Fair algorithms: the importance of transparency and collaboration

Read more about the foundation's origins here.

Fairness

Creating a fair algorithm isn't easy. It requires expertise in a range of fields, coupled with detailed control and testing processes. Algorithm Audit is concerned with both the technical and the ethical aspects of algorithms. The foundation therefore brings together people with expertise in a wide range of fields to decide what 'fairness' implies in a given context.

One of the biggest challenges is finding an appropriate balance between technical parameters and ethical principles. "In an algorithm, fairness means more than excluding prejudice," explains Jurriaan. "It also means having a process in place, in which all stakeholders have a voice. So we look at the division of responsibilities within an organisation: who decides when an algorithm should be used, and who checks that the algorithm is fair? Getting the governance right is vital."

One of Algorithm Audit's core activities is developing what the foundation calls algoprudence – a bottom-up approach to collectively determining what's fair and responsible in the context of algorithms. The underlying rationale is that the law relies on numerous open standards, whose implications are not made explicit by legislators or regulators. The foundation seeks to address that shortcoming by working with experts in the law, ethics and technology, plus other stakeholders, such as citizens' rights groups and special interest groups.

As Algorithm Audit's co-founder Ariën Voogt asserts, "How to make responsible use of AI and algorithms is one of the key questions facing society today. It's not something we can leave to a handful of technocrats working behind closed doors. We have to work together, and decide by transparent, democratic means just what fairness actually is in relation to AI."

Local authorities and European legislation

Another important field of activity for the foundation is the development of technical tools to support the legal and ethical testing of algorithms. Tools developed so far include the Joint Fairness Assessment Method (JFAM) and a range of open-source tools. Recently, for example, the foundation was involved in a project with the City of Amsterdam, creating resources to improve the documentation and transparency of the city's algorithm use.

Municipal authorities are under increasing pressure to ensure that their activities comply with recently introduced and frequently complex European rules on artificial intelligence. "With the new AI legislation come new requirements, which need to be aligned with the Dutch regulatory framework that's already in place," explains Jurriaan. "Identifying what additional requirements must be met is a daunting task for a medium-sized local authority. So we help them out by designing processes that make it clear what the legal requirements are and how they can be met in practice. An authority can then check its algorithms without getting tied up in red tape."

SIDN Fund's Mieke van Heesewijk explains why the Fund is pleased to back Algorithm Audit. "Fair and transparent algorithms are vital in a world where society is more and more shaped by technology. By harnessing the power of collaboration and dialogue, with ethics to the fore, Algorithm Audit is helping to realise fair and unprejudiced systems. With algorithms playing increasingly important roles in our everyday lives, the importance of that can't be overstated."

Council of Europe

Algorithm Audit isn't active only in the Netherlands. "We've been chosen by the Council of Europe to ensure that the equality bodies that are responsible for overseeing the use of artificial intelligence (AI) by public institutions, such as the Netherlands Institute for Human Rights, have proper support over the next 2 years," says Jurriaan.

Equality bodies have a crucial role within Europe's human rights framework. They monitor the activities of governments to ensure that the rights of vulnerable citizens are protected in the context of, for example, demonstrations and police interventions. "Ensuring that new technologies, such AI, are used appropriately is a key element of that task," continues Jurriaan. "Working with the Council of Europe enables us to share our expertise internationally. We'll be getting started over the autumn and winter, transferring lessons learnt in the Netherlands to the European stage."

Support from SIDN Fund

"SIDN Fund's support has enabled us to take care of various practical things, such as hiring office space in The Hague," says Jurriaan. "The Fund hasn't attached restrictive conditions to its support. So we've been able to use the money to finance our operations in a general sense, rather than spend it on a specific project or projects. That's just what we need in the current phase, while we're finding our feet and expanding. We've been able to get professional financial advice, for example, to help us with our bookkeeping, which became too complex for us to do ourselves. That flexibility has been very important to us. As well as helping us with those practical things, the grant from SIDN Fund has facilitated progress in our core activities. For instance, we've teamed up with experienced software developers to refine our technical tools, and we've recruited talent from the business community."

Private sector

Algorithm Audit works with organisations in the private sector as well as the public sector. For example, the foundation is currently working on a project for a company that provides car sharing services in big cities. "The company uses sophisticated algorithms to detect reckless driving," Jurriaan explains. "Data on telltale aspects of users' driving behaviour, such as fast cornering and sudden heavy braking, is collected. We're analysing the way the data is used, to ensure that the processes of pattern recognition and intervention to minimise damage are fair and reasonable."