When talking about errors arguing about error probabilities is not enough. One need to take into account costs of error. A better metric would be "expected cost

When talking about errors arguing about error probabilities is not enough. One need to take into account costs of error. A better metric would be "expected cost of error" that multiplies error probabilities by costs of errors (and sums them up). If a system has 0.1% rate of error placing a pizza order it could be deemed OK. If it kills a patient 0.1% of the time it is unacceptble.