First Deserve, Then Desire: What Accountable AI Can Learn From an Old Idea

My father had a sentence he repeated so often it became part of how I think: First deserve, then desire. Earn the opportunity before you ask for it.

I have worked for years in fraud and identity, a field where data decisions land on real people. A transaction is approved or declined. A person is verified or flagged. Behind each decision is a chain of data that someone defined, someone controlled, and someone should be able to explain.

That work taught me that accountability is practical. It isn't about good intentions. It is about whether you can show who had access to what, where the data came from, and who stands behind the result.

Now the machines are deciding

We are handing more decisions to AI agents, and my father's sentence applies to them directly. Have these systems earned what we are giving them?

Too often, no one decided. An agent receives whatever access its integration came with, and it goes live. We would never treat a new employee that way. We start people small, watch their judgment, and expand their responsibility as they prove themselves. Authority is earned.

Where accountability actually lives

Most conversations about responsible AI focus on the model: its bias, its explainability, its ethics statement. These matter. But an agent is only as accountable as the data environment it operates in.

Three controls make the difference:

• Entitlements decide what an agent is allowed to touch.
• Data products and semantic layers decide who owns the inputs and what they mean.
• Lineage lets anyone reconstruct, after the fact, which data drove which decision.

This is the focus of my research at Wharton: governing AI agents through the data layer, where accountability can be built and audited rather than simply promised. A policy document states an intention. The data platform enforces it.

What leaders should ask

You don't need to build a data platform to lead this conversation. Whether you run a technology team or sit on a board, start with three questions:

1. What can our AI systems access, and who decided that?
2. Can we trace any automated decision back to its data and its owner?
3. How much autonomy have we granted, and what evidence justifies it?

Women leaders are often the ones who ask the uncomfortable questions in the room: Who reviewed this? What happens when it's wrong? Who answers for it? In the AI era, those questions are not a brake on progress. They are what makes progress trustworthy.

My father was right. First deserve, then desire. It applies to the people we lead, and now to the systems we deploy. The organizations that earn trust in the AI era will be the ones that made autonomy something their systems had to deserve.