Beyond the cybernetic loop: “ smart” pain technology in a recursive society

Abstract

This article examines emerging “smart” pain technologies in the United States. New wearable and implantable devices aim to partly remove human decision-making from pain management, delegating it to more or less closed feedback loops that capture signals directly from the body, process them algorithmically, and administer electrical stimulation to modulate nerve response. This endeavor to create increasingly “closed loop” pain treatment reflects both the cybernetic dream of building self-regulating machines and biomedicine's discontent with the subjectivity of pain. The article conceptualizes and empirically explores the building and maintaining of loops in pain management, examining this sociotechnical process and the many elements it implicates, both within and beyond pain management. We focus on two specific devices: one implantable and the other wearable. The analysis shows how physicians and company officials imagine—and actively seek to realize—changes in the role of patients, the nature of medical expertise, the understanding of chronic pain, and the business models of companies. We argue that such activities are a type of technoscientific activity that portends potentially new forms of healthcare and social life. Finally, we propose that an analysis of “looping techniques” can contribute to recently advanced theories of a recursive society by demonstrating how building and maintaining loops shift different actors in and out, interface them at different scales, and establish new regimes of control.

Introduction

In contemporary societies, loops of data extraction and algorithmic application increasingly redistribute agency and act on the world (Hayles 2018). Such systems read people's behavior and bodies, rendering many details of their lives available for extraction, analysis, and intervention (Zuboff 2019). As these processes of data capture and algorithm-guided intervention cycle through time, they interact and feed off each other. The acceleration and intensification of such processes have led some analysts to speak of an emerging “recursive society” (Beer 2022, 2), one in which loops of algorithm-mediated interventions are increasingly “coded into the very infrastructures in which social life is lived” (Ibid., 4). This not only challenges notions of agency and intersubjectivity but also materially infiltrates human bodies. Nowhere is this infiltration more apparent than in closed-loop medical devices, where implants and wearable technologies make physiological measurements and automatically direct interventions into “the most basic conditions of human life” (Böhme 2012, 3). Examples include cardiac pacemakers, cochlear implants, and neurostimulators.

This article explores and conceptualizes looping techniques as a specific type of techno-scientific activity using the case of chronic pain management in the United States. In this national context, we examine the new generation of so-called “smart” ¹ wearable and implantable devices that capture signals directly from the patient in real-time, process them algorithmically, and administer electrical stimulation to modulate nerve response. This continuously guided stimulation forms a more or less “closed loop” system for managing pain. ²

Pain medicine has employed neuromodulation since the 1960s, using surgically implanted spinal cord stimulators (SCS) and transcutaneous electrical nerve stimulators (TENS) placed on unbroken skin at the location of pain. ³ Until recently, however, both SCS and TENS were only used in “open loop” devices, strictly regulated by people, patients or care providers. The new smart devices incorporate SCS and TENS into feedback-governed control systems that capture data in real-time to algorithmically direct interventions. Promoters of “closed loop” pain technology herald it as a way to render neuromodulation more convenient, personalized, and accurate, making it a key component of the future of pain management and even healthcare more broadly.

Below, we sketch out our approach to the analysis of looping techniques. This approach moves beyond the cybernetic conceptualization of feedback control to focus on conceptual and social changes associated with deploying smart technologies. Next, we introduce the sociopolitical milieu of smart pain technologies in the United States and present our study design and two cases, the Evoke implant and the Quell wearable. We then show how looping activities are implicated in reconfiguring the position of the subject-in-pain; the nature of pain medicine expertise; the datafication of pain; and the business models of neuromodulation companies. Finally, we argue that one direction for studying increasingly “recursive societies” is to examine looping techniques and their role in constructing new modes of interfacing at different scales and in constituting specific regimes of knowledge control.

Looping beyond the cybernetic loop

Underlying smart pain technologies is an engineering paradigm based on the cybernetic concept of feedback-governed control (Halpern 2014). Cyberneticians like Norbert Wiener believed that the ability to adapt behavior in real time to a changing environment is not unique to living entities but can be replicated in machines (Galison 1994; Kline 2015). A prominent early example is Wiener's anti-aircraft predictor, a device for predicting the location of an enemy aircraft to help gunners target German bombers during the Second World War. Wiener's system continuously measured an aircraft's position and predicted its location a short time in the future, assisting the gunner by adjusting for the time required for bullets to reach the aircraft. Hence, by a cybernetic loop we mean a particular technical setup consisting of a feedback-controlled circuit of information and goal-directed action that is designed to pursue a certain objective. Crucially, this paradigm assumes that the mechanisms which shape the behavior of the target need not be understood. The target can be treated as a black box so long as its behavior can be monitored and fed into the circuit of information and goal-directed action.

Like Wiener's system, smart pain technology frames pain management as a problem of feedback-controlled targeting. Instead of focusing on the causes or experience of pain itself, these technologies capture neurophysiological data from the body and use it to adjust pulses of electrical energy to target pain. A cybernetic loop aims to control pain by establishing continuous feedback between the patient's neurophysiology and the pulse generator's output.

Smart pain technologies that operate without ongoing human action are called closed loop systems. Those that do require human action (e.g., to manually regulate stimulation intensity) are considered open loop systems. In practice, the distinction between “open” and “closed” is not a simple binary, since the nature, frequency, and extent of human involvement varies. Some systems never require human action within the cybernetic loop; others require regular human action to correct course. Between those extremes are various intermediate, partially closed systems. By this definition, Wiener's anti-aircraft predictor is open loop, and as we will see below, the Evoke is a closed loop system while the Quell is partially closed.

For an analysis of looping techniques, purely focusing on the technical setup of cybernetic loops is too limited, because it neglects the social and political conditions that enable their operation and fails to address the broader epistemic and social changes that they entail. This article draws on previous work (Lipp and Dickel 2022) to account for activities of looping to broaden the perspective beyond a techno-centric view. First, our analytical approach does not take the operation of cybernetic loops for granted. Rather, feedback loops need to be installed, maintained, and repaired by a variety of actors, such as the patient, health professionals, and company officials. Put otherwise, to close the loop actors must engage in practices of interfacing (Lipp 2023); for example, by calibrating the device.

Second, this interfacing is not limited to the body–device interface but takes place on multiple scales. Operating the cybernetic loop may require interactions between patients, providers, and company officials; it may change the relationship between health professionals and device companies; and it may lead to new machine–machine relations, as devices communicate with and feed information into company data infrastructures (Lipp et al. 2024). Interfacing thus contributes to a process—and politics—of scaling (Pfotenhauer et al. 2022), not least because the delegation to algorithms facilitates action at larger scales.

Third, looping is not a neutral process of simply relaying information along passive data pipelines. Rather these flows are embedded in emerging “knowledge-control regimes” (Hilgartner 2017) that regulate transfer, use, and ownership. Looping is thus a political process that allocates control of automated systems, including some actors and excluding others. The redistribution of control can be contentious. For example, some proponents of closed-loop technologies see displacing humans from the cybernetic loop as highly desirable, whereas others seek to keep the “human in the loop” to maintain human agency and accountability (Monarch 2021, Shneiderman 2022).

Our analytic focus on looping activities raises a variety of empirical research questions. As closed-loop technologies are introduced, what entities are shifted into and out of the system and to serve what—and whose—ends? What activities beyond the loop are necessary to operate and maintain a cybernetic loop? What epistemic changes do specific modes of looping entail, enable, or preclude? And how, as looping unfolds in various contexts, does it reconfigure roles, rights, responsibilities, and regimes of control?

Smart pain technologies in the United States

Smart pain technologies have emerged as a promising modality to tackle what has been called the “crisis” of chronic pain in the United States (Institute of Medicine 2011). The problem of chronic pain, estimated to affect one in five Americans (Zelaya et al. 2020), receives extensive attention in public arenas. Chronic pain is considered a leading cause of disability worldwide (Cohen, Vase and Hooten 2021), but pharmaceutical development has failed to check its rising burden. On the contrary, the expansion of opioid prescriptions in the 1990s and 2000s is seen as a major cause of the US opioid epidemic, indexed by a dramatic rise in addiction and overdose deaths. In response, the United States entered a state of “opioid pharmacovigilance” (Knight et al. 2017), with government authorities aiming to lower opioid prescription rates (CDC 2016, 2022) and facilitate access to non-addictive and non-pharmacological alternatives (Institute of Medicine 2011; U.S. Department of Health and Human Services 2023; US Congress 2016). The COVID-19 pandemic also affected pain management in the United States, with limited in-person healthcare (Shanthanna et al. 2022), an unprecedented rise of telemedicine (Wahezi et al. 2020), and increased levels of social isolation, anxiety, and depression among chronic pain sufferers. Investment in digital health innovation skyrocketed. In 2021, venture capital funding for digital health reached an estimated $57.2 billion, almost double the previous year (Landi 2022). Startups in mental health and chronic illness, including some specifically focused on pain, particularly benefited from these trends.

In this context, startups and established device manufacturers have promoted neural stimulators as an innovative, patient-centered solution to an important health problem. Neuromodulation devices like SCS and TENS are promoted as enabling patients to self-manage their pain in a “drug-free” and “non-addictive” way. Especially, SCS is said to have entered a “renaissance,” with the market for implants growing almost sevenfold over the past two decades (Ross 2019).

Medical devices for modulating neural response to relieve pain are based on “gate-control theory,” which conceives of pain as governed by neural pathways that can be opened or closed like a gate (Melzack and Wall 1965). Gate-control theory draws heavily from cybernetics (Paterson 2019). It treats pain as a communication process, which can be modulated by way of stimulating signals in the spinal cord. The body is framed as electric, with its physiological processes governed by action potentials and electromagnetic waves in the nervous system (Helmreich 2013). While gate-control theory has been extended and criticized, it is still considered influential (Mendell 2014).

The “mechanisms of action” underlying neuromodulation are not fully understood (Ong Sio et al. 2023)—a fact often cited to explain why physicians have long treated SCS as a last resort and TENS as merely an auxiliary tool. Many of our interlocutors described neuromodulation's past as “crude” or even “black magic.” They expect closing the cybernetic loop to change the future, as continuous loops of data collection, algorithmic calculation, and neural interventions render a formerly “crude” technology “smart.” This transformation requires the quantification of pain, either through subjective metrics or biological or behavioral measures as proxies for pain. Attempts to describe pain are unavoidably intersubjective and relational (Middleton 2022), not least because the very act of talking about or measuring pain can reshape how it is experienced, rendering it epistemically fragile (Baszanger 1992). Modern pain medicine thus struggles to find a pathway through a terrain rift by the conceptual fissures that split mind and body, culture and nature, subject and object (Graham 2015). Smart neuromodulation technology is imagined as a promising way to navigate this subjective and ethically fraught domain.

Not surprisingly, the growing attention to neuromodulation has generated some controversy, especially about implantable devices. Spinal cord stimulators featured in the “Implant Files,” a publication of the International Consortium of Investigative Journalists (2018). While controversy has mostly centered on such implants as breast implants and contraceptive coils, spinal cord stimulators have also been criticized for insufficient testing and industry influence on regulatory policy (Woodman 2020, see also Traeger and Bero 2024). Questions have also been raised about infection risks and diminishing effectiveness over time. One study found that 7.6–30% of SCS patients had the devices explanted (Gill et al. 2022). Controversy about SCS was palpable at the 2023 North American Neuromodulation Society (NANS) meeting, where speakers presented the field as under attack. One frequently referenced example was a Norwegian study (Hara et al. 2022) that found no significant therapeutic effect of SCS when compared to a placebo. Conference speakers advanced rebuttals, pinpointing methodological flaws. As observers, we also encountered some suspicion, with some interlocutors asking whether we were part of an anti-SCS group before agreeing to talk. On the other hand, some physicians we interviewed expressed concern about overprescription of SCS or the extent of industry control.

Research design and methods

This article examines closed-loop pain management as part of a qualitative study on the promise and use of neuro-technologies of pain in the United States. It focuses on the former, the expectations (Borup et al. 2006), imaginaries (Jasanoff and Kim 2015), and scripts (Akrich 1992) of smart pain technology. Using Foucauldian discourse analysis (Khan and MacEachen, 2021), we examine how different actors, mainly company officials, sales representatives, and physicians involved in the field of neuromodulation, frame those technologies in relation to their respective clinical or business practice. This approach enables us to trace how these actors’ positions of power are altered via the introduction of closed-loop systems (their roles, rights, and responsibilities), and how these change ideas about truth in pain management. In short, this perspective uncovers the co-evolution of power-knowledge in processes of looping.

In total, the authors conducted 17 interviews in 2022 and 2023. While we asked all interviewees about their general expectations about closed-loop technology and the future of neuromodulation, we also asked those involved with the Evoke and Quell devices about them in more detail. Beyond interviews, field research data were also collected at relevant field-configuring events (Lampel and Meyer 2008), such as the annual PAINWeek conference and the annual meeting of the NANS. We also draw extensively on scientific articles as well as company websites, product manuals, and US Security and Exchange Commission filings.

The Evoke implant and the Quell wearable

The Evoke implant (an invasive SCS system) and the Quell wearable (a non-invasive TENS device) represent different cases of how more or less closed cybernetic loops are configured. Both the Evoke and the Quell combine electrical stimulation with data monitoring and analytics. While they both seek to improve neuromodulation by providing pain care automatically and continually, they differ in significant ways (see Table 1). Comparing these devices showcases the variety of looping activities emerging in smart pain technology today.

Table 1.

The Evoke implant and the Quell wearable compared.

	Evoke implant	Quell 2.0 warable
Company	Saluda medical (Australia)	Neurometrix (US)
Modality	Spinal cord stimulation (SCS)	Transcutaneous nerve stimulation (TENS)
Location	Implanted near the spine	Worn on the exterior of the calf, fastened with a band
Data collection	Neural response (ECAP measurements), usage, outcome data (in Australia)	Sleep, activity (e.g., steps and gait), weather, usage, pain levels
User interfaces	External remote for patient and tablet for programmer	Touch interface on the device, smartphone, and smartwatch app
User interaction	Control of stimulation intensity, battery charging, feedback during programming sessions	Control of stimulation intensity, battery charging, electrode management, device placement, device configuration, active tracking (e.g., pain diary)

The Evoke implant

Since the introduction of SCS, the typical system has consisted of a pulse generator, leads with electrodes, a form of remote control, and a battery charger. Doctors implant the pulse generator and leads such that the electrodes can stimulate the dorsal column of the spinal cord. More recent SCS devices often feature a wireless recharger and a digital application for remotely controlling simulation intensity. Before the entire SCS device is implanted, the patient begins with a trial phase, in which the leads are inserted but the pulse generator remains outside the body. These trials often last for 3–7 days, with success usually defined as the patient reporting at least a 50% reduction in pain. Configuring the system involves “programming” it to aim for optimal pain reduction. In the past, programming (and periodic reprogramming) took place during hospital visits, but remote programming is getting more common, especially since COVID-19. Physicians and nurses—and increasingly company sales representatives—select the appropriate electrode and adjust the amplitude, width, and frequency of pulses (Sheldon et al. 2021). Specific combinations of these four parameters are called waveforms. The therapeutic effects of waveforms differ; for example, with some stimulating muscles while others stimulate nerves. Since the mid-2010s companies have focused research and development (R&D) on exploring new waveforms, including “high-frequency” and “burst” stimulation. If the 2010s were the decade of new waveforms, many of our interlocutors expect the 2020s to be the decade of AI; that is, to continuously adjust stimulation rather than rely on a fixed level of intensity.

Until recently, all SCS systems were open loop, with a pain physician, nurse, or company representative setting the stimulation to a fixed output and patients making continual adjustments to compensate for fluctuations in intensity. In contrast, Saluda Medical's Evoke device is a closed-loop system. Rather than being set to a fixed output, it automatically adjusts stimulation in real time based on the spine's response. To accomplish this, the device measures the electrical effect of the stimulation—known as the “Evoked Compound Action Potential” (ECAP)—via one of its leads while the other administers stimulation. The Evoke captures this data about 4 million times per day (Saluda Medical 2023a). ECAP measurements allow the device to compensate for variations in stimulation caused when patients move and shift postures. Evoke's closed loop thus aims to constantly keep the effective activation of the spine within the “therapeutic window” (Mekhail et al. 2020).

The Quell wearable

TENS units are a common pain treatment method, often used in conjunction with physical therapy to make it easier for patients to exercise. Units for self-treatment are available over the counter for less than $50. There are also clinical-grade devices usually administered by health professionals. In both cases, TENS therapy requires the patient to rest and thus does not allow activity during use. In contrast, patients can wear Neurometrix's Quell device throughout a day of normal activity. In principle, the device can be worn all the time, delivering therapy even during sleep. However, some patients experience skin irritation, so the company recommends removing the Quell after 5 hours to air out the skin.

In keeping with gate control theory, TENS devices are usually placed between the location of pain and the central nervous system. By contrast, the Quell is designed to be worn at a fixed site on the upper calf to control pain in the ankle, foot, and lower leg. However, the Quell has been reported to potentially have “widespread effects beyond the site of stimulation” (Gozani and Kong 2018, p. 8). Indeed, in 2022, the company received FDA authorization to treat fibromyalgia, a chronic condition that causes pain throughout the body.

Unlike the Evoke, the Quell does not directly measure the patient's neurophysiological response. However, it captures several kinds of data, both automatically and by engaging the patient. Using a smartphone, patients can report their pain level on a scale from 0 to 10 and describe how pain interferes with their mood, sleep, and activity. The device can also passively track sleep quality, activity, and usage of the device. To continuously personalize the intervention, some of these data are fed directly back into the device. The company promises that “like cruise control on your car, this technology automatically manages stimulation intensity, while allowing the user to take manual control at any time” (Gozani 2022, p. 12). Furthermore, the Quell has added a feature intended to adjust the stimulation depending on “sleep quality.” Using an accelerometer, the device counts “leg movements” and “body roll events” (Ferree et al. 2015, p. 17), both of which are presumed to indicate compromised sleep quality. However, closing the loop between sleep data and stimulation has proven difficult, as the CEO concedes: “we just couldn’t do a good enough job of that, and it wasn’t being used.” Because the Quell engages the patient much more than the Evoke in customizing the cybernetic loop, it offers a case of a partially closed loop system and a useful contrast with the Evoke.

Below, we examine promises and tensions in the discourse of companies and physicians involved with neuromodulation devices, focusing on the Evoke and the Quell. We organize the discussion under four themes that emerged from the interviews, observations, and documentary analysis: (1) how smart pain technology (dis)engages the patient from pain management; (2) how smart pain technology is implicated in reshaping neuromodulation expertise; (3) how these technologies are entangled with epistemic shifts in pain medicine; and (4) how neuromodulation companies are using feedback loops between bodies, devices, and data to spin off assets.

(Dis)engaging the patient

Smart pain technology transforms the way patients interact with neuromodulation technology changing how they engage in their pain care, promising them control over their pain while simultaneously reducing their need to actively manage it. Unlike traditional TENS and SCS devices, which required ongoing attention to keep stimulation on course, smart pain technology learns from data passively collected from patients, adjusting stimulation in real time. Chronic pain varies over time, and open-loop devices often under- or over-stimulate. Closed-loop devices promise continuous regulation, always keeping stimulation within the “therapeutic window.” Smart pain technology strives to delegate this ongoing fine-tuning to the device, relieving the subject-in-pain from constantly making decisions. Because patients are shifted out of their care, the technology addresses them not as neoliberal subjects responsible for controlling their pain (Lupton 2013, Schüll 2016) but as passive targets of pain management within a regime of digital convenience (Brubaker 2023, p. 46). Moreover, the patient is treated as a “body electric” (Helmreich 2013) that responds to stimulation automatically and unintentionally. The goal is less to engage the patient-as-subject in the process of pain management than to interface the patient-as-body on a neuro-technical level.

Smart pain devices thus frame the patient as someone who not only wants to be free of pain but also to be free of thinking about its management. The Quell is presented as a kind of “autopilot” that delivers not only pain relief but also convenience. The Evoke implant is also presented this way:

…with our system, once patients are set up, they don’t need to even touch the device. They don’t interact with it. It's set it and forget it. So, once the device is tracking the spinal cord, then patients go home and it's like in a year, ‘oh, I forgot my remote,’ and most patients are using the device or touching their device like once a week that that's just to charge their device. They’re not playing with up and down and all this stuff. The device does that all on its own. It's faster and smarter than the patient can change things. (Saluda Medical official)

The notion of “set it and forget it” is ubiquitous in discussions of smart pain technology. Neurometrix markets the Quell as “wear it and forget it.” But while the Quell requires active tracking and management to protect the skin, the Evoke device aims to completely disappear from the patient's awareness. Instead of adjusting stimulation “up and down,” the patient lets the device adjust the stimulation at rapid speeds. Here, the promise of convenience is combined with a narrative of machine superiority common in discourses on automation (Heßler 2015). The management of pain “ideally … without any human element,” as one pain physician put it, avoids the limitations of the patient, who is seen as unable to maintain real-time optimization or unwilling to bother with it.

Nevertheless, the disengagement of the patient—and the corresponding dream of a fully automated fix—is incompletely realized (Katzenbach et al. 2024). The Quell requires the patient's active engagement to configure the system and select the level of automation. The Neurometrix CEO sees such participation as necessary to give patients the capacity to choose their level of engagement:

… we wanted to have a very feature-rich product that patients that … go in and … really interact – optimize you know people call it – hack, hack their experience, if you will. But we also wanted a system that… Other people have no interest in doing that. They just want their chronic pain, their fibromyalgia to get better. So I think … we’re talking about a complex population here (…) I get e-mail from people still (…) that are like, you know, they tell me about the nuances and they’re like engineering their therapy, and they like that. (…) These chronic diseases take away control. Quell can give them some control back. (Neurometrix CEO)

Unlike the Evoke, the Quell actively involves users in configuring it; for example, it enables patients to choose among several modes of automatically regulating the stimulation during sleep. The device lets users set it to administer higher or lower stimulation during the night, and it also allows them to disable such “personalization” features. In addition, the device lets the patient set it to automatically adjust the stimulation based on local weather data. Using the patient's location and online weather data, it can provide advance warnings about impending weather conditions that the patient previously identified as tending to increase pain. In these ways, the Quell addresses its users as “optimizers” of their therapy, who “hack their experience,” selecting from a menu of options to personally manage the complexities of living with chronic pain.

While company discourse presents such “choices” as governed by consumer preference, the flexibility of smart pain technologies also makes demands on the user. Quell users must calibrate the device to their “sensation threshold”—the intensity at which the user feels the stimulation. Quell users also need to regularly charge the device and replace electrodes periodically. Because the device should be removed every 5 hours, users who depend on continuous stimulation also need to engage in extensive planning and monitoring.

In short, smart pain technologies aim to disengage the patient from the burdens of pain management (“set it and forget it”), but they also impose new obligations on the user. However, the target of these new activities is not simply the self or the body but it is the neuro-technical interface (Lipp and Dickel 2022) linking the device and the body electric. As “engineers” of their therapy, patients can assume some control, but they exercise this control by interfacing with the device, not by acting directly on their medical condition or subjective experience. In these systems, “control” is achieved by maintaining interference-free feedback loops linking the body to the device in a continuous, real-time way. These feedback loops draw in data and materials from the body and beyond: from ECAPs and sensation thresholds to weather data; from electrodes and batteries to the user's vulnerable skin. Establishing a “closed” loop between smart pain technology and the patient-as-body-electric thus implicates the patient-as-subject in new responsibilities directed at maintaining and repairing these feedback loops. As patients are shifted out of managing their pain, they are simultaneously shifted in as engineers of their devices.

Reprogramming expertise

The introduction and maintenance of smart pain technologies also entails the redistribution of expertise and authority in pain medicine, particularly regarding the programming of devices. “Programming” in this context means adjusting the stimulation parameters—electrode selection, amplitude, pulse width, and frequency—with the goal of optimizing the therapeutic effect for the individual patient (Sheldon et al. 2021). Traditionally, health professionals have done most programming while interacting with patients during clinical visits, as Oudshoorn (2015) found with pacemakers and defibrillators. However, closed-loop pain technology increasingly shifts out the physician, concentrating programming expertise in companies and thus redistributing control over pain therapy.

Confronted with the subjectivity and heterogeneity of pain, programmers must collaborate with the patient to decipher pain relief (Baszanger 1992), and patient reports remain an important benchmark for evaluating success. However, advocates of closed-loop technology often frame dependence on patient reports as a lack of transparency and objectivity, as “putting in electrical energy blind” (Saluda official). In contrast, they view closed-loop systems such as the Evoke as providing a basis for “objective” programming, based on data, namely ECAPs. The programmer can directly see how adjusting the device affects the patient's neural response by observing changes in ECAPs. One physician explains the value of this with the help of an analogy:

If you have some sort of catapult, right? That you’ve built, and you’re trying to hit a target. The more times that you pull the lever on that catapult, and then make adjustments in order to see if you hit the target, the better a chance you’re gonna have of hitting the target, right? And so, the more times that you go back and adjust your target, or adjust your catapult, the tighter that loop is gonna be, ⁴ because you’re ­gonna have more revolutions during that time period. (Physician)

Here, both the quantity of data and the speed of adjustment in closed loop systems are valorized. The Evoke device enables the company to send the physician a report that indicates “down to hours … how the [spinal] cord was activated” between clinical visits. This information can help the patient and provider to pinpoint times when the spine did not receive enough electricity. Moreover, these data can help physicians, together with the patient, to identify specific activities in the patient's life that correspond to those insufficient dosages.

This does not mean that our interlocutors do not value patients’ feedback. Faced with the possibility that despite adequate dosing, the patient may not experience pain relief, physicians were quick to add: “Yeah, yeah. You always believe the patient”. The “human element” in programming is still considered crucial, but mainly in situations where the intervention seems inadequate. The “ideal” situation is imagined as an interference-free, data-driven, and fully integrated loop that is “faster and smarter than the patient,” and needs only occasional correction by a programmer. The promise is a future in which “objective” data, not subjective patient reports, regulate treatment. The patient is shifted out and only brought back in to compensate for occasional imperfections of an otherwise closed system.

Controversial reallocations of expertise and authority accompany this effort to make programming increasingly data-driven. Companies have created automated systems to capture physiological data, pain scores, and information about device usage and build proprietary datasets. Drawing on these data, company sales representatives increasingly participate in programming, sometimes taking it over, especially in the United States.

With commercial datasets increasingly driving neuromodulation practice, competition to develop new systems and waveforms has produced a fragmented landscape of device-specific skills, guidelines, and educational resources (Sheldon et al. 2021). Some physicians we interviewed see relying on companies as the only practical way to ensure high quality programming under these circumstances. For example, one physician argues that doctors or nurses “don’t do as well,” because ongoing training is needed to use devices that “evolve so quickly.” They see data-driven, self-regulating neuromodulation as superior to programming by fallible clinicians. Yet others see relying on companies to manage programming as threatening the professional role of doctors as the “stewards of therapy.”

I personally am a part of programming my patients. And so, I am a part of that loop, and I’m a part of that decision-making process. And that's because I have taken the time to really understand the science of neuromodulation and understand how it works. And I think that true neuromodulators should understand that piece. I don’t believe that most physicians in that situation truly understand the actual science of the therapies. I think that they’re purely technicians putting in the device. (Physician)

For him, the allocation of control over programming is about nothing less than “the care of your patient” since (re)programming is the main way to tailor therapy to the specific patient. Beyond making the science harder, the proliferation of devices, waveforms, and modalities increases the importance of maintaining his identity as a “true neuromodulator” as opposed to a “technician.” Surrendering expertise to companies means that physicians “can’t offer suggestions, or … double-check or … troubleshoot if things aren’t going as planned”.

In contrast, companies see capturing and analyzing large datasets as an opportunity to fundamentally change the future of therapy. For this reason, Saluda Medical stores ECAPs (and in Australia even outcome data from patients). The company celebrated the milestone of reaching a trillion datapoints in 2023 as “a revolution in our field” (Saluda Medical, 2023a). These data are already being used for “titrating … to success” (Saluda official). Rather than the traditional practice of monitoring patients only when something seems wrong, the company can constantly capture physiological responses and outcomes. The goal is to continuously adapt the algorithms to individuals, even as their responses change due to aging or medication. Indeed, some physicians we interviewed envision a future of devices autonomously adjusting stimulation parameters. “So, we won’t be dependent on the [sales] rep or the doctor to program. It’ll program itself.”

Closed-loop systems not only reconfigure the relationship between patient and provider, replacing patient reports with physiological data, but they also challenge the professional jurisdiction of physicians, the culturally authorized “stewards” of pain management. Put differently, these developments might translate programming (and much of “the care of the patient”) into a new knowledge-control regime: one in which corporations drive therapy using proprietary models designed and operated by their data science departments.

Changing the data gaze of pain medicine

Closed-loop technologies also have the potential to enable change in the data gaze of pain medicine (Beer 2019). Proponents of these devices seek new ways of rendering pain legible and susceptible to therapeutic intervention. They hope that accumulating behavioral and physiological data will improve understanding of neuromodulation as well as chronic pain. Saluda advertises the Evoke technology as making visible “how the body controls pain” (Saluda Medical 2023a). For physicians, closed-loop systems are “fascinating” or “really impressive,” not least because they render visible what was previously invisible. Neuromodulators are no longer “just putting in electrical energy blind,” because such data as ECAPs promise to demystify pain and suggest new designs of waveforms and systems.

However, as some of our interlocutors point out, these data do not reveal pain itself. Closed-loop systems cannot appreciate pain as an experience but rather detect a proxy which “stands in for the absence of quantifiable pain” (Middleton 2022: 130). The data that the Quell and Evoke produce are mere proxies, assumed to be correlated with pain and its reduction. Closed-loop systems make a “leap of faith” using various measurable markers, such as blood pressure, behavior, or ECAP data, to represent the patient's lived experience of pain.

ECAP does not say you have pain relief, but if the ECAP says you had a good signal, we’re providing you the therapy in a good range that we know for other people works, or maybe we expect it to work, but it's not an end all be all. (Physician)

To bridge this gap, patients are typically asked to describe their pain, for example, through standardized questionnaires (Crawford 2009). Health professionals interpret these descriptions, filtering them through their own assumptions (Johannessen 2019). In this way, closed-loop systems aim to shift out these inter-subjective negotiations, grounding neuromodulation in continuous, objective sensing of the body and its environment.

Different devices treat different types of data as relevant. While the Evoke radically narrows the information fed into the cybernetic loop to ECAPs, the Quell attempts to incorporate a plethora of measures. These include data about leg movements, bodily motion and activity, as well as gait cadence and variability. For example, increased gait variability might be interpreted as indicating a higher level of pain, thus suggesting the need to raise stimulation intensity. Lacking direct markers for pain, the Quell focuses not on pain but on pain-related behavior. Reaching beyond the body entirely, the Quell also uses weather data and the time of day to adjust stimulation. Instead of the clinical gaze of the physician, the Quell employs a data gaze that continuously measures machine-readable signs on the surface (Krasmann 2020).

The transformation of the data gaze is also changing the objectives of closed-loop neuromodulation. All our interlocutors, whether physicians or company officials, expressed skepticism about treating pain relief as the only significant outcome measure. The “fixation on pain,” one physician argued, is a relic of the “war on pain.” This echoes a renewed understanding of pain management in the United States, where under a regime of opioid pharmacovigilance, the status of pain as the “fifth vital sign” and its routine commensuration through numerical scales is increasingly put into question (Levy et al. 2018). At the same time, physicians and companies are looking beyond pain relief as the sole therapeutic goal.

So, you look at kind of neuromodulation research over the past three years or so, we’ve really started to look at functional improvements. But there's this idea of holistic responders, or holistic pain scores. Or holistic scores, and not just looking at pain, but also looking at other aspects of quality of life, depression, sleep, activity, etc. (Physician)

The data gaze of these systems is simultaneously narrow and broad; narrow, because they require specific, quantifiable, physiological, behavioral, and environmental information as inputs; broad, because in the absence of distinct markers for pain they seek to achieve outcomes that reach beyond it, addressing quality of life and function as a “whole” (Gardner 2017). Moreover, this gaze enables companies to recursively adjust the design and goals of pain management with neuromodulation systems. One might even say that closed-loop technology is only indirectly concerned with pain. Its focus is establishing interference-free feedback loops that continuously build on themselves, but which are taken as having further ripple effects in the patient's neurophysiology and experience. Because the relation between neural intervention and pain can never be fully validated, more becomes better; both regarding data collection (“1 Trillion ECAPs”) and regarding intervention itself. To repeat the quote from a physician above, “… the more times that you go back and adjust your target … the tighter that loop is gonna be, because you’re gonna have more revolutions during that time period.” Closed-loop technologies are neither about the singularity of pain nor its causes and effects. They are about the quantity of data, the speed of looping, and the recursive reconfigurations that looping enables (Beer 2022: 4).

Reconfiguring neuromodulation companies

The vision of neuromodulation reaching beyond pain is not only epistemic and clinical but also sociopolitical, with specialists imagining significant change in their profession and in company business models. Key to this promise is the notion that closed-loop technology represents a turning point in understanding “how electricity should be used as medicine.” In publications, interviews, and at the PainWeek and NANS conferences, we encountered pervasive optimism that neuromodulation will soon evolve from a “crude” and poorly understood technology to a new modality of electric medicine. Electrical stimulation might be used to re-grow nerves, make paralyzed patients walk, and regulate many bodily functions. One physician speculates that “we’re gonna rename our field pretty soon to a non-pain-based name.” For companies, this vision promises new applications and an expanding market. It also implies a re-definition of business models and corporate identities. As a Saluda official put it:

… we believe we’re really more of a neuro response and data company than we are a neuromodulation company in the future (…) The therapy is one thing, but the data that we’re collecting is gonna lead us to new innovations. We’ve already developed second generation algorithms based on seeing trends and how the device is being used and we’re just looking at automation of everything. You know, [from] programming to predicting when patients are gonna need to change the therapy over time. (Saluda Medical official)

Companies see wearables and implantables as creating a seamless connection between traditional medical devices that act on the body and smart sensing capabilities that extract data from the body. Neurometrix describes its “Quell Health Cloud” as among the world's biggest repositories of pain data, with over “100,000 patients.” The CEO calls this repository an “incredibly important” asset. Similarly, at the 2023 NANS meeting, big medical device manufacturers like Medtronic and Boston Scientific presented themselves as “database companies.” Companies imagine using “real-world” data to develop new devices, therapies, and platforms that will both benefit patients and serve business interests. Some physicians express skepticism about this vision, arguing that some products are more “fake marketing” than real innovation. Nevertheless, this “algorithmic drama” (Ziewitz 2016: 5) changes how these companies understand themselves and their assets.

Devices such as the Quell or the Evoke integrate data capture into their basic functioning, blurring the boundaries between therapeutic intervention and monitoring, as well as between treating individuals and studying populations. On the one hand, capturing data is a natural extension of what the device is already doing, since “treatments should be data-driven” (Kuch et al. 2020). On the other hand, building data repositories creates genuinely new opportunities for pain research.

Quell's weather feature provides a telling example. It assumes a correlation between pain and certain weather conditions, such as low barometric pressure. The Quell allows patients to specify kinds of weather they believe trigger pain. The device, which monitors weather in the user's location, then automatically issues warnings and asks whether the device should increase the stimulation. While the company emphasizes the value of looping the patient in, it also ultimately aims to shift the patient out of detecting triggers: “We’d like to get to the point where we monitor it for you, we monitor barometric pressure temperature, other conditions, and tell you, ‘Here are the triggers.’” Neurometrix thus sees itself not only as a producer of therapeutic devices but also as a producer of new knowledge about the conditions it treats. In the case of weather, some evidence suggests that it correlates with increased pain, but its clinical relevance remains a matter of debate (Fagerlund et al. 2019). Neurometrix is using the data it captures to study weather as a predictor for pain and exasperation, and, in turn, to evaluate the efficacy of higher stimulation levels (Gozani and Kong 2017).

Coupling intervention with continuous monitoring allows companies like Neurometrix not only to demonstrate the efficacy of their devices (e.g., in clinical trials), but also to develop “insights into various chronic pain conditions”, the CEO says. What is crucial to this strategy is not only the data, but also the construction of continuous flows between interventions and monitoring. Data drives treatments, but treatments also drive the automated capture of new kinds of data. Saluda's immense collection of ECAP data is another case in point.

Companies incorporate the data these systems capture into knowledge-control regimes that constitute them as proprietary company assets. In principle, these data could be provided as an open-access resource like the public genome sequence databases, which are made available to the entire scientific community to enable and speed research (Hilgartner 2017, pp. 157–184). The creation of proprietary assets through looping techniques raises ethical issues and requires justification. Smart pain technologies are thus a space of “ethics-in-the-making,” by which we mean a domain in which ethical choices are actively being built into emerging sociotechnical systems.

Under US regimes of contract law and informed consent, the ability to constitute these data as corporate assets depends on patients permitting data collected from their bodies to be used in this way. Both Saluda and Neurometrix offer patients the option of storing their de-identified data on company servers (Saluda 2023b, p. 35, Neurometrix 2018, p. 13). ⁵ Companies tend to frame the choice of allowing such use in altruistic terms.

People are remarkable in that I’m surprised at how much people that suffer chronic pain are willing to participate to help the broader community. It's kinda heartening actually in today's day and age to see a willingness. But that's enough for a lot of people to provide their data. If it can lead to improved treatment and better understanding, that's – they enjoy and they want to participate. They feel a kinship and it provides – that provides a strong motivation. (Neurometrix CEO)

Extending this register, companies cast themselves as benevolent actors, arguing that the data they collect never “just helps us.” They “don’t make money off people's data”; that is, they don’t sell patient data to third parties. In the end, “we’re really trying to make the product better and trying to help the chronic pain community.” Few observers would doubt that a device that can predict pain early and treat it pre-emptively would benefit patients. At the same time, such statements disclaim the underlying assetization of patient data. Stressing that Neurometrix does not sell patients’ data may be true, but it may also downplay how “people's data” is crucial to a certain business model that extracts data for R&D purposes. Indeed, framing research use of data in this way turns making money into a responsibility to help pain sufferers, as the CEO continues:

I mean ultimately, we’re a business. I mean we make money. You can only help people if you’re making money and you exist. Sometimes it's not charity unfortunately, it's – you want to try to achieve both great pace and outcomes, but you have to make money to grow and satisfy your shareholders. (Neurometrix CEO)

In the end, such statements deploy the switch between the register of a money-making business and data kinship to ethically justify the assetization of the data that looping techniques produce. Invoking a “chronic pain community”—that provides data as charity and that in turn is helped by new devices—underwrites a knowledge-control regime in which loops that capture data become an asset-generating machine.

Conclusion

Above, we have shown how introducing closed-loop technologies into pain management opens a variety of possibilities for reconfiguring social relations far beyond the cybernetic loop. As smart pain devices link bodies, information, medical intervention, and R&D into recursive flows of feedback-governed control, they strive to disengage patients from management of their pain, both for convenience and because devices are trusted to be smarter and more reliable than the patient. At the same time, the patient is shifted back in to configure, maintain, and repair the flow of information and intervention when needed (especially in the case of the Quell). Be it through “engineering their therapy” or repairing the loop, the patient's responsibility is directed toward the device–body interface.

Furthermore, smart pain technology also reshapes physicians’ authority and expertise. The physician-subject may be shifted out of the loop and replaced by company specialists or even devices themselves. For some, this challenges the profession's self-understanding as “stewards of therapy,” as clinical authority clashes with corporate control. However, physicians and company officials alike see opportunities to heighten medicine's understanding of chronic pain by both narrowing and broadening the data gaze of pain medicine. The vast amount of data that continuous data flows capture promises to render pain legible—if not directly, then through improved proxies for it. Along with broadening the focus from pain to overall quality of life, this attention to proxies illustrates that smart pain technology is less concerned with pain as an experience than with maintaining ever faster, recursive loops.

Finally, the wider data gaze of pain medicine is also seen in changing self-understandings of the neuromodulation industry and profession. Closed-loop neuromodulation is seen here as promising the possibility of addressing many medical problems, from regrowing nerves to treating paralysis. For electric medicine enthusiasts, pain becomes just one among many neurological circuits to target for intervention, and manufacturers of neuromodulation devices reimagine themselves as data companies. However, rather than simply shifting from device to data, these companies become harvesters and integrators of continuous loops between devices and data infrastructures. The Quell and Evoke are assets insofar as they enable the extraction of data and permit therapeutic experimentation—both in technical and ethical terms.

Our fine-grained analysis suggests that empirical research on the “recursive society” (Beer 2022) should examine looping techniques and the ambivalent role of different social groups as they are shifted in or out of loops of data extraction, algorithmic management, and medical intervention. These dynamics appear in the controversy among physicians, some of whom, while not fundamentally challenging closed-loop technology, strive to retain control over treatment.

In addition, and in partial tension with Beer's (2022) reflections, we contend that research should avoid focusing specifically algorithmic recursivity or on the cybernetic loops linking bodies to devices. Our analysis has shown the value of examining the dynamics of interfacing at different scales (Lipp et al., 2024). For loops to be closed, manifold actors must get involved in configuring, repairing, and maintaining interfaces, not just between body and device, but also between physicians, patients, and companies (see also Oudshoorn 2015, Jansky 2024). Creating and sustaining stable loops between data, algorithms, and intervention is precarious and requires work (Schwennesen 2019). Information technology discourse often makes this work invisible, but investigating the interfacing work involved in making automated technology operate in the first place is well worthwhile (Lipp 2023).

Finally, looping processes create openings for introducing new knowledge-control regimes and reconfiguring existing ones, as exemplified in our case by the capture and assetization of data and by the prospect of adjustments in the roles of neuromodulation professionals and companies. Here, the question of what epistemic changes closed-loop technology entails has yielded some first results. We see a shift away from pain and its causes toward the quantity and speed of recursive looping as a value in and of itself (Beer 2022, p. 4). Ultimately, our analysis goes beyond a mere focus on data and its infrastructures. As digital and medical technology converge in devices that continuously sense and intervene in the body, we also see looping techniques building new interfaces between data and devices, between technology and the body, between professionals and companies, and between clinical practice and commercial R&D. All these developments warrant further investigation.