MEASURE 2.5

Objective

The AI system to be deployed is demonstrated to be valid and reliable. Limitations of the generalizability beyond the conditions under which the technology was developed are documented.

Incidents linked to this control

From automatic embedding match (not a formal assessment).

High False Positive Rate by SWP's Facial Recognition Use at Champion's League Final

June 3, 2017

Poor Performance of Tesla Factory Robots

October 8, 2016

Amazon’s Experimental Hiring Tool Allegedly Displayed Gender Bias in Candidate Rankings

August 10, 2016

Unreliable ShotSpotter Audio Convicted Black Rochester Man of Shooting Police

April 1, 2016

Coffee Meets Bagel’s Algorithm Reported by Users Disproportionately Showing Them Matches of Their Own Ethnicities Despite Selecting “No Preference”

July 30, 2013

Police Departments Reported ShotSpotter as Unreliable and Wasteful

October 9, 2012

Unveiling Machine Bias: A Case Study by ProPublica

Can AI perpetuate racial bias? An examination of the problem

Unintended Implementation: Google's AI Model

Evaluating Google's Anti-Internet Troll AI Platform: A Test of Resilience

Google's Misidentification: A Case Study in Responsible AI

Google Addresses Algorithm Mistakenly Identifying Humans as Gorillas