This just sounds like minimizing a loss function. by this definition, aren't most machine learning algorithms in constant suffering, which they seek to reduce?