ETUC Resolution calling for an EU Directive on Algorithmic Systems at Work
Adopted at the ETUC Executive Committee of 6 December 2022
Key Messages:
- Algorithmic systems, especially AI (Artificial Intelligence), have a great influence on the work of the future. To improve working conditions and avoid negative effects, the use of such systems in the workplace must be better and effective regulated;
- The AI Act is not suitable for regulating use of AI in the workplace. An EU directive on algorithmic systems in the workplace, based on Article 153 TFEU, should define European minimum standards for the design and use of algorithmic systems in the employment context;
- Key element of the new directive is the strengthening and enforcement of collective bargaining rights of trade unions as well as information, consultation and participation rights of workers' representatives;
- Algorithmic systems at work need to be transparent and explainable. Workers and their representatives shall have the right to receive information about the used applications in plain and understandable language;
- Trade unions and workers’ representatives shall have the right to gain external expertise;
- An algorithmic impact assessment for changes in working conditions, including a fundamental rights and equality impact assessment, must be carried out by the employer, with the full involvement of trade unions and workers' representatives before any system is implemented and should be repeated regularly after implementation.
- Intrusive applications should be banned in the context of work. Applications to monitor workers shall only be allowed if their use is negotiated and agreed with trade unions and/or workers’ representatives;
- Algorithmic systems and AI should assist workers in the employment context. The human-in-command principle has to be defined and the rights of human decision makers have to be protected;
- Workers shall have the right to check and revise algorithmic decisions.
Algorithmic systems, especially Artificial Intelligence (AI) systems, offer immense opportunities for improving workplaces – for example opportunities to improve efficiency, fairness and worker safety. The increasing use of AI systems[1] in the workplace is ushering in a new wave of digitalisation that differs significantly from the previous one. Previous digitalisation was mainly characterised by technological innovations such as computerisation, automation and robotisation. This was based on automated processes through explicit rules and manually written computer programmes. Workers' experiences with this kind of digitalisation have been mixed. However, where trade unions and workers' representatives have helped to shape digitalisation in the workplace, it has led to create the necessary negotiated frame to guarantee workers’ involvement and control when operating digital tasks as well as providing additional qualifications and training to operate robots.
AI is different. It is highly disruptive, self-learning, can independently derive connections and make decisions. While logical "if-then" program steps were in principle comprehensible until now, AI can induce decisions processes that can no longer be explained by the programmers themselves after some time nor anticipated by developers. The danger of dehumanisation of decision-making processes, especially when used in works such as human resources tool for example to recruit workers, monitor work, analyse behaviour and even terminate employment, is already bitter reality.
AI poses enormous dangers to workers when unregulated.[2] In addition to massive surveillance, it can be used to recognise feelings and judge workers without considering the context or a performance. A classic example of this are AI-based applications that can determine the mood and emotions in contact centre agents' conversations and evaluate them as part of performance monitoring. Whether the respective worker is in a personally stressful situation (e.g., the death of a relative) is not taken into account in the evaluation. Thus, AI tools can quickly create a kind of the dehumanisation of the worker, transforming him/her into a purely economic commodity. AI can also make predictions about the performance of a worker. A promotion or even dismissal can therefore be based, at least in part, on a prediction of future performance of a worker and not on actual performance. Even more dangerous, such systems can be used for predictions about political attitudes, childbearing preferences and trade union membership.
AI and algorithmic management will also have a massive impact on the work of the future, individual activities in job profiles will change or disappear altogether, work organisation and work relationships will change with major implications for the work environment, working time and health and safety.
The ETUC has therefore been involved in the discussion on the regulation of AI from an early stage. Based on scientific work and evidence[3], the ETUC described the need for a comprehensive regulatory strategy in its Resolution on the European Strategies on Artificial Intelligence and Data[4] already in July 2020. In its White Paper on Artificial Intelligence - A European approach to excellence and trust[5], the European Commission itself pointed out the impact on the world of work: "In light of its significance for individuals and of the EU acquis addressing employment equality, the use of AI applications for recruitment processes as well as in situations impacting workers' rights would always be considered "high-risk" [...], the use of AI applications for the purposes of remote biometric identification and other intrusive surveillance technologies, would always be considered "high-risk" [...]".
Measured against the ETUC's requirements, the EU Commission's proposal on the draft AI Act[6] is more than disappointing from the workers' point of view. Although the EU Commission has defined AI systems used for hiring, promotion or dismissal as high-risk systems, such systems have to undergo only a self-assessment in which the software provider only certifies compliance with the law and standards yet to be defined. Neither trade unions nor workers' representatives have the possibility to get informed and consulted; nor are workers granted any protection or safeguards. A self-assessment is not enough to protect workers’ rights. In order to strengthen it, a third-party conformity assessment with trade union involvement is needed. Transparency rights are only granted to the end-users, who are the employers in the employment context.
In addition to a system based purely on self-assessment, the EU Commission has outsourced the design of everything from specific criteria to ethnic guidelines to private standard-setting organisations where companies and business-related organisations can buy influence and thus make the process purely business friendly.
The legal base, Article 114 of the TFEU, excludes the regulation of issues concerning the rights and interests of workers by definition. In principle, this would not be a problem if the EU Commission had limited itself to regulating the placing on the market within the framework of a simple product safety directive and not also the use of AI at the same time. The EU Commission makes it clear in recital 1 of the draft AI Act that national regulations that could restrict the use of AI systems are to be prevented:"[...] This Regulation pursues a number of overriding reasons of public interest, such as a high level of protection of health, safety and fundamental rights, and it ensures the free movement of AI-based goods and services cross-border, thus preventing Member States from imposing restrictions on the development, marketing and use of AI systems, unless explicitly authorised by this Regulation."
The ETUC strongly criticises this approach, as it is likely to restrict workers' rights. In the general formulation, for example, national occupational health and safety regulations or workers’ participation regulations that restrict the use of AI in the workplace could violate European law. The ETUC therefore strongly demands that the current legislative process should ensure that national regulations on the protection of workers' rights, such as the rights to information, consultation and participation or health and safety at work regulations, are respected.
Although the legislative process on the draft AI Act has not yet been completed, it is clear that the AI Act will not be suitable to ensure the effective protection of workers in the employment context due to its legal basis and its weak level of protection. The ETUC therefore calls for a lex specialis for the use of algorithmic systems in the workplace.
In its proposal for the draft Platform Work Directive (PWD)[7], the EU Commission has made the case for more worker participation and transparency in the use of automated monitoring and decision-making systems but has limited this to the scope of the Directive. However, automatic management systems are widely used outside the platform economy. Classic examples are big online retailer, that uses such systems extensively and in an exploitable manner in their warehouses. A new directive on algorithmic systems at work should build on the Posting of Worker Directive, close remaining gaps and apply to all workers.
For an EU directive on algorithmic systems in the workplace[8]
An EU directive on algorithmic systems in the workplace, based on Article 153 TFEU, should define European minimum standards for the design and use of algorithmic systems in the employment context. The scope of the directive should cover algorithmic systems with and without AI[9], must be as broad as possible by referring to "all algorithmic systems processing workers’ personal data, as well as affecting workers in the employment relationship or in matters of training or further education".
It goes without saying that one directive cannot solve all problems. It should try to address the most important issues to deal with the use of algorithmic systems at work. In addition, the social acquis must of course be adapted to the new digital challenges, be it through the enforcement of the right to disconnect or an urgently needed directive on psychosocial risks.
The guiding principle of the new directive on algorithmic systems in the workplace must be to preserve the dignity of workers and to counteract dehumanisation at work. The ETUC believes that the most effective tool is to strengthen collective bargaining of trade unions as well as information, consultation and participation rights of workers' representatives. This is the only way to ensure that the use of algorithmic systems has a positive impact on the worker's working conditions and qualifications. Therefore, the new directive does not have to start from scratch, but rather building on an already existing European acquis on workers’ information, consultation and participation rights. However, it must enforce these rights and ensure that trade unions and workers' representatives are effectively and timely informed, consulted and involved in decisions on the development, procurement, deployment, configuration and evaluation/review of algorithmic systems.
The involvement of trade unions and workers' representatives should ideally begin at the design and development stage of AI systems. Developers of algorithmic systems should consider issues of occupational health and safety and the protection of workers' rights as early as the code-writing stage, thus following a good work by design approach. Therefore, it would be necessary for developers of these systems to have prior knowledge, awareness and training not only on AI topics but also concerning ethical matters to avoid bias in the programming. Moreover, a worker and/or her/his representative must be able to redress a decision made by an algorithm or an AI system in order to correct possible bias. Furthermore, it must be ensured that algorithmic systems in the employment context have been subject to an independent fundamental rights and equality impact assessment, which explicitly includes non-discrimination and the protection of workers' rights. Independent review and complaints mechanisms must ensure compliance with the specific requirements in the workplace.
Participation rights must be designed in a process-oriented way. Trade unions and workers’ representatives must have the right to regularly review the AI system and demand adjustments or restrictions. For this purpose, the employer must be obliged to conduct an algorithmic impact assessment for changes in working conditions, including a fundamental rights and equality impact assessment, in cooperation and full involvement of the workers’ representatives, before the system is implemented. The fundamental rights and equality impact assessment must be repeated on a regular basis after the implementation.[10]
In order to meaningfully consult workers' representatives and trade unions, employers should be legally required to provide a complete and understandable overview on the algorithmic system in question, including:
- Which data it is trained in,
- Which data it processes,
- How the data is processed,
- Where, for what purpose and for how long it is stored,
- Who has access to the data,
- Which process and workplaces are affected directly or indirectly and in what form,
- How the requirements and stress profiles of jobs are changing
- Are qualification measures necessary and is sufficient time provided,
- How are requirement and stress profiles changing,
- Are new psychological or physiological risks emerging,
- Is the system non-discriminatory.
Trade Unions and workers’ representatives must have full access to the data and system information at all times, if required. In addition, they must be enabled to assess this data also by means of external expertise financed by the employer. The right to appoint external experts must be enshrined in the Directive.
Transparency and explainability are central to algorithmic systems. The ETUC is of the opinion that the directive should stipulate that the employer must inform workers in plain language about the use of algorithmic systems, such as AI, at the earliest possible stage. This should include but not be limited to information about the nature, task and scope of the systems, what data is processed, what output is produced and what the consequences are. The information should also include a reference to human involvement and details of the competent complaints body.
The ETUC is convinced that a directive on algorithmic systems in the workplace must address the different power relations between employer and worker. It cannot be assumed that the individual consent of the worker is a sufficient basis for the use of such systems in the workplace. Rather, a collective agreement with the relevant trade union and/or workers’ representative is required.
Data minimisation is a key principle of European data protection law and must apply especially to algorithmic systems in the employment context. At the same time, however, AI systems should be non-discriminatory and unbiased. The General Data Protection Regulation (GDPR) is already an effective resource. Trade unions and workers’ representatives should make full use of the possibilities offered by the GDPR, including the universal right to explainability in relation to high-risk algorithmic systems, with the right to ask for a personalised explanation. The Directive on algorithmic systems could concretise what can be considered as "specific rules" according to Art. 88 GDPR. This could serve as a basis for the implementation of this Article in all Member States. In relation to Article 8 of the GDPR, the legislative provisions that Member States have incorporated to ensure the protection of rights and freedoms in relation to the processing of personal data of workers in the workplace, in particular for the purposes of recruitment of personnel, execution of the employment contract, should be reviewed and adequately included. Nevertheless, even if the possibilities are fully exploited, there are still gaps in employment data protection. The ETUC is therefore in favour of including elements of workers’ data protection in the directive on algorithmic systems.
The prohibition on employers collecting certain data from workers must be effective enforced. This includes, for example, data outside working hours and/or the place of employment, data collected in connection with private conversations, especially conversations with workers' representatives, or in private rooms shall not be collected or processed by the employer. Strict prohibitions also apply to the collection or processing of any data that poses risks to human dignity or fundamental rights. This includes, in particular, data related to emotional and psychological state.
Algorithmic systems, especially AI, can be misused to monitor and supervise workers extensively. Abusive forms of surveillance must be prohibited. In particular, it must be ensured that, especially in times of increasing teleworking, work and private life are not mixed. Privacy, especially in one's own home, must be preserved under all circumstances. The employer must not be allowed to use algorithmic monitoring systems, especially AI-based ones, in the workplace. If a justified interest can be proven, a collective agreement and/or company agreement with trade unions and/or workers’ representatives can provide otherwise.
The ETUC advocates that certain highly intrusive AI systems should be banned in principle in the employment context. This includes, in particular applications that aim to make predictions and thus violate the fundamental rights of workers, including the right to join a union or to be judged on his/her political affiliation and participation. Fully automated decisions that affect the employment relationship of a worker should be prohibited as well. Algorithmic systems may at most be used as assistance systems. Decision must be taken by a human being according to a set of transparent and proportionate criteria, agreed upon with the trade union and/or workers’ representatives. A mere reference to the result of a software is not sufficient.
The directive must provide for a human centred decision process, as well as grant the human decision-maker the necessary protection, should he / she decide not to follow the outcome of an algorithmic system. Employers must be responsible for having effective procedures in place that fight “automatisation biases” among the human decision-makers.[11] This includes, among other things, ensuring that the human oversight receive the necessary training and have the necessary authority to carry out their role. To ensure this, it is essential that trade unions and workers' representatives are already involved in the selection process. In this context the directive could recall and specify the use of Art. 22 GDPR in the employment context.
For the ETUC, a directive on algorithmic systems must also address the issue of qualification of workers and workers' representatives. The directive must enable workers and their representatives to become 'AI literate': acquiring technical skills and using them 'at work', although necessary, is not enough and mostly serves the interests of one's employer. Becoming 'AI literate' means being able to critically understand the role of AI and its impact on one's work and occupation and being able to anticipate how it will transform one's career and role. Passively using AI systems does not benefit workers themselves - a certain distance needs to be established for them to see AI's overall influence.[12] The new directive should grant workers' representatives information, consultation and participation rights in the initiation and implementation of in-house training measures and the design of vocational training/qualification. At the same time, the directive must grant workers' representatives the right to employer-funded training in the use of algorithmic systems and AI. Additionally, employers should be trained in ethical matter related to AI systems and in the danger that the introduction of an AI system could cause in the workplace.
The directive must entail effective and dissuasive sanctions as well as an effective non regression clause and a more favourable clause. It should also take the different labour market systems taken into account and allow for national flexibility, if needed.
[1] For the purpose of readability, this resolution uses "AI" or "AI system" without specifying whether “AI” refers to machine learning, algorithmic (management) systems or another technology
[2] United Nations (2021): The right to privacy in the digital age. Report of the United Nations High Commissioner for Human Rights, Human Rights Council, 84th session, Agenda Items 2 and 3
[3] Ponce Del Castillo, A. (2020). Labour in the age of AI: why regulation is needed to protect workers. ETUI Research Paper-Foresight Brief; Ponce Del Castillo, A. (2018): Artificial Intelligence: a game changer for the world of work. ETUI Research Paper-Foresight Brief
[4] ETUC (2020), Resolution on the European strategies on artificial intelligence and data
[5] European Commission (2020), White Paper on Artificial Intelligence - A European approach to excellence and trust, COM(2020)65, p. 18
[6] European Commission (2022), Proposal for a Regulation of the European Parliament and the European Council on laying down harmonised rules on Artificial Intelligence and amending certain Union legislative acts, European Commission COM(2021)206
[7] European Commission (2021), Proposal for a Directive of the European Parliament and the European Council on improving working conditions in platform work, COM(2021)0762
[8] For the purpose of readability “AI" or "AI system" refers to algorithmic systems in general, regardless of the degree of real AI
[9] For the description of the need for regulation and the demands of the ETUC, it is irrelevant whether an algorithmic management system operates with or without real AI. However, the concrete level of protection for the worker may differ depending on the degree of use of real AI
[10] Adams-Prassel, Jeremias et all. (2021): Regulating Algorithmic Management, The Blueprint Proposals, non-paper
[11] Automation bias refers to a tendency of the human decision-maker to believe computer and their advice.
[12] Ponce Del Castillo, A (2022), Artificial intelligence: filling the gaps, Social Europe