(no subject)
Nick Bostrom, Information Hazards
Suppose that hundreds of rock fans are driving to the Glastonbury music festival. At some point each driver reaches an intersection where the road signs have been vandalized. As a result, there is uncertainty as to whether to turn left or right. Each driver has some private information, perhaps a dim drug‐clouded recollection from the previous year, which gives her a 2/3 chance of picking the correct direction. The first car arrives at the intersection, and turns right. The second car arrives, and also turns right. The driver in the third car has seen the first two cars turn right, and although his private intuition tells him to turn left, he figures it is more likely that his own intuition is wrong (1/3) than that both the preceding cars went the wrong way (1/9); so he turns right as well. A similar calculation is performed by each subsequent driver who can see at least two cars ahead. Every car ends up turning right.
In this scenario, there is a 1/9 chance that all the rock fans get lost. Let us suppose that if that happens, the festival is cancelled. Had there been a dense fog, preventing each driver from seeing the car in front (thus reducing information), then, almost certainly, approximately 2/3 of all the fans would have reached Glastonbury, enabling the festival to take place. Once the festival starts, any lost fan can hear the music from afar and find their way there. - We could thus have a situation in which reducing information available to each driver increases the chance that he will reach his destination. Clear weather creates an informational cascade that leads to an inefficient search pattern.
More information is not always better - particularly when we have mistaken ideas about how to interpret and use that information.
Suppose that hundreds of rock fans are driving to the Glastonbury music festival. At some point each driver reaches an intersection where the road signs have been vandalized. As a result, there is uncertainty as to whether to turn left or right. Each driver has some private information, perhaps a dim drug‐clouded recollection from the previous year, which gives her a 2/3 chance of picking the correct direction. The first car arrives at the intersection, and turns right. The second car arrives, and also turns right. The driver in the third car has seen the first two cars turn right, and although his private intuition tells him to turn left, he figures it is more likely that his own intuition is wrong (1/3) than that both the preceding cars went the wrong way (1/9); so he turns right as well. A similar calculation is performed by each subsequent driver who can see at least two cars ahead. Every car ends up turning right.
In this scenario, there is a 1/9 chance that all the rock fans get lost. Let us suppose that if that happens, the festival is cancelled. Had there been a dense fog, preventing each driver from seeing the car in front (thus reducing information), then, almost certainly, approximately 2/3 of all the fans would have reached Glastonbury, enabling the festival to take place. Once the festival starts, any lost fan can hear the music from afar and find their way there. - We could thus have a situation in which reducing information available to each driver increases the chance that he will reach his destination. Clear weather creates an informational cascade that leads to an inefficient search pattern.
More information is not always better - particularly when we have mistaken ideas about how to interpret and use that information.