Improve government purchase and use of algorithms (NL0050)

What is the public problem that the commitment will address? More and more government management and policy decisions are based on data and algorithms. Decisions based on algorithms are often not transparent to citizens and companies, while the consequences of applying these algorithms can be of great significance for society.

What is the commitment? In summary, the following activities are carried out for improving the supervision of the use of algorithms by the government:  Development of an algorithm and human rights impact assessment in which connection is sought as much as possible with existing assessments;  Improvements to purchasing conditions for government purchasing algorithms from companies.  Exploration with government organizations and scientists whether and how to arrive at joint definitions in the field of AI and algorithms.  An overview of available tools for the development and use of more complex algorithms.

How will the commitment contribute to solving the public problem? Providing government organizations with tools for dealing with algorithms and making them transparent and purchasing them ensures that governments can make more conscious choices and be more transparent about their actions. This makes the government's use of algorithms more understandable for residents.

Why is this commitment relevant to OGP values? The instruments that are being developed ensure that governments can be more transparent about the use of algorithms and the trade-offs that are made. In that sense, it is also a form of accountability for the use of algorithms, because it is easier for citizens to trace what happens and how (automated) decisions are made. 30

Additional information  In 2019, research was conducted into the supervision of the use of algorithms in the government: https://www.rijksoverheid.nl/documenten/rapporten/2019/11/25/rapport-toezicht-op-gebruik-van-algoritmen-door-de-overheid  A letter to parliament was published in 2019 about guarantees against risks of algorithms and data analyzes by the government. The focus is on transparency, verifiability and legal protection, BZK has tested the guidelines in practice: https://www.rijksoverheid.nl/documenten/kamerstukken/2019/10/08/tk-waarborgen-tegen-risico-s-van-data-analyses-door-de-overheid  The Court of Audit has been investigating the use of algorithms in the public sector since February 2020: https://www.rekenkamer.nl/actueel/lopend-onderzoek/zicht-op-algoritmes  The existing Guidelines for the application of algorithms by governments are being evaluated and a further legal obligation of these guidelines is being explored. Milestone Activity with a verifiable deliverable Start Date: End Date: Development of an algorithm and human rights impact assessment in which connection is sought as much as possible with existing assessments. 1-1-2021 1-4-2021 Improvements in procurement conditions for procurements of algorithms from companies by the government. 1-1-2021 1-9-2021 Exploration with government organizations and scientists whether and how to arrive at joint definitions and supported solutions in the field of AI and algorithms. 1-1-2021 1-5-2021 An overview of available tools for the development and use of more complex algorithms. 1-1-2021 1-4-2021

Action Plan Review

Commitment 12: Open algorithms

Commitment cluster #10, 11, and 12: Open Technology

(Ministry of the Interior and Kingdom Relations and/or its subsidiaries such as KOOP (Netherlands publication office), ICTU, Foundation for Public Code, Open State Foundation, Code for NL, EMMA Communicatie, Ministry of Justice and Safety, Chamber of Audit)

For a complete description of the commitments included in this cluster, see commitments 10, 11, and 12 on pages 24-30 of the Netherlands 2020 – 2022 action plan here.

Context and objectives:

The Dutch government owns vast amounts of data, which can be accessed publicly via the national open data portal data.overheid.nl. Following the completion of a pilot in 2020, Commitment 10 seeks to scale up this work by establishing five new open data communities in addition to the four currently active communities. The open data communities consist of data owners, re-users, and experts in specific domains, such as education or mobility. [58] The communities offer specific data, reference data, applications, and an opportunity to ask experts directly about the data. This commitment also involves developing indicators that can help assess the actual impact of using and re-using government datasets. These indicators will be made visible on the national data portal and are expected to sustain a structural supply and demand for open data (with the help of these data communities). The commitment also calls for developing impact assessment on open data use.

In addition, digital transformations have altered the functioning of public service delivery in the Netherlands. Engagement with the IT community, including software developers, is essential as this enhances quality and helps foster a deeper understanding of these tools between users and creators. As the government frequently commissions software, doing so in an open-source format, meaning software is free and open to modification and re-distribution, promotes essential collaboration of public organizations and the sharing of digital tools for the public good. Aside from collaborative development, open-source software can also strengthen transparency, avoid the duplication of software tools for government agencies, and prevent so-called vendor ‘lock in’. Furthermore, and in response to inquiries from MPs, the government plans to make the source code available for software that is developed by public means so that it can be publicly reviewed, improved, and re-used. [59] Government organizations, however, have limited experience releasing source codes and it is not always clear what the costs of releasing the source code are or if this adds value in all cases. Through Commitment 11, the Ministry of Interior and Kingdom Relations, together with a broad range of technology-focused CSOs, will spread the use of working open source within government, by stimulating debate, developing a toolbox, sharing best practices, and linking this theme to policy making at the national level.

Governments also increasingly rely on new technologies such as artificial intelligence (AI), machine learning, and algorithms to analyze data and inform policy making. The use of algorithms in government policies stirred debate in Dutch society; local watchdogs criticized the use of an algorithm called SyRI by the government to fight fraud due to its its lack of transparency, apparent bias, and disregard for privacy. SyRI’s use was ultimately banned in national court who ruled that the system violated the European Convention on Human Rights. [60] The government has commissioned various studies on the topic [61] and agrees that algorithms need to be transparent for reasons of oversight and legal supervision. [62] The extent to which algorithms are used, however, still appears to be largely unknown. In addition, the Court of Audit has reported that little attention is currently paid to ethical aspects or potential biases in the government’s algorithms. [63] It noted that citizens should be able to understand the use and operation of algorithms and know where to turn to with questions or objections. The Court of Audit recommends that the government secure personal data in the management of its algorithms and ensure an unambiguous common language that defines quality requirements for algorithms.

Against this backdrop, the fourth action plan continues to work on the theme of algorithms under Commitment 12. The Netherlands’ previous action plan included a commitment on developing preliminary frameworks and guidelines around algorithmic transparency, but it saw only limited completion. [64] This new commitment, on the other hand, explicitly focuses on the issue of ethics and algorithms. It aims to develop a human-rights based impact assessment tool for potential algorithms and use this to set standards in engaging with third parties (such as external software suppliers). It also plans to improve conditions for the government purchasing algorithms from companies and explore how to arrive at joint definitions in AI and algorithms. However, the commitment does not explicitly state if the human rights impact assessments for government algorithms will be made available to the public.

Potential for results:Modest

Taken together, the commitments in this cluster could improve the openness, transparency, and participation in how the Netherlands uses critical technologies and data. All three commitments seek to work with a broad variety of stakeholders outside government and facilitate external (including citizen) feedback to improve government owned or hosted data and software tools.

Open government data can only live up to its potential if, aside from being readily and publicly available in appropriate formats, it is used by an ever-growing group of diverse and experienced users. Through the open data communities under Commitment 10, government agencies will learn about new, innovative ways for the re-use of government data, while users will learn to better navigate the national portal that includes tens of thousands of datasets. The particular attention devoted to monitoring and evaluating success, by seeking to develop portal-wide indicators to measure impact of data, could yield significant results. Demonstrated impact on social issues by using open data can subsequently increase the demand for more data. Drawing on that impact assessment, this commitment can also help build broader social and political support for the disclosure of government data in an open data format.

Commitment 11 has promising potential to strengthen the government’s ability to work in open source, thanks to direct linkages to existing policy-making efforts at the central government level, coupled with strong engagement from civil society and the open-source community. Although the exact results are difficult to forecast, making open-source coding more transparent could reveal new insights into how government operates regarding digital public services.

Finally, the renewed focus on government algorithms under Commitment 12 follows earlier IRM recommendations to draw on existing bodies of domestic work in this area at the central government level. [65] Although this commitment does not call for opening up additional government algorithms, the human rights impact assessment could help government agencies to safeguard against potential biases in the underlining data of the algorithms they procure. This in turn could help reduce the possible discrimination of certain segments of the population when the government utilizes algorithms in developing policies. This commitment could also enrich other efforts at the central government level, such as the ‘procurement with impact’ strategy. [66]

It is important to note, however, that the commitment does not specify if the use of the human rights impact assessments will become mandatory for all government agencies when they procure algorithms. It also does not specify if the findings of these impact assessments will be made publicly available during the algorithm procurement process. Therefore, the results of this commitment will largely depend on the uptake of the human rights assessments among government agencies and the level of detail included. The commitment’s results will also depend on the discussions held with civil society on the impact of algorithms on society and the extent to which these discussions lead to making more algorithms publicly available for scrutiny.

Opportunities, challenges, and recommendations during implementation

Open technology is a broad topic, yet several thematic overlaps could be explored inside this cluster and beyond. For example, the commitments on open data communities and open source could strengthen the proactive disclosure of government information under Commitment 5 of this action plan. Information categories and formats are central to that work. Therefore, the IRM recommends sharing relevant insights and ideas from Commitments 10 and 11 with the stakeholders working on the proactive disclosure of government information (Commitment 5). In addition, Commitment 3 on digital democracy aims to pilot an open source digital tool and AI-powered consensus platforms (such as pol.is and openstad.org), and lessons learned could be shared with the experts and organizations involved under Commitment 11.

Regarding Commitment 12, the IRM recommends making the human rights impact assessment for government procurement of algorithms publicly available. This way, the impact assessments could provide citizens and civil society with an important mechanism to monitor how government agencies are taking human rights into account when procuring their algorithms. As a result, citizens and civil society will be able to better raise potential ethical issues in the government’s use of algorithms in its policies. The IRM also recommends going a step further by making use of the human rights impact assessments mandatory for all public agencies when they procure algorithms.

In terms of open algorithms more broadly, the Netherlands has joined a group of other countries working on this topic in the context of OGP. [67] The IRM recommends engaging international experts from other countries to share their experiences and lessons learned on algorithmic transparency. EtaLab from France, for example, has experience in disclosing to citizens how and when algorithms were used and could add significant value to the work in the Netherlands. In addition, the IRM recommends assessing where disclosure is needed the most and consider listing the high-value datasets where algorithms are currently used. For example, the City of Amsterdam, that was involved in the co-creation process, has developed an algorithm register where citizens can learn more about the use of algorithms in the city administration. [68] Such examples of public outreach and awareness raising are considered important to help increase knowledge and skills for citizens to develop ‘technological citizenship’. [69]

Finally, in anticipation of Commitment 13 of this action plan (discussed below), the IRM recommends that stakeholders involve the National Ombudsperson, when possible, in their work on algorithmic transparency, as this represent one of the key channels for people to raise concerns. In addition, the National Ombudsperson has declared it seeks to assure that algorithms used by the government are sound and citizen driven. [70]

Results Report

Commitment 12. Open Algorithms

● Verifiable: Yes

● Does it have an open government lens? No

● This commitment has been clustered as: Open Technology

● Potential for results: Modest

● Completion: Substantial

● Did it open government? Marginal

Most milestones were achieved, including the publication of the Impact Assessment Human Rights and Algorithms tool. ^[35] This tool aims to guide discussions among policymakers and other parties on whether to develop an algorithmic government application. ^[36] Moreover, the government launched an online algorithm register where government agencies publish the algorithms they use. ^[37] As of April 2023, there are 109 algorithms on the register. The IRM was unable to verify whether work to improve procurement terms for algorithms and to develop practical guidance on the procurement of algorithms had been implemented. However, the new register has resulted in a marginal impact on opening government, as it has brought government algorithms into one place.