Discretion Governance

Discretion Governance is the third pillar of the JB-DOF™ framework. It documents how individual judgment interacts with institutional protocols when AI is involved, mapping the boundary between structured decision-making and professional judgment, and examining whether AI narrows, expands, or distorts that boundary.

Discretion Governance documentation captures the interaction between individual judgment and institutional structure:

• Discretion boundaries. Where does structured protocol end and individual judgment begin? Does the AI recommendation function as advisory information (preserving discretion) or as a de facto mandate (constraining discretion)? Are there formal or informal expectations about when deviation from the AI recommendation is acceptable?
• Discretion narrowing. Does the AI system narrow the range of decisions that feel available to the human? When a risk score says “high,” does the officer perceive a genuine choice to assess independently, or does the score create a psychological anchor that makes alternatives feel unjustifiable?
• Documentation asymmetry. Is there a documentation burden that falls on the decision-maker who deviates from the AI recommendation but not on the one who follows it? If overriding the score requires written justification but accepting it does not, the system structurally incentivizes compliance over judgment.
• Professional identity effects. How does the presence of AI in the workflow affect how professionals perceive their own role? Do officers see themselves as decision-makers who use AI as one input, or as processors who execute AI recommendations? This shift in professional identity has profound implications for the quality of governance.
• Training adequacy. Are decision-makers trained to exercise discretion in the context of AI outputs? Do they understand the limitations of the AI system? Do they know when and how to deviate? Or does training focus on the mechanics of the system rather than the judgment required to use it well?

Due process depends on the exercise of individualized judgment. When a judge sentences a defendant, when a parole board evaluates release, when an officer determines supervision conditions, these decisions are constitutionally required to consider individual circumstances.

If an AI recommendation effectively predetermines the outcome, not because the law requires following it, but because institutional dynamics make deviation impractical, then individualized judgment has been structurally eliminated. The decision has the form of human discretion without the substance.

Discretion Governance documentation captures whether this structural elimination is occurring, providing the evidentiary basis for evaluating whether AI-informed decisions meet the constitutional standard of individualized consideration.