RE: SIAI's flawed friendliness analysis

From: Gary Miller (
Date: Thu May 22 2003 - 07:33:43 MDT

>>I wonder why it is that government funding has
>>been so much more effective in biomedical areas than in AI.
Government research grants typically require the principles performing
the research to be heavily degreed. A successful track record for at
least some of the research team on prior grants is important. This is
why independent researchers do well to team with universities that have
received DARPA sponsorship in the past.

They also lean towards very narrow areas of research where research
methodology has been successful in similar problems. The research
grants must be specific enough to survive an informal peer review. Most
projects sponsors will ask their favorite past researchers in the field
for at least informal approval on most of the grant applications.
Grants should be able to spell out what quantifiable benefits could be
achieved for the taxpayers, DOD, or mankind if the research was

It also does not hurt to get at least one congressman on your side to
help start the good old boy network rolling. If it's a large project
and will create jobs in your state he may push harder to look good to
his constituents. Aka PorkBarrel...

-----Original Message-----
From: [] On Behalf Of Ben
Sent: Wednesday, May 21, 2003 5:10 PM
Subject: Re: SIAI's flawed friendliness analysis

> > I wonder why it is that government funding has been so much more
> > in biomedical areas than in AI. It could be "historical accident",
> > but there may also be some kind of systematic explanation, to do
> > with the different behaviors of the different government agencies
> > involved,
> > intrinsic differences between the sciences. It's a subtle thing
> > though, because DARPA, the major US AI funder for many years, had a
> > lot of
> > in its funding experiments in many other areas...
> Looks pretty obvious to me. Biomedical areas have plenty of obvious,
> useful, fruitful angles of approach. Throw money at AI and the
> researchers just bounce off the problem.
> --
> Eliezer S. Yudkowsky

yeah, you do have a point.

Biomedical research is not nearly as easy as you seem to suggest ... it
involves a huge number of insanely difficult, closely interconnected
problems.... However, I think it is true that there are clearer
gradual, incremental paths for moving toward solutions to these insanely
hard problems, than there are in AI.

In biomedicine, the gov't research establishment has done a pretty good
job of dynamically and aggressively shifting funding to promising new
areas, and not overfunding "ingrained ideas" whose time has come and
gone. Not a perfect job, but a pretty good job.

On the other hand, the habit of gov't funding in AI has been to
repeatedly fund almost the same exact research by the same exact
researchers over and over again, in spite of repeated failures. There
has been interesting AI research, but the correlation between level of
gov't funding and interestingness of research has been surprisingly low.

Anyway, I don't think "biomedicine is easier in some ways than AI " is
anywhere near a full explanation of the historical phenomenon in
question, though I agree that it does have some truth to it.


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT