Error-riddled data sets are warping our sense of how good AI really is

Link: https://www.technologyreview.com/2021/04/01/1021619/ai-data-errors-warp-machine-learning-progress/

Paper link: https://arxiv.org/pdf/2103.14749.pdf

Graphic:

Excerpt:

Yes, but: In recent years, studies have found that these data sets can contain serious flaws. ImageNet, for example, contains racist and sexist labels as well as photos of people’s faces obtained without consent. The latest study now looks at another problem: many of the labels are just flat-out wrong. A mushroom is labeled a spoon, a frog is labeled a cat, and a high note from Ariana Grande is labeled a whistle. The ImageNet test set has an estimated label error rate of 5.8%. Meanwhile, the test set for QuickDraw, a compilation of hand drawings, has an estimated error rate of 10.1%.

Author(s): Karen Hao

Publication Date: 1 April 2021

Publication Site: MIT Tech Review

Huge vaccine error: 9,000 patients were forgotten

Link: https://stiften.dk/artikel/k%C3%A6mpe-vaccinefejl-9-000-patienter-blev-glemt

Excerpt (via Google Translate):

Jørgen Schøler Kristensen, Medical Director at Aarhus University Hospital, explains the error as follows:

– When patients with a civil registration number beginning with 0 had to be entered, the civil registration number was declared invalid if a hyphen was missing. So 26 percent of the civil registration numbers we have reported have not been entered correctly, so they have not received the invitation in their e-box to be vaccinated as they should have had, says Jørgen Schøler Kristensen.

Patients born the first nine days of any month and having been unlucky with a hyphen were forgotten.

Additional:

Author(s): Jens Christian Thaysen

Publication Date: 19 March 2021

Publication Site: Århus Stiftstidende

Missourians May Get to Keep Erroneous Unemployment Payments

Link: https://www.governing.com/finance/Missourians-May-Get-to-Keep-Erroneous-Unemployment-Payments.html

Excerpt:

Legislators are taking action after Gov. Mike Parson’s administration resisted calls to let Missourians keep unemployment benefits they were paid in error.

The House Special Committee on Government Oversight on Wednesday heard seven separate proposals to forgive the return of unemployment payments mistakenly given to people who did not intend to commit fraud.

Committee Chair Rep. Jered Taylor, R- Nixa, said the committee will likely combine several of the bills, some of which are nearly identical, into a single proposal. The committee would then hold a vote on the proposal early next week.

Author(s): MARIA BENEVENTO, ST. LOUIS POST-DISPATCH

Publication Date: 11 February 2021

Publication Site: Governing

Employers make the Canada Revenue Agency mistakes

Link: https://www.vaughantoday.ca/employers-make-the-canada-revenue-agency-mistakes/

Excerpt:

The Canadian Revenue Agency (CRA) presented a spreadsheet of Canada Emergency Wage Support (CUSS) that was riddled with errors and cut some business subsidies in half.

To qualify for SSUC – which has so far helped nearly 400,000 companies with grants totaling $ 59 billion – employers must fill out “spreadsheets”: Excel spreadsheets that ARC has been offering on its website since March 15th.

However, the spreadsheet available to companies for payment periods from July 5 to August 29 contained several major errors until last Friday, nine days before the deadline for grant applications.

…..

This mistake had the effect of drastically reducing support for businesses, but only for some owners and their families. It only exists in the French version of the spreadsheet.

Author: Alan Binder

Publication Date: 25 January 2021

Publication Site: Vaughan Today