The cost function analogy for optimization may not always be useful as systems with multiple or dynamic cost functions may not be easily optimized with gradient descent algorithms.