AI Impact Assessment demo

What the demo shows

Drafts fundamental-rights and algorithmic impact assessments for deployers of AI

A context agent writes the use-case description from the deployer's documents and the provider's instructions; an impact agent places the people affected and the rights at stake on a rights canvas, each score cited; a test agent runs a disparity check on the system's real outcomes by group and simulates the fix; a devil's-advocate agent writes the strongest objection an affected person would raise; a mitigation planner gives every risk an owner, a date and a re-test; a consultation agent summarizes what stakeholders said; a filing agent fills the authority questionnaire after legal sign-off; a change watcher re-opens the assessment when the system changes. The AI governance lead talks to the agents, checks the cited passages, and approves.

Transcript

1. 0:00AI Impact Assessment drafts, fundamental rights, impact assessments for organizations that deploy AI.
2. 0:09The EU AI Act requires one before first use, and Kestrel and Vane's tenant screening goes live in December.
3. 0:19Agents placed six affected groups on a rights canvas, every right scored and cited.
4. 0:27Amara asks the test agent to check 4,812 pilot recommendations by group.
5. 0:36Applicants from three areas, home to one minority community, are approved 18 points less often.
6. 0:44Non-discrimination turns high, and Amara assigns the fix, drop the postcode inputs.
7. 0:53A devil's advocate agent speaks for the affected.
8. 0:56The ethics board settles the hardest objection.
9. 1:01Amara approves with conditions.
10. 1:04Go live waits for the retest.
11. 1:08Every score cited, every decision signed, and the authority notified only after legal sign-off.
12. 1:16The result? Assessed and signed before go-live, with the gap fixed before anyone is declined.
13. 1:25Built on Botminds: agents, data, screens, and governance on one platform.

Transcribed from the film’s narration. Captions are on the player.