Mr. Tim Culhane:

Voice of Vision Impairment is Ireland's disabled persons representative organisation representing the human rights of blind and visually impaired people. I thank the committee for inviting us here today to give our views and perspectives on artificial intelligence as it relates to visually impaired people and their human rights. Before I start, I will thank my colleague, Martin O'Sullivan, who cannot be here today. He put a lot of work into researching this presentation. Unfortunately, he has serious ill health at the moment. I just wanted to put that on the record.

I will start with a real world example of how unregulated AI can have a negative effect on people in minorities in general. In 2023, iTutor, an American tutoring company, had to settle a case with the American employment commission for over $365,000 because it had used AI software in recruitment that rejected people based on age and gender. In particular, all women over the age of 55 and all men over the age of 60 were removed from the recruitment process. While there is no certainty that visually impaired people or disabled people in general were discriminated against, it is likely that such discrimination is occurring already.

There is no doubt that AI has the potential to be a game-changer. It is already benefiting visually impaired people significantly in many areas of our lives. I will highlight a few of those. There is the ability to read and process inaccessible material such as menus, signs and printed documents. It allows us to quickly access essential information such as emergency contact numbers and transport software updates. It makes shopping much more accessible by describing and identifying products, doing price comparisons and reading important information, such as nutritional and cooking information, that would not have been available to us thus far. In education, it unlocks accessibility to printed archives, books and other research materials.

However, there is no doubt that, if left unregulated, AI could introduce a number of significant risks for visually impaired people. We would argue that visually impaired people are proportionately more exposed to these risks. I will highlight a few of those potential issues for us. The first is privacy and data security. Many visually impaired people are using AI tools to photograph and scan sensitive documents such as bank statements, medical records and so on. This leaves us vulnerable to data leaks and identity theft. It is important that there is legislation to ensure GDPR compliance, data minimisation and the availability of secure offline AI tools.

The second major issue we see with AI, which was touched on earlier by DFI and the National Disability Authority, is affordability and accessibility. Many of the most beneficial and hence most useful features of AI are based on a subscription model. This will automatically exclude a significant proportion of people with vision impairment, given the low-levels of employment in the vision-impaired community. Affordability is already a barrier to equality. We recommend the introduction of a non-means-tested disability allowance to enable access to essential AI technology.

The next issue is bias and discrimination. This was well described earlier by my colleagues in that the AI models are using data sets that are inherently biased towards non-disabled people. If we use those, we will be perpetuating discrimination because we will not be representing disabled people and visually impaired people from that perspective. Therefore, we need to reverse that exclusivity. Otherwise AI will actively magnify the existing inequalities we see in society as a whole.

Lastly, I want to point out accountability and transparency. AI systems are often opaque black boxes when it comes to high-stake decisions such as those relating to employment, welfare and healthcare. This lack of explainability is really unacceptable from anyone’s perspective. We recommend that AI systems used in such contexts provide transparent, auditable decision-making and proof of non-discrimination on the nine grounds set out in Irish equality law.

To sum up our recommendations and views on AI as a whole and what we think needs to be done, it goes without saying that mandatory consultation is needed with disabled persons and organisations when it comes to AI policymaking, strict privacy and data protection standards, GDPR compliance and the option for offline tools, proof of non-discrimination in AI use in employment and service delivery, accessibility by design in all AI systems and not just as an afterthought, affordability measures to include a cost-of-disability allowance, transparency and explainability requirements for AI in high-stakes decisions.

We have all seen the negative repercussions of unregulated social media over the past number of years. I am sure many committee members have experienced the negative effects. It is really important that we do not make those mistakes when it comes to AI and that we get out ahead of the game and regulate.