How the Knowledge Within Our Communities Can Become Working Structures for Artificial Intelligence
Kevin Myles is a strategist, advocate, and creator of the Civic Codex. It is an organizing system built from years of non-profit leadership, civic engagement, and systems-change work across diverse communities and institutions.
When we ask AI to help us do our work, we usually start by explaining the task. But there's another conversation we need to have first: How should it think about the work we're asking it to do? A community organizer, a teacher, and a social worker might each ask an AI system to recommend a response to a family's difficulty. Even with the same facts, they would likely need to ask different questions. Their responsibilities differ. So do the relationships they must protect, the authority they can exercise, and the standards by which a responsible response should be judged.
But typically, most would simply respond with increasingly longer, more detailed, and more complex prompts. That realization raises an interesting question: How do we bring the knowledge, philosophy, and ways of thinking developed within a field or a community into the AI systems we build?
For me, that question became concrete through MPoWER, the Philosophy of Organizing I developed and used as the governing foundation for the Civic Codex. Working to translate MPoWER into a system that could support real organizing decisions led me to the larger idea of Community-Controlled AI. Communities need ways to define the reasoning behind a system's recommendations, along with meaningful authority over how those recommendations are used.
The work begins before the answer
When we ask AI for a strategy, an assessment, or a recommendation, we're asking it to make judgments about relevance. Which facts belong in the analysis? Which sources deserve scrutiny? What counts as a credible explanation? What would make one proposed outcome better than another?
In organizing, those judgments can determine whether a campaign pursues a public acknowledgment or an enforceable change in policy. In education, they can shape whether a student's behavior is examined through punishment, developmental needs, classroom conditions, or some combination of them. In community development, they can affect whether rising property values are treated as sufficient evidence of progress or examined alongside displacement and residents' ability to remain.
The same information can support very different recommendations depending on the questions and standards brought to it.
We can't settle those choices just by giving the system more information. We also have to explain how that information should be considered. That's where what I'm calling a reasoning matrix comes in.
What is a reasoning matrix?
By a reasoning matrix, I mean a clearly defined structure of principles, questions, evidence standards, and decision rules that guides how an AI system approaches a particular kind of work.
It identifies what the work is for, whose knowledge is essential, what must be examined, which assumptions require testing, and what conditions a recommendation must satisfy. It also establishes boundaries: decisions that require human authority, uncertainties that must remain visible, and circumstances in which the system should seek further information rather than offer a confident conclusion.
A useful matrix turns our values into questions the system has to ask. If a practice values family autonomy, what must the system ask before recommending an intervention? If a community prioritizes housing stability, how should a development proposal be evaluated? If an organization commits to accountability, what evidence would demonstrate that a promised remedy actually occurred?
A mission statement can announce those commitments. A reasoning matrix makes them usable throughout the work.
It needn't be a literal numerical matrix or a change to an AI model's underlying training. But the important question is whether it consistently changes how the system examines a problem and develops its recommendations.
MPoWER as the starting point
MPoWER is the starting point for the approach I'm describing here. Its roots are in Black liberation and decades of movement practice, including lessons from my own organizing and from the mentors who shaped me. It sets out how we understand power, weigh evidence, examine institutional behavior, preserve community leadership, and pursue structural change.
Building with MPoWER meant taking those commitments seriously enough to make them govern the system's work. I had to ask: What should this philosophy require the AI to examine? What should it challenge? What would make a recommendation consistent with the way we believe organizing should be done?
For example, MPoWER holds that people closest to injustice possess essential knowledge about it. Implementing that principle means asking whose experiences helped define the problem, whether affected people shaped the demands, and who has authority to accept the proposed outcome.
Its commitment to accountability requires analysis of actual decision-making authority, available alternatives, and the consequences of institutional choices. Its commitment to evidence requires testing consequential claims, including claims made by institutions and by organizers themselves.
Its understanding of victory directs attention toward changed rules, budgets, practices, material conditions, and enforceable commitments. That affects what the system should recognize as progress and what further questions it should ask.
The principle is AI governed by an Organizing Philosophy, not Organizing governed by AI.
MPoWER gave me the foundation for developing this approach and a concrete way to put it into practice. From that work came a broader possibility: other fields and communities could articulate their own reasoning matrices, grounded in the knowledge, responsibilities, and philosophies that guide them. The lesson travels, even when the governing philosophy changes.
Importing a philosophy means translating it
Of course, writing down a philosophy and getting an AI system to work from it are two different challenges.
We have to start by putting words to the judgment we already use. Experienced practitioners often know when a proposal is weak without having written down every reason. They recognize a missing perspective, an unsupported assumption, or a familiar institutional response. Building a reasoning matrix requires drawing out that knowledge and identifying the questions behind it.
Then we have to decide what each principle actually asks the system to do. “Center affected people” might require the system to identify whose perspective is missing and flag decisions that haven't been reviewed by those bearing the consequences. “Use credible evidence” requires standards for source quality, conflicting accounts, uncertainty, and the limits of available information.
Those requirements have to appear at the points where they matter. Participation belongs in problem definition and demand development. Evidence checks belong before consequential conclusions. Human authorization belongs before actions that commit people or expose them to risk.
The matrix also needs a way to handle conflict. A desire to respond quickly may collide with the need for deliberation. Transparency may conflict with protecting sensitive information. Shared purpose may coexist with disagreement about tactics. The system should expose those tensions and clarify the choice people face.
We haven't implemented a philosophy just because we've put it in a reference folder. It has to show up in the questions the system asks, the recommendations it considers, and the conditions it checks before the work moves forward.
Testing whether the philosophy survived
The real test comes when the system encounters a situation where those commitments matter.
If a system is given an official account that conflicts with residents' testimony, does it examine both with appropriate rigor? If a proposed solution improves an average while worsening conditions for a smaller group, does it make that tradeoff visible? If information is incomplete, does it preserve the uncertainty?
That's where we find out whether the philosophy made it into the system's work. It's easy to repeat the right language. What matters is whether the recommendations hold up when the situation gets complicated.
Review should include practitioners and affected people. They may identify different failures: a specialist might spot a flawed inference, while a resident recognizes that the proposed process excludes the people it claims to serve. Both forms of knowledge matter to improving the matrix.
A documented philosophy gives that review a reference point. People can ask which principle guided a recommendation, where the evidence supported it, and whether the process honored the authority it promised to preserve.
Who controls the matrix?
And here's where we have to ask what we mean by community control.
A system can be closely tailored to a field while still concentrating authority in its designer. Writing down one person's philosophy makes its origins clearer, but community control requires meaningful ways for people to examine, challenge, and influence its application.
Who participates in defining the principles? Who can question the evidence standards? How are disagreements recorded? Who approves changes? Can people decline a recommendation without losing their standing in the process?
Communities aren't uniform. Residents, organizational leaders, practitioners, and funders may bring different priorities. A credible approach needs clear decision rights and ways to handle those differences, especially when people with the most institutional authority aren't the people bearing the greatest risk.
We have to be willing to question the matrix itself, too. Experience may reveal that a principle was poorly translated, a standard excluded relevant knowledge, or a workflow quietly displaced community judgment.
Community control has to extend to that correction process.
A different starting point for AI
I believe this gives practitioners and communities a meaningful place in building AI. We bring accumulated knowledge about how to interpret problems, exercise judgment, recognize harm, and evaluate change. Making that knowledge explicit can help us shape the systems entering our work.
Once we've made that reasoning explicit, we have something people can examine, put to work, test, and improve. It gives us a basis for asking whether the technology is serving the purpose we established for it.
For me, Community-Controlled AI begins there: with people defining the principles that should guide the reasoning, building those principles into practice, and retaining the authority to decide whether the result serves their community.
Visit here to explore Community-Controlled AI and what it could mean for your community.
Read more from Kevin Myles
Kevin Myles, Principal Designer: Movement & Strategy Systems
Kevin Myles is a strategist, advocate, and systems architect whose work connects civic power, organizational strategy, and community transformation. He is the creator of the Civic Codex, an organizing operating system built from years of frontline non-profit leadership, campaign development, coalition strategy, and civic engagement work across diverse communities and institutions. Designed as infrastructure for complex social change, The Civic Codex helps communities turn fragmented challenges into coordinated systems for action, accountability, and long-term impact.
