HOS Image: Waiting as an algorithmic form: The image as a promise
/German)
/Spanish)
Marcello Mercado
HOS Image: Waiting as an algorithmic form: The image as a promise
On the threshold between generation and waiting, this installation presents an image that has not yet occurred. A system message, repeated and trapped in its own state of waiting, transforms into a generative landscape. Here the «non-event» becomes the scene: inaction as a form of agency, error as a form of aesthetics. The algorithm does not generate the image, but its ghost: an invisible performativity, a readymade suspended in latency.
.
This phenomenon can be understood as a manifestation of «differential algorithmic latency,» a mechanism structured by the system to manage computational load through access hierarchies. Waiting is not an error or an exception: it is a function of the system, a strategy of commercial optimization and differentiation of experience.
.
In this context, the image generated by artificial intelligence is no longer simply a visual result, but a computational instance conditioned by access logics, prioritization strategies, and usage policies. Each image activates a consumption of computational resources and becomes an entity with computational costs, with implications for algorithmic architecture and commercial policy.
–
This work invites reflection on how latency, waiting, and inaction can become aesthetic and critical elements in the age of generative artificial intelligence.
.
1. HOS Image: Waiting as an algorithmic form: The image as a promise:
A HOS (Held On Server) image is an image generated by an artificial intelligence model that has not yet been fully processed or delivered to the requesting user. It remains suspended in a queue within shared servers, typically assigned to non-paying or free users. Technically, this situation reflects a mechanism of resource management and computational prioritization.
Conceptually, the HOS image embodies a latent visuality, a not-yet-deployed form inhabiting an invisible architecture, in transit between the request and its visual appearance. It can also be understood as a symptom of an unequal algorithmic economy in which access to the visible is mediated by hierarchies of payment, speed, and computational privilege. The image exists as a promise, as a wait, as a structured delay.
A. General Framework:
The ChatGPT phrase «Image processing. Many people are creating images right now, so this may take a while. We’ll notify you when your image is ready« can be analyzed as an interface unit that condenses a technical architecture, a business logic, and a time-of-use modulation. Waiting is not an error or an exception: it is a function of the system.
B. System elements involved:
1. Algorithmic Queuing: The system implements a scheduled wait time to manage limited computing resources. Premium users are prioritized, while non-paying users are moved to a «low-priority zone» managed by a scheduler.
2. Load balancing: The message indicates high load. Technically, it implies saturated compute nodes and the redistribution of tasks to others. But this redistribution is not neutral: it obeys access rights policies.
3. Conditioned Temporal Experience: Users‘ experience is modulated according to their place in a hierarchy. Latency is no longer merely technical: it becomes an operational stratification mechanism.
C. Categories of Analysis:
1. Functional Latency: The difference between request time and delivery time.
2. Strategic Latency: deliberately introduced as part of the freemium model.
3. Perceptual Latency: how waiting is communicated (in this case using informal and passive language) and how attention is managed.
D. Computational and structural implications:
Waiting becomes an indirect selection operator that filters user behavior (wait? pay? abandon?).
The message is part of an algorithmic containment system designed to regulate the anxiety of waiting with a friendly phrase, while keeping its discriminatory logic intact.
From a platform architecture perspective, this is a case of differentiated scaling, where access to compute–intensive resources is regulated by the business model.
E. Conceptual Proposal:
This phenomenon can be categorized under the concept of Differential Algorithmic Latency (DAL):
It is a queuing mechanism structured by the system to manage the computational load through access hierarchies. It responds not only to technical capabilities, but also to business optimization strategies and experience differentiation.
2. The Image as a Differential Product in Generative Systems:
Latency, Business Optimization and Access Hierarchy
A. INTRODUCTION
In the context of generative artificial intelligence, the image is no longer simply a visual result, but a computational instance processed in a distributed architecture conditioned by access logic, prioritization strategies, and usage policies. This essay begins with a technical and structural analysis of the message:
«Image processing. Many people are creating images right now, so this process may take a while. We will notify you when your image is ready.»
To break down its implications into two dimensions:
(1) the redefinition of «image» as a generative entity within an economy of latency and optimization,
And second, the algorithmic design of a conceptual simulation in Python that represents this differential waiting as a structural function.
B. What is an image in generative AI?
1.1. Image as computational output (output artifact).
In generative systems, such as DALL-E or Stable Diffusion, the image is not a pre-existing object, but an artifact resulting from a series of statistical computations, latent space sampling, and pixel generation from trained models. Technically, it can be defined as
Image (AI-G): a data structure consisting of a multi-channel array (RGB matrix) derived from a probabilistic inference on a trained model, conditioned by a textual prompt and generation parameters.
1.2. Image as differentiated consumption vector.
The same image, generated with different access levels (free or premium user), presents different latency, resolution or quality conditions. Therefore, the generated image is not universal; its shape depends on the business model.
Differential image: A computational product whose availability, resolution and generation speed are modulated by the privilege level assigned to the user.
1.3. Image as a node within an economic infrastructure.
Each generated image triggers the consumption of GPU, memory and computing time. On a freemium platform, this translates into a real monetary cost for the provider. The platform manages these costs based on
Rate limiting models.
Priority queuing.
User segmentation logic (free, professional, API).
The image is therefore not a final result, but an entity with a computational cost, with implications for algorithmic architecture and commercial policy.
C. Technical analysis of the «Processing image…» message.
The message contains four layers of structural information:
Layer Content Function
1. Status: «Processing image» … Informs that the task was received and is in progress
2. Load: «There are many people creating images.» indicates system saturation and load distribution.
3. Warning: «This may take a while». Enter latency as an expectation
4. Delayed response: «We’ll let you know…» Shifts attention to the future and manages user anxiety.
This set of elements constitutes an algorithmic cognitive containment strategy designed to maintain user engagement without revealing the exact logic of the queuing system or access hierarchies.
.
3. Building a Conceptual Algorithm in Python
The next phase is to design a Python algorithm that simulates this queuing structure. The algorithm does not generate images, it simulates:
– Task queuing based on priority.
– Differential latency.
The structured response of the system based on user conditions.
This algorithm will serve as a critical analysis model for image generation platforms based on freemium models.
3.a Conceptual Algorithm: Differential Latency Simulation in AI Image Generation
(Spanish version / English version below)
Explanation:
This algorithm simulates a differentiated queue structure based on user type (free or paid). It does not generate real images, but rather:
1. Generates queues of requests.
2. Assigns wait times based on user type.
3. Prints the message «Processing image…» followed by a notification when the image is ready.
Conceptual algorithm: Its goal is to represent a simulation of an imaging system with latencies, errors, priorities, and logs: similar to the message «Processing image. Many people are creating images right now, so this process may take a while. We will notify you when your image is ready:
3b. Curating HOS Images
In the contemporary universe of AI-generated images, a new figure is emerging: the HOS (Held On Server) image. It is not an image per se, but rather its suspended prefiguration. Its existence is marked by waiting: an initiated request, a promise of visuality not yet realized. This image, held on shared servers assigned to non-paying users, has not reached the threshold of the visible. It inhabits the technical limbo of the latent.
In this context, art is no longer limited to representing what appears, but begins to incorporate the structures of access, computation times, and systemic inequalities that shape what can and cannot become an image. The HOS image is both an object of study and a critical gesture. It forces us to ask: what is left out of visuality for infrastructural reasons? How does the algorithm work as an economic and aesthetic filter? What does it mean to curate what has not yet appeared?
Curating the HOS images involves attending to the ways in which latency becomes political. Instead of exhibiting finished images, one can think of exposing dead times, unfinished processes, moments when the eye waits but does not see. This displacement transforms the exhibition into a choreography of the suspended, an archaeology of non-rendering.
4. Waiting as an Algorithmic Form
Rather than understanding waiting solely as a technical delay, it can be thought of as a computational form prefigured by access conditions, algorithmic priorities, server architecture, and business models. In the non-paying version of ChatGPT, waiting becomes a kind of computational class threshold: if you don’t pay, you wait. This is related to ideas from
A. Critical Infrastructure Theory (Lisa Parks, Tung-Hui Hu): which examines the invisible layers of digital processing and how they mediate user experience.
B. Latency Theory: latency not only as a technical delay, but as a political construction of access time.
C. Media Theory (Wendy Hui Kyong Chun): processing times are not neutral, but political forms of relation.
5. Collective power of processing
The message «many people are creating images» introduces the idea of an invisible crowd, a phantom community, producing simultaneously. Here connections are made with
A. Structured simultaneity: a form of contactless shared temporal experience that could be compared to networks like Uber or Amazon Mechanical Turk.
B. Algorithmic queues: as spaces of invisible negotiation of desire, where each user is an instance waiting its turn in an opaque architecture.
What is an «image» in the context of generative AI?
In the context of generative AI, an «image« is no longer exclusively a visual representation produced by optical or manual means. Instead, it is transformed into a computational instance generated from statistically trained models that interpret latent vectors, textual prompts, and loss functions. The image no longer represents an objective reality or a symbolic subjectivity, but becomes an output surface optimized by algorithmic architectures, computational efficiency criteria, and training patterns.
An AI-generated image is not a stable object, but a deferred process, subject to logics of waiting, commercial priorities, staggered access, and asynchronous processing. The message «Processing image. Many people are creating images right now…» reveals that the image is also an experience of latency, competition for resources, and algorithmic regulation.
Classification of definitions:
01. Techniques
02. Epistemological
03. Ontological
04. Temporal
05. Phenomenological / Perceptual
06. Political / Economic
07. Mistakes / Failures / Mutations
08. Archival / Serial
09. Archaeological / Historical–Evolutionary
1-5. Techniques
01. Image as output vector: set of numerical values in a multidimensional latent space transformed into pixels.
02. Image as functional convergence: Result of a minimally satisfied objective function during training.
03. Image as loss residual: surface that minimizes the distance between the prediction and the data set.
04. Image as weight transfer: updated state of millions of parameters after backpropagation.
05. Image as digital transduction: interpretation of a textual prompt using convolution and attention layers.
6-10. Epistemological
06. Image as an act of automated inference: what the machine deduces from a description.
07. Image as the result of a probabilistic epistemology: it represents the most likely, not the true.
08. Image as a cognitive shift: it suggests knowledge without direct human origin.
09. Image as predictive artifact: a proposition of what should or could be.
10. Image as crystallization of learned biases: visual materialization of a database.
11-15. Ontological
11. Image as non-object: it does not exist without a network, a prompt, and a generating instance.
12. Image as computational event: an action that occurs and disappears if not stored.
13. Image as interpolation surface: between n latent points without a fixed center.
14. Image as model interface: visible only as a translation of internal processes.
15. Image as threshold: boundary between code and representation.
16-20. Temporal
16. Image as waiting: what appears after an algorithmically controlled delay.
17. Image as asynchronous process: it does not respond to human time, but to the flow of demand.
18. Image as perceptual latency: a state of delayed immediacy.
19. Image as an ephemeral state in the execution queue.
20. Image as a state of postponement: always potential, never immediate.
21-25. Phenomenological/Perceptual
21. Image as an experience of delayed gratification.
22. Image as a projection of human expectation onto an opaque network.
23. Image as algorithmic representation with the appearance of creation.
24. Image as perceptual deception: it appears human but is machine-like.
25. Image as a surface for the attribution of meaning.
26-30. Politics / Economy
26. Image as a unit of value generated under payment priorities.
27. Image as a product conditioned by levels of access.
28. Image as a computational privilege.
29. Image as the result of incentive architecture.
30. Image as an algorithmic filter that decides who sees what and when.
31–35. Mistakes / Failures / Mutations
31. Image as glitch: a revealing error in the system.
32. Image as unwanted mutation of a prompt.
33. Image as productive failure: it doesn’t fulfill expectations, but generates interpretations.
34. Image as the residue of incompatible codes.
35. Image as an asymmetry between human desire and network output.
36–40. Archive / Series
36. Image as one instance among millions: a node in an infinite series.
37. Image as copy without original: each generation is first and last.
38. Image as a visual record of an interaction.
39. Image as digital trace, not preserved.
40. Image as a speculative archive: its value lies in possibility, not stability.
41–45. Archaeological / Historical-Evolutionary
41. Image as a continuation of the automated pixel of video games.
42. Image as heir to the algorithmic art of the 1960s.
43. Image as a mutation of GANs and their evolution into Transformers.
44. Image as a consequence of the computational dream of the perfect image.
45. Image as the current stage of a long history of visual automatisms.
1. Technical Definitions
01. Image as a data vector: An image is a matrix of numerical values representing visual information that can be processed by a generative model.
02. Image as computational output: The result of an inference performed by a neural network based on textual or latent input.
03. Image as optimized file: An image is a structure compressed and transformed according to efficiency parameters (weight, format, resolution).
04. Image as rendering instance: Represents a frame generated by layers of graphical processes controlled by stochastic parameters.
05. Image as minimized loss function: The visual manifestation that emerges when a model manages to minimize the error function with respect to the given challenge.
2. Epistemological definitions
06. Image as a visual hypothesis: A probabilistic conjecture that the model proposes as a valid representation of the input text or context.
07. Image as statistical knowledge: Represents a point of convergence among thousands of examples seen during training, with no direct reference.
08. Image as synthesis of correlations: It is the result of the superposition of co-occurring patterns in previous data sets.
09. mage as operational interpretation: It is a reading that the system makes of a human instruction in terms of vectors and weights.
10. Image as the result of implicit reasoning: It does not illustrate a truth, but rather a probable calculation generated by hidden relationships in the model.
3. Ontological definitions
11. Image without a referent: The generated image does not represent an existing object, but rather a formal possibility.
12. Image as operational fiction: Its existence depends on the execution of an algorithmic process and not on an empirical world.
13. Image as technical object: It has its own existence as a processed, recorded and stored entity.
14. Image as synthetic appearance: It does not arise from a physical phenomenon, but from a network of numerical transformations.
15. Image as an unstable double: It is a projection that does not refer to a thing, but rather to a set of statistical conditions.
4. Temporal definitions
16. Image as latency: An entity that does not yet exist, but whose process has begun, awaiting computation.
17. Image as process duration: Its existence is measured by the time it takes to be computed and displayed.
18. Image as a product of waiting: It is linked to a social time shared by thousands of simultaneous users.
19. Image as deferred event: Its appearance depends on a priority queue managed by servers.
20. Image as operational suspension: It is characterized by a threshold between input and visual response.
5. Functional definitions
21. Image as Visual Response: It is the functional translation of a textual instruction or prompt.
22. Image as a unit of satisfaction: It is measured on the basis of its usefulness or congruence with the user’s desire.
23. Image as a validation interface: It allows for verification of the functioning of the model or the clarity of the prompt.
24. Image as a training test: It is used to evaluate the effectiveness or biases of the system.
25. Image as Reproducible Output: It can be regenerated, adapted or transformed under new conditions.
6. Aesthetic Definitions
26. Image as an emerging style: Acquires new visual characteristics by combining diverse training.
27. Image as automatic pastiche: Recombines formal features from thousands of styles and artists without awareness of authorship.
28. Image as Perceptual Coherence: Evaluates an image based on whether it appears «plausible» or «aesthetically complete.
29. Image as Statistical Visual Pattern: Its appearance is guided by regularities in the data set.
30. Image as object without aura: It has no original or context of human production.
7. Political definitions
31. Image as infrastructure product: Dependent on global computing resources and architectural decisions.
32. Image as algorithmic curatorial decision: What appears is filtered through the prioritization, censorship, and policy mechanisms of the model.
33. Image as conditional access: It is determined by the level of subscription and permitted use of the system.
34. Image as Digital Privilege Trace: Its resolution and speed of delivery vary according to socio-economic conditions.
35. Image as a result of opaque governance: The user neither controls nor knows the exact criteria for its generation.
8. Economic definitions
36. Image as a differentiating good: It is produced as part of a strategy of exclusivity or experience customization.
37. Image as a monetization node: It can be transformed into an NFT, a commercial product or viral content.
38. Image as a by-product of a SaaS model: It is part of the value proposition that justifies a subscription model.
39. Image as a return on data investment: It is generated from years of training on massive data sets.
40. Image as an aesthetic trademark of the vendor: It implies a style, speed, or aesthetic specific to the generating system.
9. Archaeological and Evolutionary Definitions
41. Image as the current version of a genealogy: It is the heir of previous visual practices (collage, rendering, CGI).
42. Image as historical accumulation: It is loaded with layers of data from different eras and styles.
43. Image as a technical threshold: It marks a turning point in the evolution of computational imagery.
44. Image as synthetic residue: It accumulates as part of the growing archive of generated output.
45. Image as future archaeological evidence: It could be studied as a cultural trace of generative AI in our time.
10. Critical / Meta-theoretical Definitions
46. Image as visible ideology: It involves technical choices that reflect values, exclusions, and biases.
47. Image as object of critical speculation: It can be used to question the boundaries between authorship, automation, and culture.
48. Image as a form of accelerated abstraction: It is produced without consciousness, body or affect, but with visual logic.
49. Image as a reification of connections: It makes visible regularities without deep semantic content.
50. Image as a distorting mirror of desire: it does not return what is desired, but rather what the system infers it wants.
The 50 definitions of the image in the context of generative AI show that we can no longer understand the image as a passive unit of perception or as a stable representation of reality. Instead, it emerges as a technical, operational, and speculative entity. Each generated image not only manifests a process of statistical inference, but is also marked by economies of waiting, hidden infrastructures, algorithmic decisions, and a genealogy of visual techniques. The act of «waiting for an image» is a critical experience in itself, as it makes us aware of time, privilege, technical architecture, and the transformation of the image into a conditioned flow. From this perspective, the image no longer represents, but distributes, prioritizes, and calculates. AI does not produce images of the world, but of the system that produces them.
Practical applications of the definitions:
Critical Curation of Generative Images
These definitions allow for the construction of curatorial criteria for exhibitions that work with AI-generated images, focusing not on the «content» of the image, but on its technical, political, temporal, or epistemological nature.
Interface Design
Developers can incorporate these categories to create interfaces that present waiting, latency, or errors as part of the aesthetic experience, rather than hiding them.
Dataset analysis
Use these categories to classify training images and understand how aesthetic or ideological categories are distributed across datasets.
Critical digital media pedagogy
Use the definitions as a basis for courses in art, design, philosophy of technology, or visual studies that seek to problematize the image beyond its form.
Institutional critique of generative models
These definitions can inform institutional policies on algorithmic transparency and visual ethics, suggesting criteria for evaluating generative models in terms of bias, accessibility, or economic structure.
Speculative AI Architectures
System architects or artists can use the definitions to simulate alternative AI models in which time, waiting, or error are intentional and significant elements of visual production.
For example:
5. Data Set Analysis: Reconsidering the Image as a Distribution of Latent Biases and Structures
Core Concept
By redefining the image as a technical-algorithmic entity within a generative AI system, the need arises to address not only the visible result, but also the latent conditions that make it possible. The training dataset ceases to be a simple collection of images and becomes an empowered visual structure in which biases, repetitions, aesthetic hierarchies, omissions, and privileges are manifested. Thus, analyzing a dataset is not just about inspecting images, but also about mapping the patterns of visibility and exclusion they generate.
Practical application:
Using the 50 previous definitions (especially the ontological, epistemological, and political ones), a taxonomic analysis model of the dataset can be built:
For example, images can be classified according to their lighting patterns, facial proportions, or dominant artistic style.
Measure the geographic or ethnic distribution of faces.
Formal redundancies that promote «neutral» or dominant styles can also be identified.
Compare the number of images with neutral backgrounds to images with environmental context.
Evaluate implicit taxonomies: what types of bodies, spaces, or gestures are overrepresented or absent.
This can be translated into critical visual tools or interfaces, such as
Visual bias maps that show the density of certain types of images.
Latency visualizations that show which types of images appear faster or with greater certainty in generation.
Dialog interfaces that confront the user with the genealogy of a generated image and show its training «ancestors«.
6. Examples of Works, Installations and Interfaces:
A. «The Training Room (interactive installation)
An immersive space that simulates being inside the dataset. Images are projected as a floating cloud, and each visitor can «select« an image to see what 1,000 other similar images accompany it. Each selection displays metrics such as «frequency,» «stylistic repetition,» «predominant race,» «geographic origin,« and «level of detail.
Inspired by model training rooms, but visually revealing invisible structures.
B. “Latent Discriminator” (critical web interface)
A website in which the user generates any image using AI. The same interface then returns a detailed analysis of which subsets of the dataset most influenced that image, including percentages, styles, semantic classes, detected biases, and possible omissions. It would be both an educational and critical tool. The user becomes aware that the generated image does not emerge from nowhere, but from a structured and deeply unbalanced field.
C. “The Delay Mirror” (algorithmic performance and installation)
A camera captures your face, and an AI attempts to generate your portrait. But the system intentionally introduces variable delays based on statistical criteria extracted from the dataset: if your facial style is less frequent in the dataset, the image takes longer. A graph shows the system’s level of “familiarity” with your face in real time. A critique of unequal representation and visibility privileges. Not all bodies are read with the same fluency.
D. «Genealogy of an Image« (Advanced Visualization)
When generating an image with AI, the system not only gives you the final image, but also displays a genealogy of the training data: key images, dominant styles, visible/invisible authors. Each image is accompanied by a «generative transparency« score and a «complexity of inheritance« score.
This makes it possible to visualize the debt that each generated image owes to its origins.
E. «Latency Atlas (critical cartography of visual generation)
A large screen displays a map of the world divided into cultural, ethnographic and aesthetic zones. For each region, the average time it takes an AI system to generate images associated with that culture or style is measured and displayed. A bar graph shows which zones are «generated« the fastest.
Direct exposure of algorithmic inequalities. Speed is also bias.
7. Research References:
Yuk Hui – The Question Concerning Technology in China
Benjamin Bratton – The Stack: describes how platforms reorganize forms of sovereignty and time.
Mark Hansen – Feed-Forward: On the Future of 21st-Century Media: explores how 21st-century media preprocess the future.