Speaking
of progress, the great technological revolution that was supposed to have
us all riding in self-driving cars still has not arrived, and by some measures
might even be worse now than it used to be. All AI has really seemed to
accomplish is the invention of increasingly more parasitic ways to deliver
advertising to your various LCD screens.
What a waste of
great technology. We carry in our pockets computers that are manyfold more
powerful than those that landed human beings on the moon. It would be nice if
we could use these machines for something more than taking photos, checking
email, staring at advertisements, and surreptitiously looking at porn.
Hardware
manufacturers, at least, have done their part. At least, in my opinion, they
have. There is no mechanical barrier
between us and the future we might rather be living in. The binding constraint
seems rather to be that nobody's making the kind of software -- what a
tragically ineloquent word "apps"
is -- that might genuinely improve our existence.
* *
*
Using an extensive
spreadsheet full of all kinds of personal data, I was able to develop an
explanatory model for my blood glucose readings. The model is partially
auto-regressive, since one of the dependent variables for current blood glucose
is the previous blood glucose reading and the time since that last reading. The
other variables are: total calories eaten at the most recent meal, most recent
insulin bolus, and a dummy variable for cardiovascular exercise in the past
twenty-four hours.
Each of these
variables were significant at 95% confidence, and the coefficients on each
variable made logical sense with respect to how we might predict they relate to
blood glucose. That is, insulin and exercise reduced blood glucose while
calories increased blood glucose. And just as any good endocrinologist will
tell you, the higher your previous blood glucose reading, the higher your
current one is likely to be.
I had attempted to
include a sixth variable, accounting for any period of exercise between the
previous and the current blood glucose reading. I included this variable based
on the fact that glucose tests have consistently demonstrated that any period
of exercise decreases my blood sugar by about 1.1 mg/dL per minute of exercise.
(Some forms of exercise do more, and some do less, but on average it's about
1.1 mg/dL.) Unfortunately, this variable was not significant, and the model
improved slightly when I omitted it, while the significance and magnitude of
the other variables were unaffected.
This five-factor
model produced an R-squared of just 0.19. That's a low value, but it tracks the
data very closely, even if it doesn't track the variance closely. Thus, the
model isn't highly predictive, but it offers good explanatory power. It's
possible that I've left out some important factors. I certainly did not include
all the data I had. Rather, I came up with an a
priori model of what I think should
explain my blood sugar, based on my many years of living with my condition, and
tested that a priori model against the
observations I had.
One way of
interpreting these results is banal: I think I have a good idea about what
explains my blood sugar, but that doesn't translate into perfect or even
"tight" blood glucose control. A more optimistic take on the data is
that I have correctly identified a few highly significant explanatory factors
and I have empirically validated my thinking; now all I need to do is slowly
work toward incorporating more and more of my data into my model. There is
certainly no shortage of unused variables in my spreadsheet. Over time, I
should be able to improve both my model and my blood sugar control by
approaching the problem empirically.
* *
*
And so, my
dissatisfaction continues. No one's connecting the various data collection
points we have and fusing them into something actually
useful.
Here's a simple idea
that would make Amazon.com a lot of money: Wouldn't it be great if Amazon kept
track of the items purchased by me but shipped to various addresses and names
in my contacts list? For example, wouldn't it be great if I could just click on
my sister's name and find out every Christmas present I sent her via Amazon.com
for the past x number of years? Wouldn't it be great if Amazon produced
recommendations based on that product history? This is a simple, basic idea
that utilizes everything that Amazon is already doing, but grouped on one
additional variable. Amazon already knows that I sometimes buy skin cream, and
they recommend products accordingly. Why can't they figure out that I never
send that skin cream to my home address?
My smart phone can
take a photograph of anything I point to, and provide me with a location and
price of a similar item on sale somewhere close. What it doesn't do is any
level of price-comparison. This is madness. I don't merely want to know one
instance of a place that might sell something similar; I obviously want to
know: (a) where is the nearest location of such a thing, (b) what is the price
of that thing at the nearest location, (c) what is the lowest price for that
item at any location, and (d) how soon can such an item arrive if I have it
shipped, rather than going to the nearest location? Four simple data points
based on information available through simple web querying and using technology
that is already sitting on my phone.
Health and fitness
apps, ubiquitous on every smart phone, give a person ample ability to track health data, but what good comes from
"tracking?" Many of these apps don't even make it obvious how the
customer is to download his or her own data for more refined empirical
processing. I certainly shouldn't have to consult a "FAQ" just to
download a csv containing the stuff I took the time to upload in the first
place! A few of these programs do me the additional service of summarizing my
data for me on a weekly basis. That, at least, is a step beyond merely using me
as a platform to harvest data that I myself cannot use.
But so far, the
closest any of these apps has come to offering me any predictive value -- any
value at all, beyond merely capturing my
health data -- is that Garmin's "Connect" service and Strava's paid
platform both offer their own estimates of how much rest I should take before
my next workout. Near as I can tell, it's not a personalized service, it's
simply the output of a mathematical formula that uses heart rate, watts, and
calorie data as its primary inputs. It's a cut above the rest, but I still
can't help thinking that, with all the data to which these companies have
access -- health and fitness logs from athletes and non-athletes from all over
the world, of all age groups -- their algorithms ought to be able to produce
more. Only Garmin's service estimates
VO2-max, and that's just simple arithmetic based on age, BMI, and the user's
current fastest mile time!
Amazon Echo and
Google Home devices don't even play music in stereo. Think about it: monaural
home stereo systems in the year 2019, when people were recording quadraphonic
albums in the 1970s!
* *
*
"Silicon
Valley" has gone nutzo. After all these years of micro-dosing LSD,
studying cryogenics, and living 16 to a townhome, they're no closer to
producing anything close to a real technological revolution.
It's always been a
bit of a canard that software engineers have no people skills and no sense of
design; that they can write the code that will make any idea happen, but that
they have no common sense about what the end user actually wants. In my many years
working in that industry, I always held out hope that it was just a canard, that real-world people
would never be so dumb as to produce millions of lines of computer code that
offered no greater insight into my own life than I could produce with a single
spreadsheet. Alas, I have to admit that my spreadsheets are many times better
than anything Silicon Valley is producing today.
And so, the future
waits, locked in a box that can only be opened by someone who wants to solve an
actual problem, by someone who is in it for more than just upvotes at
StackOverflow.com. So much elegant code in the world, and so few elegant
solutions to real-world problems .So many predictive algorithms, and so few
ways to predict the outcomes of our most pressing concerns.
It was inevitable
that, once the computer and the telephone line had been invented, we would
eventually connect the computer to the telephone line. So it is probably
inevitable that many of these problems will be solved eventually. In the meantime, think how many resources we waste
tracking the arrival of our next package rather than predicting the day on
which we'll need to re-order.
Inventory software
applied to the common household pantry; so obvious, and yet so far away from
fruition.
No comments:
Post a Comment