The Decision-System Map

AI and Governance · From the essay “When AI Automates a Rule No One Owns”

Responsible AI governance begins with the decision process, not the model. Map the eight elements of the decision system, know the five ways an unowned rule takes hold, and answer these questions before approving an AI-enabled decision system.

The elements

• Decision point
• Applicable rule
• Required evidence
• Authority
• Discretion
• Exception path
• Consequence
• Monitoring

How an unowned rule takes hold

• Policy translated into code
• Historic operating practice
• Vendor configuration
• Data limitations
• Incremental adjustments

Questions to answer

• What decision is the system supporting?
• Which rules, models, thresholds, and assumptions shape the result?
• Who owns each consequential rule?
• What happens when required data is missing or unreliable?
• Where is human judgement required?
• Can staff override the system? Under what authority?
• What exception and appeal routes exist?
• Which institutional outcomes will be monitored?
• Who can change the system after launch?
• How will those changes be approved and documented?
• What evidence would justify pausing or retiring the system?