Ignoring Complexity

Two Versions of Complexity

Complexity has been in vogue across the social and natural sciences in the past few decades. In their introduction to the edited volume Complexities, Mol and Law (2002) observe that across many of the social sciences, a revolt against simplification has been taking place: from history to anthropology, from cultural studies to feminism, “The argument has been that the world is complex and that it shouldn’t be tamed too much …” (p. 1). In history and political theory, critiques of simplification have focused on rationalization and bureaucracy as the hallmarks of modern state power that reduce complexity by ordering, dividing, simplifying, and excluding (Scott 1998; Mitchell 2002). In science and technology studies, Mol and Law identify work that points to problems of scaling up from controlled experimental settings to large-scale technologies or from clinical trials to sick patients. The general shape of these critiques, write the authors, is to argue that “simplifications that reduce a complex reality to whatever it is that fits into a simple scheme tend to ‘forget’ about the complex, which may mean that the latter is surprising and disturbing when it reappears later on and, in extreme cases, is simply repressed” (p. 3). Such a view tends to place simplifications and reductions on the side of representations, while attributing complexity to some essential reality. We should be suspicious of the denunciation of simplification that celebrates complexity, warn Mol and Law, a position “so well established that it has become a morally comfortable place to be,” and, the authors argue, too simple of a move to boot (p. 6). Rather, the authors call for different ways of understanding, relating to, and performing complexity that do not take the relation between simplification and complexity for granted.

Miyazaki and Riles’ (2005) critique of anthropology’s analytic repertoire provides one possible path toward specifying that relationship. Miyazaki and Riles argue that complexity may be the result of the failure of analysis. For these authors, much of contemporary anthropology, with its focus on emergence, indeterminacy, and complexity, reflects analytical problems that have fallen out of the account, problems in relation to which complexity becomes an empirical category and is made to look as though it is “out there.” The diagnosed complexity of social worlds, for example, may tell us more about the failure of anthropological knowledge than about the sites anthropologists are studying. Social scientific analytical strategies, “in response to the apprehension of the endpoint of their own knowledge, [] retreat from knowing. And they also retreat from the recognition of the failure of their own knowledge by locating indeterminacy and complexity ‘out there’ …” (2005, 327). Drawing on Miyazaki and Riles’ critique, we may observe that in certain cases locating complexity “out there” may obscure relations between models or analytics and the excesses for which they fail to account. In such cases, interpreting complexity as an inherent property of phenomena naturalizes complexity as the cause of failure, rather than its effect.

Sandra Mitchell’s Unsimple Truths (2009) highlights the way complexity may also arise between representations. Arguing that the sciences and public policy often fail to take account of complexity, Mitchell advocates a position she terms “integrative pluralism,” which favors “multiple explanations and models at many levels of analysis instead of always expecting a single, bottom-level reductive explanation” (p. 13). Mitchell’s argument against reductionism and for the existence of emergent properties (properties that cannot be explained with reference to the properties of their constituents) draws on the partiality of representations. She writes,

Any representation—be it linguistic, logical, mathematical, visual or physical—stands for something else. To be useful, it cannot include every feature in all the glorious detail of the original, or it is just another full-blown instance of the item it represents. Something must be left out, and what is left out is a joint product of the nature of the representing medium (Perini 2005) and the pragmatic purposes the representation serves …. The partiality of any representation leaves open the possibility that the two representations will simplify the phenomena in incompatible ways.” (p. 31)

These views of complexity have radical implications for the commonly held normative stance that more knowledge is better. As long as we see complexity as residing in the phenomena themselves, this normative stance seems self-evident. We then find it necessary to reach for any relevant discipline to tackle the complexity of phenomena. However, this normative stance is no longer compelling on the view that complexity sometimes grows out of the theories and models themselves or out of relations between conceptual schemes. In such cases, drawing on additional bodies of knowledge may, in itself, compound complexity, while ignorance may have methodological value.

The Naturalization of Complexity

The study of biology has long relied on the simplification of life’s complexity, in order to manage and interpret experimental work. Hans-Jorg Rheinberger (1997) has argued that a finely tuned middle ground between complexity and simplicity in fact defines the strength of biologists’ experimental systems, understood as, the “locally manageable, functional units of scientific research” (p. 246). While experimental systems are “machines for reducing complexity,” that allow experimenters to exert some control, their power lies in being connected to a network of relations outside of the experimental system that keeps such systems from being trivial or closing in upon themselves and enables the generation of anomaly or surprise (p. 247). A balance between simplicity and complexity is thus built into biology’s epistemic practices. Within the experimental systems of biologists, the model organism has perhaps acted as the paradigmatic simplifying tool. Comprising species conducive to certain kinds of research in particular areas and singled out for their relative simplicity, ease of maintenance, and availability, model organisms have produced bodies of research that delineate problems and in turn enable more research (Creager, Lunbeck, and Wise 2007).

While a concern with managing complexity and its simplification has pervaded the study of biological systems, nowhere has a commitment to simplicity been more foundational than in molecular biology. In the last decade, complexity, therefore, has come as somewhat of a surprise in molecular biology and genetics, particularly in the wake of the HGP. The resurgence of complexity in molecular biology stems partly from decades of qualification with reference to the “central dogma”—famously formulated by Francis Crick in 1957 and summed up with the slogan “DNA makes RNA makes protein”—a model of “precocious simplicity” thought to underlie the relationship between genes and living beings (Rheinberger 2000, 228). The central dogma describes a relationship between meaningful information contained in the genetic code understood as a set of instructions with the “capacity to issue orders” (Keller 1995, 95) and the implementation of those instructions in the production and maintenance of life. In this account, genes figure prominently. Yet in recent decades, the central dogma, and its central actor, the gene, have yielded some ground to more “complex” understandings of the relationship between DNA and the life around us. In her book, The Century of the Gene (2000), Keller argues that it has become increasing difficult to pin down what exactly genes are, or, for that matter, what they do. Keller introduces the possibility that genes are on the verge of outliving their usefulness as ways of organizing biological research and thinking about development and heredity. Yet genes are pervasive in what Keller terms gene talk: genes are deeply entrenched in popular ideas of biology, as well as within scientific communities.

The HGP played an important role in undermining the centrality of the gene. One of the major outputs of the project was a staggering amount of information. And as sequencing technologies have become cheaper, information has become easier to amass. Eric Lander, a leader in the public effort to sequence the genome, along with Robert Weinberg (2000), wrote in a triumphant piece on the future of the life sciences over a decade ago, “biology, in the 21st century, is rapidly becoming an information science. Hypotheses will arise as often in silico as in vitro” (p. 1777). The informational terms of genomics have since normalized and become mundane. But what does all this information add up to? The answer is elusive. Thus, the paradoxical result of what has been called the mapping paradigm (Rheinberger and Gaudillière 2004), which included massive and in some ways truly effective efforts, was the realization, in the immediate years following the genome sequence’s completion, that more information had given geneticists a clearer sense of yawning gaps in knowledge. In the decade leading up to the sequence’s completion, the human genome was thought to contain somewhere in the order of 100,000 genes. By the time the human genome map was announced, the number had shrunk to around 30,000 genes and later 23,000 genes, a figure roughly equivalent to the number of genes in Caenorhabditis elegans, the modest roundworm. For many, the newly revised numbers were a shock, since species complexity had been expected to correlate with numbers of genes. Instead, the relationship between genetic code and the life we see around us has taken researchers to parts of the genome they had been keen on ignoring, and beyond, in what Erika Check Hayden (2010) terms a “complexity explosion.” They are now haunted by that once common phrase, “evolutionary junkyard” referring to purportedly noncoding regions.

The image of newly discovered “gaps” in knowledge is a mainstay of post HGP assessments, nicely illustrating Matthias Gross’ observation that “the modern idea of science as a means for turning uncertainty into certainty instead has more often led to more knowledge about what is unknown and perhaps cannot be known” (2010, 52). The interpretation of what these unknowns amount to follows a glass-half-full (of gaps)/glass-half-empty logic. For some, the emphasis is placed on how little we know and how much remains to be done. For others, the accent is placed on our newly acquired knowledge of how complex living beings really are. The former assessment tends to emphasize the epistemic side of complexity. Complexity characterizes the gap between our knowledge and some notion of biological systems in themselves. The latter, celebratory stance tends to diagnose complexity as an immanent quality of living beings, a quality so firmly rooted as to be itself available for discovery.

Ian Hacking (1983) has observed, following Norwood Russell Hanson, that theories have a generational dynamic in the sciences and are naturalized over time: “At first, an idea is proposed chiefly as a calculating device rather than a literal representation of how the world is. Later generations come to treat the theory and its entities in an increasingly realistic way” (p. 30). The naturalization of theories, models, or entities is a precursor to the naturalization of complexity, which proceeds by attributing whatever exceeds the model to the complexity of the natural world but then hides the essential relation between the model and the excess that generates the complexity. Indeed, in order to incorporate research discoveries into an understanding of how genomes work, the account has had to increase in complexity significantly. Biological systems are complex, at least partly, in so far as they exceed and undermine the model of gene action around which molecular biology was built. The epistemic aspect of complexity drops out of the equation when geneticists and molecular biologists address the gaps between our knowledge of genetics and organic systems. The result is that the partial failure of a simple biological model is now being naturalized and attributed to the complexity of organic systems. What is more, since the excess is itself produced in relation to a particular way of dividing up the world in molecular terms, it is not entirely clear that molecular biology can be disentangled from the complexity it produces and in which it is now mired.

This is where the naive hero of our synthetic biological tale gets the grist for his epistemological mill. Collins’ reflections on his own naiveté, cited above, provide a clue to the methodological role of ignorance in the early days of synthetic biology. In the next section, I describe how selective ignorance of the biological substrate was used in order to evacuate biology of some naturalized content and recolonize it with new conceptual frameworks and knowledge practices. These frameworks included the setting of standards for biological parts and the application of “abstraction hierarchies.” To dwell on these tools is not to deny their fallibility (addressed later on) or to ignore the many labs that never used them or abandoned their use. Stavrianakis and Bennett (2014) rightly suggest that many social scientists and other observers have privileged a partial picture that simply restates the grandiose ambitions of some synthetic biologists. Instead, these authors make a case for a more tempered account of the contemporary. By focusing on both the features and shortcomings of these paradigmatic tools, we can tease out relations between synthetic biology and the epistemology of biology.

Building Epistemic Barriers

In his popular essay, “Can a Biologist Fix a Radio?” cell biologist Yuri Lazebnik (2002) suggests that the sensed complexity of biological systems might have something to do with the way biologists have gone about doing their work. Writing that “complexity is a term that is inversely related to the degree of understanding,” Lazebnik argues that biologists, armed with nothing but their methods, would find a broken radio inscrutably complex, whereas engineers could figure out how it works and even how to fix it. In Lazebnik’s caricature of naive investigation, both the biologists and the engineers approach the radio as uncharted territory, with their own methods and priorities. But what happens when one approach finds its way through the output of the other? Such layered epistemic domains raise problems for the strict utility of learning, directing our attention to the way practitioners side-step existing practices and bodies of knowledge. This theme could be tracked in the lab in which I conducted fieldwork, and in some of the more programmatic tools associated with synthetic biology more broadly. The lab, located in the electrical engineering department at Princeton University, was headed by Ron Weiss, an originator and avid devotee of the engineering inspired approach to synthetic biology.

Like Collins’ comical anecdote, Ron’s early career narrative hints at the productive aspects of naiveté and lack of experience. Ron was a graduate student at MIT, when he discovered synthetic biology in the mid-1990s. Enrolled in the department of electrical engineering and computer science, Ron had been pursuing some work in amorphous computing under the guidance of a computer scientist named Tom Knight. Amorphous computing investigates two fundamental questions: “How do we obtain coherent behavior from the cooperation of large numbers of unreliable parts that are interconnected in unknown, irregular, and time-varying ways?” And, “What are the methods for instructing myriads of programmable entities to cooperate to achieve particular goals?” (Abelson et al. 1996). Not surprisingly, amorphous computing draws its inspiration from the cooperative organization of living things like cells and ant colonies, with the hope of applying some of the principles exemplified in these systems to computer science. At one point in their research, it occurred to Ron and to Tom Knight that it might be possible to reverse the arrow. Rather than take nature as inspiration and computers as the target, one could take computers as the inspiration and make nature the target. They decided to try to build controllable biological parts that conformed to the abstract language of “logic gates” (the basic building blocks of digital systems that describe relations between inputs and outputs) and construct systems by wiring logic gates together.

To get started, Ron and a fellow graduate student had to learn how to work with cells. They were completely unfamiliar with wet lab work. They began by taking some undergraduate courses in biology, reading papers and books, and talking to people. Mainly, Ron explained, “We just picked it up as we went along … and just tried things. It was probably not the most efficient way to learn, but it’s kind of the MIT way. You can figure it out by yourself and you don’t need anybody else.” MIT, in Ron’s accounts, was consistently cast as the Mecca of a do-it-yourself ethic. The institution, claimed Ron, was supportive of innovative work done under little or no surveillance. “At MIT, they are very open. They like crazy ideas. Doing something nontraditional is something that people enjoy.”

The first technical hurdle was the production of the first plasmid, a circular piece of DNA used in genetic engineering. Plasmid assembly is notoriously tedious. Weeks of work often end with the realization that one has managed to piece together the wrong genetic sequence. Ron remembered the frustration of those months and the doubts which accompanied the daily lab work: “I was thinking to myself, maybe I shouldn’t have switched to do all this biology stuff.” It took Ron six months to confirm his first plasmid. A week later, a biologist the lab had hired for assistance arrived on the scene. Ron realized the value of this new lab member quickly: “She was a biologist. An actual biologist. Which is something we should have done week one. We should have had an actual biologist.” Ron described the biologist as a technical expert who helped them figure out how to work with cells. For his doctoral thesis, Ron built several digital logic gates in vivo and programmed intercellular communication. He also produced a biological component library. Such libraries of parts are staples of electrical engineering.

In 2003, soon after Ron completed his PhD, Tom Knight proposed a set of assembly standards for bacterial parts, which he named “BioBricks”: stretches of DNA strung together in ways meant to facilitate combination for the construction of biological machines. By standardizing these parts and collecting them in a registry (informationally housed on a wiki and materially instantiated in some fridges in Cambridge, MA), Knight and others sought to streamline processes of fabrication.

The assessment of BioBricks can be divided in two, for while the performance and reliability of parts has been a matter of continuous grumbling from the outset, their symbolic hold in synthetic biology has been a miraculous success. This is perhaps because these biological artifacts are both scaffolding and product for a robust institutional life. Often introduced through a fatigued analogy to Lego sets, BioBricks have enjoyed an extended shelf life due to their inseparability from the international genetically engineered machine competition (iGEM), where university-affiliated teams of undergraduates submit synthetic biology projects that also generate BioBricks, which are then collected in the parts registry and reused in future competitions. The short time frame of iGEM along with the youth and inexperience of participants suggests a key role for amateurism and inexperience in this competition. As Alfred Nordmann (2014) has written, “The iGEM teams seek to find out through a strategic design process how much they can achieve with what little they know. They are not held back by seeking to learn all that would be needed for rationally engineering some biological structure or entity. Instead, they are invited and resolved to short-circuit the scruples of their teachers” (p. 48). Indeed, a team leader with whom I spoke in 2015 extolled the virtues of the “virgin minds” of iGEM participants.

Alongside BioBricks came a chart showing how to put parts together termed an abstraction hierarchy. Abstraction hierarchies are made up of nestled categories with different “levels” of abstraction. These hierarchies are arranged vertically so that the most abstract level is at the top, and the least abstract is at the bottom. The abstraction hierarchy for parts-based synthetic biology begins with DNA on the very bottom. The next level of abstraction is “parts,” referring to standardized genetic parts made of DNA. The next level is “devices,” a series of parts strung together to achieve signaling patterns. Above devices are “circuits,” and, finally, “systems” are usually at the very top.

In ideal form, abstraction hierarchies allow individuals to focus their attention on a particular level of abstraction while ignoring other levels. For example, a person who wants to build a biological device can design that device in terms of inputs and outputs and then plug in the appropriate biological parts without knowing much about those parts or the molecules that underlie these constructs. Thus, abstraction hierarchies are meant to organize both people and knowledge, allowing individuals working at different levels of the hierarchy to contribute to others’ efforts. Explicitly presented as tools for managing complexity, these hierarchies reflect the hope that life forms can be assembled out of separable epistemic levels.

While “abstractions” were meant to organize and streamline processes of fabrication, they also surfaced as a sensed fault line between biologists and engineers in Ron’s career narrative, to which we now return. When Ron completed his PhD, he applied for academic positions mainly in computer science departments. Princeton’s was in fact the lone electrical engineering department in the mix. The faculty brought in molecular biologists to vet Ron’s talk. Had it been solely up to these nonengineers, Ron felt that he would not have gotten the job. He reported general difficulties effectively communicating with biologists through his presentations in the early years of his work.

Especially within a scientific talk, you say all sorts of grandiose things, but then that doesn’t make a connection. And at the end of the day they say, well, regardless of what you say, what did you do? What circuit did you make? What proteins did you use? What organism are you doing this in? A lot of times they cannot break away from looking at the details to understand the main concepts … [A]fter you do the collection of the data, be able to abstract away from that about what’s really going on in the system.

Yet Ron’s contention that biologists were too bogged down in the “details” to understand some more conceptual or abstract levels of order went beyond how biologists approached engineered systems that had been designed with these abstractions in mind. Ron was more generally critical of the way instruction in biology often emphasized details which could themselves be better organized according to informational logics. Ron explained,

There’s over emphasis on the details and sometimes the big concepts are not emphasized … It shouldn’t be like a folklore kind of passing of information by the fire using smoke signals and things like that. It should be a much more, I don’t want to say rigid, but structured activity of collecting data, putting it into databases, being able to access the data, having it be accessible in machine format so that you can have software that actually looks at the data and assembles it and allows you to make all kinds of claims or hypotheses that you can then test out using computation, using modeling, using abstractions.

How Much Biology Do You Need to Know?

In 2008, there were conflicting views in Ron’s lab concerning how much biology one needed or wanted to know to do synthetic biology. On the one hand, there were signs that Collins’ bravura, cited at the beginning of this article, was still an available part of the normative repertoire in the lab. The default view, therefore, was certainly not that more knowledge of the biological substrate was necessarily better. However, a mix of practices and pronouncements suggested the absence of consensus, and creaky research progress encouraged learning about the particular biological substrate at hand in order to overcome impasses.

Ron’s lab at Princeton comprised three postdocs and four graduate students who, in the absence of a bioengineering track, were funneled into the lab from different departments and backgrounds. Ernesto and Meg were the two postdocs in the lab upon my arrival. Ernesto was a molecular biologist, and Meg was a trained neuroscientist. Ana, a Polish protein modeler, joined the lab midway through the year. Claire was a graduate student from molecular biology, who left for another lab after the summer of 2008. Ting, Saurabh, and Josh were all engineering students in various stages of PhD work. All of the lab members were bench scientists (meaning they ran experiments and got their hands dirty).

One afternoon early in my fieldwork, I ran into Josh, one of the engineering students, in a housing complex parking lot at Princeton. Josh was an applied physics major from Caltech who was taking his second-year qualifying exams in electrical engineering. Josh had thus completed the bulk of his graduate training in engineering. I mentioned to Josh that I was brushing up on molecular biology in order to follow the day-to-day goings on in the lab. He responded dismissively, “You don’t have to know much biology to do this stuff.”

In contrast, Meg, the neuroscientist in Ron’s lab, often insisted that intricate knowledge of the biological substrate should be integrated into the designs of synthetic biologists, rather than ignored (a stance that Ron fully endorsed). But she too framed a key piece of her work in terms of ignorance of the biological substrate. In her account, a lack of familiarity with cellular differentiation processes had a positive effect on her research.

When Meg’s funding ran out, Ron asked her to take over a project on insulin-producing cells called “β cells.” Switching between tissue types and pathways was an expected part of research in Ron’s lab and produced a jack-of-all-trades approach to lab work. Meg had five days to learn some endocrinology before giving a presentation to a corporate sponsor that would secure her funding. Her presentation went well, and so the β cells project became her primary project in the lab. She now had to differentiate stem cells into β cells. Through journal research, she found a five-step process that involved a host of growth factors administered at different intervals. This process took weeks to complete. Meg and Ron came up with a plan to differentiate stem cells into β cells in a two-step process. They did some initial research, got a sense for the differentiation process, made the necessary genetic constructs, and ran the experiments. The results, Meg recalled energetically, were dramatic: “We saw the cells turn from these nice little balls to what’s called a cobble stone effect,” she explained. “The cells almost look cuboidal and they’re bright green. And it is too cool. It’s just a really clear effect.”

A year and a half after she had done this work, Meg took a course on stem cells in which she encountered some incredulous colleagues: “They tell you that you can’t directly change stem cells into any of these tissues, and I’m like, yes you can. No you can’t, you have to serum starve them, make them ball up, and then you can differentiate them. And I said, no, you don’t have to do that. And they were in total shock. So our ignorance of typical stem cell biology played to our advantage.”

Meg’s was one of few relatively trial-and-error free success stories, and it was only a partial success. Her stem cells differentiated into the wrong germ layer. In the time I spent observing Ron’s lab, few projects were showing signs of advancing in the hoped-for direction in their early stages. Knowledge of substrate details was therefore often amassed at impasses, through research in scientific journals. In effect, research in molecular biology for these engineers surfaced much like the return of the repressed and often took the form of “troubleshooting”: figuring out what had happened when experiments did not produce anticipated or hoped-for results. More thorough knowledge of biological pathways and processes thus entered through the back door.

Maureen O’Malley (2009) has written of the gap between the rhetoric of rational design and the practice of “kludging” in synthetic biology. Kludging is a colloquial term used commonly in electronic and software engineering for a solution characterized by functionality rather than elegance or efficiency. Synthetic biologists have used the term to describe the dominant problem-solving mode in synthetic biology, which involves ad hoc adjustments to particular situations. As part of the kludging process, troubleshooting dismantled the barrier between synthetic biology and biological knowledge in order to slowly increase “complexity” until progress could be made at the cost of generalizability and therefore also at the cost of an engineer’s abstract symbolic order.

Conceding Complexity

In 2014, a number of high-profile synthetic biologists, Collins among them, concluded that “the most effective practitioners in the further development of the field will be those with an engineering mindset who, functioning alone or in integrated teams, understand most broadly how natural biology works” (Way et al. 2014, 159). Ron, too, delivered such a diagnosis in a similar survey of synthetic biology published that same year. “Finally,” he explained, “synthetic biology researchers are developing an ever-growing appreciation for biological complexity, which requires interdisciplinary research, new circuit design principles and programming paradigms to overcome barriers such as metabolic load, crosstalk, resource sharing and gene expression noise (and sometimes actually utilize these barriers to create more robust systems)” (Church et al. 2014, 289). These concessions link the confirmation or discovery of complexity to a need for more knowledge of the biological substrate, whether through firsthand knowledge or through interdisciplinary research.

Yet while synthetic biologists and observers of all kinds have emphasized or conceded the ontological complexity of the biological, there may still be room for appreciating inter-representational sources of complexity in the field. For example, a raging debate in and around synthetic biology in the last decade engaged in by practitioners, philosophers, and other involves the modularity of biological parts. Modularity refers to the extent to which biological parts can be separated and recombined. The debate precisely queries the boundaries between the conceptual and ontological in biological knowledge. In a review of synthetic biology, Way et al. (2014) write, “An open question is whether biology is genuinely modular in an engineering sense or whether modularity is only a human construct that helps us understand biology. To clarify, it may be useful to distinguish between ‘concept modularity’ and ‘engineering modularity’” (p. 152). The engineering modularity of biological systems seems increasingly elusive. But it is the concept modularity of biology that has been asked to play the role of engineering modularity. The question then is whether the failure of the concept modularity of biology to translate into engineering modularity signals the failure of engineering modularity as such for synthetic biology. In other words, the attempt to engineer life forms using concepts derived from biology implicates relations between differently oriented kinds of representations. As Mol and Law (2002) note,

Often it is not so much a matter of living in a single mode of ordering or of “choosing” between them. Rather it is that we find ourselves at places where these modes join together. Somewhere in the interferences something crucial happens, for although a single simplification reduces complexity, at the places where different simplifications meet, complexity is created, emerging where various modes of ordering (styles, logics) come together and add up comfortably or in tension, or both. (p. 11)

Methodological Ignorance

I conclude by turning to a couple of key contributions to what is now a sizable literature on ignorance (for detailed reviews of the scholarly literature on this topic, see Gross 2010; McGoey 2012). In his introduction to the edited volume Agnotology (2008), Robert Proctor critiques the trivializing stance that takes ignorance to be something that always requires rectification. He writes, “Ignorance is most commonly seen … as something in need of correction, a kind of natural absence or void where knowledge has yet to spread.” Instead, Proctor suggests, “[I]gnorance is more than a void, and not even always a bad thing” (p. 2). Linsey McGoey (2012) also warns that much of the work on ignorance tends “to view ignorance as a de facto negative phenomenon, something social actors have an obvious interest in seeking to overcome or to eradicate, and which sociologists have an onus to better identify so that actors are equipped to recognize and combat their own ignorance” (p. 554). McGoey analyzes ignorance from the perspective of actors’ interests within organizational settings. From the perspective of the actors she describes, far from being a negative phenomenon needing remedial attention, ignorance proves useful as a tool to combat liability and assert expertise.

In contrast to McGoey’s work on strategic ignorance, my aim has been to give ignorance more epistemological weight. In this sense, my approach is more resonant with Gross’ (2010) and Gross and Bleicher’s (2012). In their study of environmental contaminant remediation projects, Gross and Bleicher examine how practitioners’ recognition of the known limits of knowledge, which they term “nonknowledge,” allows for the pursuit of projects that are likely to involve some surprises. My own aim is to similarly move from strategy to method, while retaining some of McGoey’s emphasis on purposive ignorance, thus showing how ignorance might be recruited to combat naturalizations and mount an offensive against epistemological complexity.

At the same time, the ignorance I have described is much more ethically fraught than Gross and Bleicher’s nonknowledge, for it implies ignorance of that which is thought to be known. Thus, taking purposive ignorance as a kind of method involves acknowledging that the line between methodological ignorance and neglect is a thin one, if one exists at all. The troublingly fluid borders between different ethical formulations of ignorance are grounds for taking interest in methodological ignorance and the problems it raises, not for ignoring it. I therefore conclude by recalling a famous and often forgotten incitement to ignorance, laid out with great panache in the methodological anarchism of the philosopher of science, Paul Feyerabend. Arguing that “prejudices are found by contrast, not by analysis” (p. 22), Feyerabend ([1975] 2002) suggests that the different pieces necessary to test a new theory or conceptual system could not possibly all show up on the scene at the same time. This is because observation, including the trivial-seeming kind, carries a kernel of naturalized interpretation within it, the latent signs of a bygone era or the theory-laden tectonic marks of a prevailing view. An entire apparatus of sciences and instruments is required to bring observation into alignment with a new theory or conceptual system, but these often lag. The way to proceed, as Feyerabend famously puts it many times, is counterinductive. A new conceptual system is thus not tested by rational procedure and compatibility with observation. Instead, Feyerabend writes, the test of a conceptual system comes after we “wait and … ignore large masses of critical observations and measurements” (p. 113, italics in the original). Not all ignorance is good for Feyerabend. He laments the ignorance of anthropologists who failed to understand, much less record, the astronomical and cosmological beliefs of different groups. Ignorance is by no means a diagnostic of good practice but part of a plural world of methodological tactics tidily excluded from view by many practitioners, epistemologists, and philosophers of science (Feyerabend 1975, 113).

Feyerabend’s incitement to ignore is illuminating in another way. In ignorance’s tie with waiting, ignorance becomes nail-biting work. It requires a carefully blinkered view and the ability to bide one’s time, something that engineers, many of whom tethered their funding to technological promises, didn’t build into their research programs.