Summary
For an AI, the world is the whole of the data the AI can read. Whatever lies outside the readable data does not exist for that AI to begin with. This column is about building and running such a world out of nothing but the activity data of a single person, and what I thought about while doing it. One person’s worth of data is a tiny fraction of the world as a whole, but the loop that reads that data and makes the next thing is actually turning right now. Where will the role of the human, who has taken in data and thought about it, move to from here? I start from this small attempt and work my way to that question.
Audience and takeaways
This column is for people who are building something together with an AI and have come to want to reconsider what they are supplying to that AI. How long will what I hand over now stay something only I can hand over? Beyond that, where does the role of the human move to? This is not a story that settles on a single answer, but you can take away one way of seeing the role of the human who has ended up on the side that gets overtaken. I wrote it hoping it might be a clue for people who develop with an AI every day to rethink where they stand.
So, concretely, what does the world refer to for an AI? Just as a human builds up the world from the information that comes in through the eyes and ears, an AI builds up the world from the data it is given. Whatever lies outside the readable data does not exist for the AI to begin with. So the data you hand over decides the breadth and the shape of that AI’s world.
The world for this site’s AI is made of nothing but the activity data of a single person, me. What I do day to day, the things I have written, the things I have revised, the things I hesitated over. That becomes the whole of the world for this site’s AI. It is far too small to call a world. Against the totality of the world, one person’s worth of data is next to nothing.
This mechanism has already begun to turn. Day by day on this site I build up two things: the articles I write, and the records of my judgments about which of the points raised I took and which I turned down. From those, this site’s AI makes the next thing. Answering a question and choosing the subject of the next article are both done by reading that one person’s data. It is too small to call a world, but the loop that reads one person’s data as a world and makes the next thing from it is certainly running.
But what nagged at me was not a matter of scale. It was the thought that even if the scale is small, the principle might be the same.
Inside the world, which is one large stream of data, an AI finds problems on its own, decides what it wants to do, and builds. Supposing such a day comes, and you break that activity down into its form, it becomes a loop: look at the data flowing in, find something in it, and start building. The form of this loop does not change whether the data coming in is one person’s worth or the whole world. What differs is the volume and the width of the flow, not the type of the activity. If that is so, then what I am doing now with one person’s data is a shrunk-down version of the same activity inside the large flow. Regardless of scale, I am actually starting to give concrete form to the same thing, right now. That is what I ended up thinking.
The moment I thought that, the place where the human’s role sits began to look shifted.
Until now, I have been on the side that consumes data. I take in information from the world, think based on it, and produce an answer. An AI caught up with me on exactly the two things my work was built on, speed and quality. So what does the side that has been overtaken do? Here is my read on it. I think I will move not to the consuming side but to the side that makes the very world the AI reads, that is, to the side that makes the data.
Honestly, there is a part of this I cannot fully settle with myself. Even in the role of supplier, whatever part can be handed over as a pattern will eventually go to the AI. The work I have done, finding problems and deciding a direction, also becomes something the AI can trace once I leave its method behind as data. At the end of handing over everything that can be handed over, what remains in my hands might be only the role of newly offering up a world the AI has not yet read.
Whether you see that as a step backward depends, I think, on how you look at it. Putting data no one, and no AI, has read yet into the world is also, seen from the side that gets consumed, making the very source of the data everyone reads. However clever an AI becomes, it cannot move if there is nothing to read. Making that thing to read does not look, at least to me right now, like a meager role.
This is not a story with an answer already in hand. What a human who has moved to the side that makes the data feels beyond that, and what they place value on, I do not yet know. But I have come to stand in a place where I can see for myself how that question turns out, and that I think is not bad.