Against The Grain: Listening after Palantir

Machine listening is transforming environmental sound into data points that support surveillance infrastructure, changing what it means to hear, argues Stephen Roddy in The Wire 512

On 28 February 2026, Shajareh Tayyebeh elementary school in Minab, Iran was hit by a missile killing over 150 people, most of them children. Initial investigations have suggested that the US was responsible for the attack. While US data mining giant Palantir has not been conclusively linked to the atrocity, the wider US-led bombing campaign has made extensive use of the Palantir-supplied Maven Smart System to coordinate similar attacks.

Maven can combine large volumes of data from radar, satellite, communications and even audio sources before applying machine learning techniques to identify possible military targets. We now live in a world where our homes, schools and workplaces can become clusters of data points.

According to James Parker, academic and Earshot board member (see The Wire 499), machine listening integrates audio and machine learning systems to monitor, analyse, govern and modulate human and nonhuman systems through computational renderings of the soundscape. As these systems proliferate globally, they represent an imperial mode of decision making and environmental governance that is in conflict with the basic tenets of democratic accountability. Who owns and governs these systems? Who or what do they hear, and what remains unheard? Who acts upon the claims these systems make about the world?

Machine listening effectively expands the reach of these systems into the realm of the audible, proliferating throughout contemporary soundscapes: classifying gunshots, indicators of distress or aggression in human voices, and sounds associated with intrusion across borders and within protected spaces. Far from being neutral descriptors of a soundscape, these classifications are political, and operational categories and their detection can trigger responses from security or police forces and mark individuals for further surveillance.

Flock Safety, the surveillance tech company drawing ire for its operation of video and audio detection networks across the US, recently pulled plans to listen for indicators of “human distress” across their audio network due to strong opposition from civil liberties groups like the Electronic Frontier Foundation. Across Europe, comparable systems are already deployed in public spaces, transport lines and security networks in London, Groningen and Plzeň.

In a political moment defined by the convergence of private technological infrastructure with authoritarian state power, ubiquitous machine listening networks increasingly provide companies, states and militaries with acoustic classifications upon which they can act. What becomes of listening in such times?

Almost 50 years ago, composer and writer R Murray Schafer helped to establish the field of acoustic ecology around a central opposition between the lo-fi soundscapes of urban, industrial and mechanised life, in which dense noise acts to mask and corrupt clear and meaningful acoustic signals, and the hi-fi soundscapes he associated with the natural world. One unintended political consequence has been to allow the soundmaking activities of communities, often marginalised by race and class living and working in urban environments, to disappear into noise. This is not helped by the disturbing resonance between his oft-cited claim that “the world suffers from an overpopulation of sounds” and the eugenicist overpopulation tropes circulating throughout Western environmentalist thought of the 1970s.

The troubling concept of nature, as inherited from Schafer and his contemporaries, is usefully dismantled and reconstituted from a field recording perspective in Mark Peter Wright's Listening After Nature, in which he illustrates how nature is constituted by material interrelationships between culture, technology, place and practice. Wright’s guiding question, “What am I not hearing?”, surfaces the human lives, labour and infrastructural relations that are often concealed in the production of field recordings.

The emerging ubiquity of machine listening further complicates this problem. Decisions are now made and actions taken on the basis of features extracted from audio signals. What cannot be heard, captured and classified is excluded from the systems account of the world and becomes operationally non-existent. Conversely, audio that does not fit the systems’ predefined categories for the world can become misclassified; and as the strike on Minab illustrates, misclassification within these systems can have devastating effects.

Is this pattern of low rumbling a harmless passing vehicle or an intrusion? Is that high energy transient a firework or a gunshot? What about this repeating pattern of falling frequencies – is that a barking dog or a human in distress? Classifications like these are bundled with other data like time, location and contextual information before triggering an alert, providing a recommendation or, in some cases, initiating an automated response. The sounds involved need never be listened to by human ears before being converted into operational data to support systems of classification and control.

These questions are explored by Machine Listening, a platform for collaborative research and artistic experimentation, in Environments 12. The project is a speculative extension of a popular series of 11 environmental records released by Irv Teibel between 1969–79. It explores a future Earth in which a developing cybernetic ecology powered by planetary-scale computation has laced microphones and loudspeakers throughout the biosphere, automating the reproduction, synthesis and management of the soundscape. Dynamics of capture, identification, classification and data-driven intervention are stretched almost to breaking point across the 15 short but richly textured sonic works. Human voices vault and bounce in cacophonous chorus alongside their generative clones amid lush field recordings. Organic life and human technology are almost indistinguishable from one another as a planetary machine listening infrastructure feeds the soundscape back to itself in a continuous cycle of replication.

Where Schafer feared technology might mask nature, and Wright showed how it can constitute it, Environments 12 imagines nature to be continuous with data and deeply entangled with the infrastructures that detect and manage it. But the world imagined by the project is already being built.

Listening after Palantir cannot be about a return to some imagined innocence or a retreat into Schafer’s idealised nature. Critical listening must instead focus on machine listening itself – exposing the classifications it imposes, the infrastructures in which it is embedded, and the power relations that shape its production and deployment. It must not only ask what the machine cannot hear, but what it has misheard, and who will suffer as a result.

This essay appears in The Wire 512. Wire subscribers can also read it in our online magazine library.