Once a position is identified with a side, every argument for it becomes a soldier to be defended and every concession a casualty, so weak arguments are held rather than dropped.
Yudkowsky's point is that the military metaphor is doing the damage: on a battlefield you do not abandon your own troops, so admitting that one of your arguments is bad feels like a defection rather than a correction. The result is that the weakest supporting argument is defended hardest, because it is the one under attack.
In practice
- A colleague who will not concede a minor factual point has correctly judged that conceding it will be treated as a retreat.
- A team that drops its own bad argument is regarded by observers as losing, which teaches everyone not to.