherence. Some, such as racist works, mislead. They exemplify features that readers wrongly take to obtain in fact. Such falsehoods are infelicitous.
Others, however, may be felicitous in some contexts or for some audiences or for some purposes but not for others. Einstein’s elevator exemplifies the equivalence of inertial and gravitational frames of reference. It operates in a context where a good deal of physical theory is presupposed. It is thus felicitous for those at home in such a context. It can advance their understanding of inertia and gravity. But it is unlikely to do the same for novices or those who consider themselves novices. They have no reason to think that their inability to distinguish between being at rest in a gravitational field and moving with uniform acceleration in the absence of a gravitational field is due to anything other than their ignorance. Uncle Tom’s Cabin now strikes readers as exemplifying objectionable stereotypes and obvious moral truths. If it is not misleading, it is epistemically inert. When it was first published, however, its exemplification of the plights of the slaves was, for its intended audience—white Northern readers—revelatory. Now it is infelicitous; then it was felicitous. Even “The Elephant’s Child” may deserve a closer look. Interpreted as exemplifying something about phylogeny, it is infelicitous. Interpreted as a work about the importance of having strong friends (remember the python) or about the perils of nosiness, it may be, for young children, felicitous. The felicity or infelicity of a representation turns on its function. This is so whether the representation is a free-standing one, like a work of literature, or is embedded in a scientific, philosophical, or other systematic account.6
Notes
1. I am grateful to Georg Brun and Christoph Baumberger for raising this objection.
2. This is consonant with Platonism, but does not require it. Perhaps abstract properties and patterns exist only if instantiated, but instantiations, whether material or virtual, can be created or can naturally emerge.
3. I originally found reference to Boswell’s column in Snyder (2012), 55.
4. Carroll discusses but does not endorse this objection.
5. Carroll discusses but does not accept this argument.
6. I am grateful to Jonathan Adler, Geordie McComb, Amélie Rorty, and Michael Stuart for helpful comments on earlier drafts of this chapter.