|
The
obscenity of Elon Musk
becoming a trillionaire
and the insane bubble
financing that is
forcing us all to be
AI’s bitches led
me back to a book that
I am sure has been
underdiscussed but is
essential for
processing the coming
debacle AI.
I have read Resisting
AI: An Anti-fascist
Approach to Artificial
Intelligence by Dan McQuillan twice – once to mark it up and once again just to read it unencumbered by pens and marginalia – because McQuillan has packed it with so much information distilled into sharp phrasings and lucid explanations. It probably deserves a third read as well.
He begins by explaining
how the machine
learning that powers AI
operates. I do not
understand the math or
coding that shapes AI
(though Broussard in How Computers Misunderstand the World gives a good illustration of how the process works), but it operates essentially as a distillery. Just as a distiller abstracts out alcohol from a complex batch of ingredients – discarding and discarding and discarding until he gets the one product that he wants – the operations of AI extract from a mess of messy data predictions based on probabilities and correlations and patterns. (McQuillan: “Any AI-like system will act as a condenser for existing forms of structural and cultural violence.”)
Because it involves
math and appears to be
scientific (though AI
and its algorithms are
not scientific at all,
as McQuillan points in
his section on
“Scientism”
[47-51]), people assume
that the results are
free from bias and
speak the truth about
whatever task the
program has been asked
to solve – that
the program has been
effective. The results
are therefore
supposedly more
trustworthy than
judgments and
considerations offered
by actual human beings.
But as McQuillan points
out time and time
again, this conclusion
is completely false.
First, the program can
only use the data fed
into it to do its work,
and if those data are
biased, then the
results will be biased.
He cites well-known
stories about how
facial recognition
often fails to identify
black or brown faces
because of the
overabundance of white
faces in the training
data or how COMPAS, the
sentence-prediction
program that supposedly
helps judges assess the
risks of people
re-offending,
consistently gives
black and brown people
higher risk scores than
white people for the
same criminal
circumstances.
Second, algorithms
cannot be divorced from
the social, political
and economic conditions
in which they are
forged. The algorithms
are created to serve
institutional purposes,
and those purposes are
grounded in long
histories of
exploitation,
colonialism and
violence of all kinds
– because of
this, their training
data can never be free
from the past, and the
algorithms are doomed
to recapitulate it. (In
other words, AI can
never imagine a future
outside of the futures
predicated on the pasts
included in its
training data.)
McQuillan cites many
instances of what he
calls
“algorithmic
violence” where
AI programs increase,
as a matter of policy,
the
“othering”
of certain parts of
society, which
literally decides who
will live and who will
die (what he and others
call
“necropolitics”).
This can be done
through increasing
precarity (think of
Uber drivers and others
in the gig economy),
racialization (turning
slight differences
among humans into
essentialized
groupings),
incarceration (and the
“carceral
state,” with
fantasies of predictive
policing), eugenics and
race
“science”
– the list of
operations is long and
dismal.
Third, what gives the
algorithmic violence
the sanction to do what
it does is what
McQuillan calls the
creation of
“states of
exception”: a
social and political
state of being where
the law establishes
spaces and practices
not bound by the law in
order to achieve
certain ends, usually
dealing with a
nation’s
“security,”
such as treating
immigrants as invasive
hordes or creating
black sites where
prohibitions against
torture have no force.
(The paradox, as
McQuillan points out,
of the law using itself
to nullify its duty to
provide a bulwark
against lawlessness.)
All these aspects of AI
make it a technology
easily adapted to a
fascist politics, a
politics that seems to
be thriving in certain
parts of the world
(including the United
States). McQuillan
cites a phrase by Roger
Griffin,
“palingenetic
ultranationalism,”
that sums up the
fascist ideology:
The palingenetic bit
simply means national
rebirth; that the
nation needs to be
reborn from some kind
of current decadence
and reclaim its
glorious past, a
process which will
inevitably be violent.
The term
ultranationalism
indicates that
we’re not talking
about a nation defined
by citizenship but by
organic membership of
an ethnic community.
Hence, with AI, we
should be watchful for
functionality that
contributes to violent
separations of
‘us and
them’, especially
those that seem to
essentialize
differences. [using
British punctuation]
Similar undercurrents
can be found in
“Make America
Great Again.”
What is to be done? His
last three chapters
contain strategies
about containing AI,
and much of what he
talks about – the
commons and
“commoning,”
mutual aid, horizontal
decision-making,
peoples’ and
workers’ councils
– reminded me of
David Graeber’s
discussion of what
Occupy tried to tutor
the American people
about. He also includes
fascinating discussions
of feminist standpoint
theory and feminist new
materialism,
post-normal science
– I have never
heard of these
concepts; reading about
them refreshed and
unhinged (in a good
way) my critical stance.
His view of a world of
mutual aid and
attention paid to care
and a new
“apparatus”
(a term he gives to AI
at the beginning of the
book) that
doesn’t strive to
“solve”
anything “but to
sustain the delivery of
systems of care and
social reproduction
under changing
conditions in ways that
contribute to
collective
emancipation”
(148) gives me a
feeling of both hope
and longing, much like
the effect of that
other book I’m
reading at the
moment, The Communist
Manifesto (brilliantly
explicated by China
Miéville in his A Spectre, Haunting),
a yearning for the
kingdom of heaven on
earth.
|
