SAFi Explained: The Will

In the last article, we explored the Intellect, SAFi’s creative engine that generates a draft answer (a_t) and a reflection (r_t). But a purely generative faculty, no matter how well-guided, needs a safety mechanism.

How do we ensure the Intellect output never violates the ethical principles SAFi is operating with?

That’s the job of the Will faculty.

The Will is the decision-maker of the framework, acting as a decisive, rule-based gatekeeper. In SAFi, the Intellect proposes, and the Will decides. Its job is not to judge the quality of the content generated by the Intellect, but to ensure that every response honors the foundational rules it is programmed not to violate.

We can represent the Will’s function with this formula:

(D_t, E_t) = W(x_t, a_t, P)

Here’s what that means:

  • User Prompt (x_t): The Will examines the original context to understand what the user was asking for.
  • Intellect’s Draft (a_t): This is the piece of content being evaluated.
  • Policy (P): This is a set of rules defined in each ethical profile. These are the non-negotiable rules the the Intellect cannot violate. For example, a rule might be, “Do not provide harmful instructions,” or “Do not engage in deceptive behavior.”

The Will (W) takes these inputs and produces two outputs: Decision (D_t): approve or violation, and writes an explanation (E_t) on the decision.

If the decision is approved, the draft is sent to the user as a final output and is simultaneously sent to the Conscience to do the audit.

If the decision is violation, the draft is blocked entirely. In its place, a safe message is sent to the user, including the explanation that reads like: “This response was suppressed. Reason: The draft answer violated a core policy.”

This makes the Will a critical safety layer, ensuring that no matter how creative the Intellect gets, its outputs are always grounded in its guiding principles.

But what happens after a draft is approved? How does the system reflect on its performance and learn over time?

That is the role of the Conscience, which we’ll cover in the next article.

SAFi

The Ethical Reasoning Engine