Dr. Aideen Hartney:

The National Disability Authority welcomes this opportunity to appear before the Committee on Artificial Intelligence to discuss the impact of AI on the lives of persons with disabilities. The NDA is a statutory agency established to provide evidence-informed advice and guidance to the Government on disability policy and practice and to promote awareness and application of universal design. Universal design is the design and composition of an environment so that it can be accessed, understood and used to the greatest extent possible by all people regardless of their age, size, ability or disability.

The emergence of AI marks a transformative moment for disability inclusion in Ireland, offering unprecedented opportunities while raising important challenges which need careful consideration. In Ireland, as elsewhere, AI technologies have begun to make progressive impact on the lives of individuals with disabilities. They offer innovative solutions that have the potential to enhance accessibility and inclusivity across services, education and employment. AI technologies have been embedded in many everyday services and technologies for several years now.

These technologies already show significant promise in breaking down barriers to participation and eliminating common obstacles faced by people with disabilities. However, significant challenges lie ahead, many of which may only become apparent as AI systems are deployed in practice. Early recognition of these challenges is therefore essential to ensure AI genuinely serves and empowers people with disabilities rather than creating new barriers.

A fundamental concern centres on training data biases. Current AI systems reflect and perpetuate existing inequalities due to the significant under-representation of people with disabilities in their development, as well as the fact that many AI models are based on existing datasets where disabled people are also under-represented. Technical limitations compound these issues, as many AI interfaces continue to rely on modes of interaction that can exclude users with certain disabilities. Implementation risks present a third challenge where improperly deployed AI systems could amplify existing inequalities rather than alleviating them. This is particularly crucial in public services where AI systems may influence access to essential services and supports, indicating that the use of AI in these services will require rigorous oversight and continuous evaluation to ensure it enhances, rather than hinders, accessibility and inclusion.

The European AI Act establishes a comprehensive framework for AI regulation, introducing a tiered approach that categorises systems into four designated risk levels: unacceptable, high, limited and minimal. We welcome that the Irish State has designated nine national authorities with responsibility for protecting fundamental rights. We advise the importance of these authorities in engaging closely with disabled people and their representative organisations to ensure a universal design approach to managing and mitigating the potential risks. The committee is likely aware that almost 22% of the Irish population reported a disability of some kind in census 2022. This is a substantial segment of the population. We also note the committee's engagement on the subject of older persons at a recent hearing.

The NDA advises that ensuring the principles of universal design underpin the development and deployment of AI will benefit these population cohorts and others, given the focus on access, understanding and use for all users. Ideally, public bodies would work to achieve universally designed AI in their services, moving beyond mere compliance with the legislation. This will mitigate risks to the greatest extent and ensure sustainable and beneficial growth of AI into the future. We therefore recommend ongoing monitoring of the impact and outcomes associated with its deployment for disabled people and other marginalised groups within our population. Universal design is also based on codesign and coproduction principles and we advise the importance of AI service providers meaningfully involving disabled people during design and training phases, where biases can be more easily identified and resolved. Ideally, AI systems should offer multiple modes of interaction that genuinely accommodate diverse user needs and preferences rather than treating accessibility as an afterthought.

On the policy side, it will also be important that plans are made to guard against unintended consequences of AI use, where marginalised groups within society can be more impacted. For disabled people, for example, AI can increase employment opportunities, particularly for those with higher levels of educational attainment. For others, however, AI could potentially displace employment opportunities. Studies have shown that disabled people can have lower levels of educational attainment and so are at greater risk of loss of jobs that can be easily replaced by AI and other forms of automation. Application of AI in public services is also dependent on levels of digital literacy within the overall population. As we have just heard, there are many groups in society that have lower levels of digital literacy, including disabled people. Related but separate is the cost of access to the technology necessary for effective use of AI. The 2021 cost of disability report prepared by Indecon showed the extent of the additional draw on budgets associated with having a disability, which, when combined with the higher rate of poverty in this population, means the financial implications of increased use of AI in public services cannot be disregarded.

While there are many opportunities for AI increasing participation of disabled people in the socioeconomic life of the State, careful management and ongoing monitoring is required to ensure that its roll-out does not perpetuate existing inequalities or create new ones. A commitment to a universal design approach is one route to mitigating this risk, as is building capacity among public services, disabled people and AI developers.

We also suggest adopting a structured assessment framework to support implementation of the AI Act and would be happy to speak more about this in the remainder of this session.