By their nature AI algorithms tend to give digital answers that can appear to be very precise. However, we may have more or less confidence in the output of the algorithm, maybe because we know the training data is limited, the sensor input noisy or the algoritmm itself poor at generalisation. In some cases algorithms can provide a confidence measure alongside their output. This can be used in the user interface to enable users to make their own assessment of the outputs. It can also be used by an automated system to make a trade-off on a ROC or to ask a human for assistance.