The Application of Evolutionary Computational Methods to Mitigate Unwanted Bias in AI Systems

This work is being undertaken by Zahid Irfan.

Zahid is looking to make artificial intelligence (AI) systems more fair for people. Today AI is used in many areas like jobs, banks, and healthcare, but sometimes it can give biased or unfair results. This can happen because of problems in data or how the system is built.

In this research, the focus is to understand where this bias is coming from and how to reduce it. Instead of only fixing data or model, this work looks at how different factors are connected using  causal analysis. This helps to see which factors really affect the result and which ones are causing unfairness.

Then a new method is created by combining this idea with evolutionary computation, which is inspired by natural selection and evolution. The system tries many possible solutions and improves them step by step to find better and fairer results.

The results show that it is possible to reduce bias while still keeping good accuracy. This work helps in building AI systems which are more fair, understandable, and responsible.

Zahid's project is supervised by Dr Róisín Loughran, Dr Muhammad Adil Raja and Prof. Fergal McCaffery.

This project is funded through the Technological University Transfer Fund (TUTF) and DkIT.