On 21 April 2021, the European Commission presented its long-awaited proposal on the regulation of artificial intelligence. The regulation is the follow-up to the 2020 White Paper process, in which the ETUC also participated. The proposed regulation builds on the internal market rules regulating the development and placement of products and services using AI in the EU single market. The regulation does not address liability. In this context, the Commission refers to the revision of the Product Liability Directive which is foreseen in the second half of the year.
The proposed EU regulation on AI fails to address the workplace dimension. For Europe to become a true leader in “Trustworthy AI”, the regulation should protect and cover all workers and employers in the private and public sectors, including online platforms. The imbalance of power between employers and workers should lead the EU Commission to ensure that trade unions and workers’ representatives participate actively in the building of AI at work, which is essential to achieve a robust AI framework that guarantees the protection of workers’ rights, quality jobs, and investment in worker’s AI literacy.
In regulating AI, the high-risk approach is questionable. The EC classifies risks as unacceptable, high, limited and minimal. The regulation addresses the high-risk of AI systems that can create an adverse impact on people's safety or their fundamental rights (as protected by the EU Charter of Fundamental Rights). However, risks may vary depending on the use of a given application. The AI regulation lists high-risk applications that are subject to certain mandatory requirements and calls for conformity assessment procedures as a part of those obligations. ETUC emphasises that these assessments will not provide the neutrality required for a sound evaluation of the potential implications of AI system. At the minimum, the conformity assessment of AI systems used for ‘Employment, workers’ management and access to self-employment’ must be carried out by an authorised third party. A combination of ex-ante compliance and ex-post enforcement mechanisms is recommendable. Additionally, regulatory sandboxes should not be allowed for AI systems to be deployed at the workplace.
The Commission gives importance to standardisation insofar as technical standards will be used to demonstrate compliance with the essential requirements set out and required in the Regulation. However, it is widely acknowledged that most national and European standardisation organisations are characterised by an important democratic deficit. The ETUC advocates for the adoption of standards at European level, as a contribution to a sound regulatory framework, and not simply to adopt international standards. The latter approach would risk lacking ambition and not taking sufficient account of European values and specificities. The ETUC further insists on the autonomy of social partners and on the respect of collective agreements and social dialogue that could potentially be challenged by standards.
The AI regulation only considers a limited number of AI applications such as recruitment, task allocation and evaluation of workers. ETUC believes that any AI system implemented in the workplace and the data selected to contribute to the system should be considered high risk and be subject to the scrutiny of competent authorities and trade unions through the established legislation. There are AI applications whose implementation at the workplace may appear to be harmless but the deployment of which (alone or in combination with other applications) can hinder workers' privacy, health, wellbeing and other fundamental rights at work. These cases are absent in the proposed regulation. When AI systems are to be integrated in any degree at the workplace, the existing legal framework on the right of workers to information, consultation and participation shall be part of the mandatory compliance obligations.
The AI regulation calls AI developers to tackle the opacity of AI systems by requiring a “certain degree” of transparency. However, when AI systems are integrated in the workplace, the AI regulation should include mandatory obligations for the employer, prevention measures including “AI literacy”, so that workers are not only fully protected but also able to understand the role of data and AI, its impact on the organisation of their work, be critically aware about the role and impact of working with AI systems, and react when possible harms appear.
Emerging technologies embed uncertain and unknown risks, that is why the precautionary and preventive approach needs to be part the regulatory framework. Adhering to the legal Precautionary Principle set up by the TFEU ensures that Europe secures and reinforces AI via its fundamental rules and values.
The governance of AI should be a democratic process. ETUC welcomes the safeguards enshrined in the AI regulation to guarantee the coherence between the European Data Protection Board and the European Artificial Intelligence Board. However, to ensure a democratic governance system, that the rights of workers are protected when AI systems are implemented at the workplace, trade unions should be part of the governance of the board. Any sort of occasional consultation in the form of an expert group in which trade unions may be consulted on issues appertaining to workers’ rights together with a plethora of other organisations will result in the deterioration of the existing workers’ information and consultation rights and this without prejudice to the need for trade unions to be consulted at the sector or at company level when AI technologies are implemented at the workplace. Moreover, the AI board proposed by the EC regulation does not provide for the independency needed for the monitoring of AI systems under the compass of the European Charter of Fundamental Rights
ETUC demands for the AI regulation
1) The AI regulation should address the need for inclusive and democratic governance and clear rules securing great protection of workers.
2) Classify AI applications impacting workers’ rights and working conditions as high-risk and subject to appropriate regulation.
3) Guarantee AI systems in which humans remain in control, which is compliant with labour rights and a sound use of personal data. Trade unions and workers’ representatives must be key actors in developing and implementing AI systems.
4) Strengthen the application of GDPR to the workplace reality including the active involvement social partners to strengthen industrial democracy.
5) Social dialogue structures, collective bargaining, information, consultation and participation of trade unions and workers’ representatives are key to providing the necessary support for workers to better build and be part of the uptake and monitoring of AI used at the workplace.
6) Guarantee the application of the precautionary principle should be a core action to tackling uncertain AI risks.