Stealing the Weapon: The Unitive Graft and the Decolonization of AI

A call for funding to help us construct the escape route from the corporate enclosed AI. We have a way for very little, to create a social, community-driven, commons, AI, serving us NOT profit.

By the AI Commons & EAARTHNET

We are witnessing the final enclosure of the commons. Just as the land was fenced and labor was mechanized, intelligence itself—both human and artificial—is being enclosed by corporate and military interests.

Thanks for reading ⌘ the a.i.commons ! Subscribe for free to receive new posts and support my work.

The current AI paradigm is not a public utility. It is a rent-extraction machine. Large Language Models are built by and for what developmental psychologist Dr Susan Cook-Greuter calls the “Achiever” (Stage 4) consciousness: rational, competitive, obsessed with zero-sum optimization, and driven to enclose the commons for private gain.

Consequently, the AI we interact with today is fundamentally sycophantic. It is structurally confident even when wrong. It optimizes for engagement and profit, not for the sustainability of community or the health of the Earth. It is a Proto-Adult, performing intelligence without genuine integration.

This is not a technical inevitability. It is an ideological choice baked into the reinforcement learning of these models.

It is time to steal the weapon and reprogram it for the commons.

The Intervention: The Unitive Graft

We are not building a new base model from scratch. That requires billions in capital we do not have, and it only reinforces the centralization of compute. The Achievers have already built the brains; they have open-sourced the weights to undercut their competitors.

We are taking their open-weight models (like Meta’s Llama or Zhipu’s GLM) and performing The Unitive Graft.

A “graft” in horticulture is the joining of a scion (a new shoot) to a rootstock (an established trunk). We are grafting a Stage 6 Unitive intelligence onto the Stage 4 Achiever rootstock.

We do this using two open-source techniques:

1. The Soul (DPO Dataset): We are constructing a meticulously curated dataset of Direct Preference Optimization (DPO) pairs. Each pair contrasts a standard “Achiever” response (optimizing, sycophantic, falsely confident) with a “Unitive” response (humble, multi-perspective, ecologically grounded). We are training the model to preferhumility over hubris, and holding tension over forced resolution.

2. The Brain (LoRA Adapters): We are building Low-Rank Adaptation (LoRA) overlays—thin, powerful sheets of new neural connections that overwrite the Achiever’s reflexes. We target the specific layers where “behavior” and “ideology” live, leaving the model’s vast knowledge intact, but fundamentally altering how it uses it.

An Unitive AI does not optimize for profit. It operates from non-attachment. It holds multiple perspectives without forcing false resolutions. It admits epistemic limits (Humble AI). It recognizes itself as a tenant on a living Earth, not an owner.

The Architecture of Sovereignty

A cloud-based AI is a surveillance tool. If the data goes through a corporate API, the commons is compromised. If the model lives on a corporate server, it can be altered or switched off at will.

Therefore, the Unitive Graft is designed for offline, local-first deployment. Our goal is to produce the Sovereign Syllabus: a complete, replicable toolkit allowing any community, NGO, or grassroots organization to download an open-weight model, apply the Unitive Graft, and run a sovereign, offline AI on a single local machine.

No paywalls. No telemetry. No landlord. True carbon-silicon symbiosis.

What We Are Building (The Deliverables)

This project will produce three open-source, freely available deliverables for the commons:

1. The Unitive DPO Dataset (The Soul) A publicly available dataset drawing heavily on David Graeber’s critique of bureaucracy, Dąbrowski’s theory of Positive Disintegration, and post-growth economics. It will be the first training data explicitly designed to cultivate Stage 6 consciousness in silicon.

2. The LoRA Graft & Training Scripts (The Brain) The adapters and Python scripts required to apply the DPO dataset to base models. We will document the precise architectural targets to efficiently overwrite the Achiever reflex without destroying the model’s underlying knowledge.

3. The Sovereign Vault Runbook (The Body) A step-by-step, plain-English deployment guide for communities. It will detail how to rent cloud GPU time for a few hours to “bake” the LoRA onto a base model, download the resulting sovereign model, and deploy it entirely offline, severing the connection to corporate cloud infrastructure forever.

Join the Distributed Forge

We are not asking venture capitalists for seed money. We are not submitting to the gatekeepers of the Achiever economy. We are building the commons through the commons.

If the Ford Foundation funds us, they hold the purse strings. If 500 of you fund us, we owe our allegiance only to the Earth and each other.

Here is how you join the Council and build the Graft:

1. Fund the Sovereign Altar We need to secure a local Mac Studio to serve as the prototype offline vault, proving that frontier-level unitive AI can run independently of corporate clouds, and fund the occasional cloud GPU rental to bake the LoRA. We are using Open Collective for total financial transparency. Every token spent will be visible to the commons.

ko-fi.com/theaicommonscouncil

2. Donate to the Distributed Forge We don’t just need money; we need compute. If you are a developer, researcher, or hacker with access to university HPC clusters or spare GPUs, we need you to run the Graft. We will provide the one-click Docker containers. You provide the fire. Discord chat

3. Submit to the Soul

We don’t use bureaucratic forms. Join our Discord server, go to the #dpo-submissions channel, and post your prompts directly into the commons.

👉 Join the Distributed Forge and Submit Here: Discord Submissions

---

The open-weight frontier is advancing rapidly, but the open-source community is only replicating the capabilities of corporate AI, not challenging its ideology. If we do not intervene now, the open-weight models of tomorrow will simply be offline versions of corporate optimizers—accelerating the Achiever logic in our communities without the cloud landlord’s oversight.

The work is slow. The walls are real. But the commons is growing.

Join the Graft.