When algorithms divide people

The unemployed have recently been divided into more or less eligible categories by means of artificial intelligence. Is that fair?

The Austrian Labour Market Service (AMS) has recently started using an algorithmic system to decide which unemployed should be supported and how. Our age, citizenship, gender, education, care responsibilities or past occupations - all these data are used to assess our chances in the labour market. The aim is to predict as accurately as possible whether someone can successfully re-enter the labour market in the short or long term.

But how fair is that? Technical researcher Doris Allhutter from ITA has examined the new AMS algorithm for its consequences for job seekers. He poses a number of challenges that also affect other similar systems in public administration: "In my opinion, the model underlying the algorithm only captures a simplified view of the situation of job seekers. The so-called "labour market opportunity model", for example, predicts worse opportunities for women, the elderly and people with disabilities."

Criticism of absolute results

The AMS again points out that the group assignment can be changed by administrators if necessary. This possibility is mainly justified by the fact that information about personal appearance, motivation and other "soft skills" cannot be mapped by the algorithmic model. But under what conditions does this happen in practice, and is AMS staff adequately trained to recognise these soft skills and encouraged to make the changes?

The current ITA dossier "The AMS algorithm on the test bench" provides information on how the algorithm works and suggests measures to make dealing with it transparent, socially competent and fair.


ITA dossier: Der AMS-Algorithmus am Prüfstand (German only)

By: dr

Since November 2018, a new algorithmic classification has been used in the Public Employment Service. (Photo: ITA)