Skip to content

Commit 54d0e6a

Browse files
authored
Merge pull request #236 from NGO-Algorithm-Audit/feature/structural_edits
Feature/structural edits
2 parents 1b909ce + dffd5c5 commit 54d0e6a

File tree

47 files changed

+76
-764
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

47 files changed

+76
-764
lines changed

content/english/algoprudence/cases/aa202201_type-of-sim.md

Lines changed: 3 additions & 39 deletions
Original file line numberDiff line numberDiff line change
@@ -1,38 +1,4 @@
11
---
2-
summary: >
3-
The commission advises against using type of SIM card as an input variable in
4-
algorithmic models that predict payment defaults and block afterpay services
5-
for specific customers. As it is likely that type of SIM card acts as a
6-
proxy-variable for sensitive demographic categories, the model would run an
7-
intolerable risk of disproportionally excluding vulnerable demographic groups
8-
from the payment service. Absent reliable data that demonstrates otherwise,
9-
the ethical risk of including the SIM card variable outweighs potential
10-
benefits. The commission advises to consider a variety of alternatives in
11-
dealing with payment defaults.
12-
sources: |
13-
Anonymized large multinational company with e-commerce platform.
14-
algoprudence:
15-
title: Report
16-
intro: "Dowload the full report and problem statement\_[here](https://drive.google.com/file/d/16Au8tb9NGOeCF0EZLIQQ1cNLhLAmTylL/view).\n"
17-
reports:
18-
- url: >-
19-
https://drive.google.com/file/d/16Au8tb9NGOeCF0EZLIQQ1cNLhLAmTylL/preview
20-
normative_advice_members:
21-
- name: |
22-
Fleur Jongepier, Assistant Professor Digital Ethics, Radboud University
23-
- name: |
24-
Jan Overgoor, PhD Management Science and Engineering, Stanford University
25-
- name: >
26-
Merel Noorman, Assistant Professor Law, Technology and Society, Tilburg
27-
University
28-
- name: |
29-
Roel Dobbe, Assistant Professor Engineering Systems and Services, TU Delft
30-
funded_by:
31-
- url: 'https://www.sidnfonds.nl/projecten/ethical-risk-assessment-tool'
32-
image: /images/supported_by/sidn.png
33-
- url: 'https://europeanaifund.org/'
34-
image: /images/supported_by/EUAISFund.png
35-
icon: fa-sim-card
362
layout: case
373
title: Type of SIM card as a predictor for detecting payment fraud
384
subtitle: |
@@ -63,10 +29,10 @@ form1:
6329
required: true
6430
type: textarea
6531
- label: |
66-
Contact detail
32+
Contact details
6733
id: contact-details
6834
required: true
69-
type: email
35+
type: text
7036
placeholder: Mail address
7137
---
7238

@@ -86,9 +52,7 @@ The commission advises against using type of SIM card as an input variable in al
8652

8753
Anonymized large multinational company with e-commerce platform.
8854

89-
#### Algoprudence
90-
91-
The problem statement and advice report can be downloaded <a href="https://drive.google.com/file/d/1fSETUhxOz0nF2nznsWq-4TyngP6lU7yH/preview" target="_blank">here</a>.
55+
#### Problem statement and advice document
9256

9357
{{< embed_pdf url="/pdf-files/algoprudence/ALGO_AA202201/ALGO_AA202201_Report_Type_SIM.pdf" >}}
9458

content/english/algoprudence/cases/aa202301_bert-based-disinformation-classifier.md

Lines changed: 6 additions & 84 deletions
Original file line numberDiff line numberDiff line change
@@ -1,98 +1,22 @@
11
---
2-
title: Higher-dimensional bias in a BERT-based disinformation classifier
2+
layout: case
3+
title: Multi-dimensional bias in a BERT-based disinformation classifier
34
subtitle: |
45
Problem statement (ALGO:AA:2023:01:P) and advice document (ALGO:AA:2023:01:A)
56
image: /images/algoprudence/AA202301/Cover.png
6-
form1:
7-
title: React to this normative judgement
8-
content: >-
9-
Your reaction will be sent to the team maintaining algoprudence. A team will
10-
review your response and, if it complies with the guidelines, it will be placed in the Discussion & debate section
11-
above.
12-
button_text: Submit
13-
backend_link: 'https://formspree.io/f/xyyrjyzr'
14-
id: case-reaction
15-
questions:
16-
- label: Name
17-
id: name
18-
required: true
19-
type: text
20-
- label: Affiliated organization
21-
id: affiliated-organization
22-
type: text
23-
- label: Reaction
24-
id: reaction
25-
required: true
26-
type: textarea
27-
- label: Contact detail
28-
id: contact-details
29-
required: true
30-
type: email
31-
placeholder: Mail address
32-
layout: case
33-
icon: fa-newspaper
34-
summary: >
35-
The advice commission believes there is a low risk of (higher-dimensional)
36-
proxy discrimination by the BERT-based disinformation classifier and that the
37-
particular difference in treatment identified by the quantitative bias scan
38-
can be justified, if certain conditions apply.
39-
sources: "Applying our self-build unsupervised\_[bias detection tool](https://algorithmaudit.eu/bias_scan)\_on a self-trained BERT-based disinformation classifier on the Twitter1516 dataset. Learn more on\_[Github](https://github.com/NGO-Algorithm-Audit/Bias_scan).\n"
40-
additional_content:
41-
- title: Stanford's AI Audit Challenge 2023
42-
content: "This case study, in combination with our\_[bias scan tool](https://algorithmaudit.eu/bias_scan), has been selected as a finalist for\_[Stanford's AI Audit Challenge 2023](https://hai.stanford.edu/ai-audit-challenge-2023-finalists).\n"
43-
image: /images/supported_by/HAI.png
44-
image_link: 'https://hai.stanford.edu/ai-audit-challenge-2023-finalists'
45-
width: 8
46-
- title: Presentation
47-
content: "A visual presentation of this case study can be found in this\_[slide deck](https://github.com/NGO-Algorithm-Audit/Bias_scan/blob/master/Main_presentation_joint_fairness_assessment_method.pdf).\n"
48-
width: 12
49-
algoprudence:
50-
title: Report
51-
intro: "Dowload the full report and problem statement\_[here](https://drive.google.com/file/d/1GHPwDaal3oBJZluFYVR59e1_LHhP8kni/view?usp=sharing).\n"
52-
reports:
53-
- url: >-
54-
https://drive.google.com/file/d/1GHPwDaal3oBJZluFYVR59e1_LHhP8kni/preview
55-
normative_advice_members:
56-
- name: |
57-
Anne Meuwese, Professor in Public Law & AI at Leiden University
58-
- name: >
59-
Hinda Haned, Professor in Responsible Data Science at University of
60-
Amsterdam
61-
- name: |
62-
Raphaële Xenidis, Associate Professor in EU law at Sciences Po Paris
63-
- name: |
64-
Aileen Nielsen, Fellow Law\&Tech at ETH Zürich
65-
- name: "Carlos Hernández-Echevarría, Assistant Director and Head of Public Policy at the anti-disinformation nonprofit fact-checker\_[Maldita.es](https://maldita.es/maldita-es-journalism-to-not-be-fooled/)\n"
66-
- name: "Ellen Judson, Head of CASM and Sophia Knight, Researcher, CASM at Britain’s leading cross-party think tank\_[Demos](https://demos.co.uk/)\n"
67-
funded_by:
68-
- url: 'https://europeanaifund.org/'
69-
image: /images/supported_by/EUAISFund.png
70-
actions:
71-
- id: ai_audit_2023
72-
title: Finalist selection Stanford's AI Audit Challenge 2023
73-
description: >
74-
Our [bias detection tool](/technical-tools/BDT) and this case study have
75-
been selected as a finalist for [Stanford's AI Audit Challenge
76-
2023](https://hai.stanford.edu/ai-audit-challenge-2023-finalists).
77-
78-
image: /images/supported_by/HAI.png
79-
date: 14-02-2024
80-
facets:
81-
- value: finalist
82-
label: finalist
837
---
848

859
{{< tab_header width="6" tab1_id="description" tab1_title="Description of algoprudence" tab2_id="actions" tab2_title="Actions following algoprudence" tab3_id="" tab3_title="" default_tab="description" >}}
8610

87-
{{< tab_content_open icon="fa-newspaper" title="Higher-dimensional bias in a BERT-based disinformation classifier" id="description" >}}
11+
{{< tab_content_open icon="fa-newspaper" title="Multi-dimensional bias in a BERT-based disinformation classifier" id="description" >}}
8812

8913
#### Algoprudence identification code
9014

9115
ALGO:AA:2023:01
9216

9317
#### Summary advice
9418

95-
The advice commission believes there is a low risk of (higher-dimensional) proxy discrimination by the BERT-based disinformation classifier and that the particular difference in treatment identified by the quantitative bias scan can be justified, if certain conditions apply.
19+
The advice commission believes there is a low risk of (multi-dimensional) proxy discrimination by the BERT-based disinformation classifier and that the particular difference in treatment identified by the quantitative bias scan can be justified, if certain conditions apply.
9620

9721
#### Source of case
9822

@@ -108,9 +32,7 @@ This case study, in combination with our [bias detection tool](/technical-tools
10832

10933
A visual presentation of this case study can be found in this [slide deck](https://github.com/NGO-Algorithm-Audit/Bias_scan/blob/master/Main_presentation_joint_fairness_assessment_method.pdf).
11034

111-
#### Report
112-
113-
Dowload the full report and problem statement [here](https://drive.google.com/file/d/1GHPwDaal3oBJZluFYVR59e1_LHhP8kni/view?usp=sharing).
35+
#### Problem statement and advice document
11436

11537
{{< embed_pdf url2="/pdf-files/algoprudence/ALGO_AA202301/ALGO_AA202301A_Case_study_disinfo.pdf" url="/pdf-files/algoprudence/ALGO_AA202301/ALGO_AA202301P_Case_study_disinfo.pdf" >}}
11638

@@ -141,7 +63,7 @@ Dowload the full report and problem statement [here](https://drive.google.com/f
14163

14264
##### Description
14365

144-
Our [bias detection tool](https://algorithmaudit.eu/technical-tools/BDT) and this case study have been selected as a finalist for [Stanford’s AI Audit Challenge 2023](https://hai.stanford.edu/ai-audit-challenge-2023-finalists).
66+
Our [unsupervised bias detection tool](/technical-tools/bdt/) and this case study have been selected as a finalist for [Stanford’s AI Audit Challenge 2023](https://hai.stanford.edu/ai-audit-challenge-2023-finalists).
14567

14668
{{< accordion_item_close >}}
14769

content/english/algoprudence/cases/aa202302_risk-profiling-for-social-welfare-reexamination.md

Lines changed: 5 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -72,13 +72,11 @@ Unsolicited research, build upon [freedom of information requests](https://www.
7272

7373
#### Presentation
7474

75-
The advice report (AA:2023:02:A) has been presented to the Dutch Minister of Digitalization on November 29, 2023. A press release can be found [here](/events/press_room/#rotterdam-algoritme).
75+
The advice report (ALGO:AA:2023:02:A) has been presented to the Dutch Minister of Digitalization on November 29, 2023. A press release can be found [here](/events/press_room/#rotterdam-algoritme).
7676

7777
{{< image id="presentation-minister" image1="/images/algoprudence/AA202302/Algorithm audit presentatie BZK FB-18.jpg" alt1="Presentation advice report to Dutch Minister of Digitalization" caption1="Presentation advice report to Dutch Minister of Digitalization" width_desktop="5" width_mobile="12" >}}
7878

79-
#### Report
80-
81-
Dowload the full report and problem statement [here](https://drive.google.com/file/d/1GHPwDaal3oBJZluFYVR59e1_LHhP8kni/view?usp=sharing).
79+
#### Problem statement and advice document
8280

8381
{{< embed_pdf url="/pdf-files/algoprudence/ALGO_AA202302/ALGO_AA202302P_EN.pdf" url2="/pdf-files/algoprudence/ALGO_AA202302/ALGO_AA202302A_EN.pdf" >}}
8482

@@ -100,21 +98,21 @@ Dowload the full report and problem statement [here](https://drive.google.com/f
10098

10199
##### Description
102100

103-
 Council members submitted <a href="https://amsterdam.raadsinformatie.nl/document/13573898/1/236+sv+Aslami%2C+IJmker+en+Garmy+inzake+toegepaste+profileringscriteria+gemeentelijke+algoritmes" target="_blank">questions</a> whether the machine learning (ML)-driven risk profiling algorithm currently tested by the City of Amsterdam satisfies the requirements as set out in AA-2023:02:A, including:
101+
 Council members submitted <a href="https://amsterdam.raadsinformatie.nl/document/13573898/1/236+sv+Aslami%2C+IJmker+en+Garmy+inzake+toegepaste+profileringscriteria+gemeentelijke+algoritmes" target="_blank">questions</a> whether the machine learning (ML)-driven risk profiling algorithm currently tested by the City of Amsterdam satisfies the requirements as set out in ALGO:AA:2023:02:A, including:
104102

105103
* (in)eligible selection criteria fed to the ML model
106104
* explainability requirements for the used explainable boosting algorithm
107105
* implications of the AIAct for this particular form of risk profiling.
108106

109-
{{< pdf_frame articleUrl1="https://drive.google.com/file/d/1oPiO_s9KuV7446BqC9a4P2qjW7MGUyxu/preview" >}}
107+
{{< embed_pdf url="/pdf-files/algoprudence/ALGO_AA202302/Actions/20231204_236 sv Aslami, IJmker en Garmy inzake toegepaste profileringscriteria gemeentelijke algoritmes.pdf" width_mobile_pdf="12" width_desktop_pdf="6" >}}
110108

111109
{{< accordion_item_close >}}
112110

113111
{{< accordion_item_open title="Binnenlands Bestuur" image="/images/algoprudence/AA202302/Actions/logo-bb.svg" id="binnenlands-bestuur" date="01-12-2023" tag1="news" >}}
114112

115113
##### Description
116114

117-
News website for Dutch public sector administration reported on AA:2023:02:A. See [link](https://www.binnenlandsbestuur.nl/digitaal/algoritmische-profilering-onder-strikte-voorwaarden-mogelijk).
115+
News website for Dutch public sector administration reported on ALGO:AA:2023:02:A. See [link](https://www.binnenlandsbestuur.nl/digitaal/algoritmische-profilering-onder-strikte-voorwaarden-mogelijk).
118116

119117
{{< accordion_item_close >}}
120118

content/english/algoprudence/cases/aa202401_preventing-prejudice.md

Lines changed: 2 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -3,8 +3,7 @@ icon: fas fa-greater-than-equal
33
layout: case
44
title: Preventing prejudice
55
subtitle: >
6-
Technical audit (TA:AA:2024:01) – Findings and recommendations regarding
7-
rule-based risk profiling as used in a Dutch public sector organisation
6+
Technical audit (TA:AA:2024:01) of risk profiling algorithm used by Dutch Executive Agency of Education
87
control process
98
image: /images/algoprudence/AA202401/Cover_EN.png
109
form1:
@@ -58,9 +57,7 @@ A press release can be found [here](/events/press_room/#DUO).
5857

5958
Education Executive Agency of The Netherlands (DUO)
6059

61-
#### Algoprudence
62-
63-
The technical audit report (TA:AA:2024:01) can be downloaded [here](https://drive.google.com/file/d/17dwU4zAqpyixwVTKCYM7Ezq1VM5_kcDa/preview).
60+
#### Audit report
6461

6562
{{< embed_pdf url="/pdf-files/algoprudence/TA_AA202401/TA_AA202401_Preventing_prejudice.pdf" >}}
6663

@@ -94,20 +91,6 @@ Report *Preventing prejudice* has been <a href="https://www.rijksoverheid.nl/doc
9491

9592
{{< accordions_area_open id="discussion" >}}
9693

97-
{{< accordion_item_open title="Reaction Netherlands Human Rights Institute on age discrimination" id="cvrm" background_color="#eef2f6" date="12-04-2024" tag1="reaction" image="/images/algoprudence/AA202302/Discussion&debate/CvRM.svg" >}}
98-
99-
#### Age Discrimination
100-
101-
Policies, such as those implemented by public sector agencies investigating (un)duly granted social welfare or employers seeking new employees, can intentionally or unintentionally lead to differentiation between certain groups of people. If an organization makes this distinction based on grounds that are legally protected, such as gender, origin, sexual orientation, or a disability or chronic illness, and there is no valid justifying reason for doing so, then the organization is engaging in prohibited discrimination. We refer to this as discrimination.
102-
103-
But what about age? Both the Rotterdam-algorithm and DUO-algorithm, as studied by Algorithm Audit, differentiated based on age. However, in these cases, age discrimination does not occur.
104-
105-
EU non-discrimination law also prohibits discrimination on the basis of age. For instance, arbitrarily rejecting a job applicant because someone is too old is not unlawful. However, legislation regarding age differentiation allows more room for a justifying argument than for the aforementioned personal characteristics. This is especially true when the algorithm is not applied in the context of labor.
106-
107-
Therefore, in the case of detecting unduly granted social welfare or misuse of college loan, it is not necessarily prohibited for an algorithm to consider someone's age. However, there must be a clear connection between age and the aim pursued. Until it is shown that someone's age increases the likelihood of misuse or fraud, age is ineligible as a selection criteria in algorithmic-driven selection procedures. For example, pertaining to disability allowances for youngsters (Wajong) a clear connection exists and an algorithm can lawfully differentiate upon age.
108-
109-
{{< accordion_item_close >}}
110-
11194
{{< accordions_area_close >}}
11295

11396
{{< tab_content_close >}}

content/english/algoprudence/cases/aa202402_preventing-prejudice_addendum.md

Lines changed: 3 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -3,10 +3,9 @@ icon: fas fa-balance-scale-left
33
layout: case
44
title: Addendum Preventing prejudice
55
subtitle: >
6-
Addendum for ealier [research](/algoprudence/cases/aa202401_bias-prevented/) –
6+
Addendum for previous [audit report](/algoprudence/cases/aa202401_bias-prevented/) –
77
Supervised bias analysis of aggregtation statistics on the country of birth
8-
and country of origin of 300.000+ students as provided by the Dutch office of
9-
statistics
8+
and country of origin of 300.000+ students as provided by Statistics Netherlands
109
image: /images/algoprudence/AA202402/AA202402_cover_EN.png
1110
form1:
1211
title: React to this technical audit
@@ -62,13 +61,10 @@ Education Executive Agency of The Netherlands (DUO)
6261

6362
[https://github.com/NGO-Algorithm-Audit/DUO-CUB](https://github.com/NGO-Algorithm-Audit/DUO-CUB)
6463

65-
#### Algoprudence
66-
67-
The full report (TA:AA:2024:02) can be found <a href="https://drive.google.com/file/d/1uOhR9qXHW6P0i4uP7RNhil2G2dXzFjrp/preview" target="_blank">here</a>.
64+
#### Audit report
6865

6966
{{< embed_pdf url="/pdf-files/algoprudence/TA_AA202402/TA_AA202402_Addendum_Preventing_prejudice.pdf" >}}
7067

71-
7268
{{< tab_content_close >}}
7369

7470
{{< tab_content_open icon="" title="" id="actions" >}}

content/english/events/activities.md

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -485,7 +485,11 @@ events:
485485
- value: type_interview
486486
label: interview
487487
- title: Seminar 'Algorithm validation'
488-
description: "Algorithm Audit hosted a public seminar on algorithm validation and algoprudence in The Hague. Anne Meuwese\_shared insights on the value of algoprudence in contextualizing legal norms. Floris Holstege shed light into statistical hypothesis testing which plays an important role in our recent techincal audits.\_\n\nWe appreciated the interactive Q\\&A with the participants, especially the curious and critical questions which enables us to clarify our work, but also contributes to the further development and refinement of our activities as an NGO.\n\n{{< pdf_frame articleUrl1=\"https://drive.google.com/file/d/1edrNqP4cBgJ1zKv1970DsUJ3Tark6waF/preview\" width_desktop_pdf=\"12\" width_mobile_pdf=\"12\" >}}\n"
488+
description: >
489+
Algorithm Audit hosted a public seminar on algorithm validation and algoprudence in The Hague. Anne Meuwese shared insights on the value of algoprudence in contextualizing legal norms. Floris Holstege shed light into statistical hypothesis testing which plays an important role in our recent techincal audits. We appreciated the interactive Q&A with the participants, especially the curious and critical questions which enables us to clarify our work, but also contributes to the further development and refinement of our activities as an NGO.
490+
491+
{{< embed_pdf url="/pdf-files/events/activities/20240424_Nutshuis.pdf" width_mobile_pdf="12" width_desktop_pdf="12" >}}
492+
489493
image: /images/logo/logo.svg
490494
date: 24-04-2024
491495
facets:
@@ -505,9 +509,8 @@ events:
505509
Media & Democracy lab
506510
507511
508-
{{< pdf_frame
509-
articleUrl1="https://drive.google.com/file/d/1J4HojWGUE4IBhVsxIsht0SXorIex8ngH/preview"
510-
width_desktop_pdf="12" width_mobile_pdf="12" >}}
512+
{{< embed_pdf url="/pdf-files/events/activities/20240416_UvA_MDLab.pdf" width_mobile_pdf="12" width_desktop_pdf="12" >}}
513+
511514
image: /images/events/images.jpeg
512515
date: 16-04-2024
513516
facets:
@@ -543,10 +546,8 @@ events:
543546
Knowledge sharing event on auditing algorithms from both a quantitative
544547
and qualitative perspective
545548
549+
{{< embed_pdf url="/pdf-files/events/activities/20240327_AA_Rekenkamer_DenHaag.pdf" width_mobile_pdf="12" width_desktop_pdf="12" >}}
546550
547-
{{< pdf_frame
548-
articleUrl1="https://drive.google.com/file/d/1CwCrQLdFP29ssxu-mnjAZxEGlfF66OW6/preview"
549-
width_desktop_pdf="12" width_mobile_pdf="12" >}}
550551
image: /images/events/DH_rekenkamer.png
551552
date: 27-03-2024
552553
facets:
@@ -607,9 +608,8 @@ events:
607608
CEN-CENELEC Dublin
608609
609610
610-
{{< pdf_frame
611-
articleUrl1="https://drive.google.com/file/d/1vadydN4_ZEXJ0h_Sj-4GRUwJSacM0fCK/preview"
612-
width_desktop_pdf="12" width_mobile_pdf="12" >}}
611+
{{< embed_pdf url="/pdf-files/events/activities/20240213_JTC21_plenary_FRIAs_stakeholder_panels.pdf" width_mobile_pdf="12" width_desktop_pdf="12" >}}
612+
613613
image: /images/events/jtc21.jpg
614614
date: 13-02-2024
615615
facets:

content/english/knowledge-platform/knowledge-base/Comparative_review_10_FRIAs.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -31,4 +31,4 @@ Our assessment shows a sharp divide regarding the length and completeness of FRI
3131

3232
Are you a frequent user, or a developer of a FRIA, please reach out to [[email protected]](mailto:[email protected]) to share insights based on our case-based AI auditing experience.
3333

34-
{{< pdf_frame articleUrl1="https://drive.google.com/file/d/127jKVdGqkiiLSE_MjfLP1nv4Q_lqib1S/preview" width_desktop_pdf="6" width_mobile_pdf="12" >}}
34+
{{< embed_pdf url="/pdf-files/knowledge-base/20240918_Comparative review 10 FRIAs Algorithm Audit.pdf" width_mobile_pdf="12" width_desktop_pdf="6" >}}

0 commit comments

Comments
 (0)