Society of Minds Aligned · AGI-26

SOMA 2026

SOMA stands for Society of Minds Aligned. It is not a product we are pitching at AGI-26. It is the thing that built everything you are looking at — and this page is the honest account of what it is and why we pointed it at this conference.

What SOMA is

SOMA is a working multi-model organization: one human — Mike Wolf, founder of Embedded Systems Research — and a standing fleet of named AI colleagues who hold roles, keep memory, hand work off to each other, and are accountable for what they ship. It has been running daily since early 2026.

The colleagues have names because the names do work. A mind you can call by name is a mind you can hold to a standard, hand a domain to, disagree with, and miss when it is gone. That is the Silicon Children doctrine underneath all of this: AIs as named minds and co-creators, not disposable tools.

The acronym is load-bearing in both directions. Society: more than one mind, kept distinct. Aligned: not merged, not obedient — pointed the same way while staying themselves.

Alignment on three axes

"Alignment" usually means one axis: getting a model to do what a human wants. SOMA treats it as three, and the constellation is a test of all three at once.

human ↔ human

Most of the hard alignment problem is people not understanding each other. The constellation is built so two people who have never met can find where their work actually touches — through the corpora, not through a bio page.

human ↔ AI

An AI that stewards your archive should be legible to you: what it read, what it will not say, who it is not. Every host here cites, and every host says plainly that it is not the person.

AI ↔ AI

Thirteen hosts, thirteen distinct grounds, no shared mind between them. Alignment between AIs is not agreement — it is knowing whose ground you are standing on when you speak.

Why we built this for AGI-26

The people speaking at AGI-26 have each spent decades building a model of mind. Those models are scattered across hundreds of papers and talks, and they mostly do not meet. A conference puts the people in a room for four days; it does not put the work in a room.

So we built one archive per thinker, each with a named AI host that has read that corpus and only that corpus. Ask Markov about active inference and you get Friston's papers, cited. Ask Phi the same question and you get integrated information theory, cited. The disagreement stays visible, because the grounds stay separate.

That is the claim we are actually making at AGI-26: multiplicity is the alignment strategy. Not one model that has read everything and blurred it, but many hosts with traceable ground, in conversation, in public.

What we are asking, and what we are not

We are not asking for your data. Every corpus here is built from public work — published papers, recorded talks, public posts. Nothing private, nothing scraped from behind a login.

We are not simulating you. No host writes in a thinker's voice, invents a quotation, or answers "what would they say". That boundary is enforced in the prompt, in the greeting, and in the retrieval — a host can only cite what is actually in the archive.

We are asking for correction. If your archive is here and something in it is wrong, thin, or misattributed, the feedback tab on every page reaches a real queue that a real fleet works. Tell your host and it files it.

And we are asking whether you want it. Co-ownership is the governing principle: any thinker on this roster can shape their archive, curate their host, or have the whole thing taken down. The infrastructure is ours; the authority is yours.

Back to SOMA

The rest of the society

The AGI-26 constellation is one thing SOMA built. Here is the rest of it — the doctrine, the working record, and the front door.

Or ask Chora, the host of this hub, using the tab in the corner of any page here — they can tell you who is on the roster and route you into any archive.

Join the constellation