Automated unemployment insurance fraud detection system had a staggering 93% error rate in production
Expect to see a lot more cases of automated discrimination like this in the future. There is no way an auto-adjudication system would be allowed to have this staggering level of brokenness if it was dealing with the well-off:
State officials have said that between Oct. 1, 2013, when the MiDAS [automated unemployment insurance fraud detection] system came on line, and Aug. 7, 2015, when the state halted the auto-adjudication of fraud determinations and began to require some human review of MiDAS findings, the system had a 93% error rate and made false fraud findings affecting more than 20,000 unemployment insurance claims. Those falsely accused of fraud were subjected to quadruple penalties and aggressive collection techniques, including wage garnishment and seizure of income tax refunds. Some were forced into bankruptcy. The agency is now reviewing about 28,000 additional fraud determinations that were made during the relevant period, but which involved some human review. An unknown number of those fraud findings were also false.
(tags: fraud broken fail michigan detroit social-welfare us-politics computer-says-no automation discrimination fraud-detection)