Akrasia is weakness of will: doing what you judge you should not, while judging it. The Greek means lacking command, and the puzzle is not whether it happens but how it can.

Socrates denies it

Socrates argued that nobody knowingly does wrong. If you genuinely judge one course better and still take the other, you did not really have the knowledge — you had an opinion overwhelmed by the appearance of pleasure.

The position follows from treating virtue as a kind of knowledge, and it has the merit of consistency. Its cost is that it denies something everyone believes they have experienced.

Aristotle's analysis

Aristotle wanted to save the phenomenon. His account distinguishes having knowledge from using it in the moment. The akratic person holds the right general principle and fails to apply it to the case in front of them, because appetite has made the particular vivid and the principle abstract.

He also distinguishes the akratic from the intemperate. The intemperate person acts wrongly and endorses it; the akratic acts wrongly against their own view and regrets it. Only the second is torn, which is what makes it a puzzle about the structure of agency rather than about bad character.

Why it still matters

It is the clearest case where the picture of a person as a single reasoning agent breaks down. If you can want what you judge you should not want, then either judgment and motivation are separable, or the self is not unified enough for the question to have a clean answer.

Modern treatments frame it through discounting — see present bias — where the preference genuinely reverses as the moment approaches. That explains the behavior without requiring anyone to act against a judgment they hold at the time, which some philosophers regard as dissolving the puzzle and others as changing the subject.

In practice

Every commitment device is a response to it: deciding in advance and making the decision expensive to reverse, because the person making the plan and the person executing it cannot be relied on to agree. See the promise ratio.