Source path
content/writing/short-posts/2026-07-08-focus-post-we-are-interacting-with-each-other.md

Local AT Protocol-shaped preview used to inspect the record before an exact public cache is refreshed.

{
  "note": "Local AT Protocol-shaped preview. Run `make garden-refresh-atproto` to cache exact public records where available.",
  "sourcePath": "content/writing/short-posts/2026-07-08-focus-post-we-are-interacting-with-each-other.md",
  "value": {
    "$type": "site.standard.document",
    "title": "Using AI kind of involves interacting with other people",
    "description": "A focused note on generative AI use as mediated interaction with upstream human contributors.",
    "publishedAt": "2026-07-08",
    "site": "at://did:plc:doxvahqvyhyqf32v7wz7p5xk/site.standard.publication/3mmpcciuaj22a",
    "content": {
      "$type": "at.markpub.markdown",
      "text": "# Using AI kind of involves interacting with other people\n\nThis is a companion to a few posts and papers that discuss the current AI paradigm as creating a kind of quasi-enclosure. In those posts, I've made the point that using AI tools has some key similarities with contributing on a platform like Stack Overflow: when you use an AI tool, even a proprietary one like Claude Code, you are (1) in a certain sense interacting with people and (2) your actions are potentially helping other people. Here I just want to discuss the focused point: each time I use an AI model, I am, in an unusual but meaningful way, interacting with other people digitally.\n\nConsider the \"causal chain\" upstream of my ability to get ranked tokens from some AI system. At some point, some person \"pushed the model weights\" in a particular direction. Perhaps by writing something, answering something, labeling something, or being paid to demonstrate a task. All these individual acts meaningfully changed the responses I received from the AI system. Mechanically, this is not so different from someone choosing to post an answer on Stack Overflow that causes me to write my code a little differently, or a Reddit post about a good recipe that causes me to eat a different dinner. Here, the contribution is pooled, recombined, and stripped of its origin, but the underlying structure is the same: another person acted, that action flowed through data, and my experience changed as a result.\n\nCurrently, we are mostly blocked from forming any relationship along that chain. Platforms like Google or TikTok rank and deliver discrete _bundles_ of information, e.g. a webpage, a video, a Stack Overflow answer. Each bundle was created by some identifiable person or organization through a social or institutional process. When I consume a bundle I can usually click through to it, check where it came from, and find the human on the other end, to thank or follow or argue with. Generative AI instead ranks _grains_: sub-bundle chunks, recombined into something new. The output that helps me is a statistical blend of many grains from many sources, so it no longer maps onto any single person's discrete creation.\n\nEven so, I think we can hold onto some of the value by deliberately keeping the causal chain in view. We can't (yet) fully re-bundle on the fly (trace a given output back to the specific people whose work it draws on) but we can at least keep remembering that there are people back there, and perhaps re-bundle in a coarse fashion.\n\nI've discussed some economic implications at length in other posts, but this also has implications for understanding our emotional reactions to AI-generated token sequences. In fact, I think this is a reason to actually give a little bit more \"spiritual weight\" overall to AI-driven digital interactions (in the same sense that we might have a reflective experience when we think about the vastness of our food supply chains).\n\n**Related posts on this site**\n\n- [The Paradox of Reuse in 2026: A Case of Quasi-Enclosure](ref:post:the-paradox-of-reuse-in-2026-a-case) and [AI progress as quasi-public good production](ref:shortpost:quasi-short): the quasi-enclosure framing.\n- [Google and TikTok rank bundles of information; ChatGPT ranks grains.](ref:post:google-and-tiktok-rank-bundles-of) — the bundles/grains distinction this post leans on.\n- [Each Instance of \"AI Utility\" Stems from Some Human Act(s) of Information Recording and Ranking](ref:post:each-instance-of-ai-utility-stems): every output traces back to human acts (\"ranking all the way down\").\n- [ChatGPT is Awesome and Scary: You Deserve Credit for the Good Parts](ref:post:chatgpt-is-awesome-and-scary-you-deserve-credit) and [Is Zuckerberg right to say that your specific creative work has no value to AI?](ref:post:is-zuckerberg-right-to-say-that-your): on whether individual contributions matter.\n- [Attestation across the AI Supply Chain](ref:post:attestation-across-the-ai-supply-chain): infrastructure for reconnecting outputs to people.\n- [Augmentation is a data flow problem](ref:shortpost:augmentation-is-a-data-flow-problem): sibling focus post.\n"
    }
  }
}