[extropy-chat] Inside Vs. Outside Forecasts
Eliezer S. Yudkowsky
sentience at pobox.com
Wed Oct 12 02:03:09 UTC 2005
Adrian Tymes wrote:
> --- "Eliezer S. Yudkowsky" <sentience at pobox.com> wrote:
>>Moral 1: The outside view always wins.
> But only so long as the outside view uses relevant data.
> Example: predicting personal computing, back in the '60s and early
> '70s. Citing a large and growing number of industrial computing users,
> each and every one requiring a support staff dedicated solely to
> keeping their computers up, industry leaders felt confident in
> predicting that computers small yet functional enough for a private
> citizen to use, without said support staff, would never come to be.
> I'm typing this on an example of that prediction's failure.
Let me rephrase: The outside view always wins over the inside view.
Naturally, predictions sometimes fail. It's what they do. Have you any
evidence that, in the 60s, I could have done systematically better by
accepting "inside views" of computing - stories told about how, in
particular, computing power would grow - over outside views? It is not
enough to find one inside view that turned out to be correct. That
could be luck. If enough people make enough different guesses one of
them is bound to be correct.
Outside views fail when conditions change. The outside view does not
tell you that computers ten years later will need a support staff, it
tells you that, right now, you can do better by asking how much support
staff are *actually* required to support similar computers, rather than
drawing up a project diagram which shows that only three people should
be required. As for the idea that you can figure out how much support
computers will need ten years later, there is no evidence that this
could be done with an inside view either.
Eliezer S. Yudkowsky http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence
More information about the extropy-chat