US 12,013,874 B2
Bias detection
Brianne Boldrin, Cary, NC (US); Eliza Salkeld, Raleigh, NC (US); and Rebecca Rose James, Cary, NC (US)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Dec. 14, 2020, as Appl. No. 17/121,458.
Prior Publication US 2022/0188328 A1, Jun. 16, 2022
Int. Cl. G06F 16/28 (2019.01); G06F 16/27 (2019.01); G06N 5/02 (2023.01)
CPC G06F 16/27 (2019.01) [G06F 16/285 (2019.01); G06N 5/02 (2013.01)] 22 Claims
OG exemplary drawing
 
1. A computer-implemented method, said method comprising:
identifying a plurality of data fields of interest;
receiving data for each of said plurality of data fields of interest, wherein said data is received from a plurality of data sources, wherein said receiving data comprises:
accepting user input data from a second user;
harvesting social media for social media data, wherein said harvesting social media for social media data comprises analyzing one or more interactions with posts;
providing said second user with a prompt; and
analyzing user results from said prompt;
computing a plurality of bias scores for said plurality of data fields of interest based on said data, wherein each bias score of said plurality of bias scores is a quantitative assessment of a bias in a corresponding data field of said plurality of data fields of interest;
developing a bias matrix with said plurality of bias scores, wherein said bias matrix is associated with a member of a group, wherein said group comprises a plurality of members, wherein said bias matrix is respectively computed for each member of a group of users;
aggregating said respective bias matrix for each member of said group of users into a group bias matrix;
incorporating said bias matrix into said group bias matrix, wherein said group bias matrix is associated with a collective bias of said group; and
mitigating a model bias of an artificial intelligence model using said group bias matrix.