Dr. Michelle O'Sullivan:

The difficulty at the moment is that regulation is generally very far behind internationally. People had an assumption that AI would be biased. There is now lots of research showing that while equality legislation is useful, the information being fed into systems in recruitment is itself biased. Therefore, the algorithmic management systems are themselves biased. We know from research that AI systems are more likely to select men for certain types of jobs, for example, because they are picking up on particular words men are more likely then women to use in CVs. That is because of the information the systems are being fed and what they are looking for. Legislation is only useful if people know and can identify the source of the problem to begin with. The complexity of AI systems means we do not know what they are deciding or what information they are being fed. That is why having people involved at the development stages of technology, before it is introduced in the workplace, is important, rather than that happening after a decision has been made, when it is too late to know what is being done. There are some very useful proposals on these issues. I am happy to forward some of them to the committee if members feel they might be useful.