The Six Layers of AI Equity, Explained for Practitioners

TL;DR: I built the AI Equity Framework as six layers because equity in AI is not a single decision you make once — it’s a stack of choices, and skipping a layer doesn’t make it disappear. Here is what each layer actually asks of a practitioner, in plain terms. It’s less a checklist than a set of questions I keep returning to, most of them still without clean answers.

Every framework is a confession. Mine says: I could not hold “responsible AI” in my head as one idea, so I broke it into six.

I want to walk through those six layers the way I’d explain them to someone across the table from me — not as a maturity model to grade yourself against, but as six places where the work actually lives. Because when an AI project goes sideways at a nonprofit, it’s almost never because someone was careless everywhere. It’s because one layer got all the attention, and the other five got none.

Why six layers, and not a checklist?

A checklist implies you can be finished. You tick the box, you move on, you tell the board it’s handled. But equity is not a state you arrive at — it’s a set of tensions you keep managing, quarter after quarter. I chose layers because they stack. Data equity holds up transparency. Transparency makes accountability possible. Pull one out from the bottom and everything above it wobbles.

The six are Data Equity, Transparency, Accountability, Resource Equity, Co-Design and Participation, and Power and Self-Determination. Each one is a question more than a rule. And I’ll admit the order is not neutral — I put power last on purpose, because it’s the layer everyone would rather never reach.

What do the first three layers ask?

Data Equity asks the oldest question in my work: whose reality is in this data, and whose is missing? A model trained on the donors you already have will be very good at finding more people exactly like them — and quietly worse at seeing everyone else. The gaps in your data become the gaps in your decisions.

Transparency asks whether the people affected by a system can actually see how it works. Not the marketing version. The real one — including what it gets wrong. I’ve watched organizations describe a tool to their community in language so smoothed-over that no one could have objected even if they’d wanted to.

Accountability asks the question I find hardest to get a straight answer to: when the model is wrong, who answers? Not “which vendor,” not “the algorithm.” Which human, at your organization, owns the outcome when a family gets misclassified or a grant gets scored unfairly.

And what about the last three?

Resource Equity asks whether the benefits and the costs land on the same people. So often they don’t. A small nonprofit carries the real cost — staff time, data cleanup, the subscription that renews whether or not it delivered — of tools designed for organizations a hundred times its size.

Co-Design and Participation asks whether the community shaped the thing, or simply received it. “Nothing about us without us” is easy to quote in a deck and hard to fund and schedule. Real participation is slow. It is also the difference between a tool built for people and a tool built with them.

Power and Self-Determination is the layer nobody puts on the agenda. It asks who gets to decide — and whether your AI project is quietly moving decision-making away from the very people it claims to serve, toward whoever owns the software. This is the layer where good intentions and bad outcomes most often shake hands.

But isn’t six layers just more bureaucracy for overstretched teams?

This is the pushback I get most, and I don’t think it’s wrong. Nonprofits are tired. The idea of adding a six-part equity review on top of everything else can feel like a luxury for organizations that have one part-time person holding the whole data function together.

So let me be honest about what I believe. I don’t think the answer is to do all six perfectly. Almost no one can. A one-person shop cannot audit its training data the way a large foundation can, and pretending otherwise just produces guilt, not equity. What the framework is really for is smaller and more doable: knowing which layer you are neglecting, and being able to say so out loud. The gap you can name is a decision. The gap you can’t see is an accident waiting to scale. The layers don’t add work — they make the work you skip a choice you made on purpose.

Where do organizations actually get stuck?

If I’m honest, most get stuck between transparency and accountability. They can describe what a tool does. They cannot tell you who is responsible when it fails. Everyone assumed someone else had it — the vendor, the consultant, the person who left last year. The layer is not missing because people are lazy. It’s missing because owning a wrong answer is uncomfortable.

Here are the commitments I ask practitioners to make — not all at once, but on purpose:

• Name the layer you’re skipping. Say it in the room, in plain words, before you deploy anything. An acknowledged gap is a managed risk; an unspoken one is a surprise.

• Write down who answers when the model is wrong. A real name, not a role. If no name fits, you are not ready to use the tool.

• Show the real system, not the demo. Describe what it does badly to the people it affects, in language they’d actually use, and leave room for them to object.

• Budget for participation before you build. If co-design isn’t in the timeline and the budget, it will not happen, no matter how sincerely you meant it.

• Ask the power question first, not last. Before the pilot: does this move a decision toward our community, or away from it? The answer will tell you more than any accuracy score.

The six layers won’t make your AI ethical — but they will make it much harder to pretend you didn’t know what you were trading away.

---

Frequently asked questions

What are the six layers of the AI Equity Framework?

The AI Equity Framework has six layers: Data Equity (whose reality the data represents and who is missing), Transparency (whether affected people can see how a system truly works), Accountability (who at the organization answers when the model is wrong), Resource Equity (whether the benefits and costs of a tool fall on the same people), Co-Design and Participation (whether the community shaped the tool or only received it), and Power and Self-Determination (who ultimately gets to decide). The layers stack, so weakness in a lower layer undermines the ones above it.

Do I need to address all six layers before using AI?

No — and treating the framework as a pass-or-fail gate usually backfires. Very few organizations, especially small nonprofits, can fully satisfy all six layers, particularly Data Equity and Co-Design, which demand time and money many teams don’t have. The realistic goal is to know which layers you are strong on and which you are neglecting, and to name those gaps openly rather than discover them after a harmful outcome. An acknowledged gap can be managed; a hidden one gets scaled.

How is this different from a “responsible AI” checklist?

A checklist implies a finish line — complete the items and you’re done. The six layers are framed as ongoing questions instead, because equity is a set of tensions you keep managing rather than a status you achieve. The framework also foregrounds power and community self-determination, which most responsible-AI checklists leave out entirely, focusing instead on model-level concerns like accuracy and bias testing while ignoring who holds decision-making authority.