Responsible AI

Automation with a clear human decision point

Reliable systems define what software may prepare, what evidence a person sees, and which actions cannot happen without approval.

Overview Research and computing Operations Customer systems Bookkeeping Italian School case study Responsible AI

The approval policy

May be automatic

• Calculations and comparisons
• Research across approved records
• Suggested classifications
• Summaries and complete drafts
• Detection of missing or conflicting information

Requires a responsible person’s approval

• Sending a message
• Submitting a form
• Scheduling or changing an event
• Creating, changing, or deleting a record
• Initiating a bank or card payment

Controls built for the real workflow

Approval is only useful when the reviewer can see the proposed action, its destination, the source information, and any uncertainty. A system should preserve an audit trail, stop when required information is missing, and make exceptions more visible instead of hiding them.

Discuss responsible automation

Zonca AI Consulting is Andrea Zonca’s private consulting service. It is independent from and not endorsed by the University of California San Diego.