What Is the Cartography of Generative AI Website?

Updated

The Cartography of Generative AI (cartography-of-generative-ai.net) is a mapping project that traces what generative AI actually depends on—the extractions, agencies, and resources behind tools that write text or produce images in seconds. It is published as a set of downloadable PDFs in eleven languages, and its content is an explanatory essay rather than a software product or service. If you want a critical, systems-level account of how generative AI connects to human activities and to territories beyond the human, this site is the resource; if you need a tool, a dataset, or a technical tutorial, it is not.

What the site provides

The page presents the project under the heading "Cartography of generative AI" and offers the same material as PDFs in the following languages:

  • English
  • Català
  • Español
  • Italiano
  • Deutsche
  • Française
  • Euskara
  • Svenska
  • Português
  • 中文

The PDFs are the deliverable. There is no indication of pricing, accounts, or payment on the page, so treat access terms as unstated rather than assuming the material is free to reuse.

What the content covers

The essay is organised as numbered sections. The opening sections frame the project and then move into the material substrate of generative AI.

Framing (section 01). The popularisation of AI has produced imaginaries that "invite alienation and mystification." The project's stated purpose is to counter that by mapping how these technologies connect with human activities and with more-than-human territories. It asks a concrete question: what set of extractions, agencies, and resources make it possible to converse online with a text-generating tool or to obtain images in seconds?

How the tools work (section 01). Generative AI tools automate tasks such as writing or generating images. The automation is not achieved by programming concrete steps but by using examples: given many examples of a case, statistical networks configure themselves by analysing recurring patterns. Whether the input is words, pixels, or sound frequencies, a statistical model is obtained by analysing and exploring a training data set. The essay's formulation is that these tools "disassemble language (visual, textual) to reassemble it based on the calculation of probabilities." It also notes a shift: earlier tools were trained to produce concrete expressions (the image of a face, text in a certain style), whereas current tools generalise across many types and styles, which depends on much larger and more heterogeneous datasets. That scale of change, in turn, has required new economies and accelerated reliance on ecosystems.

Datasets (section 02). Training datasets are compiled industrially by extracting content from the largest digital archive in existence—the Internet—through automated scraping of content published and shared by millions of users. The essay makes a point worth noting for anyone researching data provenance: the original motivation for this extraction was not today's commercial exploitation by start-ups and platforms, but scholarly and non-commercial research. Now that these archives are used to generate texts and images on demand, the result is a series of paradoxes and controversies within the cultural industries. The excerpt ends mid-argument, with the ideology of big data understanding Internet content as a vast repository that can be extracted and processed.

How to use it

  1. Open the site and choose the language edition you can read most comfortably—the PDFs are parallel versions of the same text, so switching languages does not change the argument.
  2. Download the PDF rather than reading on the page if you want to follow the numbered sections in order or cite specific passages.
  3. Read section 01 first for the conceptual frame and the explanation of how generative models work; read section 02 onward for the material account of datasets, extraction, and the cultural-industry controversies that follow.
  4. Treat it as a secondary source for orientation and framing. If you need current model parameters, prices, or licence terms, go to the primary documentation of the specific tool you are evaluating—this project does not supply them.

Where it fits and where it does not

You want This site helps This site does not
A critical map of generative AI's resource and labour dependencies Yes—that is the project's core —
A reading text in a specific language Yes—eleven PDF editions —
Hands-on instructions for building or prompting a model — No tutorials or code
Current pricing, model specs, or licensing — Not covered
A dataset or data-access tool — It analyses datasets, it does not provide one

The main practical caveat is that the page evidence covers only the opening of the essay. Sections beyond the dataset discussion are not described here, so if your question concerns a later part of the argument, download the PDF and check the table of contents directly.