In the New Yorker, Lawrence Wright discusses a meeting with Hamid Gul, the former head of the Pakistani secret service I.S.I. In his time as head, Gul channeled the bulk of American aid in a particular direction:
I asked Gul why, during the Afghan jihad, he had favored Gulbuddin Hekmatyar, one of the seven warlords who had been designated to receive American assistance in the fight against the Soviets. Hekmatyar was the most brutal member of the group, but, crucially, he was a Pashtun, like Gul.
Gul offered a more principled rationale for his choice: “I went to each of the seven, you see, and I asked them, ‘I know you are the strongest, but who is No. 2?’ ” He formed a tight, smug smile. “They all said Hekmatyar.”
Gul’s mechanism is something like the following: Each player is allowed to cast a vote for everyone but himself. The warlord who gets the most votes gets a disproportionate amount of U.S. aid.
By not allowing a warlord to vote for himself, Gul eliminates the warlord’s obvious incentive to push his own candidacy to extract U.S. aid. Such a mechanism would yield no information. With this strategy unavailable, each player must decide how to cast a vote for the others. Voting mechanisms have multiple equilibria but let us look at a “natural” one where a player conditions on the event that his vote is decisive (i.e. his vote can send the collective decision one way or the other). In this scenario, each player must decide how the allocation of U.S. aid to the player he votes for feeds back to him. Therefore, he will vote for the player who will use the money to take an action that most helps him, the voter. If fighting Soviets is such an action, he will vote for the strongest player. If instead he is worried that the money will be used to buy weapons and soldiers to attack other warlords, he will vote for the weakest warlord.
So, Gul’s mechanism does aggregate information in some circumstances even if, as Wright intimates, Gul is simply supporting a fellow Pashtun.