[extropy-chat] Futures Past

[Robin Hanson]

At 08:49 AM 10/9/2005, Greg Burch wrote:
>Since I was old enough to read, I've been engaged by projections 
>about the future.  ...
>But somehow, I never seem to learn.  Seven years ago, we had a 
>discussion here on the List about what we called "near-term 
>projections" for the future to c. 2015.  I gathered together some of 
>the ideas in that discussion and put them in what I called a 
>"futurist time capsule."  Here it 
>is:  http://www.gregburch.net/writing/NearTerm.htm
>It makes for interesting and, in many instances, painful reading.  ...
>What do you take away from looking back on looking forward?

Let me suggest that we should structure and formalize this sort of 
experience enough to let us use standard statistical tools to learn 
from it.  That is, we should create a "dataset" consisting of a list 
entries each of which has fields like the following:

Date of forecast
Date of forecasted event
Forecasted parameter value
Actual parameter value (or current best estimate of future parameter value)
Author parameters: age, gender, education level, profession, ...
Relative expertize of author on topic
Soc/tech level of forecast (ranging from basic low level 
technological possibility
     to mid level device to high level change in patterns of social behavior)
Rough size of organization that required to achieve prediction
Industry of forecast: energy, transportation, computation, entertainment, ...
Supporting technology of forecast: hardware, software, genetics, chemistry, ...

With a little more thought one could probably come up with a lot more 
relevant fields.

Given such a dataset one could do multivariate regressions with 
forecast accuracy and bias as dependent variables, and all the other 
fields as independent variables.  Such statistics would then tell us 
which kinds of forecasts and authors tend to have what relative 
accuracy or biases.

Greg's forecasts would be a fine place to start.  One could add in 
forecasts by Kurzweil and others as time permits.  It really 
shouldn't be that hard to collect such a dataset. For a first cut 
analysis one needn't put that much thought into deciding the exact 
field value for each forecast.  Having two or three people 
independently fill in field values would probably be sufficient - a 
high enough correlation among their entries would show the process to 
be sufficiently reliable.

If some other people would help with putting such a dataset together, 
I'd be happy to help with the statistical analysis and with writing 
it up and working to get it published.



Robin Hanson  rhanson at gmu.edu  http://hanson.gmu.edu
Associate Professor of Economics, George Mason University
MSN 1D3, Carow Hall, Fairfax VA 22030-4444
703-993-2326  FAX: 703-993-2323