There have been questions about the reliability of computer evaluation, and famous cases in which computers miss-evaluate positions have been given. What I want to ask is more related to the practical aspect of the evaluation, rather than its effectiveness.
As humans, we may find moves that put our opponents in deep trouble, or lead to positions we may personally enjoy, even though computers may label this moves as blunders. A classical example is Tal's knight sacrifice in the sixth game of their match. Humanly speaking, the move may be considered brilliant due to the creative inspiration it shows, and the practical effects of it (for the position gets complicated and rich in tactical motives, in which Tal felt comfortable). To a computer, though, it is disastrous; artificial intelligence can not feel discomfort nor prefer certain positions above some others, and quickly finds the correct refutation to the sacrifice.
In a practical sense, for a player looking to eliminate blunders (which is a goal dictated to many beginners), what does this actually mean? Is it correct to label as a blunder all refutable moves that severely worsen one's position, as engines do? Or is it more appropriate to use the word blunder in a more pragmatical sense, by referring to refutable moves that severely worsen one's position, and whose refutation can be actually achieved by a human opponent? In other words, how pure should the concept of blunder in chess be?
I know the question is somewhat philosophical, and I do not expect definite answers.