Supporting Responsible Use of Algorithms with an Algorithm Inventory Tool

Introduction

The use of data-driven solutions and algorithms by government institutions and other organisations has increased significantly in recent years. Algorithms can provide more efficiency, effectiveness and speed in decision-making and the implementation of policies and strategies. At the same time, the use of algorithms also introduces new risks. These risks however can be minimised using an algorithm management framework.

It is crucial for a successful implementation and adoption of such a framework to create and maintain an overview of all algorithms in the organisation. This can be achieved with an algorithm inventory tool. The tool helps to create an accurate, timely and central overview of the algorithms and provides insights into the risks of these algorithms across the lifecycle. Furthermore, it creates transparency about the algorithms to internal and external stakeholders.

Algorithm risks do not only impact organisations but ultimately also citizens, businesses, and society as a whole. The European Union's plans for new legislation on algorithms (EU AI Act) confirm that attention should be paid to both the advantages and disadvantages of algorithms. In addition, Deloitte's Artificial Intelligence Center of Expertise (“AICE”) has published a view emphasising the importance of transparency in algorithms, evaluation of algorithms and ethics within the use of algorithms. The development of an algorithm management framework, including an inventory tool that guarantees the responsible use of algorithms, plays a crucial role in the view published by the AICE.

Having a clear overview of algorithms with an inventory helps to improve the responsible and ethical use of algorithms as well as to establish accountability within organisations, thereby improving trust in the use of algorithms. An inventory tool not only improves the responsible use of internally developed algorithms but also of externally developed algorithms (vendor algorithms). For vendor algorithms, the added value of an inventory tool can be considerable, specifically since the information on such algorithms (e.g. data, methodology, assumptions used in the algorithm and compliance with regulation) might be less transparent compared to internally developed algorithms. An inventory tool helps to create transparency around the decisions made in the development and use of the algorithm to different stakeholders.

Numerous policies and articles have been published worldwide on the use of an algorithm inventory to support the responsible and ethical use of algorithms, such as in the Algorithmic Accountability for the Public Sector, Amnesty International and Scientific council for government policies (“WRR”). In addition, the European Union is introducing more new and updated legislations regarding digitalization and technology the coming year (e.g. EU AI Act, and Proposal AI Liability Directive). Organisations should analyse the impact of these new and updated legislations on their business in aggregate to understand the communalities between the new and updated legislations.

How an inventory tool adds value to responsible and transparent use of algorithms

Algorithm management frameworks and inventory tools should be specific to the size and type of organisation and also consider the number and complexity of the algorithms used. Generally speaking, an algorithm management framework often contains the following building blocks, and an inventory has an essential role in each of them:

1. A lifecycle, including the steps that each algorithm goes through. The lifecycle starts with the idea for a new algorithm and the decision on what it will be used for. This is followed by the data collection, development and validation of the algorithm, but also testing, use, and monitoring of the algorithm. Establishing a lifecycle-based governance enables organisations to easily identify in which phase an algorithm is and which corresponding actions should be taken to move safely to the next phase.
   The inventory tool tracks the algorithm with the corresponding actions and tasks throughout their entire lifecycle to monitor the algorithm.
2. A clear governance structure for algorithms, with roles and (final) responsibilities for all steps in its lifecycle. The governance structure describes, for example, the role of an algorithm owner, algorithm developers, and an ethics and algorithms approval committee. Additionally, the governance structure also should describe how an independent algorithm assessment role is performed, in order to assess the quality and ethical risks of the algorithm before it goes into use.
   The inventory tool helps to apply and track systematic governance to the algorithms for each different role and lifecycle phase. In addition, each role is assigned specific rights, responsibilities and tasks.
3. Clear responsibilities and controls for every step in the lifecycle of an algorithm are essential to ensure that the governance structure described above is followed. In addition, responsibilities and controls help to consider the organisation’s risk appetite concerning algorithms by raising attention to, for example, monitoring outcomes of the algorithm.
   The inventory tool aids in efficiently executing the roles and responsibilities, including controls. Action-based responsibilities and controls are communicated to the users of the inventory tool where required throughout the lifecycle of the algorithm. The inventory tool acts as a gatekeeper for each lifecycle phase, for example, by requiring approvals at specific checkpoints.
4. Consistent policy documents and standards must be available that describe the different building blocks of the algorithm framework, including for example how data or algorithms can be used in decision making. These documents can also describe specific standards around the processing of personal data, as required by the General Data Protection Regulation ("GDPR"). In addition, an ethical algorithm framework is also essential, which describes or contains guidelines in the field of human autonomy, honesty, explainability and transparency of the algorithm, and the prevention of damage towards persons affected by the algorithm.
   The inventory tool helps to store all relevant documents and contains templates that ensure a standardised manner of documentation across the organisation.
5. Frequent monitoring of algorithms is needed to monitor whether the algorithms remain accurate, suitable and fair after development. Part of the monitoring process is determining the correct test statistics, measures and thresholds (such as statistics for transparency and fairness).
   The inventory tool helps to assess risks and impacts of the algorithms, and can help to continuously monitor the algorithms' performance and material changes. In addition, interactive dashboards facilitate a straightforward assessment of the algorithm performance for all stakeholders.
6. Reporting to (senior) management about algorithm risk increases management awareness and oversight of algorithm risks. These algorithm risks should be managed based on the risk tolerances stated in the risk appetite framework.
   The inventory tool enables organisations to communicate, report and engage transparently and understandably with both internal and external stakeholders by using systematic templates.

An algorithm inventory tool also connects internal and external information exchanges and supports transparency about the algorithms of an organisation. An important feature of an algorithm inventory is the ability to present information about individual algorithms to clients, other stakeholders and society. This can include the data and purpose of the algorithm, but also the mechanisms that are in place to prevent bias and unfair use of the algorithm.

Presenting this information to stakeholders in a transparent and clear way can help to strengthen public trust in the algorithms, and therewith ensure that algorithms can continued to be used both effectively and responsible.

Closing and contacts

At Deloitte, our mission is to help our clients be responsible organisations by enabling them to realise the benefits of high-quality algorithms and ethically deploying algorithms with a positive impact. We are convinced that algorithms should be developed, assessed, used, and maintained responsibly and safely. Operating and maintaining algorithms responsibly and safely helps algorithms be accurate, transparent, explainable, and free of bias.

Deloitte is also the implementation partner of Saidot, one of the leading platforms for AI governance and multi-stakeholder transparency. Saidot’s inventory tool provides insights and control of your algorithms. The tool integrates all algorithm information where the algorithms, lifecycle and statistics come together. The algorithm management framework and Saidot’s inventory tool help organisations use algorithms responsibly and ethically, provide insights into the quality of the algorithms, and offer meaningful transparency to stakeholders.

The implementation of the algorithm management framework and Saidot’s inventory tool supports the responsible and ethical use of algorithms while respecting the legal requirements of an organisation, being in line with the proposed regulations from the European Union. Furthermore, a clear algorithm management framework and inventory tool enable organisations to reap the benefits of algorithms in policy execution while carefully managing algorithm risks.