Making
Computers Smarter
ONStrategies, by Tony Baer, February 24, 2004
It's human nature to make computers think more like people. By that, we mean
software that can
understand what
you really want
when you submit
a query, or anticipate
your future needs
and act on them.
But most software is pretty dumb, capable only of responding to literal commands,
although gradually
we've added bits
and pieces of intelligence.
You can now type
a search phrase
into Google and
it will factor
in misspellings
or typos. Not exactly
revolutionary,
but a start.
All
this is pretty
ironic, given repeated
attempts over the
past 30 years to
make computers
understand us.
It began with top-down
approaches like
artificial intelligence,
where software
draws inferences;
fuzzy logic, where
systems process
parameters related
to what you specify;
and neural networks,
where software "learns" what you are doing to anticipate what you're looking for, or want to do.
More
recently, approaches
have been more
bottom-up. Text
parsers look for
incidence of phrases,
and in some cases,
synonyms. Voice
processors transform
sounds into words
or vice versa,
but can't yet deduce
meaning. Data mining
advances ferret
out hidden patterns
from huge troves
of data, while
web services provide
standard approaches
for software to
intelligently connect.
And, thanks to
Internet standards
for locating resources
and identifying
them, newer "semantic web" proposals could add the ability to search for content based on its meaning.
Adding icing to
the cake, we've
seen prototypes
of IBM's renowned
Web Fountain intended
to add the ability
to search for content
by its meaning.
But
where is this all
getting us? We
thought of those
questions as we
were recently briefed
by several vendors
who have fused
a variety of these
techniques. Biz360
is developing analytic
tools that apply
these techniques
to dissect what
other people (e.g.,
the press, stock
analysts, etc.)
are saying about
your products.
Fair Isaac, best
known for the FICO
credit ratings,
has integrated
its rules and predictive
analysis engine
to help financial
services companies
automate more credit
risk decisions.
Meanwhile, we've
seen pretty cool
data mining tools
from Clear Forest
that can help government
anti-terrorist
agencies ferret
out hidden links
to Al Queda on
the web, or pharmaceutical
firms identify
promising new drugs.
None
of this is brand
new of course.
What is new is
that, in place
of general-purpose
AI approaches,
we've seeing the
emergence of business-focused
solutions that
use various smart
processing techniques.
Not a bad idea.
Still
we wonder, how
smart are computers
really getting?
Take Biz360's product,
that shows not
only how much press
a company is getting,
but whether the
coverage is favorable.
We wonder how they'd
rate a recent CNET
article onDell's
efforts to improve
its sagging customer
service levels.
Would it reflect
positively on Dell's
efforts to fix
something or cast
a negative spin
on how Dell got
to this sorry state
in the first place?
We think that it
would still take
a human to answer
that one.
|
