Big point: We want to start with what
we have and get to what we want.
So, how the heck to do that?
Maybe what we are doing is just using computing
to automate what we have long done manually.
Then we know how the heck to do that and
mostly just need to program it.
Maybe we can think of a reasonably good way
to do that using just intuitive heuristics
and just need to program those.
But, what else can we do?
Here's an example: Suppose at a server farm
we have a system and want to monitor it for
any problems not seen before. Suppose
the system can report 100 times a second
numerical data on each
of 12 variables. Suppose the server farm is
relatively stable and we can collect such
data for three months, let the data 'age',
and be fairly sure that the system was
operating 'normally' (correctly, without
problems) during those three months.
Now what? In real time, how do we monitor
the system? AI 'expert systems'? Been
there; done that; don't think so.
And our customers are highly concerned
about both detection rate and false alarm
rate. The 12 variables used carefully seem
to promise a good detection rate, but what
the heck to do about keeping false alarm
rate down? Intuitive heuristics? I don't
think so. Something in 'machine learning'?
I doubt it. Now what?
Okay, it's a research problem in the applied
mathematics of applied probability.
Okay, what's the lesson here? The applied math
now should be seen as the uniquely powerful
crucial methodology for much of the progress
we seek, say, in the RFS. Why? Because
in a strong sense, math has by far the
most powerful methodology in civilization.
How? Why? Because applied math can take
what we have, use that as hypotheses in
mathematical/logical deductions and,
with theorems and proofs, come up with
mathematical consequences that are rock
solid logically and give us what the heck
we want, far beyond anything we ever did
manually or could ever guess with intuitive
heuristics.
Net, the key methodology for much of the
progress we want, that is, taking what
we have and getting to what we want,
is some applied math, with theorems and
proofs, possibly in part or largely original,
and sometimes with some advanced math
prerequisites. Such math promises to be
one of the most important pillars for the
exploitation of computing in the future,
assuming that we do get such exploitation.
So far I have yet to see "applied math"
mentioned on the Web site of any venture
firm.
Another example? Sure: Take the data we
have and get to what we want in, say,
ad targeting. So, sure, there are lots of
techniques for ad targeting, but what are the
more powerful techniques, say, that give
much more effective ad targeting, and how
do we find and confirm the power of such
techniques? Well, in some cases, maybe most,
the best methodology will be some math
such as I described.
VC, YC, SV, entrepreneurship, computing,
etc. will have to take applied math
quite seriously or miss out on a lot
of what might have been and struggle
for decades in nearly useless mud
wrestling instead.
You've been warned. Be warned. You
neglect applied math and you can't win
and must lose compared with what is
possible, big time. Did I mention
that the theorems and proofs of applied
math form the most powerful methodology
in civilization? Believe it.