build the universe on the bit
- j.a. wheeler
. . .
. .
.
operative term in wheeler's prescription
is "build".. . for the concept
of "information" is
open
.
bit
is only
as good as
the yes/no question
whose answer it represents
and shannon's theory can't say which
.
consider the information content of the
human iris: ~3.22 bits/mm^2
according to
.
john daugman
"the importance of being random: statistical principles of iris recognition"
pattern recognition, 2003

. .

iris.jpg

. .
major tests of daugman's recognition algorithm:
sandia labs 1996 . ~20k irises . no false+
british telecom labs 1997 . ~223k irises . no false+
sensar corp 2000 . ~500k irises . no false+
joh.enschede 2000 . ~20k irises . no false+
eyeticket 2001 . ~300k irises . no false+
nat physics lab uk 2001 . 2.73m irises . no false+
j daugman 2003, 9.1m irises . no false+
iridian technologies 2003 . 984m irises . no false+
.
iris recognition works because
iris patterns are invariant
throughout a person's
lifespan
.
outstanding
but hides that
3.22 bits per sq mm
is not an intrinsic property
of [the distribution of] iris patterns
but a joint property of the interaction of iris,
measurement protocol, eg selected optical spectrum
and daugman's specific wavelet encoding & related processing
. .
"information content" of messages only makes sense in context
that context, far-from-eq, supplies the mappings
relating the measure spaces
to semantics
else:

msg.1.jpg

.
. .
.
previous attempts to solve end-to-end artificial intelligence problems
have failed in one of two ways:
some oversimplified the problems to the point that
solutions did not transfer to real world
others ran into a wall of engineering complexity
too many things to put together
too many interactions
too many bugs
to do better we need
a new mathematical language for a.i.
at the least we need a language that combines
logic and probability
pedro domingo / "solving a.i" mit tech review "energy tech issue" 5.2009
.
in math, we need to average over many similar objects
thus if we have too few objects, we cannot perform the averaging
if we have too many objects, non-standard behavior
which rarely occurs could be washed away
bj miller + rf bighash
.
the old hope of bourbaki
to see mathematical structures arise naturally
from a hierarchy of sets, from their subsets
and from their combination, is doubtless
only an illusion
r thom
.
independence
is a dynamic notion
one may find 2 events independent
at some point, but then find them dependent
after obtaining some evidence
a darwiche
.
either math is
too big for the human mind
or the mind is more than a machine
k goedel