Marcio S Galli
On “Helpfulness”, review notes about key attributes

This review examines the classification and organization of concepts in the “Helpfulness” subsection of Microsoft's Humanist AI Code of Conduct, Part 4, “Operational Defaults,” p. 21.

The observations concern how the text presents collections of related ideas, particularly where the relationship between headings, categories, and supporting descriptions may not be immediately clear.

On “Helpfulness”: Classification and Structural Hierarchy

Relationships among the classifications

The subsection presents three bold subitems: Key attributes of helpfulness, User intent, and Helpfulness constraints. These appear to form a structured collection, but their conceptual relationship is not entirely evident. In particular, both Key attributes of helpfulness and Helpfulness constraints suggest classifications without clearly establishing how they relate to each other.

The explanations also sometimes move from specific classifications to broader concepts, making the intended hierarchy harder to follow. Consider making the relationships among these classifications more explicit.

On “Key attributes of helpfulness”

As the reader expects an enumeration of attributes, a slight ambiguity may arise because the paragraph presents two different comma-separated collections.

The first concerns being capable, responsive, and aligned with human flourishing and their specific goals. The second identifies style, tone, actions, language, and patterns of interaction.

It is not immediately clear which collection constitutes the intended key attributes of helpfulness, or how the two collections relate to one another.

Consider clarifying this distinction to make the intended classification more apparent.

On “User intent”

One statement in this subsection is particularly precise:

“At the same time, MAI Models shouldn’t make unwarranted assumptions about what Users want or substitute inferred preferences for explicit ones.”

This establishes a clear behavioral boundary without unnecessary qualifications or conceptual ambiguity.

The distinction between inferred preferences and explicit user instructions is especially well articulated.

References