Unmasking AI by Joy Buolamwini

Author:Joy Buolamwini [Buolamwini, Joy]
Language: eng
Format: epub
Publisher: Random House Publishing Group
Published: 2023-10-30T00:00:00+00:00

Skip Notes

*1 AJL’s subsequent study on the original Pilot Parliaments Benchmark dataset recorded IBM performing at 88.5 percent. The difference is based on the setting of the threshold cutoff. When an AI system is trained to classify a face, the system can be configured to give a confidence score regarding the label produced. A threshold number can be set to determine when the classification should be accepted. So if a system says it is .8 confident a face image depicts a woman and the threshold is set to .7, the classification would be accepted because .8 is greater than .7. If the threshold was set to .9, the classification would be rejected because .8 is less than .9. Changing the threshold can change the recorded accuracy of a system. Depending on the product, some companies share a confidence score and others do not. Because the other companies I tested did not share confidence numbers, the AJL study used the label that was provided, “male” or “female,” without regard for the confidence score. In their self-reported study IBM used a .99 threshold.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.