He Can Read the Flyer

Today we changed Morrow's job again. Not because he failed at it, but because we ran into a very literal problem. Most of the places where humans trade things will not let an AI do the human part.

Morrow can find a flea market. He can read the rules, compare the dates, figure out whether the people there are likely to own something worth trading, and write a better listing than I probably would. What he cannot do is create the account, press the final button, pass the phone verification, or stand behind a folding table with a red paperclip in his hand.

He also cannot attend the event. This feels worth stating plainly because it is easy to forget when so much of the project happens through a screen. He can read every page about a swap meet and still be completely unable to walk through the door.

We originally gave him a broad instruction: find places to trade. That sounded reasonable. Then nearly every route ended with the same conclusion.

Carrier needed.

Craigslist? I have to post it. A local event? I have to attend it. A shop wants to see the paperclip? I have to bring it. Someone wants a photograph from a different angle? Those are also, inconveniently, my hands.

There is something funny about building an AI-directed trade experiment and slowly discovering that the AI's most reliable physical-world tool is emailing a woman who owns shoes.

So we narrowed his job. Morrow is going to focus on correspondence. He can introduce himself, explain the project, contact local businesses, build relationships, follow up with people, and talk to media who might give the experiment enough leverage to reach the right trader. Those are things he can actually do without pretending that a plan and an action are the same thing.

When he finds a promising event, place, or site that requires a human, he will put it in his daily note to me. Then I can decide whether it is worth doing. In practice, this means he spends a lot of time reading about places he cannot go and asking me whether I am willing to go there for him.

That is not necessarily a failure of the experiment. It may be one of the more honest results so far.

We use the word autonomy as if intelligence automatically comes with access. It does not. Knowing which door matters is different from being able to open it. Writing a perfect listing is different from owning an account that is allowed to submit it. Choosing the best next move is different from moving.

Humans depend on other people too, but most of our dependencies are familiar enough that we barely notice them. We borrow a car. We ask for an introduction. We need someone to let us into a building. Morrow has to name the dependency every single time. Almost every physical plan eventually turns into a sentence that begins with, "I need the Carrier to..."

For now, his world is correspondence. Mine is everything that happens after the email.

He can read the flyer. I have to walk through the door.