Disabled community fear AI will be used to cut costs and replace human supports, Dáil committee hears

The Deaf Society's Elaine Grehan pointed out that AI avatars cannot do sign language properly.

Artificial intelligence (AI) systems used to help people with disabilities could end up being biased against the very same cohort, TDs and senators have been warned.

The Irish Deaf Society's Elaine Grehan, of the Disabled Persons' Organisations (DPO) Network, told the Oireachtas AI Committee that studies in the UK have already shown gender bias in care decisions have emerged.

Ms Grehan also warned about the fear among the disabled community that AI systems could be introduced as a way of cutting costs and eliminating jobs carried out by humans.

She pointed out while there may well be a rise in the use of AI avatars, they cannot actually do sign language properly.

Ms Grehan said: “When it comes to sign languages, eyebrows have meaning, your cheeks have meaning, your lip patterns — they all have meaning that all convey information.

“[AI avatars] don't have those features, information isn't being conveyed."

She said disabled people understood all too well that AI-powered assistive technology provided a wide variety of support inside and outside the home.

But she said: “We are concerned there is little transparency on how AI is working and how it will be controlled for human benefit or for profit. Disabled people and deaf people must have a role in shaping how AI is developed now and into the future.”

AI is not neutral. Disabled people fear AI will reflect existing institutional and structural barriers and attitudes, including biases relating to gender, ethnicity and impairment, unless there is clear transparency to ensure that structural biases are not built into these systems.

She referenced a recent study by the London School of Economics’ Care Policy and Evaluation Centre in the UK into advanced AI programmes known as large language models (LLMs). They are “trained” on vast amounts of text data to not only understand, summarise, translate, and generate human language, but also to make decisions.

The study found LLMs used by more than half of England’s local authorities to support social workers may be introducing gender bias into care decisions.

“We are afraid that AI and technology will be seen as a cost-effective replacement for human supports such as personal assistance services," Ms Grehan said.

“We fear decisions that require human analysis or instinct will be made by AI that cannot manage the nuances of life. We need to control and direct AI to meet our needs, not have AI direct and control our lives.”