Published 14 April, 2020; last updated 15 September, 2020
The only times gross world product appears to have doubled in n years without having doubled previously in 4n years were between 4,000 BC and 3,000 BC, and most likely between 10,000 BC and 4,000 BC.
A key open question regarding AI risk is how quickly advanced artificial intelligence will ‘take off’, which is to say something like ‘go from being a small source of influence in the world to an overwhelming one’.
In Superintelligence1, Nick Bostrom defines the following answers, seemingly in line with common usage:
However the specific criteria for takeoff having occurred are generally ambiguous.
Paul Christiano has suggested2 operationalizing ‘slow takeoff’ as,
There will be a complete 4 year interval in which world output doubles, before the first 1 year interval in which world output doubles. (Similarly, we’ll see an 8 year doubling before a 2 year doubling, etc.)
We were interested in whether anything faster than a ‘slow takeoff’ by this definition would be historically unprecedented. That is, we wanted to know whether whenever the economy has doubled in n years, it has always completed a doubling in 4n years or less before the beginning of the n year doubling.
We took historic gross world product (GWP) estimates from Wikipedia3 and checked at each date how long it had taken for the economy to double, and whether it had always at some point doubled in as few as four times as many years prior to the start of that doubling.4
We found two apparent examples of faster takeoffs, so defined:
The 300 year period immediately after 1300 saw a doubling of GWP growth, and the 1200 years beforehand did not see a doubling, however there was an earlier doubling within the 1200 years ending at 1200AD. So this is not technically an instance, but was a case of briefly accelerating growth. GWP between 1100 and 1300 actually declined though, so this is perhaps a different kind of case to the ones we are interested in.
Corresponding author: Daniel Kokotajlo