Ms Jean Carberry:

We may be conflating two different things. The social media piece is under the Digital Services Act. It is not AI per se, and the Digital Services Act is about what is legal or illegal, and it has to be that way to give certainty because it would be very difficult to have the grey areas of what is good and what is bad. That is the way the Digital Services Act 2024 was constructed.

On the AI side, the high-risk areas are areas in which fundamental rights are set out. To boil it down very simply, as Ms Patricia Scanlon sometimes does, it is a common-sense thing of one's right to life or well-being. A lot of public services are in a high-risk sphere. In the context of justice and your right to liberty, gardaí are in a high-risk space using AI because they can impede on people's right to liberty. In the private sphere, recruitment is in this space because of your right to a job and your right to access employment. Bias is inherent to AI because there is bias from the programme or from the data set. A good way to illustrate this is by looking back at history and crunching the numbers around what makes a good doctor. AI would probably tell you, looking back over those 100 years, that being male is one of those things because 99% of doctors in the history of time have been men. The algorithm would conclude this is one of the criteria. This is a simple example of how bias can happen and why some human intervention is needed. Recruitment is one example in the private sphere in which a human right could be breached. Medical devices are another example as they impact your health and are therefore high risk if they use AI. Health insurance and life insurance are other areas. To make sense of it, it concerns things that actually impact.