On Tue, Apr 27, 2021 at 09:04:53AM -0400, Nico Kadel-Garcia wrote:
> The movie is, itself, profoundly biased. It didn't explore at all why
> a public housing project might benefit from cameras on the door of a
> densely populated building with numerous poor, old, or unhealthy
> tenants. 

On Sun, Apr 18, 2021 at 7:24 AM LaToya Anderson <[log in to unmask]> wrote:
> Data does not remove bias. And one can and should both read the article and watch the movie.

I imagine the camera was helpful for young white men
entering the building.  But how does that help the old
non-white women who are locked out or their apartments
because the software fails 30% of the time for them?

An automated camera makes sense if software is perfect,
but the point of the film and the paper is that the
software is not perfect.  A automated camera should
work BETTER than adept (and unbiased) human being paying
attention to a monitor, not just CHEAPER.

"Dancing bearware" that only pretends to do the job should
result in prosecution and hefty penalties for the software
designers and decision makers if their "cost saving"
replacement of trained human security guards results in 
a crime ... letting a criminal in or locking a tenant
out, to be robbed on the doorstep.

I too was bothered by the film's seeming "lefty bias",
but "my side" is *human achievement* ... engaging all
8 billion of us.  Leaving people out is economically
suboptimal, but most organizations are insensitive to
the costs they impose outside their organization.  We
create (often bad) laws to internalize those costs so the
organizations MUST pay attention.  Sadly, laws usually
just make organizations pay attention to loopholes.

If the automated cameras are redesigned to do their job
perfectly, I would love that.  If diligent security
guards are trained and employed for the task, I'm for
that as well. 

What I am not for is replacing quality human effort with
slapdash "cost cutting", which often means "job cutting".
In this case, putting human security guards on the dole,
or not hiring enough competent software designers to
properly design and properly TEST recognition software,
that works for everyone, not just software designers
and ethnically similar product purchasers.

I imagine a room full of $10/hour Chinese programmers 
designing this software for Chinese customers.  I bet
their software would do a good job recognizing blacks
in the US if there were enough blacks in China to test
their software with.  Offshoring has its costs as well,
and the point of the film is that the costs are imposed
on those least able to pay them.  

I also imagine devolving software purchasing decisions
downwards to the people who are affected by them.  In
my ideal world, some of those tenants would be involved
in testing and selecting the software. 

Or training tenants to look at security cameras as a
part-time job; perhaps for a rent reduction.  Software
might be used to insure that those "informal employees"
are doing the task they are paid for, but that could
have bias as well.  

Quality is hard work.  Nobody is perfect, and some folks
are quite imperfect - thugs in Armani suits.  Automating
imperfection is the opposite of quality, while designing
to compensate for imperfection is the path to continuous
quality improvement.

But hey, I'm a chip designer.  I invented a circuit that
is used to ultra-cheaply identify individual electronic
devices, WITHOUT identifying the individuals using them.
I wrote A LOT of Linux software to test and improve those
designs; we got the failure rate down below 30 parts per
million, and we designed fallbacks for the unhappy 30.
The customers would have accepted worse - but I would not.

At the end of the day, our professional satisfaction rests
on what we have accomplished, not on what we are paid to
do it.  If you are just doing it for the money, please go
into investing, not into engineering.

Keith

-- 
Keith Lofstrom          [log in to unmask]