Goods and services (Directive 2004/113)

• Author: Jenny Julen Votinius
• Pages: 45-48

45

9Goods and services (Directive 2004/113)49

9.1General (legal) context

9.1.1Surveys and reports about the difficulties linked to equal access to and supply of

goods and services

No surveys have been conducted concerning the difficulties linked to equal access to supply

of goods and services.

9.1.2Specific problems of discrimination in the online environment/digital

market/collaborative economy

There is no national research on the frequency of gender-based discrimination resulting

from the use of algorithms or technologies using artificial intelligence. However, in 2018,

the Equality Ombudsman initiated a project to map and estimate the risks of discrimination

in algorithms, automated dataprocessing and artificial intelligence. Based on existing

international scholarship, the project identified a number of typical risks of discrimination

in automatised decision-making. These risks may relate to: the data (non-relevant, biased

or low-quality), the programmer(intentional discrimination, unconscious prejudices, or

bad understanding of the effects of certain types of programming), or the organisational

context in which the algorithms appear and are used (one example is that programs from

different producers which are not discriminatory when used separately may have a

discriminatory effect when combined in a program structure, another example is when the

context in which the programs are used does not provide for checks or follow-ups). The

risk may also relate to machine learning and black box systems.

The aim of the project on algorithmic discrimination carried out by the Equality

Ombudsman was to increase the level of knowledge in this field, mainly at the authority

level. The project had the character of a mapping exercise and concluded that there is a

need to increase the public awareness of artificial intelligence and the risk of algorithmic

discrimination, a need for competence in this field at authorities and organisations, and a

need for increased cooperation and exchange of knowledge between authorities. The

opinion of the national expert is that these observations are relevant; the need for research

and increased competence on algorithmic discrimination appears to be urgent.

In 2017, the Government tasked the Swedish Social Insurance Inspectorate (Inspektionen

för socialförsäkringen, ISF) to analyse the use of ‘selection profiles’ at the Swedish social

insurance agency. The ISF has delivered two reports on the matter.50 An important aspect

of the analysis carried out by the ISF is the matter of equal treatment of men and women.

The ISF explains how results are generated by machine learning algorithms, and why such

results run the risk of being biased on the ground of, inter alia, gender. The conclusion is

that the use of ‘selection profiles’ displays deficits in the area of equal treatment, but the

ISF also underlines that this does not necessarily mean a violation of the Discrimination

Act (2008:568). The reports concern the statutory social security scheme, but the results

are equally relevant for similar products on the private market.

49 See e.g. Caracciolo di Torella, E. and McLellan, B. (2018), Gender equality and the collaborative economy,

European network of legal experts in gender equality and non-discrimination, available at:

https://www.equalitylaw.eu/downloads/4573-gender-equality-and-the-collaborative-economy-pdf-721-kb.

50 Swedish Social Insurance Inspectorate, Profilering som urvalsmetod för riktade kontroller (Profiling as

selection method for targeted controls), ISF Report 2018:5, and Riskbaserade urvalsprofiler och

likabehandling (Risk based selection profiles and equal treatment), Working Report 2018:1AR, available (in

Swedish only) at: http://62.13.72.13/fb/ISF/Rapport%202018-5/HTM/22/ and

https://inspsf.se/download/18.6ce5045216a58f96d2f56214/1565330431885/Riskbaserade%20urvalsprofile

r%20och%20likabehandling%202018-1ar.pdf.