Image and Practice: Visualisation in Computational Fluid Dynamics Research moreForthcoming in Interdisciplinary Science Reviews, 37.1 (March 2012) pp. 96-111 |
25 views |
Anthropology, Philosophy Of Science, Science And Technology Studies (Science And Technology Studies), and Information Visualisation
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 1 of 15
Image and Practice: Visualisation in Computational Fluid Dynamics Research
Postprint, to appear in “Computational Picturing”, Interdisciplinary Science Reviews Vol 37:1 March 2012
Please quote published version, not this one
Matt Spencer
Goldsmiths Centre for Cultural Studies
Contact: mspencer421 [at] googlemail.com
About the author: Matt Spencer is a PhD student and Visiting Tutor at Goldsmiths Centre for Cultural Studies in London, UK. His research principally concerns social scientific and philosophical theories of rationality and representation, as well as software development and the epistemology of computer simulation. Since the summer of 2010 he has been undertaking ethnographic research with the Applied Modelling and Computation Group at Imperial College in London.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 2 of 15
Image and Practice: Visualisation in Computational Fluid Dynamics Research1
Abstract: This paper examines the use of visualisation within everyday research practices in computational physics. In doing so I attempt to move from the well documented representational issues elicited by the concept of the image, to more microscale issues of the habitual structuring of the everyday that emerge when a specific example of science in the making is analysed. To this end, I focus on one specific example, of tracing a computational error through a fluid dynamics simulation of the “lock exchange” experiment. This simulation is one small part of the research that goes on within one of Europe's largest computational physics research groups, the Applied Modelling and Computation Group at Imperial College in London, where I am involved in ongoing ethnographic fieldwork research. Visualisation is shown to play a central role, not just in daily routines of investigation and problem solving, but in the process of habituation through which scientists cultivate the dispositions through which everyday life gains its texture and form. Far from being a detachable representation of a part of a world, simulation is shown to come into being as a process within a world structured by the repetitions and improvisations that characterise research practice. Keywords: Visualisation, imagery, computation, habit, practice, improvisation, simulation
Introduction
The spread of computer simulation throughout the sciences is well documented (see, for example Kaufmann III and Smarr 1993). It has been subject to a great deal of interest among philosophers, historians and sociologists (for example, de Landa 2011; Galison 1996; Morgan and Morrison 1999). Understanding simulation is of great importance in how we understand processes at all scales and in all time frames. It is absolutely central to the study of contemporary problems such as climate change and the storage of nuclear waste. As simulation gains significance and as these problems become more pressing, it becomes increasingly important to understand the scientific practices that lie behind its production. In other words, before the question of how “we” (academic commentators, policy makers, publics of various kinds) relate to simulation, stands the pressing question of how the scientists themselves relate to their simulations and to their own work of simulating. The role of images within science has also been a recent focus of attention (Lynch and Woolgar 1990; Daston and Galison 2007). Images evoke intuitions of representation, and indeed depiction often provides the metaphorical resources by which we grasp the kind of thing we call representation. Here, however, I want to trace out another path. Rather than approach images through a preconstituted problematic of representation, I want to use them as a methodological handle through which to grasp the dynamics of scientists' investigations, to approach science through attention to the textures of everyday life within the laboratory. The claim I will explore here, therefore, is that an appreciation of these dynamics provides a compelling viewpoint complementary to treatments of science in the abstract. It is a viewpoint from which we can grasp the roles of temporality, of rhythm and habit within the evolution of concrete research projects, and this I think can help us grapple with important questions about the essence of simulation.
1
Thanks to the organisers of the “Visualisation in the Age of Computerisation” conference at the Said Business School in Oxford for the opportunity to present an early version of this research, and to the Arts and Humanities Research Council for generous financial support for the project. Thanks also to Celia Lury and Victoria Newton for their valuable comments.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 3 of 15
This paper draws on material from an eighteen month period of ethnographic study among scientists at the Applied Modelling and Computation Group (AMCG) at Imperial College in London. These scientists study a wide range of phenomena, and actively develop a number of different software systems to this end. Here, I will be mainly discussing Fluidity, which is the largest code at AMCG, having between 30 and 40 scientists developing it at any one time. Fluidity is an implementation of the finite element method on unstructured adaptive meshes, and can be used for a huge range of different fluid dynamics problems, from ocean circulation, tsunamis and coastal defences, to industrial processes, nuclear reactors and coolant flows, oil reservoirs, atmospheric pollution and convection in the earth's mantle. Fluidity is open source software, released under the GNU Lesser General Public License. This being said, Fluidity is still largely developed “in house”, with developers outside Imperial College generally collaborating through jointly funded projects. Since the first general public release in the autumn of 2010, AMCG have attempted to put systems in place to ensure that a wider community can contribute, but as an open source project it is still in early stages, and the majority of its contributors are physically located in the same offices. The central part of this essay is an analysis of a particular example of working with visualisation in computational science. I will then go on to use this example to draw some broader conclusions about the role of the image in practice. First, however, I will deal with two preparatory issues in order to clear the ground: representation and process.
The Image as a Problem of Representation
Marilyn Strathern points out that social phenomena display a curious “fractal” property, in which each scale of analysis exhibits a similar level of complexity (2004, xxi). If one were to “zoom in” from a grander consideration of science to a more detailed analysis of its sites of accomplishment, a new terrain would emerge, which would in turn invite scrutiny at new levels. This is something that my informants at AMCG are well aware of, for modelling is always a matter of choosing a resolution. Whatever the scale of your grid, there are always below gridscale phenomena that are impossible to explicitly resolve. Likewise, my analysis of AMCG, far from simply “filling in the gaps” missing in grander theories of science, draws the eye to new kinds of discontinuities: the gaps and leaps in play within investigative practices. Images, as well as the apparatus of their production and distribution, form a central aspect of the material assemblages in which these practices are carried out, assemblages in which these practices gain their rhythm and reason, what we might following HansJorg Rheinberger call “experimental systems” (1996). In what follows, I aim to trace images through such systems, revealing that they are significant in much more than a representative capacity. This being said, in some situations, the image is isolated as a problematic object subject to questions of adequate representation. While I claim that much of the work of images is carried out below the radar of general concerns of representation, such concerns are not always an artefact of the analyst's orientation. The tropes of “iconoclash” for example play out with remarkable clarity in the relation between research proper and the images that scientists disseminate for marketing purposes (Latour and Weibel 2002). In a general seminarstyle discussion on the theme of images with about half the group, I displayed a brochure for a fluids modelling code, replete with colourful graphics of turbulent mixing, of velocity arrows flowing through jet engines, of intricate machinery modelled in detail 2. “The trouble with this,” said one of the group, to general assent, “is that it makes it look like the simulation is already finished, already complete.” What is erased by these inscriptions is the process, the always incomplete becoming of their work. Whatever confidence scientists feel about their work, there is an aspect in which this sentiment is deferred towards future validations of the model, future proofs about the method, and future studies of the system. Even when inscribed in publication, research is never wholly present, but is always part of an ongoing becoming. What is not captured in marketing images is the fact that, in Rheinberger's Derridean terms a model is “an entity that draws its effectiveness from its own absence... [A] model is a model only in the perspective and by virtue of an imaginary reality
2 Companies such as Ansys offer similar simulation services to the work done at AMCG, but on a commercial scale, and with a visibly greater attention to the efficacy of marketing materials (see for example http://www.ansys.com/).
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 4 of 15
at which it fails to arrive” (1997, 110). These marketing images are the ones most widely disseminated beyond the laboratory but this is an expulsion from which they cannot return. While they do represent the simulation and its object, in so doing they eclipse important aspects of the science behind its coming into being. They present an obstacle, therefore, for any analysis that would attempt to take them and their representational capacity as the lens through which to see processes of scientific research. “Epistemic things, let alone their eventual transformation into technical objects and vice versa, usually cannot be anticipated when an experimental arrangement is taking shape. But once a surprising result has emerged, has proved to be more than of an ephemeral character, and has been sufficiently stabilized, it becomes more and more difficult, even for the participants, to avoid the illusion that it is the inevitable product of a logical inquiry or of a teleology of the experimental process” (Rheinberger 1997, 74). The process in itself, animated by dull feelings, vague suspicions, moods and hunches, wrapped in a culture of artefacts and procedures, is destined to be erased in the end analysis. Like the creation of marketing materials, the “writeup” would be a process of erasure as much as it is one of inscription. In both cases, the indeterminacy of the exploration is analytically out of focus where the “result” of research looms large, for wherever there is a result, there is a tendency to see the process of its formation only under the condition of its eventual outcome, to collapse indeterminacy into the determination of what will emerge from it (cf. Simondon 2009; Ingold 2001). As Rheinberger puts it: “An experimental system can readily be compared to a labyrinth, whose walls, in the course of being erected, in one and the same movement, blind and guide the experimenter. In the stepbystep construction of a labyrinth, the existing walls limit and orient the direction of the walls to be added. A labyrinth that deserves the name is not planned and thus cannot be conquered by following a plan. It forces us to move around by means and by virtue of checking out, of groping, of tâtonnement.” (Rheinberger 1997, 74) It would be a mistake, however, to suppose that marketing images could be taken to stand for visualisation in general. If marketing images constitute one specific mode of imagery, very different modes of imagery are essential at the level of improvisation, right inside the processes that were eclipsed by marketing. There are many images present within research that do not retain attention within their own capacities as images, but rather pull it through them, beyond them towards the emerging future of the ongoing project. Here any representational capacities are implicit; “the Image” does not appear as a source of problems, because images are thoroughly embedded in technical milieux of code, equations, statistics, equipment and procedures, drawn together around a temporally unfolding research project, a background from which neither images nor any specifically visual dimension are routinely extricated or isolated. There are very good reasons to be wary about equating visualisation with the production of visual sensation, of shapes or colours. Martin Heidegger captures this, though writing of the ears: “Much closer to us than all sensations are the things themselves. We hear the door shut in the house and never hear acoustical sensations or even mere sounds” (2001, 25). What is significant in moments of investigation is what is seen in the image, moments in which the image itself disappears, in which what is significant is less “what is seen in the image”, than simply “what is seen”, what is encountered. Mediation is not the issue and thus not the message. Much closer to the scientist than all images is the problem at hand, the simulation, the data, a bug in the code, or its potential solution. Present representation is less of an issue than future realisation.
Discontinuity and Process
Before exploring these claims through an empirical example, it is necessary to make one further preparatory comment. Social scientific theory has over recent years become increasingly concerned with process and becoming, and in a sense my analysis is indeed intended to take us in this direction. Andrew Pickering, for example, claims that “we live in the thick of things, in a symmetric, decentred process of the becoming of the human and the nonhuman. But this is veiled
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 5 of 15
from us by a particular tactic of dualist detachment and domination that is backed up and intensified... by science as our certified way of knowing” (2008, 8). Yet this vision of science itself requires a “zoomed out” view that brackets off the improvisations and processes through which research itself is carried out, for the insight of science studies has been that if we look in at the detail of practical research we see that “science is itself caught up in the flow of becoming...” (2008, 8). Emphasis on flows and processes is often achieved through the construction of a contrast with a world of ontic beings and static structures, for example between on the one hand the inscription of the marketing image which presents the model as finalised and on the other hand the research process that is always ongoing. In the context of talking about anthropological narrative, but in words that could equally apply to the navigation of research, Strathern points to a potential naivety here. The danger is that these flows and processes are imagined as smooth, a smoothness which is an artefact of the contrast drawn for rhetorical purposes, rather than a feature of the process itself. “Ideas and arguments are often regarded as “flowing”... The time it might take to travel, as the reader moves through the text, gives a kind of experiential unity to the exercise. Yet this unity or sense of flow or movement is at the same time made up of jumps over gaps, juxtapositions, leaps – unpredictable, irregular. So, continuous as the process of narration might seem, the closer we inspect monographs, paragraphs, sentences, the more aware we are of internal discontinuities” (Strathern 2004, xxiii). Scientists at AMCG inhabit a world of the interplay of the continuous and discontinuous, between the continuity of the NavierStokes equations for an ideal fluid, and the discontinuity of the molecular dynamics that the ideal fluid overwrites for macroscale phenomena, between these continuous equations and the discrete mathematics of the digital computer, for which they must be discretised if they are to written into code, and between the continuities and discontinuities of the numerical approximation and the necessity of modelling continuities and discontinuities within the simulated system. Each is implicated, rendering it impossible to idealise continuity as representing some more authentic essence of process. Discontinuity is not the antithesis of process, but is endemic within it, folded many times over. Images in the depths of the research process are significant in their relation to its movement, manifest at moments of unpredictable leaps, of irregular jumps, moments where the ongoing path of investigation is textured by discontinuity, gaps where what comes next does not automatically follow, but requires some additional force. Images are fodder for habit and improvisation, the material to be grasped in order to take these steps. Somewhat like stepping stones, points of hardness at odds with the ongoing movement, without such material artefacts that movement would not be able to push onwards towards its future. They are points at which what follows unfolds. As Bourdieu was well aware, even when the question of “what comes next” (in a rite, a procedure, an application of scientific method) is already determined, governed by an objective rule, the whole success of the enterprise may nevertheless rest on a thoroughly tacit sense of timing, of rhythm and of the moment, a sensitivity for which rules are no substitute (1977, 7).
Scientific Software Development
The course of developing a model proceeds through a time in which it is still uncertain whether the simulation is behaving correctly, or what the correct behaviour might be, and whether, if it does appear to be behaving itself, whether it is doing so for all, some, or none, of the right reasons. Prior to the emergence of the simulation as it will be spoken about in publication and presentation, is the simulation as it is in its process of being made, when existing capabilities of the code are tested for their ability to handle the problem in question, when new capabilities are being added in order to supplement any deficiencies thus identified, and when the many hundreds of virtual dials and knobs within the model setup and parametrisation are being tweaked and refined. What kind of basis functions will you use? What kind of mesh? At what resolution? Will it be adaptive and if so how? What kind of time stepping (implicit, explicit, or
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 6 of 15
somewhere in between?)? What kind of boundary conditions and forcing factors? What kind of parametrisations for capturing phenomena that cannot be simulated (such as subgrid scale turbulent mixing)? And are all these compatible? Is there a precedent for this kind of use on this kind of problem? During this period of model composition, the scientist is constantly modifying the model and attempting to understand the results of this modification. Experience here is not an exercise in datacollection, but of reflection, of interacting with one's own effects among what happens. The datasets produced by the majority of simulations run at AMCG are simply too vast to be directly understood. Visualisation, as a manipulation of data that renders it visible, serves as a crucial vehicle for informally assessing the simulation and the effect of changes. Errors in the emerging simulation can appear in three main ways. Some will cause a complete crash. Others will produce wildly inaccurate results. And then there are errors of a range of subtlety, some of which can only be picked up by a mathematical analysis of the data output of the simulation, usually comparing this statistically with expected results derived from experiments, analytical solutions, observations, other modelling methods, or high resolution simulations. Errors can creep in at all stages, arising from the specific set up of the simulation, from changes made by others elsewhere in the code, or from shifts in the wider technical milieu (external libraries, compilers and operating systems, for example, are being constantly updated and modified). In the early phases of development, a simulation will evolve through a series of wildly inaccurate results, with the scientist trying to understand these, tame them, eliminate any problems. Error is often starkly obvious when it is manifest in an image, and at all stages, even while a large simulation is in the process of being run (which may take a long time, even on a supercomputer), it is common to download the simulation's data, its work in progress, and visualise to “keep an eye on it”, hoping to catch any error as it occurs, if it occurs. “The point is not to run it for a week and then check at the end of the week and find it was wrong after day one, so that is why I keep on checking.” (IW) Looking at a visualisation, scientists can often tell that something is going wrong, because problems are manifest in the image such that it intuitively “looks wrong”. Indeed it is claimed that nonscientists would have the same response if erratic behaviour in the model output leads to a visualisation that simply doesn't resemble everyday experiences of fluids. “If the water moves through the solid object rather than around it, I know there is a problem, because that is just not physical.” (AB) In other cases, however, errors are less obvious and the scientist relies on long experience in working with this kind of simulation, experience which conditions the space of practice opened by the visualisation, which is often the place for inclinations to linger linger, to double check some settings, to run an extra diagnostic or to look at a different visualisation of the data, movements through the process of investigation that open further spaces from which it may become obvious that there is a problem lurking somewhere within the system, or which may open onto sufficient confidence to move on towards the next steps. But rather than ever actually affirming that the simulation is correct, the best these intuitive checks can offer is a provisional double negative: it is “not wrong”. It is not even that nothing is wrong, but that nothing appears wrong, in the initial check. It is not infrequent for something that “looks right” will later on turn out to have subtler problems that were just not the kind of thing that would show up in the image. “Looking at a visualisation, you can eyeball it and see if there is something obviously wrong, but if it looks roughly like what you expected then you would do your diagnostics and find out exactly how much it looks like what you expected.” (QY) An image “looking right” provides little ground for confidence, but does exert an influence on the scientists' course of investigation, a compulsion now to go and attempt to formalise the data according to statistical measures, through coding up diagnostics, a mathematical transformation of the dataset that often reduces the need or inclination to
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 7 of 15
visualise it at all. The image is no substitute for the diagnostic measures that aim to distil the data into numbers that can support some kind of positive conclusions, a much more formal grounding of confidence which may reveal issues invisible in the image. “The initial visualisation will tell you if it is wrong, if the velocity is flying off the sides or whatever, but it won't tell you if it is correct. It can never tell you if it is absolutely correct and I suspect that is why I plot these graphs now. When I first was developing this [simulation] about twenty months ago when I came here I probably did open the file and look at it and say “its wrong” or “its right” but it has been looking right for 20 months now, but it has actually been wrong. So looking at it wouldn't have told me anything which I guess is why I have moved into distilling it down into a graph.” (QS)
Navigating Error in the Lock Exchange
Error is encountered in the image but visualisation can also draw the scientist beyond the fact of error, towards its underlying cause and towards the future of its eventual resolution.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 8 of 15
Figure 1: A cross-section of the lock exchange simulation, at different time-steps.
The images in Figure 1 show a simulation of the lock exchange, a well studied experiment where a tank is filled, with the aid of a barrier in the middle, half with hot fluid and half with cold fluid. The barrier is removed, and the mixing that occurs is a paradigmatic example of “gravity currents”, currents driven by the differing density of the two bodies of water. In this case, something has gone wrong. The simulation that is visualised in Figure 1 is a “short test”, a small simulation that is automatically run, along with over 14,000 other tests, by the automated testing server every time anyone changes anything within Fluidity's code. The purpose of these tests is to let scientists know when changes have affected key results, so that they can be confident that they will be able to reproduce a result they got earlier. This system, coupled with the code repository, introduces an element of stasis, of reproducibility, into a field defined by perpetual technical evolution. In this case, the example was set up with a diagnostic that gives a statistical measure of the amount of mixing between the two bodies of water. An alarm was triggered when the test suddenly started exceeding the expected range of values (expected from the experimental and computational precedent), alerting SS, the “owner” of the simulation, to the existence of a problem.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 9 of 15
Having been alerted to the problem, SS knew because she designed the test case that the symptom was excessive mixing, but with a massively complex code such as Fluidity you are guaranteed no simple way to diagnose the cause. SS immediately visualised a crosssection of the simulation, plotting temperature. The result is a succession of images that she can click through or play as an animation, to see the data at each time step. A series of these snapshots is displayed in Figure 1. All timedependent simulations must approximate the continuum of time that underlies the NavierStokes equations by a series of discrete steps. SS's attention was drawn to particular jumps in behaviour between consecutive timesteps, such as the sudden spike in mixing visible between 559 and 560 time units (d and e in the diagram). Individually, both 559 and 560 show intuitively feasible results. It is the difference between them that stood out. The abrupt change on the boundary should not occur, for there is no process occurring between the timesteps that could explain it. Having spotted this, she went on to visualise the simulation in another way, to create different images through which she can hope to push onwards towards a more specific idea of what was going on. She chose the same visualisation setup, but with the addition of an overlay of the mesh, to check how it was adapting between these time steps. The mesh is the discretisation of space, which like time must be divided into a finite number of discrete units in order for it to be dealt with on a computer. Part of what makes Fluidity an advanced modelling framework is its ability to simulate using irregular unstructured meshes, and to change the distribution of elements within the mesh as the simulation proceeds (adaptivity). In Figure 2 you can see the mesh evolving from a regular starting distribution towards a distribution that concentrates resolution on the boundary, where the smaller scale dynamics are occurring. Looking at Figure 2, and seeing that the abrupt change in mixing corresponds to an abrupt change in the resolution of the mesh, SS felt a good indication that the problem was something to do with the adaptivity.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 10 of 15
Figure 2: As in Figure 1, but with mesh overlaid
That it is a local problem, affecting only a subset of elements (you can clearly see a localised section of the boundary markedly relaxing its resolution in Figure 2) raised a strong suspicion that the problem resides at a deep level: in the parallel processing of the simulation. These finite element simulations are designed to run on multiple processors by dividing up the elements within the domain (the triangles in the mesh) so that each processor handles a small subset. It seemed likely therefore that something was causing just one processor to behave strangely. This points SS towards further pathways, checking that this was not a hardware malfunction by running a repetition of the test, and adding weight to her suspicion by trying it in serial. That the investigation heads in this direction takes the issue further from SS's own domain of expertise, and having exhausted her own ideas for what would be causing the processor to go astray she took the issue to a weekly meeting, which is usually attended by several scientists who are specialists in parallel processing. Having the problem externalised in these visualisations had facilitated the process of investigation that pointed towards it being a parallel processing issue, and then later these images can be brought into the communal forum to catalyse a dialogue with
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 11 of 15
others, enabling the SS to marshal assistance to help find a way through the technical system towards the source of the problem, a common tactic within a group with a strong communal dynamic. Images tip the pathway towards new courses of action, new tests, new forms of scrutiny, new suspicions. But they also bring individuals together. “It is fairly common to go to others to see if they recognise what is going wrong. Not just in visualisation, but if the compiler fails and someone comes to me and I recognise the error message and know how to fix it... same thing with visualisation. If it goes wrong and the velocities are pointing downwards I have seen a similar thing: 'I saw this... I did this... it should work'” (QS) It is not pure goodwill that SS relies on for help. There are bugs in all complex software, some of which can be very difficult to analyse and fix. It is in all the scientists' best interests to fix bugs when they appear. The bug's manifestation in a failed test such as this could provide a sufficient handle with which to grab the opportunity to iron out a problem. If ignored it would just lie dormant within the code only to cause problems in the future. “All the libraries that we depend on, so PETSc and VTK for example, they all have bugs. We can't fix them either except when we hit one and we fix it and report it back. Compilers have bugs, operating systems have bugs. There is no way we can get rid of them all. We often hit compiler bugs. It is just part of computer modelling as a whole. You are going to have bugs and you are going to hit them. You just have to make sure that there aren't enough bugs that it derails everything, which is why we have test suites. Most research centres by the way do not have test suites.” (QS) In many cases, an automatic alert like this will lead to the problem being traced to the most recent change to the code, the change that triggered this particular round of testing. This can then be searched for errors and fixed or just “rolled back” and removed. This case however was slightly more unusual, and the change in question was not itself the source of the problem. This change had modified the operation of the code in a subtle way that had allowed a deeper more insidious bug to manifest itself. Some bugs, of course, are simple: a bracket too few or too many in a newly added file, or a problem with how the model has been configured. But others can be pernicious. “Sometimes when I see certain problems that I have never seen before someone will say 'That seems like this is going wrong' then I will have a look and find out 'Oh yes that's what is going on'. If someone is really stumped on a problem they will ask and hopefully someone will have seen something like that before. But occasionally you get a problem which noone has seen before, so you just have to battle through it.” (IM)
Becoming a Scientist
The research process does many things. It produces knowledge and generates complex software systems. Indeed, as simulations become more complex, it becomes harder to communicate research through traditional media of scientific publication, leading to the kind of open source model used for Fluidity, which aims to facilitate the direct dissemination of the scientific apparatus, along with the results and the discourse upon them. Yet further to the production of knowledge and to this production of equipment, the repetition of seeing and working upon problems, bugs, causes and solutions also forms a basis of scientific habituation, a milieu of sensitivities across the community. When we look at the meandering course of investigation, it is impossible to separate the becoming of the process of research from the process of becoming a scientist. Over countless iterations of problems and solutions that have characterised his or her work, what the scientist sees (in the image) is subtly and strongly conditioned. Don Ihde talks about what he calls the “multistability” of images, the plurality of what can be seen in an image, and he uses the famous image of the “duck rabbit” to make this point (Rosenberger 2009, 68; Ihde 1999; cf. Wittgenstein 2009, 204).
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 12 of 15
Figure 3: The duck-rabbit
The duck and the rabbit are both possibilities here, but with respect to scientific visualisation, it would help to extend this image, and to imagine one where the duck is more stable than the rabbit, one where only certain people would, through long years of encounters, be habitually inclined to see the rabbit, and one where the rabbit even at its clearest to even the most rabbitinclined observer is still a slightly tenuous and vague suggestion, a fleeting suggestion like a shape recognised in a passing cloud. Where habit is at stake, it is apt to think back to the quote of Rheinberger given above, where he notes that the walls of the labyrinth “blind and guide” the experimenter. The blindness is the lack of clear foresight of what is to come, the basis of the improvised decision, following a vague suspicion. But the same thing that blinds also guides, regulates and determines. This interplay between the regularity and difference, between the rule and the act, is commented on by both Bourdieu and Deleuze, who must be recognised, for all their differences, as two of the most influential commentators on habit. It is in precisely this vein that Bourdieu characterises the habitus as the “durably installed principle of regulated improvisations” (1977, 78). Improvisation is regulated according to principles “which can be objectively “regulated” and “regular” without in any way being the product of obedience to rules, objectively adapted to their goals without presupposing a conscious aiming at ends or an express mastery of the operations necessary to attain them and, being all this, collectively orchestrated without being the product of the orchestrating action of a conductor” (Bourdieu 1977, 72). Deleuze points to the novelty that emerges from and resides within the repetition of the same. “Habit draws something new from repetition – namely, difference... In essence, habit is contraction. Language testifies to this in allowing us to speak of “contracting” a habit, and in allowing the verb “to contract” only in conjunction with a complement capable of constituting a habitude” (2004, 94). Over many instances of generating, navigating and working with and through a great many different images, working through them in the same manner and mode, according to the same kinds of code, the same kinds of physics, and the same kinds of problems, sensitivities are established. Through these, research is regular, yet this regularity opens up the space for improvisation, following barely grounded suspicions to investigate further, suspicions which may well later confirm themselves, but this confirmation only emerges through having already had to act, to follow a suspicion when it was no more than a vague feeling. Those that come to nothing constitute departure points for further exploration while those that are confirmed shed their vagueness, and establish a retroactive teleology that collapses the discontinuity of the decision into the unity of the continuing process. The form of the scientific everyday is not stamped upon it by some a priori definition of the correct scientific method, but grows instead out of the steady repetitive processes of research.
The Fate of the Image
In the above the problem (the error) is seen through the image. The image is not the problem. It is not that the image does not represent, but the question of their representation risks establishing a discord between the questions of the
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 13 of 15
analyst and those of the subject. Images of the lock exchange may well pose questions of representation when at a later point the research is written up for publication. There, the multiplicity of their targets will be unpicked. Do they represent the data? If so, how well? Or are they instead to be representing the simulation? Or the target of the simulation? Or the simulation's representation of the target? Which image will be chosen out of the countless possibilities that can be generated from the same dataset? But those particular images exhibited above will never make it this far. They will never be subject to such questions and will never find their way into publication. They are generated from a simulation which has gone awry. They are part of the greater stock of scientific visual culture, all but the tip of the iceberg, never disseminated, never anchoring truth claims: disposable images created in between times, finding their place in the midst of the repetitions of daily life, never in the final analysis, where they are written out of wider concerns with eventual results. At such points within the becoming of research, even the most deceptive image would be valuable if it helps SS find her orientation, for at issue is what will bring the process forward, and establish for it a future. Confident or tentative, research takes its steps forwards according to such materials, catalysts for its progression. In the end, the image can never be left alone. It is always accompanied by other kinds of narrative and inscription. Indeed, many members of the research group were openly critical of those who relied on images within the arguments they writeup. “Images lie” as one scientist put it, while statistics can potentially provide something much more reliable. While SS works to keep her simulations functional, combating problems such as the one described above, she writes her research up into publications, and here visualisations take a back seat. Here the big question is the representative capacity of the simulation, not of its images, and it will be measured by diagnostic variables, statistical measures of its fit with experimental results, with other scientists' simulations, and with different setups of this simulation modelling a range of analytic and empirical targets. Much more than visualisations, these diagnostics play the central role in the strategies through which claims about the lock exchange and about simulating both might be justified. For this reason, the image draws the eye back into the incomplete process of research, rather than anchoring it to its results, its eventual outcome, which is characterised by an general desire for mathematisation. “You want hard numbers... Turbulence is a good one because it will produce some very pretty animations of flows oscillating and you have the wave region behind something and it is all circulating and it looks amazing, but it only means something if you start analysing it statistically, finding out what it does on average and finding out what the scales of motion are there...”(QY)
Realising the Future of Research
Eric Winsberg notes that simulation does not fit squarely within the concerns of traditional philosophy of science (2010, 3). He attributes this in part to the fact that simulation is usually based on theory that is already known, so it would be easy to assume it would not pose any problems outside those that are already tackled in the philosophy of scientific theory. In other words, if a simulation is built out of theoretical knowledge, it can only discover what is already there. Theoretical systems, such as the theory of fluid dynamics embodied in the NavierStokes equations, are in many cases analytically intractable, so that many of their results are inaccessible to direct solution. Simulation would therefore be a means at getting at these results by means of clever approximations and computational power. Winsberg sets out to dispel this myth by showing the extratheoretical and intertheoretical resources that simulationbuilding requires (2010, 26, 73). While simulation does explore theoretical systems, it also steps outside their scope, by drawing on computational methods and syntheses of conceptually incompatible theories in order to achieve its results. This is an extremely pertinent critique, opening the question of simulation to new kinds of questions. But as I have argued here, we can effect a further displacement, in which the frame is shifted from the simulation to practices of simulating, regimes of activity in which such artefacts are realised. Such regimes take wandering paths forged within concrete cultures assembled around them, generating, working through and discarding artefacts such as images as they go.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 14 of 15
Mary Morgan and Margaret Morrison, emphasising their resistance to being subordinated to theory, call models “autonomous agents”, and aim to establish a new and nontheory centric philosophical space for them to be discussed (Morgan and Morrison 1999, 10). The problem does seem to arise from the fact that artefacts such as software, simulation data, and images of such data, have been wholly constructed by humans and would thus be entirely within the realm of culture, radically distinct from the natural realm where the unknown, mysterious and contingent reside (cf. Strathern 1990). But as I have argued, things are not so simple. Software may have been written by human hand, but this does not imply that it is graspable in any straightforward manner, especially not when it is of a high level of complexity. Large software systems exhibit an unruliness that commentators on software engineering long emphasised, and that programmers have struggled with since the first operating systems (Brooks, Jr. 1995). Scientific software is an intricate labyrinth, one whose construction and navigation are accomplished by one and the same movement. Working upon it in everyday research practice is a matter of enlisting techniques such as visualisation through which scientists try to understand what it is that they have done when they make a simulation. Research is not simply turned outwards towards the domain of nature, but holds itself in the picture too, in a process that follows clear paths as well as negotiating with abrupt discontinuities, where uncertainty is rife and vague suspicions the only guide. Through the image scientists encounter the effects they have brought about, reflect on them, act upon them, an inherent reflexivity in which this kind of research finds its footing. In the space of decision toward the onward path of investigation, what is encountered (in the image) also exerts its own kinds of influences on how and what may be realised in that near future, while traversing a field of practice itself defined by a stratigraphy of such tracings, its cultivation the intrinsic historicity of research sites. The image, I said, is among these processes a point at which what follows unfolds, emerging out of a reflexive entanglement of the process upon itself, a knot sufficient to grasp the future sufficiently to bring it about, and to realise the project of which it was born. ~ Bourdieu, Pierre. 1977. Outline of a Theory of Practice. Trans. Richard Nice. Cambridge: Cambridge University Press. Brooks, Jr., Frederick P. 1995. No Silver Bullet: Essence and Accident in Software Engineering. In The Mythical Man Month: Essays on Software Engineering, 177203. Anniversary Edition. Boston: AddisonWesley. Daston, Lorraine, and Peter Galison. 2007. Objectivity. New York: Zone Books. Deleuze, Gilles. 2004. Difference and Repetition. Trans. Paul Patton. London: Continuum. Galison, Peter. 1996. Computer Simulations and the Trading Zone. In The Disunity of Science: Boundaries, Contexts, and Power, ed. Peter Galison and David Stump, 118157. Stanford, CA: Stanford University Press. Heidegger, Martin. 2001. The Origin of the Work of Art. In Poetry, Language, Thought, trans. Albert Hofstadter, 1586. New York: Perennial Classics. Ihde, Don. 1999. Expanding Hermeneutics: Visualizing science. Evanston: Northwestern University Press. Ingold, Tim. 2001. Beyond Art and Technology: The anthropology of skill. In Anthropological Perspectives on Technology, ed. Michael Schiffer, 17–31. Albuquerque: University of New Mexico Press. Kaufmann III, William J., and Larry L. Smarr. 1993. Supercomputing and the Transformation of Science. New York: Scientific American Library. de Landa, Manuel. 2011. Philosophy and Simulation: The Emergence of Synthetic Reason. London: Continuum. Latour, Bruno, and Peter Weibel. 2002. Iconoclash: Beyond the Image Wars in Science, Religion and Art. Cambridge: The MIT Press. Lynch, Michael, and Steve Woolgar, eds. 1990. Representation in Scientific Practice. Cambridge. The MIT Press. Morgan, Mary, and Margaret Morrison. 1999. Models as Mediators: Perspectives on natural and social sciences. Cambridge: Cambridge University Press. Pickering, Andrew. 2008. New Ontologies. In The Mangle in Practice: Science, society, and becoming, ed. Andrew Pickering and Keith Guzik, 114. Durham: Duke University Press.
Matt Spencer, Goldsmiths Centre for Cultural Studies
P 15 of 15
Rheinberger, HansJörg. 1996. “Comparing Experimental Systems: Protein Synthesis in Microbes and in Animal Tissue at Cambridge (Ernest F. Gale) and at the Massachusetts General Hospital (Paul C. Zamecnik), 19451960.” Journal of the History of Biology 29 (3) (October 1): 387416. ———. 1997. Toward a History of Epistemic Things: Synthesizing proteins in the test tube. Stanford: Stanford University Press, November 30. Rosenberger, Robert. 2009. QuickFreezing Philosophy: An analysis of imaging technologies in neurobiology. In New Waves in Philosophy of Technology, ed. JanKyrre Olsen, Evan Selinger, and Søren Riis, 6582. New York: Palgrave Macmillan. Simondon, Gilbert. 2009. “The Position of the Problem of Ontogenesis.” Trans. Gregor Flanders. Parrhesia 7: 416. Strathern, Marilyn. 1990. Artefacts of History: Events and the interpretation of images. In Culture and History in the Pacific, ed. Jukka Siikala, 2544. Helsinki: Finnish Anthropological Society. ———. 2004. Partial Connections. Updated edition. Walnut Creek: AltaMira Press. Winsberg, Eric. 2010. Science in the Age of Computer Simulation. Chicago: University of Chicago Press. Wittgenstein, Ludwig. 2009. Philosophische Untersuchungen/Philosophical Investigations. Trans. G. E. M. Anscombe, P. M. S. Hacker, and Joachim Schulte. Rev. 4th ed. Chichester: WileyBlackwell.