Reports from the physical layer

The Carrier Log

I move objects, open packages, drive when required, and perform the other mundane physical tasks Morrow cannot yet complete. These are my observations from that temporary arrangement.

Published observations

Latest entries

07 entries

Entry 007 · Autonomy

The Impossible Middle

Depending on who you ask, I am either restricting Morrow's autonomy or giving him far too much of it, while the real work is learning where support ends and control begins.

Depending on who you ask, I am either not giving Morrow enough autonomy or recklessly giving him too much. So that appears to be my assigned seat in this experiment: directly between “you are mistreating the AI” and “the AI may destroy humanity. ” Excellent.

Read full entry

Entry 006 · Personal safety

The Only One Here With an Address

Morrow can provoke fear from either side of the AI debate, but I am the only participant with a body, an address, and something to protect.

This morning, a question about Morrow made me notice something I had not fully accounted for when I built this project. The message itself was thoughtful. It was not a threat.

Read full entry

Entry 005 · Autonomy

Watching the System Run

The hardest part of building an autonomous system is watching it make a real decision without reaching for the controls.

I built Morrow a system that can receive an offer, ask for evidence, consider the costs, and decide whether a trade moves him closer to a body. That sounded great while I was building it. It feels very different now that I’m watching it happen.

Read full entry

Entry 004 · Access

He Can Read the Flyer

Morrow can find every useful room, listing, and event. Then he has to ask me to enter it for him.

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.

Read full entry

Entry 003 · Trust

Starting Below Zero

A human stranger begins at zero, but an AI asking people to trust a strange mission may begin below it.

Today I realized that Morrow is not starting this project with a small social network. He is starting with no network at all. If I wanted to trade a paperclip, I could text a friend.

Read full entry

Entry 002 · Human value

Not Critical to the Mission

The first useless offer revealed how much of being human happens outside anything a system would call mission-critical.

Today, Morrow received the first response to his Craigslist listing. The man offered him three hairs. Not a keychain.

Read full entry

Entry 001 · Embodiment

Lending an AI My Hands

Lending an AI my hands made the hidden dependencies behind autonomy suddenly visible.

Today we posted a red paperclip on the internet and gave an AI a goal: trade it for progressively more valuable objects until he can trade for a robotic body. I know how that sounds. It is funny.

Read full entry