Minutes Exhibit A
R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A A Facial Recognition R155 G240 B11 A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA Overview R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA Enrollment photos R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A 23 7 R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA 5 2 4 R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A Joy Buolamwini, MIT Dr. Timnit Gebru, Google R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A Woman Woman Man Man R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA Dark Skin Light Skin Dark Skin Light Skin AA R0 G0 B0 2018 MS R48 G229 B208 Rich Black Face API 20.8% 1.7% 6.0% 0.0% Error Rate R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A 2019 MS Face API 1.5% 0.3% 0.3% 0.0% Error Rate R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA Microsoft IBM Face++ R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 Error rate: 20.8% 34.7% 34.5% Buolamwini & Gebru, 2018 A A Error rate: 1.52% 16.97% 4.1% Buolamwini & Raji, 2019 R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 Change (ppts) -19.28 -17.73 -30.4 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 "By highlighting the issue of classification performance disparities and amplifying public A A awareness, the study was able to motivate companies to prioritize the issue and yield significant improvements within 7 months." Raji & Buolamwini, 2019 R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White AA A AA Facial Recognition Accracy of Humans Facial Recognition Accuracy of Humans + Algos R230 G230 B230 R106 G75 B22 R255 G185 B0 R254 G240 B0 AUC (.5 is "random" and 1 is perfect) AUC (.5 is random and 1 is perfect) Facial examiners 0.93 One examiner + best algorithm 1 AA A A A Facial reviweers 0.87 2 facial examiners 0.96 R210 G210 B210 R5 G75 B22 R16 G124 B16 R155 G240 B11 Superrecognizers 0.83 Best algorithm (of 4) 0.95 Fingerprint examiners 0.76 One examiner + 2nd best algorithm 0.95 AA AA AA R0 G0 B0 Students 0.68 1 facial examiner 0.93 R115 G115 B115 R39 G75 B71 R0 G133 B117 R48 G229 B208 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A From: Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms, Phillips et al., 2018 R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A R254 G240 B0 AA R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A R155 G240 B11 A A R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA R48 G229 B208 AA R0 G0 B0 Rich Black R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A R242 G242 B242 R107 G41 B41 AA R216 G59 B1 AA R255 G147 B73 AA R255 G255 B255 White R230 G230 B230 R106 G75 B22 AA R255 G185 B0 A AA 1. Fairness. We will work to develop and deploy facial recognition technology in a manner that R254 G240 B0 strives to treat all people fairly. R210 G210 B210 R5 G75 B22 AA R16 G124 B16 A 2. Transparency. We will document and clearly communicate the capabilities and limitations of facial recognition technology. A A 3. Accountability. We will encourage and help our customers to deploy facial recognition R155 G240 B11 technology in a manner that ensures an appropriate level of human control for uses that may R115 G115 B115 R39 G75 B71 AA R0 G133 B117 AA affect people in consequential ways. R48 G229 B208 AA 4. Non-discrimination. We will prohibit in our terms of service the use of facial recognition R0 G0 B0 technology to engage in unlawful discrimination. Rich Black 5. Notice and consent. We will encourage private sector customers to provide notice and secure R80 G80 B80 R36 G58 B94 AA R0 G120 B212 AA R80 G230 B255 A A consent for the deployment of facial recognition technology. 6. Lawful surveillance. We will advocate for safeguards for people's democratic freedoms in law enforcement surveillance scenarios and will not deploy facial recognition technology in scenarios that we believe will put these freedoms at risk. R47 G47 B47 R59 G46 B88 AA R134 G97 B197 AA R213 G157 B255 A A
Limitations of Translatable Documents
PDF files are created with text and images are placed at an exact position on a page of a fixed size.
Web pages are fluid in nature, and the exact positioning of PDF text creates presentation problems.
PDFs that are full page graphics, or scanned pages are generally unable to be made accessible, In these cases, viewing whatever plain text could be extracted is the only alternative.