Take It With a Grain of Salt
- Mar 26
- 8 min read
Commissioned by The Shed, New York
“What is the pattern that connects?”
— Gregory Bateson, Mind and Nature (1979)
Gregory Bateson’s work sits at the intersection of biology, cognition, communication, and ecology. He was not concerned with isolated phenomena, but with the relational structures that persist across systems. His statement points to the recurring logic of feedback, symmetry, and recursion that appears in living systems. Understanding, in this sense, is not about analyzing parts in isolation. It is about recognizing the relationships that hold systems together. This lens becomes particularly useful when examining societal fears and perceptions regarding artificial intelligence.
AI as System, Not Object
Artificial intelligence is often framed as a discrete technological breakthrough. Public discourse tends to isolate it as a thing, which makes it easier to attach narratives of control or takeover. Through a Batesonian lens, this framing becomes a category error when applied to its broader effects. AI does not operate as an isolated object but as a system of relations embedded within data, institutions, and human behavior.
At its core, AI functions within a recursive structure in which data informs training, training informs deployment, deployment shapes behavior, and that behavior generates new inputs that influence subsequent retraining. While not all systems update continuously, they tend toward this feedback pattern once embedded in real-world environments. The process is not linear but cyclical, extending beyond the model itself into the surrounding social and economic systems.
From this perspective, the unit of survival is not the model but the system of interactions in which it participates. Models are replaced, updated, or discarded, while the broader system persists and adapts. When viewed in this way, AI is less a rupture than an acceleration of patterns that already exist in modern systems, including abstraction, optimization, and feedback-driven adaptation. At sufficient scale, however, this acceleration can produce effects that resemble structural change.
The current anxiety surrounding AI is often framed as fear of the technology itself, but that framing is incomplete. While some concerns are tied to tangible risks such as job displacement or misuse, the intensity of the response is amplified by structural conditions that alter how humans understand, interpret, and act within the world.
One source is the erosion of epistemic grounding. AI produces outputs that are coherent and precise, yet fundamentally probabilistic. Under conditions of speed and cognitive load, there is a tendency to treat these outputs as reality rather than interpretation. As this tendency scales, the boundary between map and territory becomes increasingly unstable.
Another source is opacity. Many AI systems are not easily interpretable, even by those who build them. This does not eliminate feedback, but it degrades it. Outcomes remain visible, while the causal pathways that produce them become difficult to trace. As a result, the loop between action, understanding, and correction becomes less reliable, making trust harder to sustain.
A third source is the rate of change. Technological systems are evolving faster than social and cognitive adaptation. This imbalance is not uniform, but it is persistent across institutions, individuals, and governance structures. Humans require time to integrate new patterns into stable frameworks of meaning, and AI compresses that timeline, creating ongoing disequilibrium. Together, these conditions generate anxiety not because AI is inherently threatening, but because it destabilizes the processes through which we know, decide, and adapt.
The Double Bind
Gregory Bateson introduced the concept of the double bind as a condition in which conflicting demands create a situation with no clear resolution and no clean exit. Artificial intelligence presents a contemporary version of this dynamic. Individuals and organizations are simultaneously encouraged to adopt AI to remain competitive while exercising caution due to its risks. The expectation to move quickly exists alongside the expectation to avoid harm, and these pressures are reinforced by external conditions.
This is not merely a contradiction in messaging but a structural feature of the current environment. However, it is important to qualify that the condition is not absolute. Actors do have choices, but those choices are constrained and often carry tradeoffs that cannot be fully reconciled. The tension persists because the system rewards acceleration while penalizing missteps. As a result, participants may operate under continuous pressure without a stable resolution, navigating competing demands rather than resolving them.
Jobs as Patterns, Not Positions
Much of the conversation around AI focuses on job loss, but a Batesonian perspective reframes the issue by shifting attention from roles to patterns. Jobs are not static entities but patterns of coordinated behavior embedded within economic systems. From this perspective, AI does not eliminate work in a simple binary sense. Instead, it reorganizes the structure of work by redistributing tasks within and across roles.
Tasks that are routine and rule-based are more susceptible to automation, while value tends to shift toward judgment, synthesis, and relational coordination. This pattern aligns with broader economic trends, including job polarization, in which middle-skill roles decline while both high-skill and lower-wage service roles expand. However, this transition is uneven and sector-dependent, and displacement can occur faster than new roles are created, particularly in the short term.
At the same time, artificial intelligence is often described as existing within a digital domain, yet it depends on extensive physical infrastructure. Data centers, energy systems, water management, and semiconductor manufacturing form the material foundation that supports AI systems. What appears abstract is grounded in physical and industrial systems, reinforcing the idea that technological change is always embedded within broader economic and ecological constraints and opportunities.
The Map Is Not the Territory
Bateson’s observation that the map is not the territory is directly applicable to artificial intelligence. AI systems operate on representations, including data abstractions, encoded patterns, and statistical relationships. These representations are useful for inference and prediction, but they are not reality itself. The primary risk is not limited to technical error but extends to a gradual shift in how knowledge is interpreted.
Under conditions of speed, scale, and reliance, there is a tendency to treat outputs as facts rather than as interpretations. This can lead to increased dependence on model-generated proxies in place of the lived context. As a result, nuance, contradiction, and context are compressed into simplified patterns that do not fully capture the complexity of real-world systems. This constitutes a category error, not because models are incorrect, but because they are misclassified as substitutes for relational experience.
The relationship between a bee and a flower illustrates this distinction. The interaction is shaped by the environment, sensory input, and biological processes unfolding in real time. While aspects of this interaction can be modeled, the full relational process cannot be fully captured within a representational system. Human experience operates in a similar way. Once translated into data, it becomes partial and selective rather than complete, and its meaning depends on the context that cannot be fully encoded.
Learning from Social Media and Smartphones
Recent technological history provides a useful precedent. Social media and smartphones were not inherently harmful, but they were deployed at scale without sufficient constraints on how they would shape attention, behavior, and social interaction. Systems designed to maximize engagement produced reinforcing feedback cycles that amplified certain signals while suppressing others. Over time, these dynamics altered how information was consumed, how attention was allocated, and how individuals related to one another.
The lesson is not to reject technological innovation, but to recognize that systems optimized around narrow objectives can generate broader unintended effects. However, it is important to note that these outcomes were not uniform and have since prompted adaptive responses, including changes in user behavior, platform design, and emerging regulatory efforts.
Artificial intelligence operates at a deeper level by influencing not only attention but also cognition and decision-making. The appropriate response is therefore not to embrace or resist, but to constrain and integrate. The pace of adoption, the boundaries of use, and the surrounding ecosystem are shaped through distributed decisions across individuals, institutions, markets, and policy frameworks. Outcomes are not predetermined by the technology itself, but by how it is embedded within these systems.
Designing for System Health
The central task is to design systems that prioritize feedback. These conditions define whether a system can sustain learning and coherence over time. When feedback remains visible, systems can detect errors, surface unintended consequences, and adjust accordingly. When feedback is obscured or filtered, systems may continue to function while gradually losing the ability to recognize degradation. This does not eliminate feedback entirely, but it reduces its quality and reliability.
When judgment remains situated, decisions are made by individuals or institutions embedded in the context of those decisions and accountable for their outcomes. When judgment is displaced or overly delegated to systems, decision-making risks becoming detached from the conditions it affects, even if it appears efficient. When adaptation remains possible, systems retain the capacity to revise behavior in response to new information and changing environments. When adaptation is constrained, systems may converge too quickly on optimized patterns, reducing flexibility and resilience over time.
These are not abstract principles but practical design requirements. The challenge is not to control artificial intelligence as an isolated technology, but to preserve the conditions under which human systems can continue to perceive, interpret, and respond effectively within an increasingly mediated environment.
Artificial intelligence will continue to shape society, but that is not the central question. The more consequential question is whether we shape the conditions under which it operates. Those conditions are not determined by the technology itself, but by how it is integrated into systems of decision making, incentives, and human behavior. If AI is integrated in ways that preserve feedback, maintain situated judgment, and support adaptation, it can function as part of a system that remains responsive and capable of learning. If those conditions are weakened, systems may optimize for immediate outcomes while gradually eroding long-term viability.
The pattern that connects is not the technology itself, but how it reorganizes perception, decision making, and feedback across systems. In that sense, artificial intelligence does not determine outcomes on its own. It participates in systems that are continuously shaped through distributed choices across individuals, institutions, and governance structures. The responsibility, then, is not singular but shared. It lies in how those choices are made, and whether they preserve the conditions necessary for systems to remain open, adaptive, and grounded in the realities they are meant to serve.
This analysis, including the research and assumptions, was developed and tested through interaction with an artificial intelligence system. It is itself a product of the system it examines. As such, it reflects both the strengths and limitations of working through representation, shaped by my own lived context and inherent biases. It should be read with discernment. Take it with a grain of salt.
©1982 Agnes Denes
"Wheatfield—A Confrontation, which the scholar and curator Jeffrey Weiss has called “perpetually astonishing … one of Land Art’s great transgressive masterpieces” (Artforum, September 2008), is perhaps Agnes Denes’s best-known work. It was created during a four-month period in the spring and summer of 1982 when Denes, with the support of the Public Art Fund, planted a field of golden wheat on two acres of rubble-strewn landfill near Wall Street and the World Trade Center in lower Manhattan (now the site of Battery Park City and the World Financial Center). Among her many other artistic achievements is Tree Mountain—A Living Time Capsule, a monumental earthwork, reclamation project and the first man-made virgin forest, situated in Ylöjärvi, Finland. The site was dedicated by the President of Finland upon its completion in 1996 and is legally protected for the next four hundred years."
Born in Budapest, Hungary in 1931, Agnes Denes was raised in Sweden and educated in the United States. Since her exhibition career began in the 1960s, she has participated in more than 450 exhibitions at galleries and museums throughout the world including, among others, solo shows at the Corcoran Gallery of Art, Washington, D.C. (1974); the Institute of Contemporary Art, London (1979) and retrospective surveys at the Herbert F. Johnson Museum of Art, Cornell University, Ithaca, N.Y. (1992); the Samek Art Gallery, Bucknell University, Lewisburg, Pa. (2003); and the Ludwig Museum, Budapest, Hungary (2008). Her work has also been featured in such international surveys as the Biennale of Sydney (1976); Documenta 6, Kassel, Germany (1977); the Venice Biennale (1978, 1980, 2001), and more recently The Last Freedom: From the Pioneers of Land Art of the 1960s to Nature in Cyberspace, Ludwig Museum, Koblenz, Germany (October 16, 2011); Systems, Actions & Processes: 1965–1975, PROA Foundation, Buenos Aires (through September, 2011); Erre: Variations Labyrinthiques, Centre Pompidou, Metz (September 12, 2011 – March 5, 2012); and Light Years: Conceptual Art and the Photograph: 1964 – 1977, Art Institute of Chicago (December 11, 2011 – March 11, 2012).
Agnes Denes is represented by Leslie Tonkonow Artworks + Projects, New York.
Comments