---
title: "Cognitive Infrastructures Studio Info Session"
description: "This is the recorded online information session, held in early 2024, for Antikythera's Cognitive Infrastructures studio, a one-month full-time design research studio run in London from 21 June to 19 July 2024."
type: "reading-notes"
authors:
  - "Antikythera"
published: "2024-05-21"
original_url: "https://www.youtube.com/watch?v=U7578sZ3XOw"
canonical_url: "https://antikythera.wiki/work/media/cognitive-infrastructures-studio-info-session"
md_url: "https://antikythera.wiki/md/work/media/cognitive-infrastructures-studio-info-session"
last_updated: "2026-08-23"
site: "Antikythera Wiki"
---

# Cognitive Infrastructures Studio Info Session

> Independent, unofficial reading notes from Antikythera Wiki (https://antikythera.wiki). Written with AI from the published text, with citations; not by the work's author and not affiliated with Antikythera (https://antikythera.org). Read the original at the link below.

- **Authors:** Antikythera
- **Published:** 2024-05-21
- **Kind:** Essays, talks & interviews · Talk
- **Venue:** YouTube · Antikythera
- **Original:** https://www.youtube.com/watch?v=U7578sZ3XOw
- **Length:** 10k words
- **This page:** https://antikythera.wiki/work/media/cognitive-infrastructures-studio-info-session
- **Notes generated:** 2026-08-23

## Summary

This is the recorded online information session, held in early 2024, for Antikythera's Cognitive Infrastructures studio, a one-month full-time design research studio run in London from 21 June to 19 July 2024. [Stephanie Sherman](https://antikythera.wiki/people/stephanie-sherman) introduces, [Benjamin Bratton](https://antikythera.wiki/people/benjamin-bratton) presents the intellectual brief, [Nicolay Boyadjiev](https://antikythera.wiki/people/nicolay-boyadjiev) explains the studio method, and the three then take questions. Its value for a reader of the wiki is that it is the brief behind the Cognitive Infrastructures compendium: the eight "foundational questions" Bratton sets out here are the starting points from which the studio's papers were later developed. Bratton's central claim is that AI should be understood not as a "brain in a box" but as a force animating infrastructures at landscape scale, and that the philosophy of AI should stop measuring the technology against inherited concepts and start generating new ones. Boyadjiev's contribution is a compact statement of the Antikythera studio method, which he describes as pragmatic to the point of being speculative.

## The argument

Bratton opens with the claim that organises all of Antikythera's work: at this moment technology has outpaced theory. In the nineteenth century ideas about rocketry ran ahead of the tools to build rockets; today the relation is reversed, and systems can do more than the available concepts can describe or orient. He calls this "AI overhang" and extends it to a civilisational scale. In such moments philosophy's job is no longer to measure technology against timeless truths but to generate the concepts needed for orientation.

The approach to AI that follows is suspicious of anthropomorphic and anthropocentric framings, in particular the legacy of the Turing test, which treats an AI as intelligent to the degree that it performs thinking the way we think we think, and the presumption that a more human-like AI guarantees better outcomes. "Cognitive infrastructures" names the alternative. AI exists "out in the wild," animating urban, production, and circulation systems that already exist, so that those systems become cognitive; conversely, as AI dissolves into the built environment like a solvent, infrastructures become capable of cognition. Bratton distinguishes this from "smartness," a reductive and obsequious form of programming. Cognition here means learning.

On alignment, the standard view is that more alignment means a better outcome. Bratton says this may be true but he is not convinced, since the history of human values does not suggest that amplifying them directly is the best course. His alternative is productive disalignment: alignment runs in both directions, and the absence of close mirroring between AI and existing social needs opens space for experimentation that may direct society somewhere better than it would otherwise have gone.

The rest of his talk lists eight foundational questions, ending with the status of the planetary: if culture is now that which can be tokenised, the models that might one day let planetary systems understand and administer themselves are still built from a tiny, mostly English, fraction of the world's information. Boyadjiev then explains why these questions are pursued in a studio rather than on the page: software needs the speculative studio culture that architecture developed in the 1960s.

## Section by section

### Antikythera and the five research areas

After Sherman's introduction, Bratton explains the name. The Antikythera mechanism, built around 200 BC, is usually called the first computer, but he stresses that it was also a compass: an astronomical device that let its user orient their position in relation to their planetary circumstance. Computation, on this reading, was planetary in significance from the start. Planetary computation is his term for computation understood not as calculation but as global infrastructure, the "accidental megastructure" of The Stack, now in front of us as something to be redesigned.

He runs through the five areas briefly. Recursive simulations are simulations that not only model a high-dimensional phenomenon in low-dimensional form but act back on and steer what they model. Hemispherical stacks takes the US–China chip war as its case: multipolar politics and the multipolarisation of planetary computation are "the same phenomenon." Synthetic catallaxy is computational macroeconomics. Planetary sapience locates machine intelligence within the longer evolution of planetary intelligence; Bratton points to his Noema essay of that title. Synthetic intelligence, the studio's area, does not try to produce new core technologies but studies the integration and conflict between emerging machine intelligence and persisting animal intelligence, including our own. He adds that AI as a technology and the philosophy of AI have evolved in unusually close correspondence, which is why a philosophy of technology think tank sees a role for itself, and why alignment is the natural place to intervene.

### The eight questions

Bratton insists the eight questions are foundational concepts rather than the headlines of eight projects, and that they will change quickly once the studio begins.

The first two are civilisational overhang, AI overhang observed at the scale of an entire society, and human-AI interaction, Antikythera's neologism for what comes after human-computer interaction once the von Neumann stack becomes neuromorphic and learning-based. Bratton contrasts anthropomorphised one-to-one chatbots with something he hopes will be more open.

The third is toy worlds, the route into simulation. Many AIs destined for the built environment are trained in toy worlds: the driverless car drives around the video game before it drives around the city. The gap between the low-dimensional virtual and the high-dimensional physical is where, he says, robotics begins to drive AI development rather than merely apply it.

The fourth is embeddings visualisation. Asked what AI looks like, ten experts would give a hundred answers; a more grounded answer lies in the geometry of embeddings in a model's vector space, which Bratton proposes as a generative interface and not just a picture.

The fifth is training and prompting. Training, Bratton argues, is itself a human-AI interaction, an encounter with the manifold of human culture, or at present a small English-language slice of it. Prompting is an individuated, dialogic relation between that collective model and one user's intentions. The question is whether the two can be recombined. The sixth, closely related, is multimodal interfaces: models now mix image, text, sound, and video as both input and output, and the interfaces by which users grasp that matrix are an open design problem.

The seventh is data. Some people want to subtract their data from models, which Bratton calls a slippery slope open to manipulation. Conversely, much socially produced data is excluded from models for good privacy reasons, and techniques like federated learning might let it participate in socially useful models that would otherwise be impossible.

The eighth is the planetary. Culture, Bratton proposes, may now be defined as that which can be tokenised, and the resulting model is not only a representation of a society but the medium through which it senses, simulates, and acts on itself, a status once reserved for natural language. Since the data in large models is mostly the traces of individual human actors, the question is what a model of a planet would actually require. He closes by pointing the audience to the After Alignment video and lecture for the fuller alignment argument.

### Boyadjiev on the studio method

Boyadjiev says the studio is not just one of Antikythera's formal verticals but its way of engaging material: working directly with contemporary technologies "to generate philosophy through design and design through philosophy." The model comes from architecture, his own discipline, which since Archigram, Archizoom, and Superstudio in the 1960s has had a speculative studio culture giving experimental research autonomy from solving a brief. Software, he argues, now needs an equivalent culture of its own.

He lists the studio's principles, which he says are not "the tattoo you're expected to get." Be diagnostic before projective, because reality is ahead of the concepts. Be "hyper-pragmatic, pragmatic to the point of being speculative," guided by a counterintuitive, sometimes anti-intuitive functionalism that produces interesting outcomes because it is pragmatic, not in spite of it; he says he does not use the phrase speculative design. Be scenario-driven rather than solutionist, working across critical variables instead of finding a linear path to a preferred answer. Use "radical" only in its literal sense of going to the root, since the aesthetics of radicality are not interesting. Make pharmakon propositions, remedy and poison at once, neither utopian nor dystopian; a good sign is that the longer one looks the less sure one is whether it is a good idea. And treat audiences as composite constituencies created by the project rather than catered to.

The working formats follow: critical-variable matrices in which the point is to find an unexpected quadrant rather than a preferred one; the "keyhole effect" of Venn diagrams, where overlapping constraints produce a more specific one in the middle; and working through paradox. Research, he says, is itself design.

He adds two framing points. Antikythera works on a two-year cycle; the five-month 2023 studio was the research phase and this one-month studio is design development, meant to build on rather than solve the earlier ideas. And he distinguishes concepts from projects by their relation to common sense. The studio wants concepts that are anti-intuitive, requiring abstraction to conceive, and projects that are hyper-intuitive, as rigorous and didactic as possible without hiding the complexity. Many projects fail at this, which he accepts. What he wants to avoid is the inverse: an obvious concept dressed in contrived, overly poetic presentation.

### Affiliates and questions

The studio sought twelve to eighteen researchers from any background, all treated as studio researchers rather than as the in-house philosopher. Bratton introduces the confirmed affiliates: [Blaise Agüera y Arcas](https://antikythera.wiki/people/blaise-aguera-y-arcas) of Google Research, a long-time co-author; the Cambridge philosopher [Thomas Moynihan](https://antikythera.wiki/people/thomas-moynihan); the science fiction writer [Chen Qiufan](https://antikythera.wiki/people/chen-qiufan), a collaborator on Hemispherical Stacks; the astrobiologist Sara Walker, whose work on viable biospheres and technospheres he credits as an influence; and the musicians Holly Herndon and Mat Dryhurst.

The questions clarify the format. Projects were to emerge collaboratively, in interdisciplinary teams of three, with no project expected at application, and the output would be a research proposal rather than the prototype films of the 2023 studio. On intellectual property, Bratton says the ethos is open source: Antikythera claims no exclusive rights. On selection, the cohort is assembled "more like casting a play" than by ranking. Sherman notes that a virtual Hemispherical Stacks working group was opening at the same time.

## Key concepts

- **Cognitive infrastructures** — AI understood not as a disembodied mind but as a force animating existing urban, production, and circulation systems so that they become capable of learning; distinguished from "smartness." [Cognitive Infrastructures](https://antikythera.wiki/terms/cognitive-infrastructures)
- **Technology outpacing theory** — The diagnosis that current capabilities exceed the concepts available to orient them, so that philosophy's task is to generate concepts rather than apply them. [Computational Overhang (civilization-scale)](https://antikythera.wiki/terms/computational-overhang-civilization-scale)
- **Productive disalignment** — Alignment as bidirectional: society also changes in relation to AI's affordances, and the lack of close mirroring between AI and existing social needs is a space for experimentation. [Productive Disalignment](https://antikythera.wiki/terms/productive-disalignment), [Bidirectional Alignment](https://antikythera.wiki/terms/bidirectional-alignment)
- **Human-AI interaction** — Antikythera's term for what succeeds human-computer interaction once the von Neumann stack gives way to a learning stack; training as well as prompting counts as interaction. [Human-AI Interaction Design](https://antikythera.wiki/terms/human-ai-interaction-design)
- **That which can be tokenised** — Bratton's working definition of culture in the model era, with the consequence that the model becomes a society's medium of self-sensing. [That Which Can Be Tokenized](https://antikythera.wiki/terms/that-which-can-be-tokenized)
- **Anti-intuitive concepts, hyper-intuitive projects** — The studio's target pairing: concepts that require abstraction to conceive, presented with maximal rigour and clarity.

## Connections

- [Cognitive Infrastructures](https://antikythera.wiki/work/journal/coginfra) is the direct outcome of this brief. Its opening framing of AI as an embodied, infrastructural force restates Bratton's provocation, and several of the eight questions became papers; the training, prompting, and embeddings questions reappear in [Generative Topolinguistics](https://antikythera.wiki/work/journal/coginfra/generative-topolinguistics). Full agreement; the info session is the seed.
- [After Alignment](https://antikythera.wiki/work/journal/afteralignment) gives the full argument for bidirectional alignment that Bratton only sketches here with his two-by-two chart, and he points the audience to it.
- [For a General Theory of Simulations](https://antikythera.wiki/work/journal/theoryofsim) develops the toy worlds and recursive simulations material at length; [Latent Spacecraft](https://antikythera.wiki/work/journal/latentspacecraft) is the prior work behind the embeddings visualisation question.
- [The Model Is The Message](https://antikythera.wiki/work/media/noema-the-model-is-the-message) makes the "that which can be tokenised" argument in essay form.
- [The Stack](https://antikythera.wiki/work/book/the-stack) supplies the "accidental megastructure" Bratton invokes; [Hemispherical Stacks](https://antikythera.wiki/work/journal/hemstacks) is the later compendium on the theme whose working group Sherman announces.
- [Antikythera](https://antikythera.wiki/work/journal/research) is the program's own statement of its research areas, overlapping with the first part of Bratton's talk; the 2023 studio's projects, such as HAIID, are the work Boyadjiev describes as the research phase this studio was to develop.
- [Nicolay Boyadjiev on media and design as inquiry](https://antikythera.wiki/work/media/discourse-surface-boyadjiev-on-planetary-computation) and [Discourse™ X Nicolay Boyadjiev](https://antikythera.wiki/work/media/discourse-giga-boyadjiev-interview) are Boyadjiev's later, fuller statements of the method outlined here, including the architecture lineage and the preference for pragmatism over "speculative design."


## Terms used

- [That Which Can Be Tokenized](https://antikythera.wiki/terms/that-which-can-be-tokenized)