Skip to content

Commit 208a176

Browse files
authored
Merge pull request #128 from NGO-Algorithm-Audit/feature/structural_edits
Publication white paper 'Public standard profiling'
2 parents c6f22f6 + 4890680 commit 208a176

File tree

4 files changed

+72
-0
lines changed

4 files changed

+72
-0
lines changed

content/.DS_Store

0 Bytes
Binary file not shown.
Lines changed: 36 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,36 @@
1+
---
2+
icon: fas fa-search
3+
featured: true
4+
layout: article
5+
type: knowledgebase_item
6+
author: Algorithm Audit
7+
summary: Public standard profiling algorithms
8+
facets:
9+
- value: type_white_paper
10+
label: white paper
11+
- value: subject_FR
12+
label: fundamental rights
13+
- value: subject_standard
14+
label: standard
15+
weight: -12
16+
title: Public standard profiling algorithms
17+
subtitle: |
18+
Public standard profiling algorithms
19+
image: /images/knowledge_base/Public_standard_profiling.png
20+
---
21+
22+
### Public standard profiling algorithms
23+
24+
Based on our case-based experience with auditing risk profiling systems, Algorithm Audit publishes a public standard providing qualitative and qualitative safeguards for responsible use of these type of algorithms.
25+
26+
This standard provides a concise step-by-step guide for the responsible use of profiling algorithms in the public domain, but can also be utilized in the private sector.
27+
28+
Deployers of profiling systems should not wait for European harmonised standards, as developed in support of the AI Act, to regulate profiling systems, because:
29+
30+
1. AI Act standards have a broad scope (all type of AI-systems) and are therefore ineffective to regulate specific algorithmic applications, such as profiling
31+
2. Classic rule-based profiling systems, that are based on rules defined solely by natural persons, do not fall under the definition of an AI-system and are therefore not in scope of the AI Act
32+
3. AI Act standards will only be available behind the paywall of standardisation organisations and therefore impede public knowledge building about responsible algorithms.
33+
34+
In the Netherlands, ‘simple’ rule-based profiling systems have been deployed irresponsibly for over the last 15 years. This public standard aims to contribute to concise risk management measures that can be implemented today to deploy profiling more reponsibly.
35+
36+
{{< pdf_frame articleUrl1="https://drive.google.com/file/d/1_Bbeo-R7Mn3_om3hiwRCuqq_hppNwqmi/preview" width_desktop_pdf="6" width_mobile_pdf="12" >}}
Lines changed: 36 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,36 @@
1+
---
2+
icon: fas fa-search
3+
featured: true
4+
layout: article
5+
type: knowledgebase_item
6+
author: Algorithm Audit
7+
summary: Public standard profiling algorithms
8+
facets:
9+
- value: type_white_paper
10+
label: white paper
11+
- value: subject_FR
12+
label: fundamental rights
13+
- value: subject_standard
14+
label: standard
15+
weight: -12
16+
title: Publieke standaard profileringsalgoritme
17+
subtitle: |
18+
Publieke standaard profileringsalgoritme
19+
image: /images/knowledge_base/Public_standard_profiling.png
20+
---
21+
22+
### Publieke standaard profileringsalgoritme
23+
24+
Voortbouwend op onze ervaring met het auditeren van risicoprofileringsmethoden heeft Algorithm Audit een publieke standaard opgesteld met kwalitatieve en kwantitatieve waarborgen voor de verantwoorde inzet van dit type algoritmes.
25+
26+
The standaard biedt een concreet stappenplan voor de verantwoorde inzet van profileringsalgoritmes voor toepassing in de publieke sector. De standaard kan ook in de private sector worden gebruikt.
27+
28+
Ontwikkelaars van profileringsmethoden moeten om verschillende redenen niet wachten op Europese geharmoniseerde normen, die ter ondersteuning van de AI Verordening worden ontwikkeld, om profileringssystemen te reguleren:
29+
30+
1. AI Verordeningen geharmonizeerde normen hebben een brede reikwijdte (alle type AI-systemen) en zijn daarom ineffectief om specifieke algoritische toepassingen te reguleren, zoals profilering
31+
2. Klassieke regelgebaseerde systemen, waarvan regels door de mens zijn opgesteld, vallen niet onder de definitie van een AI-systeem en vallen daarmee buiten de reikwijdte van de AI Verordening
32+
3. AI Verordeningen geharmonizeerde normen zullen enkel beschikbaar komen achter de betaalmuur van standaardiseringsorganisaties en hinderen daarmee publieke kennisopbouw over de verantwoorde inzet van algoritmes.
33+
34+
In Nederland hebben ‘simpele’ regelgebaseerde profileringsalgoritmes de afgelopen 15 jaar tot veel ophef geleid. Deze publieke standaard beoogt een bijdrage te leveren aan risicobeheersmaatregelen om de methoden op een meer verantwoorde manier in te zetten.
35+
36+
{{< pdf_frame articleUrl1="https://drive.google.com/file/d/1lwaPipWJT_plsryO7QtZqRw5A7zGzqix/preview" width_desktop_pdf="6" width_mobile_pdf="12" >}}
299 KB
Loading

0 commit comments

Comments
 (0)