Between September 2nd and 5th, the University of Washington at Seattle, WA, USA, the 50th Annual Meeting of the Society for Social Studies of Science (4S) took place, with the theme “Reverberations.” Laura represented the Redistributed Computing Systems Group (with an assist by Chris) with a paper “Why are Passive Acoustic Monitoring systems not truly passive?,” for the panel “Discovering Sound in Technical Society: Method and Practice,” organized by Qiushi Xu.
Why are Passive Acoustic Monitoring systems not truly passive?
Machine learning is swallowing the world. A less-attended aspect of this ingestion are the listening infrastructures that endlessly capture and analyze environmental sounds, known as Passive Acoustic Monitoring (PAM) systems. PAMs continue expanding into nearly every domain—from domestic spaces equipped with audio-driven security devices, to marginalized urban neighborhoods surveilled by gunshot detection technologies, and tropical forests monitored by bioacoustics sensors. In this talk, we took a critical approach to argue that they are not passive tools for neutrally listening and documenting passive environments; rather, they actively shape, sustain, and expose political dynamics.
The World Wildlife Fund describes PAM systems as those “surveying and monitoring wildlife and environment using sound recorders (acoustic sensors) (…) for hours, days, or weeks” (Browning et al., 2017). Recordings are later processed with the aid of Artificial Intelligence (AI) to extract useful data. The acoustic sensors used in the collection process are small, affordable, and less-invasive, qualities that underwrite their classification as “passive”: they emit no obvious signals, only receive.
While PAMs are frequently deployed in natural landscapes in the Global South, such as tropical forests (Ritts et al., 2024), these environments are far from being the only environments “automatically heard”. In 2018, Walmart patented Listening to the Frontend, a system that analyzes checkout sounds—such as the beeps of register machines—to monitor and control employees’ performance (Jones et al., 2017). In 2023, Amazon released Alexa Emergency Assist, which claims to detect “alert” sounds in the home, such as breaking glass or smoke alarms, and trigger notifications (Amazon, 2024). Meanwhile, ShotSpotter—by SoundThinking—markets itself as capable of detecting any gunshot. Yet, tellingly, its data has never been admitted as legal evidence when the gunfire came from a police officer’s weapon (Abu Hamdan & Parker, 2020). These cases illustrate how the boundary between scientific monitoring and surveillance is becoming increasingly blurred.
PAMs are scientific instruments.
Drawing on Shapin and Schaffer (1985), we propose that PAMs function as scientific—and therefore political—instruments. They discipline what and how we hear, shaping what gets counted as evidence, proof, and truth. In this case, the “experimentalists” are not just individual scientists, but rather corporations, governments, NGOs, universities, and other institutions that use these sound-capture and analysis systems to produce scientific “facts” and legitimize political decisions. In this process, environments are treated not as living, contested spaces, but as passive entities “amenable to scientific dissection.” (Ochoa Gautier, 2015; p. 187).
Ritts, Simlai, and Gabrys (2024) examine the political effects of these instruments—what they call Digital Acoustic Monitoring—through the concept of environmentality to argue that such systems give rise to new spatial formations of power. In this way, sensed sounds and algorithmic logics reshape relations between bodies and spaces, enabling decisions about security, land distribution, and population management. By foregrounding these dynamics, they complicate narratives that portray these systems as straightforward success stories of conservation or environmental crisis management.
PAMs are classification systems.
PAMs are classification systems that, as Bowker and Star (1999) argue of such kind of infrastructures, create temporal, spatial, and conceptual segmentations of the world, determining which phenomena are rendered perceptible and which are not. Moreover, the signals PAMs capture are rarely interpreted in terms of their acoustic qualities alone. Instead, they are read as proxies for other categories—such as legal versus illegal occupation, punishable versus legitimate actions, or alarming versus normal sounds—extending beyond the ecological metrics often invoked as justification under the banner of biodiversity loss and climate change (Ritts et al., 2024). This, reconnecting with Shapin and Schaffer, brings us to the question of what an experiment and the way it is recalled through literary and social technologies actually provide evidence of—the phenomenon itself or its cause? Likewise, we need to ask: what are these instruments listening for, and whose purposes are ultimately served by their modes of sensing?
For example, Parris-Piper and colleagues (2023) examine an acoustic monitoring system installed on northern Palawan Island in the Philippines, which relies on global technologies—smartphones powered by solar panels—to study wildlife and detect ‘threatening’ sounds, such as chainsaws. Yet, because it was installed without regard for local conditions, the system deploys sound in ways that are, as they put it, anti-political. It does so through two interconnected mechanisms: first, by imposing external notions of legality—distinguishing categories of “legal sounds” and “legal actions”—onto Indigenous and local peoples, undermining their rights to access forest resources and ancestral lands; and second, by obscuring the larger economic and political agendas such as their interest in giving land for industrial mining or large oil palm plantations.
What is most striking is that these decisions are made despite the systems themselves facing numerous technical problems. They are treated as evidentiary, even with their limited reach and weak performance metrics. For example, while cameras can capture a wide visual field, the earshot of a microphone is much more restricted. And although recorders are said to be placed in “strategic” locations, “where they can hear better,” these placements do not follow “pure” technical or scientific reason. Rather, they reproduce political biases, as in the case of ShotSpotter, whose locations in the US disproportionately target communities of color (Nast, 2024).
Also, the labor involved in deploying PAMs is often overlooked. Installation demands careful and complex logistics to ensure devices withstand the harsh conditions of environments like tropical forests. The operation requires frequent visits to retrieve memory cards and replace batteries. Furthermore, hours of human listening are needed to sift through recordings in search of acoustic signs. Likewise, false positives are a contested metric. A researcher from the Elephant Listening Project explained to us how easily the system can misinterpret the buzz of a wasp on a microphone for the sound of a gunshot. Likewise, while SoundThinking—ShotSpotter—claims its false positive rate is below 0.5%, independent research has reported places where more than 80% of its alerts are false alarms (Stanley, 2021).
Conclusion
Describing acoustic monitoring systems as passive obscures their political dimensions behind walls of technical reasoning. It neglects their infrastructural configurations and their real impacts. In this sense, we argue that embarking on machine listening projects requires at least four commitments:
- To recognize that there are no universal ways of hearing. Any intervention—even something as seemingly neutral as sensing an acoustic signal—should be conducted with communities, and ideally, originating from them.
- To question the why of these systems. Lawrence Abu Hamdan (2020) points out that ShotSpotter is justified on the grounds that people don’t call the police when they hear a gunshot. But this is not because populations are deaf. Instead of trying to replace the human ear, the real question should be: why don’t they call?
- To acknowledge that PAMs, as classification systems, always imply exclusion. Defining a set of categories also defines what remains unheard. Asking what is excluded is therefore essential. And,
- To recognize that the placement of microphones and the generation of acoustic information inevitably redistribute power. So, we must ask: which entities or actors are empowered by these systems? If the answer points to those who are already powerful, then we need to return to the earlier questions—why, from whom, and with whom—and, if necessary, begin again.
References
Abu Hamdan, L., & Parker, J. (2020). Forensic Listening as Machine Listening: Lawrence Abu Hamdan in conversation with James Parker [Interview]. https://disclaimer.org.au/contents/forensic-listening-as-machine-listening
Amazon. (2024, September 18). Everything you need to know about Alexa Emergency Assist, an Amazon service that can keep you and your family safe. https://www.aboutamazon.com/news/devices/alexa-emergency-assist
Bowker, G. C., & Star, S. L. (1999). Sorting things out: classification and its consequences. MIT Press, Cambridge, MA, USA.
Browning, E., Gibb, R., Glover-Kapfer, P., & Jones, K. E. (2017). Passive acoustic monitoring in ecology and conservation. https://repository.oceanbestpractices.org/handle/11329/1370
Jones, N. A., Vasgaard, A. J., Taylor, R. J., & Jones, M. A. (2017). Listening to the Frontend. https://patentscope.wipo.int/search/en/WO2017184971
Nast, C. (2024). We Tracked the Secret Police Microphones Hidden Everywhere. WIRED. https://www.wired.com/video/watch/we-tracked-the-secret-police-microphones-hidden-everywhere/
Parris-Piper, N., Dressler, W. H., Satizábal, P., & Fletcher, R. (2023). Automating violence? The anti-politics of ‘smart technology’ in biodiversity conservation. Biological Conservation, 278, 109859. https://doi.org/10.1016/j.biocon.2022.109859
Ritts, M., Simlai, T., & Gabrys, J. (2024). The environmentality of digital acoustic monitoring: Emerging formations of spatial power in forests. Political Geography, 110, 103074. https://doi.org/10.1016/j.polgeo.2024.103074
Shapin, S., & Schaffer, S. (2011). Leviathan and the Air-Pump: Hobbes, Boyle, and the Experimental Life (Revised edition). Princeton University Press.
Stanley, J. (2021, August 24). Four Problems with the ShotSpotter Gunshot Detection System | ACLU. American Civil Liberties Union. https://www.aclu.org/news/privacy-technology/four-problems-with-the-shotspotter-gunshot-detection-system