Most of us learned to read charts in school. A bar chart compares amounts. A line chart shows change over time. A pie chart splits a whole into parts. Here we use a different kind of chart asking a different question:

Who is talking about AI and work, and whose words are they using?

That question needs a different picture because a conversation has no single axis. People do not line up from left to right. They cluster, borrow each other’s phrases, argue past each other, and sometimes reach for the same word while meaning opposite things. A network graph is built to show exactly that kind of structure. This post explains how to read one, using the live graph that anchors the series.

 

Open the graph full screen →

Open the interactive graph

The basics: dots and lines

Every small dot is one public post about AI and work: a Reddit thread, a Hacker News discussion, a TikTok or YouTube video, an Instagram post, a GitHub discussion. Click a dot and its text, source and date appear in the side panel.

A line between two dots means the two posts use similar wording. Each post is connected to the few posts whose vocabulary is closest to its own. No one drew these lines by hand. They come from comparing the words in every post with the words in every other post.

Three things follow from that, and they are the key to reading the picture.

  • Closeness means similar wording. Connected posts pull toward each other and unconnected posts drift apart, so posts that talk alike end up near each other.
  • Clusters are the main signal. A dense knot of dots is a group of posts sharing a vocabulary. That knot is a topic the conversation keeps returning to.
  • Exact position means nothing. There is no up, down, left or right in this picture. The layout is produced by a physics simulation that settles the dots into place. Drag any dot and the others will shift and settle again. A cluster on the left is no different from the same cluster on the right.

Dot size carries one more piece of information. Larger dots sit between clusters: they are the posts that connect one group’s vocabulary to another’s. If you want to find where two conversations touch, look at the big dots on the edges. (Network analysts call this measure betweenness.)

Why it is called a graph

The picture can look like a map, and the resemblance is worth correcting. On a map, position is the information. Every point stands for a real place, distance follows a scale, and north means something. Draw the same territory twice and you get the same map.

A graph records something else: which things are connected. Here the things are posts, and the connections are shared wording. Where a dot lands on the screen is a by-product of the layout, which settles a little differently each time.

Distance still tells you something, loosely. Posts that sit close together tend to share meaning, because they share the words that carry it. Meaning is what this project is after. Wording is the graph’s stand-in for it, and the two can come apart. Two posts can use the same word for opposite things, and two posts can say the same thing in different words. So read closeness as a hint of shared meaning, to be checked by reading the posts. The Boundary objects lens, below, is built for the first case.

Linking texts by the words they share has a history in the sociology of science. Michel Callon and colleagues introduced it in 1983 as co-word analysis, a way to follow how a research field ties its problems together.

Two layers: what people say, and what the experts call it

The graph holds two layers of information, and keeping them apart is the point of the whole project.

The first layer comes from the posts themselves. The colors in the default view, called Topics, mark groups of posts that share vocabulary. A clustering method found these groups from the wording alone. Each group is labeled with its three most distinctive words. The largest group’s words, for example, are “trump · signed · executive.” Another reads “job · lost · writing.” These labels are machine-made and sometimes clumsy, but no outside category shaped them.

The second layer comes from published taxonomies of the debate. Over the past few years, journalists and scholars have offered ways to sort the AI debate into camps. This graph uses two. NPR described six factions in the AI safety debate: accelerationists, populist skeptics, safetyists, effective altruists, AI ethics scholars and pragmatists. Timnit Gebru and Émile P. Torres named a cluster of related ideologies they call the TESCREAL bundle (transhumanism, extropianism, singularitarianism, cosmism, rationalism, effective altruism and longtermism). In the graph, the two taxonomies appear as squares, the camps they name appear as diamonds, and each camp’s signature vocabulary (terms like “e/acc,” “p(doom),” “tech oligarchs” or “longtermism”) appears as a labeled circle sized by how often posts use it.

Here is the move that makes this graph different from a summary of who believes what. The published taxonomies are treated as part of the data. They are claims about the conversation, made by participants in it, and the graph tests how far those claims reach into what people say. Sometimes a camp’s vocabulary is everywhere. Sometimes nobody uses it at all.

Lenses: one question at a time

A graph with 767 posts is too much to read all at once. The buttons on the left are lenses. Each one asks a single question and highlights the part of the graph that answers it. Each lens comes from a body of scholarship on how knowledge and technology are shaped by society, and each will get its own post in this series.

  • What are people talking about, grouped by what their posts have in common? This is the bottom-up view, with no taxonomy applied.
  • Boundary objects. Which words do different camps share while meaning different things by them? A word like “superintelligence” can be a warning in one camp and a promise in another. (The term comes from Susan Leigh Star and James Griesemer, who studied how one object can serve groups with different goals.)
  • What do the published taxonomies reach, and what falls outside them? Posts that use no camp vocabulary at all are marked as unclaimed. (Geoffrey Bowker and Susan Leigh Star showed that every classification system leaves something out, and that what it leaves out has consequences.)
  • How much of this picture was made by the searches that built it? Each search appears as its own node feeding its posts. The four searches barely overlap: two searches found the same post only 33 times. (Sheila Jasanoff’s work on co-production holds that the tools used to know something shape what becomes known.)
  • Has camp vocabulary seeped into ordinary talk about AI and work? Plain-language posts that use any camp term are highlighted. (Antonio Gramsci argued that a worldview wins when it becomes common sense and no longer needs its own name.)
  • How is the White House’s Super Intelligence order being received? On September 29, 2026, an executive order directed federal agencies to replace the term “AI” with “Super Intelligence.” This lens tracks posts that invoke the order, from any camp or none.

The week player: watching the conversation move

Below the graph is a row of weekly bars and a play button. Press play, or drag the slider, and the graph lights up week by week from August 1 to October 5, 2026. Posts from the current week glow, and earlier posts fade like an afterglow. It works a bit like a satellite loop of a weather system: the point is to watch where activity forms and where it moves.

One caution matters here. Earlier weeks hold fewer posts, and most of August comes from Hacker News, because the search tool reaches further back on some platforms than others. Compare the shape of each week; the raw counts across weeks are uneven by design of the sources.

Three things to try first

  1. Open Classification and look for the empty circles. Four terms from the published taxonomies never appear in any post we collected: “EA,” “longtermism,” “human-centered AI,” and “extropian / cosmist.” The taxonomies name these camps, but in this sample nobody speaks in their vocabulary.
  2. Open Diffusion. Of the 245 posts found by searching ordinary language about AI and jobs, 17 use any camp term at all, about 7 percent. Most people talking about losing or changing work do not use the debate’s vocabulary.
  3. Open Reception and press play. Watch what happens in the week of September 29. 123 posts either use the order’s two-word spelling, “super intelligence,” or talk about the renaming. Two appeared before the order’s date and 111 after it (10 carry no date). Some of the early matches only share the spelling, and sorting those out is the next step for this lens.

What this graph cannot tell you

A picture this detailed can look more certain than it is. These are its limits, stated plainly.

  • It is a sample. The posts come from four searches run with a social-listening research tool (last30days) across Reddit, Hacker News, TikTok, YouTube, Instagram and GitHub. What the searches asked for shapes what the graph can show. The Instrument lens exists to make that visible.
  • It leaves platforms out on purpose. X (Twitter) and Truth Social are excluded by choice. Chinese-language platforms are not covered.
  • It is English-language and mostly American.
  • Topic labels are machine-made. They are the most distinctive words in each group. Read the posts before trusting a label.
  • Individual creators are anonymized. Posts by individual people on TikTok, YouTube, Instagram and GitHub show a generic label in place of the creator’s name. News organizations and institutions keep their names.
  • It is a snapshot that will be updated. This version covers August 1 to October 5, 2026. The graph will be refreshed as the conversation moves, and each update will be dated.

Why build it this way

When a debate is new, the words for it are still being fought over. Who gets called a “doomer,” whether AI is “AI” or “Super Intelligence,” whether a job loss is described as “replacement” or “efficiency”: these are not neutral labels. Each one is a move. A taxonomy of camps is a move too, however carefully it is made.

This series looks at the conversation about AI and work as it takes shape, week by week, with the method in the open. Each upcoming post takes one lens and reads what it shows. The graph is the evidence and the starting point. The posts are the reading.

Sources

  • Bowker, Geoffrey C., and Susan Leigh Star. 1999. Sorting Things Out: Classification and Its Consequences. MIT Press.
  • Callon, Michel, Jean-Pierre Courtial, William A. Turner, and Serge Bauin. 1983. “From Translations to Problematic Networks: An Introduction to Co-word Analysis.” Social Science Information 22 (2): 191–235.
  • Gebru, Timnit, and Émile P. Torres. 2024. “The TESCREAL Bundle: Eugenics and the Promise of Utopia through Artificial General Intelligence.” First Monday 29 (4).
  • Gramsci, Antonio. 1971. Selections from the Prison Notebooks. Edited and translated by Quintin Hoare and Geoffrey Nowell Smith. International Publishers.
  • Jasanoff, Sheila, ed. 2004. States of Knowledge: The Co-Production of Science and Social Order. Routledge.
  • 2026. Guide to the six factions in the AI safety debate. September 26, 2026.
  • Star, Susan Leigh, and James R. Griesemer. 1989. “Institutional Ecology, ‘Translations’ and Boundary Objects.” Social Studies of Science 19 (3): 387–420.
  • The White House. 2026. “Fact Sheet: President Donald J. Trump Inaugurates the Era of Super Intelligence.” September 2026.
  • Data: teKnoculture, last30days pulls on AI and work discourse, August 1 to October 5, 2026 (four searches, 767 posts, 734 distinct).