Researcher: Dr Nigel Crook
Understanding how an AI system comes to a result is essential if people are willing to trust an AI system. Nigel Crook has been working on systems that can help people interpret the outputs of neural networks for over 30 years. He is currently working on a tool that can assess the amount of skewness and kurtosis (sharpness) of peaks in probability distribution curves of the answers arising from deep learning neural networks. This tool will enable people to make assessments of the following:
- what factors are influencing the decision
- if the distribution of the decisions are unbalanced or skewed to certain factors
- whether the base of a particular decision parameter is too narrow, making the decision likely to be unstable and the system brittle and lacking in robustness.