EU Artificial Intelligence Act – recommendations on accessibility
April 2022
Why accessibility of Artificial Intelligence (AI) technologies is important for persons with disabilities.
Supporting inclusion of persons with disabilities instead of creating barriers through AI
Emerging technologies such as Artificial Intelligence (AI) can support equal participation and inclusion of persons with disabilities, helping them overcome societal barriers. AI-enabled tools are already being used to help people with vision, hearing, mobility and learning disabilities. Areas such as predictive text, visual recognition, speech-to-text transcription, and captioning have undergone great advances in the last few years, allowing greater access and independence for persons with disabilities.
Especially in an increasingly digitalised world, affordable and accessible technologies become a gateway to participate for persons with disabilities on an equal footing with others in employment and education, to access public and private services, communicate with others and access information, and enjoy culture and leisure. At the same time, inaccessible AI systems and their deployment can lock persons with disabilities out of social participation, infringe on their human rights and further increase inequality gaps. Examples of inaccessible AI-based technologies include certain solutions for speech recognition, smart assistants, or crowdsourcing websites for remote working not developed with persons with disabilities and accessibility in mind.
Importance of accessibility of AI systems and their use for persons with disabilities
The High-Level Expert Group on AI has noted in its 2019 Ethics Guidelines for Trustworthy AI “AI systems should not have a one-size-fits-all approach and should consider Universal Design principles addressing the widest possible range of users, following relevant accessibility standards. This will enable equitable access and active participation of all people in existing and emerging computer-mediated human activities and with regard to assistive technologies.” It is important here to note that accessibility of AI extends beyond the needs of more than one hundred million persons with disabilities in Europe, and is also of great relevance for rapidly ageing societies in the region.
Considering the great potential of digital technologies for persons with disabilities, the UN Convention on the Rights of Persons with Disabilities (CRPD) has been the first international human rights instrument that recognises the need to promote equal access to information and communication technologies and systems as a fundamental right for persons with disabilities (article 9). As State Party to the Convention, the European Union and all Member States are bound by this international treaty to develop and enact laws that advance accessibility of digital technologies for persons with disabilities, including through the EU Artificial Intelligence Act.
Building on existing Union accessibility legislation to ensure accessible AI systems
Currently, the most comprehensive Union legislation on accessibility is the European Accessibility Act (Directive (EU) 2019/882). The Accessibility Act (EAA) applies to a limited scope of products and services[1] and does not address AI systems directly. In the meantime, the accessibility requirements detailed in Annex I of the EAA provide the legal clarity on how to ensure accessibility of digital products and services, and can be easily applied to AI systems. In fact, the EU legislators have foreseen usefulness of the EAA for ensuring accessibility of services and products under other (including future) EU acts, therefore have established in Article 24 of the EAA that any product or service, the features, elements or functions of which comply with the accessibility requirements set out in Annex I of the EAA shall be presumed to fulfil the relevant obligations set out in Union acts other than the EAA, as regards accessibility, for those features, elements or functions, unless otherwise provided in those other acts. Since then, reference to the EAA is a well-established practice in EU law, examples of which can be found in Regulation (EU) 2021/782 (recast Rail Passengers’ Rights Regulation); EC proposal for amending Regulation (EU) No 910/2014 as regards establishing a framework for a European Digital Identity; EC proposal for a regulation on the deployment of alternative fuels infrastructure; EU Funds Regulations, among others.
The gaps in the Artificial Intelligence Act (AIA) and suggestions for improvement.
In its current version, the AIA lacks mandatory accessibility requirements for AI providers and users.
The proposal states that providers of non-high-risk AI systems may create and implement codes of conduct which may include voluntary commitments, including related to accessibility for persons with disabilities (Recital 81, Art 69. (2)).[2] There are no accessibility requirements for users of AI. This is an inadequate approach to disability; it falls short of obligations laid out in the UN CRPD and is inconsistent with existing EU legislation such as the European Accessibility Act and Council Directive 2000/78/EC on equal treatment in employment and occupation, which ensures non-discriminatory access to employment for persons with disabilities, including in sectors where AI technologies are used.
Accessibility should be required for all AI systems and their use, irrespective of perceived or actual level of risk (‘high-risk’ or ‘non-high-risk’) because lack of accessibility can jeopardise life, independence, and well-being of persons with disabilities. For example, if a human oversight system and interface of an AI-based self-driving car is not accessible, this can put the passenger with disability in great danger. Similar risks can occur when trying to operate home appliances through smart home and Internet of Things (IoT) technologies, which would normally be considered as “non-high-risk”.
It is crucial to refer to the legal accessibility requirements of the EAA in the AI Act to ensure accessibility of AI systems through legal coherence and certainty for developers and users of AI.
[1]See article 2 (Scope) of Directive (EU)2019/882
[2] Important issues such as environmental sustainability, stakeholders’ participation in the design and development of AI systems, and diversity of development teams are also suggested in the AIA as merely voluntary measures.
This paper was drafted by and with the support of the European Disability Forum (EDF), Access Now, AlgorithmWatch, Bits of Freedom, European Center for Not-For-Profit Law, European Digital Rights (EDRi), Fair Trials, Homo Digitalis, and Panoptykon Foundation.