Responsible epistemic innovation: How combatting epistemic injustice advances responsible innovation (and vice versa)

ABSTRACT

Epistemic resources, including concepts, categories, and metrics, are invented regularly. Yet this process of epistemic innovation has not been recognized in responsible innovation (RI) scholarship. I argue that it should be: epistemic innovation can foster central goals of RI, including anticipatory governance and alignment with the goal of epistemic justice. An RI framework can help direct and evaluate epistemic innovation, as shown in three examples of epistemic innovation in communities adjacent to oil refineries: initiated without RI in mind, not all were well aligned with epistemic justice and each would have been strengthened by a stronger commitment to deliberation and foresight. These examples highlight challenges to achieving responsible epistemic innovation: having innovation be mistaken for error; coalescing experience and data into intelligible epistemic resources; and structuring inclusive deliberation. These challenges can be addressed by developing new forms of material deliberation and including resources for responsible epistemic innovation in RI policy.

KEYWORDS:

• Epistemic resources
• epistemic injustice
• environmental regulation
• environmental justice
• air monitoring

Introduction

In 2015, the U.S. Environmental Protection Agency issued updated regulations for oil refineries. For the first time, the rule required refineries to monitor ambient air levels of benzene at their perimeters. It also required companies to calculate a new quantity: ΔC, the difference between the highest and the lowest concentrations of benzene measured in any sampling period. In the context of the regulation, the ΔC is taken to represent a refinery’s contribution to benzene concentrations, above and beyond background levels.

In the San Francisco Bay area, two oil refineries had already been monitoring ambient levels of air toxins. Following major accidents and community activism, Phillips 66 in Rodeo and Chevron in Richmond had both installed continuous monitors that reported results in real time to a web site. From 2015–2018, I worked with an interdisciplinary group of researchers and residents of communities adjacent to Bay area refineries to try to make sense of the voluminous data produced by these monitors. One of the outcomes of the project was another new quantity: the Toxic Soup Index, or the number of chemicals detected by a monitoring station at any given time. Inspired by residents’ concern about the possible synergistic effects of chemical exposures, the Toxic Soup Index aims to represent refinery impacts in the aggregate.

Both the ΔC and the Toxic Soup Index are examples of epistemic innovation, or the creation of new epistemic resources—concepts, categories, and metrics that help us understand the world and our experiences of it. Because epistemic resources are both incomplete and unevenly distributed, epistemic innovation is important for making more aspects of human experience legible. For marginalized groups, whose experiences tend to be poorly represented by dominant concepts and categories—an aspect of what philosopher Miranda Fricker (2007) has dubbed epistemic injustice—epistemic innovation can be vital to changing the conversation about social and environmental injustices. The Toxic Soup Index, for example, speaks to the sheer volume of exposures experienced by communities on the frontlines of pollution, in a way that environmental risk assessment—an epistemic resource favored by regulators—cannot.

Despite its importance, epistemic innovation has received little explicit attention in studies of responsible innovation (RI). Greater attention is warranted because epistemic innovation has significant potential to contribute to the central goals of RI. Responsible innovation seeks to push sociotechnical development in the direction of meeting societal needs, among them the need for social and environmental justice (van Oudheusden 2014). Because it entails the creation of more adequate epistemic resources, epistemic innovation can help enable greater recognition for marginalized groups, an important ingredient in the pursuit of social justice (Fraser 2000). Responsible innovation also calls for using foresight, scenario planning, and other techniques of anticipation to manage the emergent consequences of new technologies (Owen et al. 2013). But these consequences may not be fully or accurately knowable using existing epistemic resources, making epistemic innovation integral to effective anticipation.

Yet if epistemic innovation is to fulfill its promise to contribute to the goals of RI, that innovation must itself embrace the core components of responsible innovation, including societal alignment, inclusive deliberation, and anticipation. This paper takes as examples three efforts to expand epistemic resources for representing air quality in communities adjacent to oil refineries, known as fenceline communities: Levels of Concern, ΔC, and the Toxic Soup Index. Epistemic injustice is rife in this domain, in that environmental regulators’ categories and metrics for understanding air quality fail to capture fenceline communities’ everyday experiences of pollution (Ottinger 2022). Using an RI framework, I assess how well each of these efforts served to combat epistemic injustice, and suggest ways that adopting RI principles could have better aligned these innovations with societal needs for epistemic resources.

Attending to epistemic innovation within an RI framework also demonstrates obstacles to epistemic innovation that have yet to be explored by RI researchers or addressed in RI-promoting policy recommendations. Efforts to develop new epistemic resources for fenceline communities meet three challenges likely to be found in other attempts at responsible epistemic innovation. Marginalized groups struggle to establish the legitimacy of new epistemic resources in the face of hegemonic ones (the error problem). Methodologies and information infrastructures that enable diverse groups to collaboratively coalesce lived experience, social priorities, and available data into intelligible epistemic resources are scarce and underdeveloped (the data deliberation problem). Finally, the places where epistemic innovation is most needed may also be places where there are the fewest resources for engaging diverse populations in deliberation (the inclusion problem). To overcome these challenges, RI practitioners will need to become adept at recognizing epistemic innovation and articulating its value in expanding epistemic resources. To foster responsible epistemic innovation, practitioners will also need to create new forms of material deliberation (Davies et al. 2012) well suited to thinking through data. Finally, responsible epistemic innovation will need to be included explicitly in policies designed to foster innovation that meets societal needs.

Epistemic resources and innovation

Philosopher Gaile Pohlhaus explains that ‘knowing requires resources of the mind, such as language to formulate propositions, concepts to make sense of experience, procedures to approach the world, and standards to judge particular accounts of experiences’ (Pohlhaus 2012, 718). These concepts, procedures, and standards are examples of ‘epistemic resources,’ or ‘hermeneutical resources,’ the tools that we use to know the world and to communicate about our experiences to one another (Pohlhaus 2012; Fricker 2007). Other kinds of epistemic resources include classification schemes, numerical indicators, and techniques of quantification (c.f. Bowker and Star 1999).

Epistemic resources are collectively held. Because they facilitate knowing and understanding among people, they are effective tools only when they are widely acknowledged, accepted, and used (Pohlhaus 2012). The epistemic resources that we do hold in common have been shown to be a powerful force in coordinating action and shaping social life. Bowker and Star (1999), for example, show how classification schemes make some practices (and people) available for study and intervention, while rendering others invisible. Tsosie (2012) argues that, because US law lacks concepts for understanding cultural harms wrought by environmental degradation, it leaves indigenous peoples without recourse when their lands are violated.

As these examples suggest, epistemic resources are partial and may be lacking in consequential ways. This creates the possibility, and even the mandate, for what I am calling ‘epistemic innovation,’ or the creation of more adequate epistemic resources (see Ottinger forthcoming). Pohlhaus describes how the need for epistemic innovation arises:

our epistemic resources must answer to our experiences even while we must answer to them. Good epistemic resources put us in particular relation to our experiences (for example, noticing more or certain kinds of details about the experience or anticipating what will follow from the experience). If our language, concepts, or standards don’t do that, then we need to develop new resources that do. This in fact is what we do when faced with such situations (Pohlhaus 2012, 718).

Although scholars acknowledge the need for, and the fact of, epistemic innovation, very few have documented its processes. Fricker (2007) describes the emergence of the term ‘sexual harassment’ as an example of expanding hermeneutical resources, but does not dwell on how the term became a widely shared concept. Meadowcroft and Fiorino (2017) offer perhaps the most detailed study in this area, documenting how new concepts such as ‘sustainable development’ and ‘biodiversity’ have come to structure environmental governance. Yet in focusing on conceptual innovation, they tend not to thematize the myriad other epistemic resources, such as thresholds and ranking systems, that had to be invented in order to institutionalize these important concepts.

Nor has there been purposeful research about why particular concepts succeed or fail in becoming part of our shared pool of epistemic resources. ‘Slow violence,’ a concept advanced by Rob Nixon in his 2011 book, has been an influential idea among environmental scholars but has seemingly not been taken up in environmental policy, at least not yet. While we might speculate that this has partly to do with difficulties in quantifying environmental degradation that takes place over the very long term (Mah 2017), other potential factors determining the fate of epistemic innovation have yet to be explored.

Scholarship on responsible innovation has also been largely silent about the processes of epistemic innovation. New developments in knowledge infrastructures such as digital agriculture and sea ice services systems would seem especially likely to involve epistemic innovation. These technologies draw on new data collection and processing capabilities to provide actionable information to a variety of users and are necessarily characterized by their categories and metrics: studies like Bronson’s (2019) examination of the ‘uneven engagements’ of small and large farmers with digital agriculture show how the concepts and assumptions embedded in knowledge infrastructure have consequences for social outcomes. Yet in conducting deliberative foresighting and scenario planning exercises to anticipate the effects of these knowledge infrastructures, RI researchers have so far tended to treat epistemic resources as fixed and stable (e.g. Blair, Lee, and Lamers 2020; Bruynseels 2020; Fleming et al. 2021).

RI and epistemic innovation

Although RI scholarship has not thus far attended to epistemic innovation, the development of new epistemic resources is a form of innovation that merits attention within the RI framework. The RI movement centers on two concerns: aligning innovation with societal needs and guarding against undesirable consequences of innovation, even before evidence of impacts is available (von Schomberg 2013; Owen et al. 2013; Stilgoe, Owen, and Macnaghten 2013).

Theorists of epistemic injustice show how societal alignment is at stake in the development of epistemic resources. Shared epistemic resources are not evenly distributed, they argue; gaps are most likely to occur around the experiences of marginalized groups, a subcategory of epistemic injustice that has been termed ‘hermeneutical injustice’ (Fricker 2007; Tsosie 2012). Among the reasons for this are that those who are marginalized have structural incentives to understand the experiences and perspectives of those who oppress them—indeed, their survival may depend upon it—while dominant groups may be actively interested in remaining ignorant of what those on the margins live through (Pohlhaus 2012; Tuana 2006). New epistemic resources advanced by marginalized groups may also be dismissed by dominantly situated knowers, especially if they characterize experiences not shared by dominant groups, because they appear ‘to attend to nothing at all, or to make something out of nothing’ (Pohlhaus 2012, 722).

The development and uptake of new epistemic resources thus has consequences for social justice. Epistemic innovation that pushes dominant groups to acknowledge the unique experiences of marginally situated knowers can help to foster a more equitable society. The RI commitment to societal alignment should thus promote innovation that fills epistemic resource gaps around the experiences of marginalized groups, no less than it encourages innovations that meet the material needs of low-income populations.

RI’s approach to avoiding the undesirable consequences of innovation also intersects with the development of epistemic resources, in two ways. First, enacting ‘a collective commitment of care for the future,' a defining feature of RI (Owen et al. 2013, 36), depends on having appropriate epistemic resources. These might include categories for the various sorts of harm that could be done by new technologies, so that care may be taken to avoid them, or metrics that enable societal outcomes to be reliably tracked. To be sure, many such categories and metrics already exist and could be used to learn about the impact of innovation. However, these existing epistemic resources may or may not prove adequate to apprehending the potentially transformative consequences of new technologies. For example, social media platforms have proven hard to regulate in part because they defy existing categories. Facebook is not a media outlet, but neither is it a public square. The contestation over what it is highlights a weakness in our epistemic resources for understanding communicative forums. Here, epistemic innovation could aid in efforts to care for the undesirable effects that social media platforms have had. In addition, making epistemic resource needs an explicit subject of deliberation and scenario planning could invite greater attention to the different kinds of impact new technologies have on differently situated populations. It also stands to make deliberation more inclusive by highlighting the possibility of multiple frameworks for sense-making, and by extension the recognition of diverse ontologies that already exist among participants (c.f., Valkenburg et al. 2020).

Second, the need for forward-looking care that RI emphasizes extends to epistemic innovation, whose consequences, like those of other innovations, cannot be fully known in advance. The new concepts of environmental risk and risk assessment, for example, helped to justify environmental regulation in the deregulatory climate of the 1980s but ultimately worked to the disadvantage of environmental justice communities (Kraft 2017; Kuehn 1996). Innovative techniques for measuring air pollution have aided communities in communicating how petrochemical facilities’ accidents and flaring affected their everyday life (Ottinger 2022), but measurements of high pollution levels also risk tarnishing the image of a neighborhood and its people (Ottinger 2013; Phillimore and Moffatt 2004). RI principles of anticipation, reflexivity, deliberation, and responsiveness (Owen et al. 2013) could help ensure that epistemic innovations do not end up undermining the social goods that they aim to promote.

Scholars and practitioners of responsible innovation, then, potentially have much to learn from incorporating epistemic innovation into their research and practice. Making gaps in epistemic resources a focus of RI can help innovation align with social needs by fostering the recognition of marginalized groups’ experience. And approaching emergent technology with the idea that new epistemic resources will be required to fully understand its consequences can strengthen anticipatory governance. Realizing these gains, however, requires developing practices of responsible epistemic innovation. This is all the more challenging because research to date has so rarely focused on the processes through which new epistemic resources are made.

Methods

To better understand how epistemic innovation occurs, how RI frameworks can be applied, and what obstacles might hinder efforts at responsible epistemic innovation, I examine three cases of new epistemic resources being developed to understand air quality in fenceline communities adjacent to oil refineries. The cases are drawn from a long-term ethnography of advocacy by environmental justice (EJ) activists for ambient air monitoring in fenceline communities, where epistemic resources for understanding air quality have long been inadequate to the lived experiences of residents. The measures of ‘average’ or ‘representative’ chemical concentrations favored by environmental regulators capture neither the temporal nor spatial variations that characterize communities’ encounters with air pollution (see Ottinger 2022). As a result, although the research was not originally designed to investigate epistemic innovation per se, one of its key findings has been that activists’ and regulators’ strategies for expanding air quality do entail the development of new epistemic resources (Ottinger forthcoming). Although none of the cases of epistemic innovation that I found were RI by design, examining them critically through the lens of RI offers insight into how epistemic innovation can be effectively incorporated into the larger RI framework.

Research began in 2001 and continued through 2018, and I remain engaged informally with activists around various air monitoring and data interpretation projects. Research activities have included interviews with fenceline community residents and EJ activists involved in community air toxics monitoring, in campaigns for more comprehensive monitoring by industrial facilities, or both; participant-observation in three EJ organizations who were leaders in providing technical assistance with monitoring to fenceline communities; and examination of new regulatory programs to expand or require monitoring in fenceline communities, through attendance at public meetings and review of public documents. Most of my research has focused on the San Francisco Bay area, where fenceline communities and their EJ allies pioneered new monitoring techniques in the mid-1990s that were subsequently adopted by communities around the country and the world, and where regulators were the first to require continuous ambient air monitoring at refinery fencelines (see Ottinger 2016). Southeastern Louisiana, where fenceline communities invested heavily in, and advocated vigorously for, better air monitoring, was also included in the study, as were federal regulatory programs. Finally, my research includes a participatory design project, in which I collaborated with residents of Bay area fenceline communities, software engineers, and data scientists, in an effort to create better interpretive tools for continuous air monitoring data.

Epistemic innovations in understanding air quality

In the air quality domain, it is always a heterogeneous assemblage of epistemic resources that determines how the air, and fenceline communities’ experience of it, is understood. The epistemic resources involved include concepts for understanding air-related phenomena, monitoring devices, and interpretive techniques. These resources are intertwined and mutually reinforcing: concepts shape measurement and interpretation, even as the ability to collect certain kinds of data shape the conceptual underpinnings of air quality assessment.

The epistemic innovation discussed here—Levels of Concern, ΔC, and the Toxic Soup Index—act differently with respect to concepts, measurement technologies, and interpretive frameworks, inventing here, repurposing there. Yet each constitutes epistemic innovation in that the distinct assemblage that it creates offers a new way of understanding and communicating about the state of the air in fenceline communities. None exemplify all of the principles of RI; rather, they offer a window into how societal alignment, inclusive deliberation, and anticipation might function—or founder—in the context of epistemic innovation in particular. The degree to which these innovations have or have not become widely shared epistemic resources also suggests some of the challenges responsible epistemic innovation will have to overcome.

Levels of Concern

Fenceline communities experience intermittent, semi-frequent bursts of strong odor from nearby facilities. As a class, these ‘odors,’ or ‘releases,’ are not easy to talk about, nor is their impact on communities. There is no category for them in environmental regulation. Individual instances may be attributable to particular sources—flaring, a fire, a broken valve—which may or may not occasion regulatory intervention. Other odors go unaccounted for. Because they do not exist as a category, there are also no standards for, for example, how many odors are allowable each month, or how bad of an odor is too bad to be tolerated. In short, there has been a paucity of epistemic resources for understanding communities’ experiences of odors, especially in the environmental regulatory context.¹.

In the mid-1990s, activists in northern California developed the ‘bucket’ air sampler for use by fenceline communities (Kullenberg 2015). Buckets are an example of a well-established kind of technology, the ‘grab sampler.’ Grab samplers suck air into an evacuated vessel, which is then sent to a laboratory for analysis. Buckets, however, were designed specifically for the odors experienced by fenceline communities. The vacuum is of moderate strength, allowing the bucket’s sampling bag to fill over a period of several minutes. This makes bucket sampling long enough to be unaffected by small variations in the wind but short enough to not to dilute the strongest odors with cleaner air. The device is inexpensive to build, and only its easily replaceable sampling bag is sent to the laboratory. These features enable communities to have several on hand to capture unpredictable odors as they occur.

The invention of the bucket was accompanied by new techniques for sampling and interpreting sample results. The activists who train communities to use buckets (some of them residents of fenceline communities themselves) advise taking a sample when an odor is a at least a six or seven on a scale of one to ten, with ten being the worst smell the community experiences. This heuristic helps to solidify odors as a category. It matters not whether the smell is of rotten eggs or burning tires, its intensity can still be rated. Nor does it matter whether the source of the smell is a visible flare or an unidentified leak; its effect on the community exists independent of the nature of the industrial malfunction.

Further innovation was necessary to make sense of sampling results. The information that an air sample contained 3 parts per billion (ppb) of benzene means little on its own. And because ‘odors’ are not regulated as such, government agencies offered no ready-made categories or thresholds that would allow activists to determine whether 3 ppb in a five-minute sample was ordinary or egregious. Activists had to invent their own ways of putting bucket data in context.

They did so by appropriating the guidance that agencies do publish about chemical levels that could be hazardous to human health. A few U.S. states, including Louisiana, have statutory limits on toxic air contaminants in the ambient air. Other agencies have non-enforceable, health-based guidelines, such as the Agency for Toxic Substances and Disease Registry’s (ATSDR’s) ‘minimal risk levels.’ In all cases, the guidance for each toxicant includes a time period (e.g. eight hours or one year) along with a concentration. Regulators consider air quality to be of more-than-minimal risk or out of compliance when the average level over that time period exceeds the published value.

Working with early bucket users, Louisiana chemist and MacArthur Fellow Wilma Subra compiled several sets of these guidelines into ‘Levels of Concern’ for bucket results.² The Levels of Concern retain only the threshold chemical concentrations and not their associated averaging times. Thus transformed, the levels are compared to bucket results, regardless of whether the original standard or risk level was for one hour or a whole year. Activists justify this practice by arguing that odors are a chronic problem for fenceline communities. As such, levels developed for chronic exposures are applicable to chemical concentrations measured in bucket samples.

Drawing on Levels of Concern, bucket users have yet another way to quantify odors. In addition to their original, embodied assessment of the odor’s intensity (7 out of 10), and the chemical concentrations measured in the bucket sample, activists speak of odors in terms of multiples of Levels of Concern. They might say, for example, that benzene levels in the community were four times the Louisiana state standard and hydrogen sulfide levels were ten times the ATSDR’s minimal risk level. All of these modes of quantification constitute new epistemic resources, or tools for understanding the experience of odors in fenceline communities. The bucket air sampler is integral to these epistemic innovations, but, importantly, the innovation extends beyond the instrument to include how it and its data are deployed. The concept of ‘odor’ or ‘release’ existed prior to the development of the bucket; in that sense, it is not a wholly new epistemic resource. Yet the ability to compare and quantify odors does help to establish them as a category, and an idea that could shape strategies for environmental protection—against a prior set of concepts that focused on industrial events as causes of excess emissions.

Levels of Concern and other bucket-related epistemic resources for understanding odors embody some of the principles for responsible innovation. These epistemic innovations are clearly aligned with the societal goal of environmental justice: they help to produce knowledge of the chemicals to which the communities on the front lines of petrochemical pollution are exposed. Further, they help to remedy the hermeneutical injustice these communities face, by quantifying ‘odors’ and allowing community claims about them to circulate more easily.

The RI principle of inclusive deliberation, on the other hand, is difficult to apply to these epistemic innovations. No deliberative process was established to create the buckets or Levels of Concern. Rather, they were developed by scientists and activists who had extensive experience working in fenceline communities, then they were modified and refined in use by communities. In this sense, the process of creating these resources could be considered one of ‘material deliberation’ (Davies et al. 2012). Activists did not aim for inclusivity in the sense of bringing all possible stakeholders to the table—indeed, had they done so, their attempts to create new epistemic resources might have been quashed by environmental regulators and industry supporters. Instead, they attempted to work broadly with those most affected by odors.

Finally, anticipation is notably absent from epistemic innovation here. I have found no indication that activists considered possible down sides to these technologies—among them, that bucket samples showing high levels of pollution could tarnish a community’s image (Ottinger 2013) or that quantifying odors could create a ‘data treadmill’ for EJ activists (Shapiro, Zakariya, and Roberts 2017). This would have been an extraordinary step them to take; nonetheless, it does distinguish this epistemic innovation from what responsible epistemic innovation would look like.

Developed more than 20 years ago, Levels of Concern and buckets have not been widely adopted or institutionalized. These ways of thinking and talking about odors have been influential in fenceline communities around the world, but they have not become part of the pool of epistemic resources shared by people without first-hand experience of industrial pollution. Environmental regulators have continued to connect industrial releases to specific events or malfunctions, undermining the idea of them as a category of harm. Moreover, they have rejected the validity of Levels of Concern. Pointing to the averaging times associated with the original screening levels, they scold activists for comparing apples (5-minute sampling results) to oranges (8-hour or annual average standards). From their point of view, appropriating screening levels to communicate both the intensity and the chronic nature of odors is not an innovation; it is an error.

ΔC

In 2015, the US EPA issued a long-overdue update to their regulations governing petroleum refineries. For the first time, the updated regulation included a requirement for fenceline monitoring. Under the new rule, refineries must measure ambient air levels of benzene at multiple points around their perimeters. They do so using sorbent tubes to collect the gasses in the air over a 2-week period. Laboratory analysis determines the average concentration of benzene at each sampling site—usually a few dozen per refinery—over those two weeks.

In addition to prescribing a monitoring method, the EPA rule specifies how sampling results are to be interpreted and acted on. For each sampling period, a refinery subtracts its lowest measured benzene concentration, taken to represent local background levels, from the highest measured benzene concentration. The resulting value, the difference in concentration or ΔC, becomes the basis for determining compliance and enforcement. If a refinery’s six-month rolling average of ΔC exceeds a certain threshold, they are required to investigate the causes and take corrective action.

The ΔC is an epistemic innovation that expresses the idea that refineries bear only a fraction of the responsibility for air pollution. Refinery officials—and even some residents of fenceline communities—are quick to point out that toxicants in the air can come from numerous sources, including traffic, small businesses like dry cleaners, and large industrial facilities other than the refinery. Accordingly, when refineries concede to community demands for ambient air monitoring, they take pains to adopt methods that can help allocate responsibility. An air monitoring program in Norco, Louisiana, for example, placed one sampling site next to a busy road as a point of comparison. Initiatives that feature continuous monitoring usually also monitor wind direction, in order to facilitate the laying of blame. The ΔC parses responsibility by normalizing the lowest benzene levels measured at a refinery’s fenceline as ‘background’ from other sources and attributing to the refinery only the benzene that remains after the background levels are subtracted.

While ΔC can be seen as an epistemic innovation, it can hardly be seen as a responsible one, judged by the principles of RI. It could perhaps be said to meet the social need of quantifying refineries’ impact on air quality. However, it does not allow regulators to distinguish between a facility (call it Refinery A) whose lowest measurement is 3 micrograms per cubic meter (µg/m³) and whose highest is 3.5 µg/m³ from a facility (Refinery B) whose lowest measurement is 0.6 µg/m³ and whose highest is 1.1 µg/m³. Both refineries have a ΔC of 0.5 µg/m³, yet it would be hard to justify the assumption that emissions from Facility A make no contribution to the 3 µg/m³ of benzene measured at its fenceline—it is simply too high to be written off as ‘background.’ Moreover, the need to quantify refineries’ impact on air quality arises as part of the more fundamental need to protect the health of the people living nearby. From a public health standpoint, the benzene levels measured at Refinery A would be of much greater concern than those at Refinery B. Yet the innovation of ΔC renders the facilities equivalent in terms of their risks to neighbors. In so doing, it forestalls action to protect fenceline communities and undercuts the societal goal of environmental justice.

It would also be hard to claim that any sort of inclusive deliberation was involved in the development of the ΔC. EPA staff sought public comments only once they had drafted the updated rule, including the methods for monitoring and data interpretation. The EPA received hundreds of unique letters—and hundreds of thousands of form letters—urging them to reconsider the structure of the fenceline monitoring requirements. EJ activists called for continuous monitoring of multiple pollutants; industry criticized the requirements as too burdensome and unhelpful in pinpointing the source of emissions (within and external to a refinery). The public comments, however, did not change the EPA’s approach. The monitoring methodology in the draft rule, including the innovative ΔC, carried over to the final rule intact.

Anticipation also seems to be absent from this instance of epistemic innovation. There is no evidence that the EPA considered, for example, how a refinery’s ability to articulate its share of local pollution problems might change the balance of power between large polluters, small polluters, and exposed communities, or even how it might give refineries new kinds of leverage in interactions with regulators.

Nonetheless, the ΔC can now be counted among the collective epistemic resources used to understand refinery pollution and its impact on fenceline communities. It is institutionalized in regulation and, likely as a result, it is central to how regulators and refinery representatives now talk about fenceline monitoring and compliance. As part of the regulatory structure, the ΔC is also available to fenceline communities and EJ activists who wish to talk about the effects of refinery pollution. Indeed, the Environmental Integrity Project used ΔC values—calling them ‘net concentration’—to rank the dirtiest refineries in its analysis of fenceline monitoring data collected under the updated EPA rule (Kunstman, Schaeffer, and Russ 2020). I have not, however, heard residents of fenceline communities refer to the ΔC; instead, they generally dismiss the 2-week sampling as a bankrupt endeavor and continue (I would say) to look for more adequate resources for characterizing refineries’ impacts.

Toxic Soup Index

In 2016, the Bay Area Air Quality Management District (BAAQMD) adopted a rule that requires all northern California refineries to conduct fenceline monitoring in keeping with what activists had asked the EPA for: continuous monitoring of multiple chemicals, with measurements reported to a website in real time. The Unocal refinery in Rodeo, California,³ had set a precedent for this kind of monitoring when, under pressure from neighboring communities, it installed optical remote sensors following a 1994 accident (Ottinger 2016). In 2013, the Chevron refinery in Richmond followed suit with both fenceline monitoring and three community monitoring stations that also provided real-time data about air quality.

The new fenceline monitors expanded data without expanding epistemic resources for making sense of it (Ottinger and Zurer 2011; Ottinger 2016). In 2016, with monitors in place at two Bay area refineries and expected to come online at the other three,⁴ I initiated a participatory design process with the goal of developing tools that would enable fenceline communities and EJ activists to better access and mobilize the data being produced by these new monitors. The process included residents of several, but not all, of the Bay area’s refinery-adjacent communities, long-time advocates of fenceline- and community monitoring, computer scientists, and data experts. It deliberately excluded regulators and refinery representatives, who, in my experience, tended to participate in conversations with community members and activists by instructing them on the ‘right’ ways to think about pollution and air quality (see Ottinger 2013).

While at the time I was not thinking explicitly in terms of ‘epistemic resources,’ I was very aware of activists’ critiques of these ‘right’ ways and wanted to create conditions under which we might collectively find alternatives.

One of the outcomes of this process was the Toxic Soup Index, a measure of the number of pollutants detected in the air at any given time. This new epistemic resource builds on a concept common in environmental justice circles: activists living on the frontlines of pollution often claim that they are living in a ‘toxic soup’; that is, they are continually exposed to a mixture of toxicants. The ‘soup’ concept is in part a rebuke to toxicological studies that assess health risks by studying the effects of individual chemicals in isolation. The potential for synergistic effects between the chemicals makes activists worry that the soup could be much more dangerous than the sum of its ingredients (Tesh 2000).

The Toxic Soup Index was also shaped by the nature of the monitoring data.⁵ When we began to look for patterns in the data, we imagined that we could see in the measurements some trace of what residents experienced. For example, many community members say that smells from refineries are worse at night and on the weekends, so we tried grouping measured levels by day and hour to see if their observations were reflected in the data. We were thwarted, however, by the fact that most of the time, most of the pollutants that the sensors looked for were at levels below what the monitors could detect.⁶ In other words, monitors reported ‘not detected’ far more often than they reported a numeric value for concentration. Statistical analyses of such spotty numerical values would have made little sense. But counting detections—the strategy that the TSI is based on—proved both feasible and meaningful.

Using data from community monitoring stations in Richmond, we found meaningful variation in the average Toxic Soup Index in different neighborhoods and at different times of the year.⁷ This suggested to us that the Toxic Soup Index could be useful to activists, as a metric for assessing refineries’ relative performance, as well as to researchers, as an indicator that could help identify fruitful areas for investigation.

The Toxic Soup Index is an epistemic innovation in that it expands the epistemic resources available for understanding and communicating about the multiple, simultaneous exposures that fenceline communities experience. Toxic soup is not a new concept, but with no associated metrics, it was easy to dismiss as rhetoric or hyperbole. Having an index adds to the concept’s potential to be useful in new contexts and be taken up by groups who may not have experienced the toxic soup themselves.

As an elaboration of epistemic resources in an area that is generally under-resourced— namely, fenceline communities’ experiences of pollution—the Toxic Soup Index is well aligned with societal goals of social and environmental justice. Like the other epistemic innovations discussed here, its development did not include any form of anticipation. The degree to which it exhibits the RI principle of inclusive deliberation is debatable. While we excluded regulators and refinery officials, we did attempt to include a large diversity of EJ-oriented participants, including participants from each of the refinery-adjacent communities in the Bay area. In the end, our self-selected group of community members and activists was overwhelmingly white and mostly college-educated.⁸ Further, when it came down to analyzing data to see what sort of innovation was possible, it became hard to find ways to include participants, because the data were so voluminous that even testing out ideas required programming skills. Deliberation often took the form of feedback on what software developers and data experts had created.

Because the Toxic Soup Index emerged in the eleventh hour of our project, we did not have a chance to pursue the activities that would have made it available as a broadly shared epistemic resource, such as automating the calculation so that fenceline community residents could check on their Toxic Soup Index each week. We also hesitated because the data most suited to this calculation were from the community monitors in Richmond, whose residents were not well represented among our participants. In the communities that were most involved in the project, limited monitoring and high detection limits make Toxic Soup Index calculations less viable.

Were we to advance the Toxic Soup Index, it seems likely that it would, like Levels of Concern, be dismissed as error rather than accepted as an epistemic innovation. The Toxic Soup Index only counts the pollutants detected at a monitoring station in each sampling period (usually one minute for the continuous monitors we worked with). It discards information about which pollutants were detected. A Toxic Soup Index of 4 could include any four of pollutants being monitored—or, in a weekly average, a changing mix of them—regardless of their relative toxicity. The Toxic Soup Index also does not retain the detected concentrations; those four pollutants could be sky-high, or just barely above the monitors’ detection limits. There seems reason to expect that this approach would seem misinformed and scientifically unjustifiable to scientists and regulators trained to understand chemical health effects as dependent on exposure levels (‘the dose makes the poison’) and accustomed to assessing them one chemical at a time. Although we in turn could point out that their own models are replete with comparable simplifications (see e.g. Vogel 2008), it would likely be an uphill battle to have something like the Toxic Soup Index be accepted as a legitimate epistemic resource by regulators and other dominantly situated knowers.

Challenges for responsible epistemic innovation

These three epistemic innovations did not explicitly adopt a responsible innovation framework, although all of them arguably sought to act on a problem of broad social concern: the lack of epistemic resources for representing the on-going exposure of refinery-adjacent communities to toxic air pollutants. Nonetheless, observing them through an RI lens sheds light on the kinds of challenges that epistemic innovation, pursued as a deliberate part of a responsible innovation agenda, would be likely to face.

The error problem

Epistemic innovation will be most aligned with societal goals—and arguably most responsible—when it addresses areas in which epistemic resources are systematically lacking or, put another way, when it confronts hermeneutical injustice. This means working to invent, amplify, and extend concepts that more adequately capture the experiences of structurally marginalized groups. New metrics and techniques for quantification, to the extent that they are effective expressions of marginalized experience, can be especially powerful components of epistemic innovation, in that they may allow new concepts to circulate more easily in spaces of power, including policy processes.

As theorists of epistemic injustice have pointed out, epistemic innovation that takes seriously marginalized experience—that is, responsible epistemic innovation—is vulnerable to being dismissed by dominantly situated knowers (Pohlhaus 2012; Dotson 2014). The case of Levels of Concern demonstrates one mechanism through which this can happen, especially where epistemic innovation involves quantification or exposes the limitations of scientific concepts. Established epistemic resources can be used to claim that new ones are somehow in error. That is, if hegemonic concepts and methods are taken by dominantly situated knowers as the only appropriate foundation for making new knowledge, epistemic innovations built on alternative ways of looking at the world become unrecognizable as resources for knowing. Any knowledge made with these resources can then be said to be simply wrong, because it rests on a failure to understand or appropriately apply the ‘correct’ epistemic resources.

Avoiding the error problem in responsible epistemic innovation requires that dominantly situated knowers acknowledge the problem of hermeneutic injustice and actively commit to epistemic innovation. In practical terms, this would involve listening for the experiences to which one cannot relate, and recognizing others’ attempts to convey them as epistemic contributions. It would also involve questioning the epistemic resources that one takes for granted, even when they are well institutionalized and seemingly neutral ‘scientific’ tools. It would involve admitting the possibility that the available quantitative resources can be made better and more inclusive—or more diverse and heterogeneous. Theorists of epistemic injustice have tended to consider this a problem of virtue: they would argue that scientists and regulators need to cultivate such qualities as humility, curiosity, and open-mindedness in order to recognize and epistemic innovation for what it is (c.f. Medina 2012). Those involved in the promotion of responsible epistemic innovation, however, might find that they need not only to cultivate these epistemic virtues themselves, but also to play a translational role by articulating the limitations of hegemonic epistemic resources and the value of particular epistemic innovations in filling gaps.

The data deliberation problem

The epistemic resources of environmental regulation intertwine concepts, measurements, and interpretive frameworks; certainly this intertwining characterizes other domains as well. Where epistemic resources are heterogenous and interconnected, however, the most successful epistemic innovation must synchronize diverse elements. New concepts that cannot be linked to data may be valuable epistemic innovations, yet they are unlikely to have as much influence in bureaucratic or policy contexts as new ideas accompanied by new metrics to express them. Similarly, new measurement techniques are unlikely in themselves to shift the ways we think about social problems if they are not coordinated with new concepts and interpretive frames.

In practice, the need for synchronization creates challenges for inclusive deliberation. Concepts in isolation can be innovated and iterated discursively. But discursive practices are likely to be inadequate to testing measurement methods that could potentially support a new concept, or examining new data sources to see what under-resourced stories they could tell. The creation of ever-larger data sets intensifies the problem, as we found in the development of the Toxic Soup Index. There, it felt like a necessity to let programmers and other experts crunch numbers without the active participation of fenceline community residents. A generous read of the EPA’s development of the ΔC could attribute their lack of attention to activists’ calls for more extensive monitoring to a related problem. Had they required continuous fenceline monitoring at each of the nation’s refineries, they would have been swamped with data whose significance could have been a subject of on-going contention. Including community members and refinery representatives in interpreting such a large data set would have been challenging and resource intensive, to say the least. By establishing both a narrowly circumscribed monitoring procedure and a technique for data interpretation that resulted in a concrete metric, the ΔC, they avoided the gridlock that more extensive monitoring could have created.

To say that deliberations on the meaning of large data sets are difficult is not to excuse a lack of inclusive deliberation—on our part or the EPA’s. Rather, the practical difficulties point to the need for RI advocates to develop and theorize forms of inclusion that can better support epistemic innovation. Davies et al. (2012) offer one set of possibilities with their work on ‘material deliberation,’ an example of which is an interactive simulation that enables participants to visualize the consequences of policy decisions. Further experimentation is necessary to identify techniques of material deliberation that facilitate collaborative sense-making with large data sets in a way that permits participants to try out new concepts. It seems likely that such techniques would demand greater inputs of time, money, and expertise than a purely discursive deliberation. Responsible epistemic innovation will thus require new kinds of investment from policy makers and other RI proponents.

The inclusion problem

The idea of responsible epistemic innovation embeds a particular political commitment, to the reduction of hermeneutical injustice. The development of epistemic resources that further express the standpoints and experiences of those who are already well recognized in society may constitute epistemic innovation, but it would be hard to argue that it is responsible epistemic innovation in situations where epistemic resources are unevenly distributed. Aligning epistemic innovation with social goals and values demands creating concepts, categories, and metrics more adequate to the experiences of marginally situated people, and helping those resources to become part of our shared pool.

This justice-oriented notion of responsible innovation is potentially in tension with certain ideals of inclusion. If ‘inclusive deliberation’ is taken to mean a process that includes all stakeholders and treats dominantly and marginally situated knowers equally, there is great danger that emergent epistemic innovations will be shut down as error or dismissed, to paraphrase Pohlhaus (2012, 722), as making something out of nothing. Yet it is challenging to justify, within an RI framework, purposefully excluding certain groups of dominantly situated knowers, as we did in the process of developing the Toxic Soup Index. Nor are such exclusions necessarily desirable. The expertise of those groups could potentially be harnessed to help craft powerful expressions of marginalized experience that can travel in regulatory and policy circles—as long as they were willing to use their expertise in service of others whose standpoints they do not share.

Redefining inclusion to incorporate ‘epistemic inclusion,’ as Valkenburg et al. (2020) do, would certainly help clear the way to responsible epistemic innovation, as would the purposeful cultivation of epistemic and republican virtue in deliberative forums. However, orienting an inclusive process to actively recognizing and addressing the epistemic resource needs of its least powerful participants may require adopting some form of deliberation expressly designed to address structural inequalities, such as Mansbridge et al.’s (2010) ‘deliberative negotiation.’

Conclusion

Epistemic innovation is an on-going process. New concepts come into common parlance regularly, and some of them expand our resources for understanding the experiences of marginalized groups. Hearing the term ‘intergenerational trauma’ for the first time, for example, made me aware of a phenomenon that had previously been invisible from my standpoint. Epistemic innovations like these can change how a society thinks about the world, about inequality, about what is worth producing knowledge about, and about what warrants policy action.

Responsible innovation has largely neglected the processes through which new epistemic resources are made. I argue that this has been to the detriment of both RI and epistemic innovation. Promoting responsible epistemic innovation is one aspect of aligning innovation with societal needs. Creating epistemic resources that better represent the experiences of marginalized groups helps to counteract epistemic injustice, foster recognition, and advance social justice. Responsible epistemic innovation is also necessary for robust anticipatory governance. New technologies create new experiences, new harms, and new inequities. Developing language, categories, and metrics to capture these is a necessary aspect of foresight.

How epistemic innovation happens, and how it can be made more responsible, is only beginning to be understood. The case studies here show that epistemic innovation can just as easily express ideas about environmental protection that limit the accountability of polluters, as they can express the concerns and experiences of those living on the front lines of pollution. Further, efforts to align epistemic innovation with needs for epistemic resources felt by marginalized communities are susceptible to being dismissed as error when they do not conform to dominant ways of knowing. Nor is this the only challenge that efforts at responsible epistemic innovation face. Weaving data and experience into meaningful epistemic resources requires new methods for deliberation, and the ideal of inclusion in deliberative processes needs to be rethought to avoid quashing potential innovations from marginalized knowers.

RI practitioners have an important role to play in fostering responsible epistemic innovation: by learning to recognize it, by actively pursuing it in foresighting and scenarioplanning processes, by becoming articulate spokespeople for new epistemic resources, and by advocating for research and innovation policies that promote epistemic innovation as one important target of RI. Further, by developing practices to meet the challenges of data deliberation and inclusion that attend responsible epistemic innovation, RI practitioners can add to the repertoire of tools available more generally to responsible innovation and anticipatory governance.

Acknowledgments

My thanks to all of the fenceline community residents and environmental justice activists who have spoken to me over the years about their experiences with air pollution, air monitoring, and regulatory characterizations of air quality. I am indebted to them for their insight. I am also deeply grateful to all of the participants in the Meaning from Monitoring project, which gave rise to the Toxic Soup Index. Dawn Nafus and Amos Akinola deserve special thanks for their work. An ACLS-Burkhardt Fellowship and a year at the Center for Advanced Study in the Behavioral Sciences greatly aided in the preparation of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by United States National Science Foundation: [Grant Number 1352143].

Notes on contributors

Gwen Ottinger

Gwen Ottinger is Associate Professor at Drexel University, in the Department of Politics and the Center for Science, Technology, and Society. She directs the Fair Tech Collective, a research group that uses social science theory and methods to promote social justice in science and technology. She is author of Refining Expertise: How Responsible Engineers Subvert Environmental Justice Challenges (2015 Rachel Carson Prize, Society for Social Studies of Science). Ottinger has been an ACLS-Burkhardt Fellow, a Fellow at the Center for Advanced Study in the Behavioral Sciences at Stanford University, and the 2022 Fulbright Research Chair in Science and Society at the University of Ottawa.

Notes

1 Activists’ efforts have added to epistemic resources surrounding this problem. However, because their innovations have been largely rejected by environmental regulators, gaps in collective epistemic resources remain. The use of present tense in this paragraph reflects that lack of uptake in the regulatory context.

2 As a volunteer for Communities for a Better Environment in 2001 and 2002, I elaborated Subra’s original Levels of Concern by incorporating levels from additional government agencies and creating a website to automate the process of looking up Levels of Concern and comparing them to bucket sample results.

3 At the time of writing, this refinery is owned by Phillips 66.

4 One of the five Bay area refineries closed in 2020.

5 Dawn Nafus helpfully characterizes this as the data’s ‘topology.’

6 This does not necessarily suggest that they were below levels of health concern. For some chemicals, like benzene, the detection limits of fenceline monitoring are significantly above levels which are considered to pose a long-term threat to human health.

7 Amos Akinola, former student at Drexel University, performed the data cleaning and analysis to come to these findings.

8 The demographics of the five refinery-adjacent communities in the Bay area are quite different from one another, and not necessarily typical of how one imagines fenceline communities in the United States. Three (Benicia, Crockett, and Martinez) are nearly three-quarters white, and more than 40% of residents in each place hold at least a bachelor’s degree. Richmond and Rodeo are significantly more diverse: less than half of residents are white, and Black, Asian, and Latinx residents each make up a significant portion of the population. Educational attainment is lower, with 28.2% of Richmond residents and 23.9% of Rodeo residents holding at least a bachelor’s degree. (Source: data.census.gov, accessed 7/29/2021).