Stochastic Media

Jacob Weinberg read bio

In 2023, I traveled to Miami as part of a research trip to the Everglades. We spent little time in Miami itself, traveling most days out to the wetlands while staying in Miami Beach. Early in the morning on the day we were set to leave, I sat on South Beach just before dawn with a radio designed to detect very low-frequency signals produced by atmospheric noise and lightning. The time of listening was crucial, as signals are best heard at dawn and late at night. To avoid interference from the hum of the power grid, I stayed close to the water. The radio detected loud clicks and pops emitted by a series of thunderstorms just off the coast that were interspersed between hundreds of fainter chirps from lightning strikes further away. As the sun began to rise, additional noise was introduced as solar radiation energized the ionosphere. These sources were far removed from my location, with the ionosphere beginning 40 or 50 miles above and the signals from lightning traveling from dozens of miles offshore to thousands of miles away, oscillating between sky and earth.¹ The beach became a kind of non-location, referring less to the sand, sun, water, and clouds in its immediate vicinity and more to what it offered beyond its horizons, as a position to detect everything exceeding my sensory boundaries.

As I listened to those louder cracks and pops generated by the approaching storm, I envisioned the water in front of me also approaching, not because of the tide but due to the global rise in sea levels in the coming decades. I was reminded of a conversation I had that week in the Everglades, where it was mentioned that insurance companies had been stopping their coverage of homes on Florida’s shores and even inland, as the market could not withstand ongoing shocks due to damages from increasing storm frequency and intensity. Additionally, insurance companies themselves were facing difficulty paying increased premiums to reinsurance companies (insurance for the insurers). Amidst this fallout, the market had been increasingly leaning on Florida’s own state-backed company to provide coverage. On the beach that morning, it seemed that the housing market would not recover and that there would be an eventual collapse.

But I learned later that the market at that time had been moving past its most dire conditions, with local and state governments already working to address their structural problems. And it became clear that, while this predicament was entangled with the inevitabilities of weather-related disaster, it was much more about disaster’s secondary effects: impacts to risk assessment, financial challenges, and opportunistic loopholes, rather than primary damages to livelihood. Since then, Florida has reconfigured its rules and regulations of property insurance law, adjusting in some ways for the certainties of increased risk but more so addressing the market’s financial and legal technicalities. Prior to these changes, contractors would have a homeowner’s insurance policy signed over to them using an Assignment of Benefits agreement, a since-banned practice that allowed contractors to receive payments from insurers directly. They would then sue the insurer for more than the actual estimated repair costs, often citing unrelated storms or past weather events as a cause. The most common claims involved roof repair or replacement, even when a roof’s condition was the result of aging without damage from any particular weather event. These claims were padded with exorbitant litigation fees, all situated within a legal structure where lawyers were paid regardless of outcome and where contractors took on little financial risk. The practice was so commonplace that by 2022, Florida had been responsible for just over 75% of all property insurance lawsuits in the entire country, with 70% of settlement payouts going to lawyers, and less than 10% going to policyholders.² Insurance companies would then sharply raise premiums as their hemorrhaging losses threatened their own financial solvency.

The impact of climate change and weather-related catastrophe in Florida becomes more oblique and almost ironic in light of this crisis. Though legal loopholes were the primary driver, the crisis was amplified and accelerated as losses were processed through actuarial models, which did not discern related and unrelated or real and unreal weather events. Rather, the models had incorporated a series of fictionalized weather events as part of insurers’ loss histories. The prospect of risk and the mere potential for damage proved to be valuable on its own. This was all occurring against a backdrop of increasing storm frequency and intensity, which was making a real impact on the market. However, this backdrop served more as a contextual support on which spurious claims were made plausible, with their basis being a conflation of the real and the imagined, rather than pure fiction.

Florida’s insurance crisis became a failure of deterrence of its own making, but not through faults of ecological preservation, at least not yet. Rather, it foreshadowed a future of uninsurability in bureaucratic form through a technical feedback loop. This was a failure of the model and its regulators to account for how abstractions and stories of disaster can be leveraged and applied beyond correspondence with real circumstances. It presents itself as an imaginary run amok. It is only fitting that this dilemma took place in Florida, given its history of concentrating the imaginary and doing so at the expense of its ecology. Sitting at its inland center is Disney World, the monumental counterpart to Disneyland, which as Jean Baudrillard observes, is a classic example of the imaginary deployed as a deterrence tactic. Baudrillard here takes deterrence from its Cold War context to describe a cultural operation at large. He identifies Disneyland, and I would argue, Disney World, as a way to deter attention from the hyperreality of America by concentrating the imaginary within a boundary against which everything beyond is contrasted.³ Disney World was presented as Disneyland built at scale; it functioned as its own self-governing municipality from 1967 until 2023 and was initially conceived as the utopian city Experimental Prototype Community of Tomorrow (EPCOT) by Walt Disney before his death. Though its construction was ultimately managed by his brother, Roy, Disney World was Walt's answer to the outside world encroaching upon Disneyland's borders and was the promise of a total environment of control. It was built atop wetlands that were drained and displaced, though Disney World’s municipality adopted the name of the wetland area, Reedy Creek, as its own.⁴ In this way, Florida itself might be thought of as a contested site of deterrence, where ecological buffer has been substituted with a buffer for the imaginary, for replicating images of fantasy and paradise. This logic stretches outward to the minutiae of property assessment near Florida’s coasts, where its insurance market concentrates imagined catastrophes in lieu of the real ones, not only to mitigate impact to its assets but also to deter belief in future threats to Florida’s habitability and its images of perpetual sunshine and leisure.

For Baudrillard, deterrence is central to the functioning of hyperreality, and it illustrates a temporal condition that the hyperreal brings to bear: a swapping of the notion of real-time into an ever-produced and fragmented, perpetual-present.⁵ Despite this lending itself to feelings of timelessness, its mechanisms are deeply timeful, operating across modes of futurity (forward) and nostalgia (backward), displacing and mitigating the possibility of a historical or material accumulation of meaning. Here, I am most concerned with its future mode, which operates through an anticipatory logic and informational maximalism, of “...substituting the signs of the real for the real, that is to say of an operation of deterring every real process via its operational double, a programmatic, metastable, perfectly descriptive machine.”⁶ An accountant’s ledger provides a sequence of debits and credits that constitutes a record of what is or was. But it is downstream of the actuary’s model, which is less concerned with what actually is but rather what might be or will be. It itself contains a concentration of the imaginary, of a maximal number of possible outcomes, situations, and sequences. These are not held deterministically, but probabilistically. This is especially the case for Florida’s property insurance market, which was a catalyst in the development and spread of catastrophic modeling (or CAT modeling) using stochastic analysis, following devastating property loss in the wake of Hurricane Andrew in 1992.⁷ The model does not eliminate contingency so much as simulate it, attempting to provide a complete description of what could be the case. In this way, the model takes a preemptive stance, strategically positioned to deter catastrophe from arriving as a surprise, not by direct prevention, but by ensuring its arrival is already known and accounted for.⁸

The actuarial model functions as a form of media, working to extend perception forward temporally. Its stochastic operations are not unique but ubiquitous today and lend themselves to thinking through other kinds of media where stochastic processes are central. These kinds of media can be called stochastic because they operationalize randomness and noise as a means to consider and describe the world across a totality of possible states. They present situations less as a set of known or unknown material conditions and more as a field of potential meanings and signs to be interpreted, decoded, and shaped. Most notably, consumer prediction markets like Kalshi and Polymarket come to mind as a form of stochastic media, where users abstract significant and insignificant world events into gambling opportunities. Stochastic media are also constituted within computational forms of text and image, such as many large language models, which hone probabilistic weights in latent space by iterating through feedback cycles, shaping noise into meaning. These media forms are defined less by the content they produce and more by their platform, interface, and structural operation, as exemplified in the case of the algorithmic content feed, which works to test and anticipate user engagement.

Across stochastic media, the common value or impulse is that of being early. It is no surprise that this value is championed by venture capital and across the finance sector, where early adoption or buy-in of promising securities yields the most return. This is evidenced in the strategies and tactics of finance: the collection of privileged data and analytics, large multi-monitor terminals, closed chat threads, private tip-offs, and speculative research. Crucially, being early is a self-reflectively understood value of those operating within the stochastic mode, shown recently in an essay published by Andreessen Horowitz, which announces prediction both as the current zeitgeist and the successor to postmodernism.⁹ Being early can be placed in contrast to the avant-garde impulse of being ahead, of severing the present from the past, producing an opening, discontinuity, rupture, and a sense of progression. The mode of being ahead is embedded within the narratives and tensions of modernism and postmodernism, concerned with urgency, coherence, meaning, and priority of narrative. Instead, being early concerns itself with latency, position, capture, and scalability. The impulse of stochastic media to be early shifts the mechanics of futurity from generating the unpredictable in real-time to administering and controlling what is yet to come.

Although stochastic media present a shift in temporal attitude, they are nonetheless shaped by a long and storied history of probability in the West. Uncertainty, initially managed through juridical frameworks of expectation, shared risk, and contract law, was gradually absorbed into mathematical treatment during the period of classical probability in the 17th– early 19th centuries.¹⁰By the mid-late 19th century, probability entered rigorous mathematics, which laid the groundwork for developments that would rapidly accelerate the mathematical treatment of stochastic processes in the 20th century, including the "random-walk" in stock prices and Brownian motion in the physical sciences and statistics. Amidst these changes, the notion of randomness began to shift from a merely epistemic condition, a limitation of human knowledge within a determined universe, to one where random events could meaningfully affect the world and inform decision-making.

It was soon understood that the randomness or unpredictability of certain physical processes could be interpreted as random information, or in the case of communication theory, noise standing in contrast to a signal. Randomness and noisy sources became useful in and of themselves, as exemplified in the case of the Monte Carlo method developed during the Manhattan Project. It was demonstrated that a sampling of a large number of random values could be used to approximate the results of complex problems. The source of randomness here matters, as those which appear random but are not may introduce unwanted correlation in a data set. To address this, researchers turned to physical processes as a source. For example, the radio static monitored at Miami Beach is one of many physical processes that can be used to create highly random numbers of varying degrees, which are difficult to predict due to the chaotic circumstances of their formulation. More often used (and more random) examples are radioactive decay, thermal noise, and shot noise in electronics.

The logic of rationalized prediction found in the Monte Carlo method continued to find an institutional home in the postwar United States, where deterrence and anticipation were becoming organizing principles for statecraft and public policy. Formed in Santa Monica just after the end of World War II, the RAND Corporation became a preeminent example of these developments, conducting interdisciplinary research that was often based in quantitative methods. Throughout the Cold War, many of its core preoccupations centered on problems of prediction, including nuclear deterrence, intelligence, simulation, and game theory. Noise played a central role in this research, both conceptually and practically. Conceptually, it worked to shift the United States towards embracing uncertainty as a fundamental aspect of military strategy, as illustrated through the writing of Roberta and Albert Wohlstetter, who both conducted research for RAND. The Wohlstetters drew on the vocabulary of signal and noise introduced by Claude Shannon’s “A Mathematical Theory of Communication,” which was first published in the Bell System Technical Journal in the same year that RAND was established. During her time at RAND, Roberta wrote on intelligence challenges, including a comparative study of intelligence during Pearl Harbor and the Cuban Missile Crisis, where she emphasized the difficulties of identifying opponents’ signals amidst noise, writing that “...true signals were always embedded in the noise or irrelevance of false ones.”¹¹ Albert, in his writings on nuclear strategy, stressed that “no bit of noise is unambiguously noise; it is always possible to hypothesize that some apparently random series of events contains a piece of information, deliberately or actually concealed.”¹²

The effects of these conceptual practices resonate most clearly when set in relation to other cultural frameworks, such as that of signs and signification. In Think Tank Aesthetics: Midcentury Modernism, the Cold War, and the Neoliberal Present, Pamela Lee traces the seemingly lateral and sometimes surprising connections in the milieu of RAND during the Cold War, one of which was the relationship between the Wohlstetters and Meyer Schapiro. Lee casts Schapiro’s writing on non-mimetic elements in visual art in light of the relationship between signal and noise posed by the Wohlstetters at RAND. Schapiro stresses in “On Some Problems in the Semiotics of Visual Art: Field and Vehicle in Image-Signs” that the smoothed and bounded ground of the picture has been taken for granted as a mere space for marking and figuration and should rather be considered a semiotic element, constitutive rather than incidental to the meanings it supports. He describes the ground or surface of early image-making in his historical progression as noisy, such as that of cave walls which could compete visually with images. His analysis suggests that the ground, eventually becoming smoothed and bounded, suppressed its own articulation as a precondition for illusory depth.¹³ A parallel can be made between this relationship and that of signal and noise, where the substrate of signal transmission, like the prepared ground, is constitutive rather than incidental to signal and is also worthy of consideration in the interpretation of meanings. Lee’s discussion weaves in glimpses of the Wohlstetters’ relationship with Schapiro, one of which is the last correspondence in Schapiro’s archives from Albert to Schapiro. In that message, Albert mentions that Roberta, when accepting the Bancroft Prize for her 1962 book on the intelligence failures of Pearl Harbor, Pearl Harbor: Warning and Decision, had identified Schapiro as a chief inspiration for her work, years before Schapiro’s essay was published in Semiotica. Lee also highlights their institutional overlap, such as Schapiro’s residency at the Center for Advanced Study in Behavioral Sciences in Palo Alto in 1962-63, which overlapped with RAND in personnel and research scope. Lee’s depiction of this relationship, while painted as sharing intellectual ground, also suggests a possible forking of viewpoint between the two parties, especially as RAND’s influence became increasingly visible during the Vietnam War.¹⁴

The point here ultimately is to demonstrate that this methodological framework of taking a noisy substrate as constitutive of signification is critical when considering wider social and cultural ramifications. When absorbed into the work of the intelligence analyst, Lee presents these frameworks as potentially formulating an agenda to “read and thus, control, an expanding empire of signs.”¹⁵ To extend Lee’s discussion slightly, we can take the smoothed ground to operate as a kind of noise suppression. The means or substrate by which the signal is communicated holds a reservoir of latent signification, but this may be interpreted merely as irrelevance or noise. However, in being constitutive of the signal, the noise of the substrate can rather be interpreted as that which is not yet relevant, a site of constant potential. This opens up the scope of what can be interpreted significantly and recasts the interpretation of signs as not only being a question of what, but also of when, emphasizing a future tense. This approach to signal, noise, and signification presents a world where a maximum number of events are interpretable. An analyst might be driven toward maximum vigilance and maximum interpretation as a means of insurance against being caught unprepared, and they might employ a maximum number of techniques to tactically signal or produce noise to obtain a desired outcome, a positioning of oneself strategically against a range of possibilities. This kind of operational maximalism can accelerate the collection of information with the intent of producing meanings or identifying latent signals in a noisy mass. It anticipates the massive scale of information-aggregating systems at present, such as those found in large language models.

Alongside its conceptual applications, RAND used randomness towards practical ends in modeling and simulation. To be able to accurately run simulation games, researchers needed an abundance of high-quality, random numbers to sample from and sought out physical processes to generate these numbers. In 1955, RAND published A Million Random Digits with 100,000 Normal Deviates, an expansive text filled with tables of numbers generated from a complex electronic roulette wheel whose chaotic electric pulses were sampled at regular intervals. Consideration was not only taken towards the production and checking of statistical biases in the number tables but also in the selection of the numbers from the book directly. It was recommended that users mark the number they start with, as they had a tendency to select the same page in the book across multiple uses and also had a tendency to pick numbers at the center of the page.¹⁶ Though now obsolete, the book provided enough statistically random information for large-scale Monte Carlo simulations, where numbers were sampled and processed to approximate likely and unlikely real-world outcomes. In this way, the incorporation of unpredictable information, and importantly, information minimally affected by human intervention, served to produce a predictive picture of a given situation.

While simulation at RAND formed abstracted and quantified images of future scenarios within its institution, very similar techniques are embedded in concrete and everyday encounters with contemporary media. For instance, when rendering a scene with a path-tracing engine, Monte Carlo methods are used to simulate the behavior of light as it falls, bounces, and scatters across surfaces and through volumes. This is most noticeable when rendering scenes with low sample counts, in which a minimum number of rays are cast and calculated. These resulting images are grainy or noisy, exposing the points of light being sampled across the scene at random. In the example rendered in Blender’s Cycles engine, objects appear more solid where we see a clustering of like values at a given position, emphasizing a more uniform color and light value. In middle or transitional areas, the results of the traced paths are less clustered with similar values, giving a mixture of light and color that can be visually approximated as gradation. It is important to note that this is more apparent because the scene being rendered contains enough contrast to discern these changes. At extreme brightness, darkness, or lack of contrast, values become perceptually singular and indiscernible. This process also exposes the path-traced image as the result of predictive calculation. The render is more an artifact or after-effect of the stochastic process prior to it. Though it works to approximate a real scene, the process itself is a mathematical simulation of general principles and properties of light and material. The render conveys an implied correlation between the clarity of a given prediction and the amount of sampling afforded to it: the more that randomness can be employed, the clearer the outcome. But this also conveys what is made clear in this process: an attested power of the predictive model itself as a stand-in for the real.

Across both the conceptual and practical applications of noise at RAND, there was a necessary maximalism in practice. Just as the intelligence analyst benefited from a maximally broad interpretation of information as holding meaning, a scientist computed a simulation across a maximal number of samples, each purporting to contribute to the overall clarity of a future scenario. These approaches have only become more pervasive since then, reflected now in the maximalism of data centers, platforms that act as vast networks to ingest user behaviors, and networked supercomputers that run billions of trials. The scale at which randomness is used today is difficult to convey or contain in thought. At the time Baudrillard was writing, he argued that simulation had already scaled enough to create a condition of irrationality, since it no longer was concerned with measuring itself against external reference points.¹⁷ The proliferation of media within and contributing to a stochastic environment has only accelerated this. First used as a means to contend with uncertainty, the stochastic mode blurs the border between possibility and determination and increasingly develops into a form of certainty in and of itself: a belief that signs signal possible futures and that a mastery of signs might unlock or even determine that. This approach, and the thinking it can induce, is urgent at a time when rising sea levels and extreme weather are now fate rather than possibility and when human-created climate change, modeled in various forms since the latter half of the 20th century, now appears as prophecy or revelation. This exists alongside an economic environment that can appear to reward risk in speculative markets as an alternative to stagnant wage labor. Probabilistic uncertainty saturates American life, and the mode of being early, of anticipation, is being cultivated and embedded as an individual practice. It appears in more obvious and literal ways, such as the expanded access to and use of new retail investing apps, but it also permeates how individuals use pre-existing forms of media, such as photography, in an age of platforms, which swap qualitative aspects for quantitative measures, such as frequency, engagement, and positioning in relation to content of a similar kind.

Not long after dawn broke, I saw these dynamics unfold as more people gathered on the shore. Some sat alone, a few meeting others who had already been there. Many used the sunrise to take photos of themselves and their friends, rushing to catch the right angle for a soon-to-be post before it was too late. And some livestreamed themselves to an audience, giving good morning messages in an attempt to be a channel of positive energy, motivation, and vibes. The scene here was less a matter of spectacle and more an example of its abolishment. There was no singularity of vision, perspective, or unifying representation, but rather a fragmentation and reprocessing of angles and views to be distributed and indexed across online platforms. Everyone was facing different directions, with many people taking images literally having their backs turned towards or oblique to the coast so that their phone could capture their faces in front of the sky and the Miami Beach skyline.

Photography is often discussed as contending with time passed, a perpetual departure from a sampled moment, its futurity a form of hindsight as a reminder or remainder. But amidst a noisy environment of infinite content and aggregating algorithms, the act of image-making here suggested an inverse relationship with time. Each image was a construction projecting forward, perpetually reaching towards individuals’ desired representations or futures, operating as a form of manifestation and assertion. They did not record as much as propose states of affairs, and they did so as samples among thousands of other images that had been taken collectively on the beach and individually elsewhere by each person prior. As I looked across the shore, I could not help but see the signals they sought to generate, which, in turn, determined what I was looking at. In this way, the coast appeared to stretch and warp to meet the contours of itself as content made legible as aggregation rather than unity. Before leaving, I also took a photo. It was automatically grouped by date, location, and likeness to other images I had taken, including a generated collection of “sunsets by the water.” I imagined these scenes taking place on the beach each day until it could no longer render itself as paradise.