From: Phil Goetz (firstname.lastname@example.org)
Date: Tue Aug 23 2005 - 09:58:08 MDT
> Novamente LLC's application for funding under this
> program was rejected, as I learned a few weeks ago. That
> particular rejection came with zero explanation.
The solicitation, in thrust B, seemed to ask for models
of brain areas, whereas in private email Tom Armour told
me that they absolutely don't want any models of brain
areas. I think you told me you wrote to thrust A? I
didn't ask about thrust A, but it might also be that what
they were looking for was not really what was described
in the solicitation.
It is very important to talk to the program managers
before sending in your proposal, especially on a BAA.
DARPA IPTO is aware that neuroscience has something to
offer, but they want somebody else to deal with that
aspect, and DARPA money to go to more traditional
AI-people who abstract and apply that knowledge.
Remember, the head of IPTO is Ron Brachman.
My application was also rejected. Good reviews, but
no acceptance. I suspect the budget was a problem;
my subcontractor's cost multiplier was a little too high.
Using a Canadian subcontractor may also have cost
me points. But mainly, my company's lack of a track
record is always the one big strike against it quoted
by reviewers. I now think it's better to get JUST ONE
contract, anywhere, from anyone, before going after
grants. And that is what I'll do when I re-start my
> This second rejection drove home the well-known fact that
> such funding is allotted on political grounds as much as
> scientific/technical grounds. (And my own "political"
> connections are not null, but nor are they the best in
> the business....)
I worked for IAI for 5 years, and won 6 grants, and had
no connections. IAI has few connections, and has
won hundreds of grants.
I suggest looking at SBIRs, where connections count
for very little. The kind of connections you need
for bigger contracts aren't political connections,
but name recognition, a proven track record, and
a history of working with some of the bigger contractors.
Ben, you probably know this already.
I think having a recognized name, good references
from people at companies with recognized names, or being
a silver-tongued devil, count for more than connections,
at least when your projects are less than eight figures.
> I note that I have successfully achieved government
> grants and contracts on many occasions, including
> recently -- but for narrow-AI work (such as
> bioinformatics or language processing), not AGI work.
The DARPA program managers sometimes want AGI-like stuff,
but they have to sell their programs. When you talk
to the people who the DARPA program managers have to
sell their programs to about what they really need,
what comes up is that they want something like what
they've got, but easier to use. They already think
Soar is too complicated. The Army likes finite-state
automata. Create a better interface than One-SAF for
building finite-state automata, make it compliant with
One-SAF and HLA, and you've got a sale.
DARPA is, in any case, committed to ideas about
modular design, re-usability, interoperability, and
backwards-compatibility, that you have to operate within.
Some good work can be done this way. IAI is still
working with DARPA to improve the MAGIC (Multi-Agent
Generic Infrastructure for Cognition) system that I
co-designed for DARPA. Now Lockheed Martin, and Jeff
Hawkin's Redwood Institute, are also involved. I'm so
proud of my baby! Even if it is a modular
design incompatible with dynamic systems theory.
It's still a large improvement on the state of the art.
> My application
> referred to various prior publications and also material
> available online, but I doubt the reviewers took the time
> to dig through all this stuff enough to fully appreciate
> the underlying ideas....
Not only won't reviewers do that;
you're lucky if they read past the second page.
Their eyes will glance over it, but if you don't
grab them in the first couple of pages, you're lost.
Start your day with Yahoo! - make it your home page
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:52 MDT