I tend to use 50 years as my rough "long term" definition. "Facebook won't be around in 50 years" is not very impressive. "Facebook won't be around in 2020" would be a bold prediction, but doesn't seem very "long term". "Facebook won't be around in 2025" might be starting to get into "long term", but is also definitely starting to feel vacuous if we think of it as "The probability of Facebook being around as a distinct entity in 2025"... well, no duh it's not going to be 100%, but the probabilities get fairly uninteresting quickly. 50/50 is a decent initial guess for that, and it isn't until you get into the 90s on either side that it starts becoming an interesting prediction.
I don't know how to turn "Facebook won't be around in the long term" into an interesting prediction without a time horizon.
Here's a non-vacuous sort of example: There's a proof somewhere that the maximal probability of the expected future lifetime of a given organization like a country or a company is its current lifespan, projected into the future. So the maximal probability of when Facebook will be gone if it is currently 14 years old is 2032. However, I can not seem to Google up the discussion of this; all the search terms I can come up with are flooded by actuarial discussions of human mortality where this most assuredly is not the case.
If you make the naive assumption that you are observing the company at a random point in its overall lifespan, the median of the distribution for how much longer it will survive is how long it has survived already.
More than that, if it has survived N years already, the odds that it will survive at least M more is N/(N+M). If you work out the probability density and then integrate it, you come up with an arithmetic mean that is also N. However the point where it is most likely a priori to die is tomorrow, and that probability steadily decreases.
> There's a proof somewhere that the maximal probability of the expected future lifetime of a given organization like a country or a company is its current lifespan, projected into the future
I tend to use 50 years as my rough "long term" definition. "Facebook won't be around in 50 years" is not very impressive. "Facebook won't be around in 2020" would be a bold prediction, but doesn't seem very "long term". "Facebook won't be around in 2025" might be starting to get into "long term", but is also definitely starting to feel vacuous if we think of it as "The probability of Facebook being around as a distinct entity in 2025"... well, no duh it's not going to be 100%, but the probabilities get fairly uninteresting quickly. 50/50 is a decent initial guess for that, and it isn't until you get into the 90s on either side that it starts becoming an interesting prediction.
I don't know how to turn "Facebook won't be around in the long term" into an interesting prediction without a time horizon.
Here's a non-vacuous sort of example: There's a proof somewhere that the maximal probability of the expected future lifetime of a given organization like a country or a company is its current lifespan, projected into the future. So the maximal probability of when Facebook will be gone if it is currently 14 years old is 2032. However, I can not seem to Google up the discussion of this; all the search terms I can come up with are flooded by actuarial discussions of human mortality where this most assuredly is not the case.