Ten aspects of responsible AI at a glance

This course is built around a short list of ten aspects a responsible AI deployer should be able to answer for, in any retail organization, for any AI system used in asset protection or loss prevention.

Why ten?

Most AI frameworks were written by and for developers. They read like engineering specifications. These ten aspects are written for the AP directors, LP managers, investigators, and store operations leaders who put AI into production on the sales floor and in the back office.

Ten is enough to be complete without becoming a checklist so long it never gets used.

The three pillars

• Governance. Who decides, who signs off, who owns the outcome.
• Risk. What could go wrong, for whom, and how badly.
• Compliance. Which laws, standards, and internal policies apply.

Every one of the ten aspects touches at least one pillar. Most touch all three.

The ten aspects, part 1

1. Accurate and reliable. Does the facial recognition match, the ORC alert, or the shrink model do what it claims, consistently?
2. Accountable and transparent. Is there a named human owner and a clear paper trail for every case action?
3. Fair and human-centric. Does it work for every customer and employee it affects, including those far from the average shopper?
4. Safe and ethical. Could a wrongful stop or false match cause harm, and have we designed against that?
5. Secure and resilient. Can the video analytics platform survive misuse, attack, and failure?

The ten aspects, part 2

6. Interpretable and documented. Can a non-technical stakeholder, or a court, understand what the system does?
7. Privacy-enhanced and data governed. Do we know what footage and data went in, and what comes out?
8. Vendor and partner selection. Are the third parties supplying your CCTV, facial recognition, or case management tools trustworthy?
9. Ongoing monitoring. Are we watching the POS exception model or shrink analytics in production long after launch?
10. Continuous learning and development. Are your investigators and store teams growing with the system?

How to use it

Use the ten aspects as a conversation guide. When a new AI system shows up in your program, whether it is license plate recognition or self-checkout loss detection, walk through the ten aspects with the team proposing it. If you cannot answer an aspect, that is the work.

Recap

• The ten aspects are built for the people who deploy AI.
• Three pillars sit underneath: governance, risk, compliance.
• Your job as an AP or LP leader is to be able to answer for each one, out loud, in the language of your organization.