A Property-Based Account of Observation Vincent Israel-Jost IHPST, 17, rue du Four, 75005 Paris France ENST/GET, d´ epartement TSI, 46 rue Barrault, 75634 Paris, France contact : vincent israel [email protected]

March 10, 2008

Abstract Keywords : Observation, Empirical Sciences, Reliability.

1

Introduction

Observation in the empirical sciences has long been defined by direct access to entities or phenomena through unaided perception. On the basis of this definition alone, we are also able to categorize which entities and phenomena are considered observable, namely those able to be perceived without the aid of instruments. Consequently, the use of any basic optical instruments which do not modify the category of ‘observable’ has generally been accepted as an extension of human perception. Usage of such instruments as glasses, binoculars, or magnifying glasses do not allow us to perceive the unseeable; rather, they merely permit us to observe what is already visible from further away, in a process that easily compares to naked-eye vision. On the other hand, instruments such as microscopes, x-ray scanners, or magnetic resonance imaging (MRI) are able to produce data concerning entities or phenomena that could never, under any other circumstances, be perceived. The original notion of observation has thus gradually become unsuitable for scientists who now use the latter group of instruments to acquire data while still claiming to “observe” microscopic organisms or cerebral connectivity. The debate led by Bas van Fraassen [9] (1980), Dudley Shapere [8] (1982) and Ian Hacking [4] (1983) concerning the theoretical term of “observation” centered largely 1

around the status of data acquired through the use of such devices, and demanded whether data provided by instruments could be considered equally reliable as data collected with the naked eye (provided that in both cases adequate conditions of observation were met: lighting, scale of the object, etc). It remains to be decided whether or not a broader application of the term “observation” can legitimately be used in the sciences. Among the many instruments that are used to gather data, imaging devices have attracted the particular attention of philosophers, who often place them at the center of the observation debate. For many philosophers, the reliability of an instrument is therefore largely determined by whether or not it can safely establish correspondence between the various features of the image (lines, surfaces, patterns) and the geometric properties of an existing entity. For example, physicians’ suspicions are aroused when a very regular geometrical pattern appears on an MRI image of the brain, since none of the structures of the brain, either anatomical or pathological, could be related to the pattern. They conclude that the pattern has been created by the instrument, and thus exclude it from their interpretation of the image. Indeed, these sorts of difficulties arise from the fact that every single source of data is also a potential source of artifacts: objects may have been duplicated on the image, or else distorted. Imperfections– such as scratches or dust– of the lens in optical instruments, as well as electronic or even algorithmic problems within computer-assisted technologies, can significantly adulterate the final image. It is thus unsurprising that most discussions concerning observation have tended to emphasize the role of the instrument, as well as, to a lesser extent, the transmission of information; these two aspects are of considerable importance to the preservation of the object’s geometry in imaging technologies. If the instrument does not alter the information directly transmitted from the object, the points of the image and the points of the object must share a clear correspondence. Moreover, if the size of the object has been adapted to the spatial resolution of the instrument, then the whole process will yield valuable information about the object. In most of the writings on observation from the last two decades, reliability criteria have thus revolved around the notion of scale and structure; according to these writings, the spatial resolution of the instrument 2

must somehow correspond with the size of the object of enquiry, and the overall imaging process must be shape-preserving. In a number of cases, shape is a matter of crucial importance. Physicians, for example, can often identify a tumor from the shape of a spot in a mammography image, and even tell whether it is malignant or benign. However, geometric properties do not alone comprise all that a scientist must know when studying entities. For example, human organs– both in normal and in severely diseased conditions– often keep the same shape; therefore, the study of an organ’s condition cannot be successfully carried out on the basis of shape alone. In most philosophical analyses of observation, however, geometric structure– more than any other property– has remained at the center of the stage. An overly simplified notion of object– premised on geometric figure alone– leads to a number of issues that must be resolved. The first issue (I1) is that scientific questions deal with the study of specific properties more often than with existence or identification of entities, thus decreasing the importance of shape, while involving properties such as elasticity in mechanics, redness in astrophysics, or perfusion in medicine. The second issue (I2) concerns the role of instruments in observation which do not produce an image, as is the case with electroencephalography or electrocardiogram; since they provide us with virtually no spatial information, should they thus be eliminated from the domain of observation? The third issue (I3) arises from the fact that, nowadays, multiple techniques are used to study a given object. In the case of the human heart, more than five different imaging techniques are routinely performed, including ultrasound, X-Rays, MRI and photons emitted by radiotracers. Yet, if there already exists an instrument that is able to provide us with the precise shape of a given object, why develop new imaging technologies? The “geometric view” of observation, which is primarily concerned with shape preservation, does not explain why new means of investigation– with poorer performances in terms of spatial resolution– continue to be developed. In Section 2 of this paper, I will discuss the positions of several authors on observation which fail to take the three different issues into account. In Section 3, I will elaborate on Paul Humphreys’ brief proposal to replace the observation of a vague notion of entity with the observation of properties, a change based on two factors: one, 3

that our instruments are property– not entity– detectors, and two, that scientists are often interested in gathering information about various properties of an entity. However, since the detected physical property may be different from the object’s property that is studied, a distinction must be made between these two kinds of properties to later understand how they relate. This leads us to Section 4, wherein I propose a property-based concept of observation.

2

The geometric view of observation

Among authors who have taken sides in the debate concerning the role of instruments in observation, many tend to have an inordinately vague notion of what exactly is observed. This is usually communicated by giving greater importance to the shape or geometric structure of objects– to the detriment of other properties. Hence, in order to accept the use of imaging instruments in observation, they emphasize that the instruments must be determined not to alter shape. The preservation of shape is stated either explicitly or implicitly as the main reliability criterium for imaging instruments in the views of Sarah Vollmer [10], Ian Hacking [4] and Megan Delehanty [3]. Accordingly, their views on observation fail to answer all of the three issues laid out in the introduction, as we shall see later in this section. Only Dudley Shapere [8], in his discussion of the core of the sun, departs from the other authors by providing an analysis of the term ‘observation’ which does not simplify the notion of object. I will, however, begin with Bas van Fraassen [9], since his analysis alone is not aimed at modifying the classical notion of observation so as to include the use of instruments. Despite this notable difference, it still does not account for all of the three issues (I1 , I2 , I3 ). 2.1

van Fraasen, Vollmer and the ‘observable’

In van Fraassen’s account of observation, observable entities are defined as entities that can be directly observed by means of unaided perception. Something may be out of reach and still be observable, so long as there exist conditions under which we might observe it, or equivalently, so long as we can possibly access the ground truth, in Seager’s terms [7]. van Fraassen gives the example of the moons of 4

Jupiter, which, despite their distance from Earth, could be observed (by naked-eye vision) by astronauts in a spacecraft. On the other hand, we have no possibility of shrinking so as to be able to observe microscopic entities, which are thus relegated to the realm of the unobservable. van Fraassen argues that, despite the vagueness of the term “observable,” it is effective so long as it provides us with clear cases of both observable entities (trees, planets, etc) and unobservable entities (electrons, subatomic particles, etc). His argument at this point is clear; indeed, many useful notions are often vague or subject to indeterminacy. For example, while there is no clear definition of the term“tall,” it is still a meaningful adjective, one which enables us to delineate people who are “tall” and people who are “not tall.” However, I would like to point out a difficulty in his analysis (and in the empiricist vocabulary in general), due to the use of the term ‘observable’, which I think oversimplifies the notion of entity. Following van Fraassen, we understand that whenever an entity can be directly observed, under certain circumstances, the whole entity is then qualified ‘observable’, even parts of it which still remain unvisible. This reasoning surely applies to van Fraassen’s example of the moons of Jupiter. The converse side of the moons, for example, is as observable as the side that is currently being observed, since astronauts have the possibility to send their spacecraft over it. The interior of the moon is observable as well, because astronauts do have the possibility to drill into the moon after landing, and to directly observe samples of its interior. Therefore, the whole entity can be qualified ‘observable’, let alone the insurmountable difficulties that the astronauts would probably encounter if they wanted to get samples from the very core of the moon. However, the problem is revealed when we try to apply van Fraassen’s concept of observability to a radically different type of entity: a functioning organ in a living body. Again, the human heart is a clear case of an ‘observable’ entity: not only are we able to directly observe a human heart (if we were to open up a man’s chest, we know that we would see it), but it is actually observed by physicians while proceeding to heart surgery. As such, the ‘observability’ of the human heart suggests that we can potentially access information about any part of it; but this time, the necessary condition that would permit us to observe the inner part of the heart is to cut it into slices and then 5

observe each slice individually and directly (as is done in histology), yet in this case, we are forced to admit that what we would observe is very different from the original functioning organ. Therefore, in this example, the knowledge gained about the structural properties of the heart and the composition of its tissues is balanced by the irremediable loss about any of its original functional properties: perfusion, ejection fraction etc. In this sense, the predicate ‘observable’ is ambiguous because its extension is not clearly defined, and seems to only apply to the structural properties of an entity. The confusion between the limited number of properties of an entity we actually receive information about and the observability of a whole entity is only compatible with scientific questions regarding geometric structure, which is precisely the problem evoked in (I1). Moreover, it becomes extremely difficult to decide whether we see through imaging instruments. MRI, for example, permits us to obtain images of the human heart, which we qualified observable. Following van Fraassen, who states that “the moons of Jupiter can be seen through a telescope,” should we then say that the heart can be seen through or with MRI? On the one hand, MRI serves as only another way to obtain information about the heart, just as telescopes serve to obtain information about the moons of Jupiter. On the other hand, the information obtained with MRI (which basically measures the concentration of nuclei of hydrogen atoms) is so vastly different from the information obtained when we look directly at the heart, that we can hardly compare MRI with naked-eye vision. At any rate, van Fraassen’s analysis, which posits entities as merely geometric volumes or structures, fails to take into account the striking differences between several ways of “looking at” entities, and thus does not answer (I3). My criticism of the qualification of a whole entity as ‘observable’ ultimately suggests that entity may not be the right level to define observation, as we shall see in Section 3. Vollmer’s attempt to modify van Fraassen’s criterion of observability is based on her intention to pinpoint the exact source of human perception’s unerring dependability and to demand that all types of data production share this key attribute. According to her, what distinguishes naked eye observation from indirect inference of objects on the basis of observation is that signals scattered from the target are recombined by the eye lens so as to create an image of the object or system seen. This 6

principle of diffraction from an object and recombination to display a representation of the object determines the kinds of properties of ordinary entities that are observed. Specifically it allows observation of their shapes and orientations. For Vollmer, an instrument can be used in observation as long as it generates data with the same physical principle as in naked-eye vision. Demanding that an instrument shares qualities with our perception is always subject to objections: to what extent should the instrument be similar to our visual system? Why not ask, for example, that the instrument be composed of different pieces that mimic the cornea, lens, retina etc.? Why not require that the radiation used to form the image be visible light only? Vollmer responds to these objections by stipulating that the radiation be diffracted and then recombined to display a representation of the object, which, according to her, is sufficient to ensure the preservation of shape and orientation. By stating explicitly that shape and orientation (geometric properties) characterize observation, she answers (I2) by the negative: any instrument that does not provide information about shape and orientation should be kept out of the range of observation, thus eliminating instruments such as stethoscopes. However, she does not answer (I1) and (I3) any more than van Fraassen, since she uses the same term of ‘observable’ to qualify entities as wholes: only by simplifying the notion of object can she infer the observability of a whole entity from the observation of only one of its aspects. 2.2

Scale, shape and ‘good-mapping’: Hacking and Delehanty

Opposed to van Fraassen’s anti-realism concerning what we see with imaging technologies or optical instruments, Hacking has emphasized the example of the microscope, which, he argues, produces data as reliable as that gathered with unaided vision, as long as instruments and vision are used in adequate conditions of light, scale, etc. The first element that supports Hacking’s thesis is the fact that microscopes produce images through a series of physical events. Unlike Vollmer, who demands that radiation be diffracted and recomposed to safely obtain the conservation of shape and orientation, 7

Hacking suggests that any property of any kind of wave can make a system reliable enough for its use in the activity of observation. Still, his conception of reliability is very close to Vollmer’s, as an instrument is ultimately aimed at delivering images which are “good mapping,” that is, images in which the spatial relationships, 2-D or 3-D, in the structure of the object are accurately reproduced. However, the condition of producing images through a chain of physical events is not sufficient to guarantee good mapping of the object. Instruments are sources of artifacts; errors, distortions, or features added on the image need to be distinguished from the features corresponding to the object. Hacking surveys two possibilities to test the instrument so as to make this distinction. The first one is to experiment by observing an object that has been expressly designed to test the instrument. The example provided is a micro-grid, whose structure is known and can be seen with a microscope. Since the microscope produces an image in which the structure of the micro-grid is accurately reproduced, one can infer that the instrument has the property to deliver a “good mapping” of any object. Again, Hacking’s discussion is limited to geometric properties; only the conservation of shape is tested, while other properties of entities are not mentioned. Hacking thus fails to answer (I1) by restricting concerns about instruments to geometric aspects. The second possibility of testing an instrument that is discussed by Hacking is to make use of one or several other instruments, in order to check if the same structure will be obtained by different means. If so, he says, it could not be a mere coincidence, for different kinds of physical processes, associated to various instruments, would be unlikely to produce the same artifact. Therefore, while Hacking recognizes the existence of a variety of observation techniques, and acknowledges that they are based on highly different physical processes, he reduces their function to providing an accurate structure. He does not mention that different sorts of physical processes lead to different sorts of images, each one corresponding to a single property, therefore failing to answer (I3). In Hacking’s account, it is especially difficult to establish why instruments providing a worse mapping than preexisting instruments (with a poorer spatial resolution for example) are developed if they are only to give equally poor information about the structure of objects. Instruments which do not provide any spatial information are not discussed by Hacking, who thus gives no 8

clear answer to (I2). It can be argued that it is not Hacking’s intention to account for the variety of properties that are accessible through instrumental techniques. Rather, Hacking declares that he is only convinced of the existence and accuracy of structures that are observed through a variety of instrumental techniques and this is certainly sufficient to ground some form of realism that goes beyond what empiricists allow. Yet, I believe that Hacking’s realism concerning geometric properties of entities can be further prolonged to other properties and I will try to build such extension in the next sections.

The tendency to reduce an object’s properties to geometric properties can in part be explained by the examples of entities van Fraassen, Vollmer, and Hacking use in their discussions, which, in a certain sense, are “simple”: particles, which cannot be decomposed, and micro-grids, which are homogeneous objects with a very regular structure. The questions that we may pose concerning these entities are thus limited to their existence or to their structure. It is therefore interesting to compare their analyses to Delehanty’s, as she is one of the few authors to have discussed observation within a medical imaging framework. Of course, physicians are concerned with shape, orientation and relative position of organs in this domain as well, for many questions deal with anatomy. Yet, organs are also often examined to check properties related to their function. For this purpose, a set of imaging techniques has been designed to evaluate brain activity, heart perfusion and trace processes such as the growth of tumors. These techniques, which are grouped under the label functional imaging usually make use of agents (contrast agents, radioactive molecules or tracers), which follow a specific path in the organism. Since these agents participate either in the emission or to the interaction of the radiation used to form an image, they give very valuable information about processes that happen in the organism. Delehanty analyses the role of a specific medical imaging technique, the positron emission tomography (PET) technique, which is now widely used in observation, especially in oncology. For this purpose, her concept of reliability is split up into requirements concerning the relevance of a given technique to address a scientific question and those requirements relative to only the detecting part 9

of the instrument. The latter category of requirements is reminiscent of those belonging to Hacking: both involve comparison with a different apparatus, intervention and prediction. The former category of requirements takes into account the fact that the imaging process is embedded within a broader scientific investigation, and asserts that the information provided by the image should somehow or another correspond with the needs of the investigator. Delehanty uses the concept of ‘granularity’ for this purpose. According to her: Granularity is a characteristic of representations. We can refer to both the granularity of the representation of the world that a particular question is directed at and to the granularity of the representation of the world (data) that an instrument generates. For the sake of ease, I will refer to these as the granularity of the world and the granularity of the instrument respectively. The granularity of a representation is the smallest object or unit required to address the question of interest. She later states that “the granularity of an instrument matches or is sufficient for a question if it is capable of providing evidence about the smallest objects needed to address that question.” This requirement is, therefore, doubly insufficient: first, it makes use of a notion which is virtually indistinguishable from that of spatial resolution, and second, it only translates the needs of the investigator in terms of the size or scale of the object, which again presupposes that scientists are only concerned with the shape or structure of entities, ultimately resulting in the same problems as the previous authors. 2.3

Shapere’s theory of the source

The last author I would like to discuss is Dudley Shapere, whose primary intention is to understand how the use of the term ‘observation’ in the context of scientific experiments departs from its traditional usage, which associates it to perception. His analysis centers on the specific case of the “observation” of the interior of the sun, which, according to scientists, is only possible by the detection of neutrinos emitted from the core of the sun. One of the key ideas is the contrast between the path of neutrinos, which, due to the extremely weak character of the interaction of neutrinos with

10

other matter, is direct, and the path of light, which is highly indirect. A photon emitted from the core of the sun would therefore take between 100,000 and 1,000,000 years to reach the surface of the sun; following so many interactions, any information about the location of the point of emission would be lost. Neutrinos, on the other hand, reach the detector after a straight trajectory, thus permitting us to know where they come from. In order to eliminate cases for which information is subject to alteration, Shapere proposes the following analysis, stating that: x is directly observed (observable) if: (1) information is received (can be received) by an appropriate receptor; and (2) that information is (can be) transmitted directly, i.e., without interference, to the receptor from the entity x (which is the source of the information). These two conditions seem to sufficiently account for what happens during transmission (which must not be affected by any interference), as well as detection, since an appropriate detector is naturally required. As such, however, it fails to enrich the notion of object with other properties than geometric properties, since it only provides a constraint on the path of radiation, similar to Hacking’s good mapping. Moreover, providing a condition on the observability of an entity (entity ‘x’) contradicts the alleged distinction between the observability of the surface of the sun, which is possible with light (photons), and the observability of the interior of the sun, which is only possible by detecting neutrinos emitted from the core of the sun. Therefore, these two conditions do not sufficiently escape the pitfalls of van Fraassen’s analysis, since both suffer from the same overall lack of distinction concerning the observability of one aspect of an entity and the observability of the entity as a whole. Unlike van Fraassen, Shapere makes a very valuable distinction between observing the surface and the interior of the sun, demonstrating the author’s intention not to reduce entities to mere structure. More specifically, Shapere develops the idea of demarcating the processes of observation into the theory of the source, the theory of transmission, and the theory of the receptor, departing from other authors who do not give enough attention to the object (the source). In the theory of the source, he explains how detected neutrinos re11

late to the production of energy in the core of stars. By doing so, he asserts that the transmitted information deals with something other than only structure – which, in the case of stars, would be rather uninteresting, since we know that they are approximately spherical – but to a more specific property of stars: their production of energy. This directs us to an analysis of the term ‘observation’ that goes beyond the problems mentioned concerning the previous authors, and that can answer the issues (I1, I2, I3). Only by establishing a notion of object that is not reduced to mere geometric structure can we begin to do so. For this, we need to center on the properties of an entity, rather than the entity as a whole.

3 3.1

The property-based position on observation Observation of properties in Humphreys’ Extending Ourselves

We can draw from the preceding overview of most of the current views on observation that concentrating on an idealized notion of entity rather than the entities’ properties circumscribes scientific questions to the existence of entities and to the accuracy of their structure. Humphreys has long recognized this problem and has consequently endeavored to be more accurate in describing “what” is actually detected in scientific investigations. In his 2004 Extending Ourselves, Computational Science, Empiricism and Scientific Method, he writes: Disputes about what is real frequently turn on objects or kinds of objects: Do prions or cartels really exist? Is there a planet beyond Neptune? What evidence was there in the late nineteenth century for the existence of atoms? Do emotions exist separately from their associated brain states? And so on. Stock examples can be misleading, for issues in realism also concern existence of properties: How do we know that quantum spin exists? Is there such a property as alienation in society? Is covalent bonding a specifically chemical property? [...] The entire debate on observables has been biased by an excessive focus on size and relative distance as surrogate measures for the degree of difficulty of observing objects. 12

The last sentence identifies precisely the same problem discussed in the preceding section: the overwhelming presence of geometric issues in observation. Humphreys’ interpretation is that those aspects should indeed play an important role in questions concerning existence or identification of entities. At the same time, Humphreys acknowledges that there are many other scientific investigations regarding entities or types of entities whose existence has been previously proven. For these entities, a partial knowledge is already established, but scientists still seek to learn about their other properties, therefore ceasing to concentrate solely on size and relative distance. Humphreys’ demarcation between questions concerning entities as a whole – mostly existence – and those concerning properties of entities seems to result in two kinds of observation, adapted to each case. There is, however, another argument in favor of an account of observation exclusively based on properties, and which Humphreys also pointed out: namely, our instruments are property – not entity – detectors. All sorts of physical properties may be detected: luminosity, radioactivity, rate of x-rays emission, redness, etc., but there is no tree or brick detector. Hence, Humphreys’ account of observation focuses entirely on properties. While he acknowledges that it is sometimes the entities themselves that are in question, we invariably learn about them through properties. This gives us a better idea of what constitutes an object. Instead of considering it to be a certain shape, composition, mass, or location, we now have access to an entirely new set of properties, most of which were erstwhile out of reach of our unaided senses. As Humphreys writes: The discovery process for “objects” consists in specialized kinds of instruments detecting one or more of the properties that constitute what we consider to be that object with, subsequently, more and more properties constituting that entity being detected through further instrumental techniques. Here, Humphreys develops the idea to explore entities as property clusters. Every new instrument is used to detect another property, leading to a more complete knowledge of the entity. At the limit, the concept may be pushed until we have detected all sorts of physical properties regarding an entity, which is then completely explored. 13

There remains a problem in this conception, which is the lack of distinction between physical properties than can be detected (as mentioned above: radioactivity, luminosity etc.) and properties of entities, which may be detected, but are not limited to physical properties. Scientists are also interested in “non-physical” properties (which we will call “higher-level” properties), as opposed to physical or “low-level” properties. Hence, just like we recognized that there are no entity detectors, we are forced to admit that myocardial perfusion detectors do not exist, let alone detectors of ‘alienation in society.’ Yet, these properties are claimed to be “observed” by scientists.The next section is aimed at establishing, when possible, the link between properties that can be detected (physical properties) and higher-level properties and to extend the notion of reliability to this link. 3.2

Property-based observation revisited

In Section 2, I reviewed the positions of several authors that all emphasized reliability criteria in observation, but dealt solely with structure preservation. At this point, while I have argued that this demand is ultimately defective, I also want to assert that structure preservation through direct transmission and accurate detection of information – or the obtention of a “good mapping,” in Hacking’s terms – is also necessary in many cases. However, if we obtain a “good mapping,” we still have to ask ourselves what the map represents. In the preceding section, I suggested that it represents a certain physical property that can be detected. Since researchers are sometimes not primarily interested in physical properties, but rather in “higher-level” properties, we must now turn to an account of the reliability of the relationship that stands between physical properties and properties of entities. Shapere provides us with our first example, which concerns what we learn about the production of energy in the core of the sun from detected neutrinos. According to him, “physicists and astronomers have developed a theory which, on the basis of a great many diverse considerations, appears to give an excellent account of the production of energy by stars”. Since this theory relates the rate of emission of neutrinos (low-level or physical property) to the production of energy (higher-level property) through a quantitative law,

14

physicists and astronomers can reliably deduce the amount of energy produced in the core of the sun (higher-level property) from the number of neutrinos that have been detected (low-level property). An important point here, is the independence between the so-called “theory of the source”, which constitutes the best available knowledge concerning the mode of production of neutrinos in the solar core - and fosters a link between low-level and higher-level properties -, and the theories of transmission and detection, which describe the path of neutrinos and their detection by a given instrument, regardless of the conditions of their emission. This first example should not suggest, however, that theory is required in the observation of higher-level properties. A parallel can be drawn between the case involving the core of the sun and, for example, oncological PET studies. In both cases, radiation, composed of neutrinos in the sun and positrons in the organism, is primarily detected. The next step is to correlate what has been detected to what has been emitted within the region of interest. In PET imaging the image is a map of radioactivity inside the body, after the administration of a radioactive molecule (a tracer ). However, physicians are interested in other properties than radioactivity, from tumoral activity to heart perfusion, and can learn about such properties, for radioactivity corresponds with a higher-level property, which is revealed by the type of tracer used. 18 F-fluorodeoxyglucose, for instance, is a radioactive tracer similar to glucose, tending to attach to tumoral tissues. In the image formed with this radioactive tracer, a tumor is thus seen as neither a source of gamma photons nor as a mere distribution of matter, but as the site of an expanding process involving a significant glucose consumption when compared to surrounding tissues. If a radioactive tracer such as 82 Rb ammonia is used instead, the same physical property of radioactivity is detected. However, it relates to a different biological property: perfusion (of the heart, for example). Thus, in the case of PET imaging, one single instrument, which detects only one physical property, permits us to access many more higher-level properties, according to the tracer used. As for Shapere’s example of the sun, physicians have good reason to think that the relationship between the radiation detected and higher-level properties is reliable, even if unsupported by theory. The following definition of observation endeavors to take these elements into account: 15

We say that we observe x if one of its relevant properties can be reliably related to a physical process (natural or artificially provoked) that occurs in it, and if information about that physical process is transmitted and detected in a manner that is accessible to us. The first part of the definition states that observation of an entity x is necessarily restricted to one of its properties. The only way to obtain information about this (relevant) property is to safely relate it to a physical property. The reliability of this link may be ascertained by a well-accepted theory or by well-tested empirical means, as illustrated by the examples of the core of the sun and of properties of radiotracers within the body. In the latter case, we pointed out that the relationship between physical properties and higher-level properties is not univocal: the same radiation may be used to observe properties of different kinds. The second part of the definition aims at establishing the reliability of the transmission and of the detection in terms of our knowledge of the process by which information is obtained. By doing so, I reformulate reliability criteria reviewed in Section 2 in a more general manner, considering Hacking’s “good-mapping” or Shapere’s direct transmission (with no interference) as conditions to guarantee our cognitive ability to correlate what is detected to a physical process (emission, attenuation, diffraction, etc.) that occurs within the studied object (see also Alspector-Kelly (2004) [1] and Linden (1992) [6] on reliability in terms of our knowledge). Together, the two parts of the definition aim to build the bridge – but emphasize at the same time the distinction, largely discussed by Bogen and Woodwad [2] – between data and phenomena. Finally, I left out the word ‘observable,’ which I think is misleading when applied to entities, as was shown in my criticism of van Fraassen’s position. By invoking what I called “higher-level” properties, I could reach what seems to be the adequate level to account for the variety of things that scientists claim to observe, therefore answering to the first issue I1 . In addition, on the course of describing how scientists access higher-level properties, I insisted on taking instrumental diversity into account, adressing the issue I3 . Finally, the reformulation of the reliability of the transmission and the detection of information avoids any reference to the preservation of geometric properties, which permits me to answer I2 in the negative: the previous analysis of observation states that we can observe with a 16

stethoscope or an electrocardiogram. The fundamental point is the distinction drawn between two kinds of properties involved in observation. Only by distinguishing between physical properties that are detected and properties of the object can we unify the concept of observation in empirical sciences. But this same distinction explains at once the striking differences that occur in scientific practices, for investigating different kinds of objects, or their different properties, leads to very disparate approaches to observation.

References [1] Alspector-Kelly, Marc (2004) “Seeing the Unobservable: van Fraassen and the Limits of Experience,” Synthese, 140:331-353. [2] Bogen, James and Woodward, James (1988 )“Saving the Phenomena,” The Philosophical Review, 97:303-352. [3] Delehanty, Megan (2005), Empiricism and the Epistemic Status of Imaging Technologies. Ph.D. Dissertation. Pittsburgh, PA: University of Pittsburgh [4] Hacking, Ian (1983), Representing and Intervening: Introductory Topics in the Philosophy of Natural Science. Cambridge: Cambridge University Press. [5] Humphreys, Paul (2004), Extending Ourselves: Computational Science, Empiricism and Scientific Method. Oxford: University Press. [6] Linden, Toby (1992), “Shapere on Observation,” Philosophy of Science, 59:293-299. [7] Seager, William (1995), “Ground Truth and Virtual Reality: Hacking vs. van Fraassen,” Philosophy of Science, 62:459-478. [8] Shapere, Dudley (1982), “The Concept of Observation in Science and Philosophy,” Philosophy of Science, 49:485-525. [9] van Fraassen, Bas (1980), The Scientific Image. Oxford: Clarendon Press.

17

[10] Vollmer, Sara (2000), “Two Kinds of Observation: Why van Fraassen Was Right to Make a Distinction, but Made the Wrong One,” Philosophy of Science, 67:355-365.

18