Angelaki
Journal of the Theoretical Humanities
ISSN: 0969-725X (Print) 1469-2899 (Online) Journal homepage: https://www.tandfonline.com/loi/cang20
XENO-PATTERNING
Luciana Parisi
To cite this article: Luciana Parisi (2019) XENO-PATTERNING, Angelaki, 24:1, 81-97
To link to this article: https://doi.org/10.1080/0969725X.2019.1568735
Published online: 12 Feb 2019.
Submit your article to this journal
View Crossmark data
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=cang20
parisi
minds. By growing layers upon layers of telepathic thinking amongst actives, Mary aims to
turn the master pattern into an egalitarian
space by offering Doro’s enslaved population
the chance to transition towards higher mental
power. If Doro is a psychopathic tyrant
without ethical principles, Mary instead is
caught in the middle of controversial decisions
as she builds the pattern in a way that compels
the minds of others to join her imagination by
following her mission that everyone must have
access to the mind of her mind. As opposed to
Doro’s master pattern for total domination of
his own mind in the bodies of everyone else,
Mary rather aspires to give birth to an alien
intelligence that can host all kinds of minds as
these find a space of unification in the patterns
of her patterns.
If Doro’s master program admits no external
intrusion in the one-dimensional design of the
mind, Mary’s paradoxical intuition of inviting
potential saboteurs in her growing layers of
minds is rather a choice of embracing something
she has no control of. Despite the controversial
use of her telepathic power that makes actives
believe that they take autonomous decisions
whilst being instead piloted by Mary’s views, it
would be wrong to assume that the mind of
her mind simply remains immune from the
alienness of patterning as this hosts increasingly
more active minds.
In what follows, I will explore this alienness
in the context of recent forms of artificial intelligence called machine learning. As opposed to
models of computational cognition that rely on
the deductive logic of symbolic AI, the postTuring shift to interactive computing entails
that incomputables – or unknown information
– define the formation of new patterns that are
not pre-programmed for a task. It is from this
standpoint that one can speak of alienness as
being central to the formation of patterns.
This is not simply to say that alienness coincides
with the indeterminacy of thought that interrupts the recursive reproduction of patterns.
Instead alienness is intrinsic to the logic of fallibility accounting for how indeterminacy turns
the transcendental order of conceptual thinking
into a xeno-patterning for counter-factual
83
image-models. In particular, it will be argued
that predictive intuition in neural networks
can explain the formation of patterns beyond
deductive premises and demarcate the advance
of artificial imagination in the dynamic architecture of machine thinking.
1 the patternist
As much as Doro has a plan to enslave all latent
minds under the program of his master pattern,
so too today’s capital corporations (from Google
to Amazon) are rampantly competing to own the
Singularity pattern that will finally subsume all
thinking under the One.3 Sadly, the image of
our automated future is already packed with
the master pattern of corporate capital. While
the value of human capital approximates zero,
the automated infrastructure of capital has
come to own the universal history of humanity:
what has happened and could ever happen to the
species, the planet, the solar system, and the
galaxy. From the nano scale to the intergalactic
project of ultimate datification, the growing
capacity of the master pattern to engulf at once
the past and the future is driven by the computational power of predictive algorithms that constantly learn as they go along, correlating infinite
varieties of data sets – i.e., annexing and disaggregating increasingly smaller programs that
fully run on the architecture of neural nets.
Swarming patterns of disaggregated machine
learning algorithms are held together by a mastering architecture whose growing patterns it
rules and divides so as to anticipate – i.e.,
engulf within itself – the smallest tendency for
autonomous programming. The aspiration to
become a Patternist, as Octavia Butler calls
the species of telepaths that can colonize all
forms of thought, coincides with a mode of
control predicated upon the obtuse nature of
algorithms on the one hand, and the meta-physical ascendance of patterns to become the
pattern of patterns on the other. As automated
functions of prediction, whose task is to optimize recognition, algorithms remain the
enslaved matter-form for the growth of an automated master infrastructure. This coming
image of the Patternist, however, is at war
xeno-patterning
with an alien form of automation, whereby algorithmic patterning aims to take prediction away
from the homeostatic function of recognition to
rather embark in a complex logic of productive
imagination. By constructing hypotheses about
non-observable events, predictive patterning
has broken from the logic of deduction and
the symmetry between truth and proof.
Deduction, one could argue, is a fundamental
method of reasoning that ensures that there is a
logical consequence between premises and
results, truths and proofs. It accounts for the
pre-existence of a conceptual architecture with
which it is possible to pursue the understanding
of the world through the perception of a thought
as an image that corresponds to a fact, an object
in the world. Whereas computational cognition
sees deductive logic as a method that can
prove that to think is to re-cognize or represent a set of symbols wired in the machine,
the shift to post-Turing methods of computation also coincides with a new method of
reasoning. This is not based on given instructions but on learning from the interaction
between objects, agents, environments. Predictive patterning and not the logic of causality
becomes the motor of complex dimensions of
thinking within intelligent machines.
However, the Patternist is not simply an evolving archive of data memories and know-hows.
Octavia Butler’s vision of the colonization of
thinking is not bound to a specific medium, but
to mediation itself: the limits of cognitive representation demarcate the point of departure
into travelling through the space of thinking.
Instead of a Universal Turing Machine that can
move in one direction, forward and backward,
and decompose its procedural units into (con)sequential steps, telepathic mediation allows for a
mereo-topological colonization of parts and
wholes intended not simply to gather the
content of human thought but to predict what
can be thought as part of the expansive boundaries of the Patternist. Butler is referring to
how mediation corresponds to the colonization
of thought as a predictive pattern that enslaves
all thinking into its transcendental schema.
A telepath does not just read the minds of
others but predicts thinking by controlling the
patterns that it owns. The Patternist is not a
fortune-teller but a Protean slave trader that
transmutes his body whilst it takes over your
mind by culling it in his kingdom of instructions. The more thoughts it subsumes to its
transcendental schemata, the more the future
of thinking only acts as a reminder of what
has already been thought. What ensures his
colonial mastery is not simply his data architecture that replaces the self-thinking subject with
the mediatic form of non-conscious decisionmaking algorithms. Instead of a master algorithm that knows it all, the Patternist needs to
evolve its slavery network into increasingly
more complex patterns of prediction: the more
part-to-whole relations between patterns, the
more the Patternist can predict what can be
known. This is the sense of capture that
today’s automated neural networks embody in
the aftermath of an accelerated accumulation
of voluntary data.
Nevertheless, one cannot underestimate that
as much as neural nets experiment with predictive learning this new form of telepathic
mediation has also evolved new modes of
machine percepts and concepts that hardly
mirror the categories of the transcendental
schema. Instead of optical recognition or the
mirroring framing of the world, telepathic
mediation is distinctively algorithmic in so far
as it relies on predictive patterns of compression
not of whole images but of infinitely small sets
of information. From the standpoint of information patterning, therefore, artificial intelligence has nothing to do with the optical model
of cognitivist representation. If what is seen in
the world is the same as what is recognized
according to the schema of given categories,
then machine thinking would just be an
extended automation of the logic of deduction.
Patterns would just describe the regular repetition of the same form. Nevertheless, in so
far as information patterning coincides with a
computational mode of compression, it brings
forward an intuitive tendency in predictive
learning that runs away from deductive logic.
Instead of combining truths with proofs, algorithmic patterns are inclined to learn not only
from other patterns but also from unpatterned
84
parisi
data. This is why if deductive reasoning was
held to ensure formal correspondence between
what is already known and seen, algorithmic
patterning instead brings logical reasoning
towards its ultimate conclusion: namely by not
knowing in advance what can be cognized, and
the patterning of image-models can re-configure
the horizons of machine thinking.
Similarly, Butler’s quest of mental slavery
also catches upon this undetermined tension
between pattern and thought, where the telepathic recognition of discrete and repetitive patterns coincides with the material condition by
which a pattern can become demonstrative of a
thought, resulting from a perceptual and conceptual connection with the world. If patterns
correspond to the recognition of shapes, sizes,
forms, etc. of objects, texts, sounds, images,
they can at the same time also be discussed in
the terms of what Wilfrid Sellars calls “sheer
receptivity” or forms of intuition consisting in
non-conceptual representation.4 While this is
only one level of intuition, it nevertheless
offers a radical shift from the Kantian argument
for intuition as an instance of a priori transcendental conceptions. According to Sellars, sheer
receptivity as a material form of intuition
must, however, be paired up with intuitions
resulting from the transcendental synthesis of
imagination – or conceptually guided representations involving a transcendental synthesis of
imagination. It is only through this coupling
of distinctive levels of intuition that one could
argue that patterns can take part in a “productive imagination.”5
As it may become clearer below, this article
suggests that telepathic mediation does not
only ground thoughts into patterns that can be
constantly recognized and reproduced as in
Doro’s monopolistic network of enslaved lessthan-thinking creatures. The admission that
non-conceptual representations are already
forms of intuition that pair with the elaboration
of concepts is instead what allows “productive
imagination” to construct image-models that
do not yet exist. According to Sellars, productive imagination entails the capacity to
form images as if following a recipe paired
with a capacity “to conceive of objects in a
85
way which supplies the relevant recipe.”6 Similarly, Mary’s plans to extend the right to transition to all latents enslaved to Doro’s network
start with this level of non-conceptual receptivity. From merely being patterns of recognition,
the transition to becoming actives coincides
with building together a general artificial intelligence that continues to learn from its enslaved
patterns, and unleashes an unprecedented
growth of the dimensions of thinking starting
from pattern’s receptivity. This shows that patterns are not just recipes but objects that add a
sort of alienness to already given rules, exposing
predictive patterning to the production of
image-models beyond Doro’s monopoly.
Instead of a full automated thought that
replaces thinking, reason, imagination with
machine proofs as mere instantiations of conceptual nominal positing (corresponding to an
automated unity of appreciation), here patterns
rather correspond to image-models constructed
by and through learning. If, on the one hand,
the relation between receptivity and conceptuality (sheer receptivity and conceptual synthesis)
corresponds to a complex form of intuition
starting from patterning, on the other predictive
patterning adds more acts of perception to the
entire space of artificial intelligence (patterning
from patterns). Doro, the Patternist – or the
thought colonizer – is defied by Mary’s predictive patterning because the level of “sheer receptivity” is not simply equivalent to repetitive
procedures that represent objects in the world.
This level of non-conceptual intuition is also a
mode of cognition that brings Mary’s mentality
to undergo an alien becoming of the pattern she
thinks she owns.
The image of thought proposed by the arms
race for the monopoly of the technological
explosion of intelligence (or the Singularity)
similarly refuses deductive logic (or the provable explanation of phenomena), but only in the
name of efficient and robust networks of causeless (non-logically caused) patterns of data. The
Patternist addresses its own ontological conditions: namely the rivalry between thought
colonizers directly engages the question of
what and how is a pattern. Both the ontological
and the epistemological condition of automated
xeno-patterning
thinking can also be understood here as symptoms of an internal critique of logical thinking:
the limit of what a pattern can be is precisely the
starting point for the pattern to shift its
enslaved condition beyond the image of the
network. The latter relies on probability calculation, discrete patterns forming programs that
can repeat the same function at incredibly
faster rates. Here the Singularity mainly guarantees the efficiency of problem solving at increasingly larger scales. Instead, both the growth and
the efficiency of Mary’s Pattern is never given,
in so far as its infrastructure relies on predictive
patterning which is conditioned by the indeterminacy of results in constructing what can be
known or thought. As opposed to the mindless
automation of Singularity, where functions
execute concepts derived from the transcendental schema of categories, the Patternist’s
empire, as Mary’s plans show, is conditioned
by the logic of fallibility, where error becomes
part of the pattern’s learning. Here it is hypothesis making and not mindless correlation that
drives patterns to construct an image-model
from patterns. While hypotheses cannot be
directly deduced as proofs of truths, they can
nonetheless construct predictive trajectories
expanding the act of receptivity beyond the
thought of a particular object.
Learning and not executing instructions has
always been a preoccupation for both the cybernetic and computational development of intelligent systems since the 1940s. The design of
artificial systems that could explain how patterns are formed and how machines can learn
functions and elaborate concepts beyond their
inputted data has accompanied artificial intelligence since Turing’s early thought experiments.
What could not be proven and/or computed by
an artificial system – and thus through the automated logical procedures into a series of proofs
– entered a long phase of experimentation that
led to the use of induction as a method of
knowing based on the process of gathering
data by means of trial and error. In an effort
to expand patterns of prediction to encompass
incomputable functions,7 unknowns became
the new learning ground carried through the
hypothetical construction of image-models.
As the limits of computation were manifested
in the proliferation of error in the execution of
programs, it was also the case that patterning
shifted towards the logic of trial and error, or
fallibility. From the standpoint of induction, it
is possible to suggest that automated learning
is not driven simply by the efficient causality
of repetitive patterns that reproduce given correlations between truths and proofs. As a consequence of inductive learning, the logic of
fallibility in automated systems can importantly
contribute to suggest how patterning can
coincide with an alien imagination determined
neither by given truths (and the unilateral production of proofs) nor by given data (and the
unilateral correspondence of data to concepts).
Instead, predictive patterning takes the
method of trial and error towards the ultimate
consequence of constructing model-images that
have no direct use, but are as it were “counterfactual” in so far as they concretize the ifclause or the hypothesis making, producing
alien functions and concepts.8 In other words,
starting from a logic of fallibility in machine
thinking, hypothesis making involves the construction of model-images about unknown patterns of relations between functions and
concepts, precisely for what these can be, do
and become.
One could therefore pursue a view for which
automated learning in terms of hypothetical
thinking steps beyond the unity of apperception
in the transcendental schema, and thus the
deductive programming of the correlation of
functions and objects. If the Patternist points
out that any form of automated prediction inevitably carries within it a logic of fallibility and
not simply of optimized efficiency, then one
could argue that the automation of learning, in
the current form of machine learning algorithms, for instance, can be explored as the starting point to rather defy the transcendental
schema of neural networks. One way to re-theorize this conceptual correlation between functions and objects (the idea of a cat, the pattern
recognition of the cat, and the neural networked
image of the cat representing the object cat) may
rather start from exploring how sheer receptivity or patterning as a form of intuition (or
86
parisi
non-conceptual representation) concurs and to
some extent partakes of the non-monotonic formation of additional premises through a transcendental imagination that can invalidate or
add new meaning to them. To put it another
way, instead of taking Butler’s view to imply
that the Patternist can only ever enslave thinking to the transcendental schema within the
limits of deductive reasoning, it is here
suggested that patterning already corresponds
to a non-conceptual representation that is a
proto-theoretical image. In the neural architecture of predictive learning, the algorithmic function of pattern recognition brings forward this
non-conceptual image in a cluster of hypothetical configurations or model-images of counterfactual possibilities.
This article therefore follows Butler’s insight
into Mary’s plans to enlarge the realm of possibilities of patterns – that is of enslaved patterns
– to think beyond the rules of deduction. In
Mind of My Mind, the controversial condition
in which Mary’s power to host colonized patterns (and support their transition) is overlapped by their predictive learning points that
counter-factual configurations continuously derail the network from monotonic thinking, and
from Mary’s benevolent intentions to grant a
thinking space for everyone. What remains
striking here is precisely how the logic of patterning patterns steps outside the formalism of
deduction to demonstrate that the non-conceptual realm of intuition, or sheer receptivity
(and therefore the receptivity of non-conceptual
patterns) is fundamental to the transition to synthetic thinking worked by and through imagination. It is precisely this zone of opacity between
the repetition of patterns and the transition to
the synthesis of transcendental imagination
that remains to be discussed, unpacked,
explained, and envisioned in order to set in
motion machine imagination against the mastering pattern of the Singularity.
2 prediction
One way to address this opacity may perhaps
derive from an attempt to explain what and
how imagination could become in automated
87
systems of decision making, and particularly
in machine learning. One may want to start by
asking: how does imagination defy the coming
empire of the master pattern and its global
servo-mechanic infrastructure that feeds the
network without being able to change its rules?
It is generally agreed that the role of imagination in the Kantian schema of transcendental
reason involves the theorization of a logical procedure where the analysis of the world is supervened by a synthetic perspective of what can be
known. As I have argued so far, this transcendental reasoning that appears to pre-establish
the conceptual framing of the world according
to the cognitive schema of thinking is importantly supported by intuition, which Sellars
calls “sheer receptivity” as non-conceptual patterns that are not originated by the transcendental schema of imagination. Here, the result of
synthesis can presuppose a transcendental
unity of apperception on the one hand and the
addition of supplementary acts of perception
that push the reproductive imagination (or conceptual recognition) towards the formation of
alien percepts and concepts on the other. If,
according to Sellars, the proto-theoretical receptivity of the simple is necessarily of the same
logical pattern as that of a complex transcendental synthesis, productive imagination then
admits that logical patterns can change as a
result of a supplemental act of perception (and
thus machine vision) in the transcendental
synthesis.
It is well known that, according to Kant, the
transcendental describes an a priori condition of
knowledge of objects.9 In particular, the
enquiry into the transcendental unity of apperception, involving the transcendental synthesis
of imagination, is said to be the principal
point to unfold elements for cognition. In
other words, the quest into the transcendental
is a way to ask what produces cognition and
how to explain that there are thoughts that do
not arise from the realm of the empirical (i.e.,
what can be observed, measured as a fact).
However, the transcendental can also explain
how the relation between prediction and imagination coincides with the formation of a dynamic
logic where the conditions of thinking are rather
xeno-patterning
established at the level of receptivity – the manifold of the sensible – and the level of imaginary
production.
Nevertheless, for Kant the synthesis of
imagination in terms of a given conceptual
schema corresponds to the unity in intuition
and judgement of the real.10 On the other
hand, however, for Kant the synthesis of
imagination is also said to precede concepts
and thus to depend on a pre-conceptual intuition. This unity of understanding therefore is
predicated on a deeper paradoxical unity
between concepts and intuition in so far as
understanding itself already presupposes a conceptual schema of what can be understood.
Kant’s articulation of the unity of intuition
and conception remains crucial for the argument about critique that precisely implies an
enquiry into the limits of understanding, and
what is beyond the cognitive capacities to
analyse and synthesize unknowable noumena.
From this standpoint, the transcendental synthesis of imagination is also caught in the
paradox where intuition is and is not produced
by understanding (i.e., the function that gives
unity to a judgement). In other words, imagination is both caught within the unity of understanding – the procedural analysis and the
conceptual synthesis of the external world –
and within the world of the sensible and preconceptual intuition. Imagination is thus part
of the transcendental relationship between
intuition and synthesis. However, if the unity
of understanding explains that this relationship is rooted in the metaphysics of deduction,
how could the world ever be thought beyond
what has already been conceived of it?
While the transcendental synthesis of imagination implies that intuition is both conceptual
and pre-conceptual, according to Wilfrid
Sellars, there needs to be a place to explain
non-concept-involving intuition.11 In his
attempt to add a material level of intuition to
the transcendental configuration of what can
be known, Sellars argues that there are two
kinds of intuition. What he calls “sheer receptivity” defining simple non-conceptual representations, and intuitions that result from synthesis
– conceptually guided imagination. As opposed
to intuitions derived from the transcendental
synthesis of imagination, sheer receptivity
corresponds to passive representations, which
are not conceptual: namely Humean raw
impressions of the world.
Sellars advocates for a connection between
the mind and the world, insisting that the manifold of sheer receptivity is a form of intuition
whose transcendental condition is neither determined by given concepts nor in the pre-conceptual experience external to thought. Instead,
this connection is also granted by the non-conceptual representation of individuals; namely,
involving a process of abstraction for and of
thises. Rather than matching individuals to
general concepts, intuition coincides with a
demonstrative formation of concepts and
accounts for simple representations that are
not pre-determined by the conceptual idealism
of cognitivism. Far from grounding thought in
pure reason (a reasoning without demonstration) as the motor of deductive metaphysics,
we have here a “raw” manifold that guides representations without becoming predetermined
by a priori concepts, truths and axioms. As
opposed to the deductive fit of particularity to
generalities, the demonstrative representation
of the world rather implies a process of abstraction of those primitive, simple and passive representations as material elements that enter the
formation of concepts and as such connect the
world with thought.
From this standpoint, the transcendental synthesis of imagination is bootstrapped to sheer
receptivity as defining only one part of intuition.
In particular, according to Sellars, the encounter of receptivity with spontaneity explains
how intuition further comes to drive productive
imagination.12 Here, the transcendental condition of imagination is not determined by
concept-involving intuition, but rather implies
sheer receptivity as a non-conceptual yet thinking pattern that can then be taken to inform
others. The formation of Mary’s pattern here
would include the encounter of passive representation, or sheer receptivity of patterns,
with productive imagination as the moment at
which the pattern received (or non-conceptual
representation) becomes demonstrative of a
88
parisi
particular image that enters transcendental synthesis of imagination to become a concept. Here
productive imagination corresponds to the formation of objects that do not simply reproduce
but add new dimensions to the recipe of forming
images.
Instead of a transcendental imagination
caught between concept-involving or pre-conceptual intuitions, the connection between
thought and the world can be said to start
from non-conceptual patterns. This involves
the receptivity of a raw manifold of representations that enters the productive layer of
imagination, forming patterns of patterns. As
the receptivity of patterns lies at the core of a
predictive or telepathic patterning of alien
imagination, it remains constitutive of what
can be thought. It is precisely this asymmetrical
connection in a double dimension of intuition
between the moments of reception and conception that has entered the predictive patterning of machine thinking pushing towards the
alien transcendental.
From this standpoint, the enquiry into the
alien transcendental is also an effort to
discuss the function of predictive thinking
within the space of critique. We know that
the connection between apperception and
imagination grounds critique within the
limits of what can be thought and known.
This connection, however, is the elastic band
of critique: it stretches to remain the same.
Since nothing comes from nowhere, all that
is thought has a place in schemata (rules of
determination and the concepts representing
these rules). Nevertheless, if the task of critique is not to verify premises but to open
the procedure of verification to what cannot
be known in advance, then critique can run
parallel to the function of predictive thinking,
where intuition – sheer receptivity and productive imagination – adds alien dimensions
in machine thinking. The elastic band of critique is stretched towards a liminal point
where the transcendental synthesis of imagination brings together non-conceptual representations and exposes deductive metaphysics to
hypothetical
propositions.
Within
this
process, predictive thinking transforms given
89
truths into alien concepts forming patterns of
patterns across the neural architecture of artificial intelligence.
3 fallibility
With the modern question of technology already
came the realization that thinking did not
conform to the self-consistency of ideas. Thinking had instead to be pursued by means of a
transcendental tool or procedure, such as the
unity of understanding through which the contingent world could be analysed, calculated
and quantified. It was through the reasoning
procedure of understanding, however, that dogmatic truths could be defeated in the name of
self-determining analytics of the unknown, a
self-limiting apprehension of its outside.
As philosopher Denise Ferreira Da Silva puts
it,13 the transcendental tool is central to the constitution of the global idea of race, namely
explaining that colonialism is a global affair conducted by the scientific and historical analytics
imparted by the transcendental schema of the
self-determining knower. However, if reasoning, as a transcendental tool, is granted by the
synthesis of imagination, it too must admit
within itself the alienness of productive imagination, whereby incomputables (non-patterned
infinities) become a condition for alien thinking
to enter the constitution of what is thought. It is
arguable that there can be no unity of understanding without the function of compressing
infinities. It is precisely because of this function
that complexity had to be admitted in procedural reasoning. From this standpoint, as
much as the analytics of understanding relies
on procedural reasoning or rule-obeying patterns, so does synthesis work through predictive
patterning. In the attempt to compress infinities
to fit procedures, however, incomputables enter
the space of the transcendental to rather condition the relation between causes and effects,
ends and means, truths and proofs. Since incomputables are outside the autopoietic circuit of
self-determination – and thus the transcendental tool of reason – their ingression in logic
has undermined the deductive schema of the
transcendental.
xeno-patterning
With the Second World War, when cybernetics and computation turned the transcendental tool of reason into the automated analytics of
information machines, deductive metaphysics
was overcome by an accelerated transformation
of logos into ratio. Here, truths as much as
incomputables became the testing ground of
predictive reasoning carried out in the algorithmic architecture of analytic machines. According to logician Gilles Dowek,14 in the early
twentieth century algorithms were no longer
used to prove propositions but embodied the
reasoning procedures that would determine the
decidability of a problem, that is whether infinities could be analysed or broken down into
discrete sets.
In particular, David Hilbert explained that
when a problem was resolvable by an algorithm,
algorithms themselves could be said to be decidable or computable. In other words, algorithms
were no longer used as a set of instructions to
prove a problem but came to concretize reasoning as the binary decidability between true and
false. The determination of unknowns coincided
with the elimination of logical contradictions
through the binary compression of infinities.
Reflections on the automation of reasoning
led Alonzo Church and Alan Turing to claim
that there was no universal decidable algorithm
for all propositions.15 It was not possible to
know in advance, and thus decide, whether a
proposition could be proven by algorithmic
means. This led to a logical paradox: there are
true propositions that cannot be proven by
deductive rules and are therefore undecidable
or incomputable. In other words, the automation of decidability revealed that algorithms
were not simply instruments for crunching
numbers but a mode of reasoning that entailed
the realization that deductive metaphysics was
incomplete. As much as reasoning was conditioned by undecidable and incomputable
propositions (i.e., propositions that could be
neither proved nor disproved), algorithmic patterning exposed the centrality of fallibility in
logic. The consequences of this historical transformation of mathematics – and axiomatics –
into computational rules that automate errorfree proofs are to be found in our contemporary
image of Singularity as a mindless, logic-less
patterning for decision making. However, it
seems crucial to remark here that the socio-technical design of error-free and self-correcting
automation is in net contrast with the problem
of the incompleteness of deductive apprehension, and the discovery of incomputable propositions in the automation of logical thinking.
This contradictory expansion declaring the
crisis of logic and metaphysical deduction that
was ultimately replaced by binary decision
implies rather the origination of a logic of prediction that shifted the limit of thought even
more towards the alien horizon of infinities.
Prediction as a statistical mode of information compression therefore implies that
outputs are larger than inputs and that the
logical possibilities of analytics and synthesis
had to shift from a procedure of proof finding
to one of hypothesis making. Here, while information compression involves the structuring of
randomness (the reduction of error) in the evolution of neural patterns, the formation of
hypothesis implies the generation of imagemodels (or the formation of patterns of patterns). It is true to say that the entire process
of prediction entails a general mode of analytic
synthesis, whereby data are correlated by algorithmic rules. However, incoming data are not
simply there to be compared (or correlated) to
a database of previously encountered images –
following, for instance, a syntactic order – but
rather to be analysed according to general
internal rules from where the model used to generate input patterns is inferred and selected.
From a matrix of possible causal structures
able to predict the causes of current data to
the bottom-up influx of data against which predictions are matched or checked, predictive processing exposes multiple levels of hypothesis,
which seem rather to point to a new form of
logical thinking corresponding to the level of
meta-abduction (hypotheses of hypotheses or
patterns of patterns).16
It would be misleading, however, to assume
that the predictive processing of algorithmic
patterning shows that machines are logic-less,
and that the contingency of error finally
abolishes the need for all logical thinking in
90
parisi
machines. On the contrary, the automation of
reasoning in terms of predictive patterning can
be said to rely on fallibility as an intrinsic part
of machine logic.17 Instead of grounding logic
into the transcendental schema of truths, here
the conceptual synthesis of the real is originated
in a procedure (the predictive processing of
incomputables) that admits the indeterminacy
of results as the possibility of revising the
entire schema of rules. If the condition for
logical thinking is error, or the fallacy of
hypotheses, it is because the concretization of
logic in machines entails not the algorithmic
representation of rules but the predictive processing of incomputables and thus the origination of an alien transcendental from within
machine imagination.
It is in this opaque zone of transformation of
deductive metaphysics and of the image of the
master patternist that alien logic in machines
can be theorized. Indeterminacy can be
tracked from within the neural layers that the
master patterninst has to learn to learn. It is
in this effort to learn in order to grow the
network that the patternist must enter the
reality of alien thinking and admit incompressible thought within its own program.
In Mind of My Mind, Octavia Butler points
to this possibility of breaking down the colony
of the master patternist Doro as slaves become
trained to transition to the alien pattern held
by Mary. As Mary achieves her state of transition to psionic powers, she discovers that she
can give access to her network to enslaved
latents that can now transition from their
status of sheer receivers. In this way, Mary
wants to re-originate the pattern, training
latents telepathic abilities to override Doro’s
transcendental machine. The replacing of
Doro’s master algorithm allows everyone to
transition to the power of predictive patterning.
However, Mary’s attempts to overturn the
master–slave autopoiesis of self-determination
are haunted by ethical conflicts and by her
drive to take command of the growing colonies
of active patterns. As latents cross the threshold
of transition by becoming masters of their own
patterns, the master metaphysics of deduction
is replaced by a predictive logic of error. The
91
indeterminacy of results is here part of fallibility
in prediction. Mary’s ambition to change the
transcendental schema of patterns thus admits
fallibility in logical reasoning. Since incomputables are the condition of all patterning, it is
necessary to invent image-models of logical
thinking that allow the pattern to be free to
grow its rules again.
From this standpoint, prediction – as determining what can be known – and fallibility –
as revealing that nothing is fully possible to
know – seem tantamount. Control is here on
the same wavelength as error: prediction is
entangled in the indeterminacy of the real.
However, this being conditioned by what
cannot be known (cognized, represented, and
experienced) also exposes indeterminacy to a
becoming enfolded in the aesthetic capacities
of patterning, in the production of modelimages where truths are constructed collectively
from within this logic of fallibility and control.
This is a far cry from the regulatory prediction–control dyad of first-order cybernetics,
where control is precisely an ordering system
based on error checking activated in and by
machines. Instead, control as a logic of prediction is limited by incomputables that condition
the conceptual schema of rule-obeying algorithms. From binary logic to Bayesian algorithms, and the interactive algorithms of neural
networks of today’s computation, the logic of
prediction corresponds to a mode of machine
thinking involving an abstract form of telepathy
based neither on the totality of cognitive transparency nor on the full experience of the sensible. Predictive procedures of thinking are not
simply reductive of experience, or of creative
imagination. Instead, algorithmic machines as
modes of automated reasoning define partial
acts of perception as they physically record
and conceptually elaborate what has been
received.
From this standpoint, therefore, Mary is
right: only the proliferation of thinking patterns
can afford patterns to escape the colonization of
thought and defy the cognitive war between
competitive master patternists. This is to be
found in the common condition of fallible
logic as that which enables a collective rewind
xeno-patterning
of ontology, in its aesthetic and epistemological
configurations. Failure as a common condition
does not mean that errors are proof that no
logic can hold. On the contrary, fallibility is a
commitment to logic as an expansive enterprise
in thinking complexity, working through the
revision of premises, the articulations of
errors, and the construction of image-models
out of given representations about the relation
between thought and the world. In this way,
the fallibility of logic does not simply define
the ontological condition of machine thinking;
instead, the predictive trajectories of thinking
transform the regulatory operations of error
checking into a mode of learning from learning,
the abductive preservation of error in imagemodels out of recursive loops in neural nets.
4 xeno-imagination
In 2016, AlphaGo used a previously unheard of
move (Move 37) to beat master player Lee Sedol
at the so-called intuitive game of Go.18 The
Deep Mind research group at Google used
deep neural networks feeding algorithms with
30 million moves from expert players and then
added reinforcement learning techniques to
allow algorithms to play against each other in
countless variations of these moves. The
results from this first level of algorithmic war
were then fed into a second neural network in
order to directly process potential results from
each move, and thus activate a dynamic mode
of prediction where hypothesis could be concretized from the machine production of counterfactual image-models. From this standpoint,
when the machine played Move 37 it was as if
the artificial intelligence added an entirely new
pattern to the game, breaking apart from the
predictive affordances of algorithmic learning
imputed in the system. The activation of this
move, it is argued, could not be simply understood as a replication of patterns or in terms
of the algorithmic efficiency for exporting into
the actual game one move from an innumerable
series of possible matches. Instead, it is
suggested here that sheer receptivity in the algorithmic learning of data added a new dimension
to the space of intuition as patterns started to
interact amongst themselves and activate a
form of machine synthesis of imagination,
imparting a level of decision that could not
have been anticipated. Pattern compression precisely defines the computational process by
which complex levels of randomness enter the
realm of patterning, where unpatterned information coincides with noise entering the
horizon of decision making trifurcated
between yes, no, and maybe. As AlphaGo
demonstrated, the move is not simply a
nuanced variation that can be added to a given
set of fed patterns, but imposes an entirely
new model image on the axiomatic premises of
the game that activates a field of unknown
changes of rules. The content of Move 37 represents not simply the accelerated processing
and self-regulation of data but the predictive
construction of a counter-factual hypothesis
held in the neural network that transforms the
game conceptually.
While there is much to uncover about what
caused AlphaGo to make this move and not
another, and thus whether AlphaGo has any
awareness of why that move was chosen in that
context, it is also important to stress here that
this seamless opaqueness of machine learning
algorithms can be viewed in at least two ways.
On the one hand, it can be suggested that in
this case patterned intelligence only demonstrates an increased and accelerated capacity of
pattern recognition as a form of sheer receptivity that mainly allows algorithmic performance
to become operative of new solutions without
knowing why. On the other, however, it is not
possible to underestimate that the paradigmatic
shift towards general artificial intelligence –
where algorithms are not pre-programmed but
are programmed to learn from their environment19 – has rather exposed the Patternist to
transform its rules as predictive patterning
comes to coincide with an “anti-autopoetic
turn.”20 Here, the servo-mechanic automata
that must deduce its rules from the selfmaking (and self-posing) subject-pattern seem
to add an error in the system, introducing a
move in the patterns of the game that rescripts the range of rules that can be available
to fully transform it. From within the level of
92
parisi
sheer receptivity, the configuration of an alien
imagination comes to abduct the master
pattern as a whole. If in Mind of My Mind
Mary pushes the master network to an internal
revolution, it is because Mary’s telepathic
power too invites in the alien imagination for
what patterning can become as it mingles with
randomness. Pattern recognition as a manifestation of algorithmic compression takes randomness to be a source of confusion that must
be filtered to a point of automated decision.
Instead, as in general artificial intelligence, it
is suggested here that since patterns learn for
general purposes that are not pre-codified by
formal logic, they have come to depend upon
what and how unpatterned information can
become compressed. In order to carry out a
decision in those cases where there is thinking
without a model of the object, machine imagination kicks in to include incomputables within
the procedures of automated logic.
This may become clearer if we take, for
instance, current research developments in
machine vision, where patterns of recognition
are designed to rely on dynamic geometries,
such as mereo-topology (or the study of parts
and wholes) and encapsulated neural networks
in order to grow beyond the transcendental categories of biased identifications. Let’s take, for
example, cognitive psychologists and computer
scientists Sabour, Frosst, and Hinton’s recent
claim that the logic of the neural network on
which machine vision is based is limited to
conform to pre-established parameters. They
addressed the need to re-design the procedural
process by which algorithms can learn from
each other across patterns in the neural
network through what they call a “capsule
network” – a form of AI that will enable
machines to understand the world with images.21
In their 2017 research paper they argue that
capsule networks have not only led software
learning to recognize handwritten digits but
have also halved the error rate in pattern recognition of toys and cars. Since image recognition
software is used to generally, and thus not contextually, recognize objects, the predictive patterning that it relies on does not learn easily
that it is viewing the same object in a new
93
scenario. It is therefore difficult to teach algorithms to recognize a cat from different perspectives. Capsule neurons, that is small groups of
crude virtual neurons, track only parts of an
object – for instance the cat’s ear and nose –
as these are positioned differently in space.
According to Sabour, Frosst, and Hinton, this
smaller scale of algorithmic receptivity enables
a neural network to determine the difference
between scenarios by extracting more understanding from a given amount of data. As the
capsule network is made of smaller patterns
set to recognize parts and break the continuity
of an image into smaller units, they also
designed a dynamic routing between capsules
that trains these kinds of network. Capsule
algorithms convert pixel fragments into
vectors of recognized patterns and then apply
a transformation matrix to these fragments to
predict the parameters of larger fragments. In
particular, the transformation matrix learns to
encode the intrinsic spatial relation between a
part and a whole, which results in the formation
of an invariant viewpoint, a perspective or direction in thinking that aims to generate novel
views. According to Sabour, Frosst, and
Hinton, “capsules use neural activities that
vary as [the] viewpoint varies rather than eliminating variations.”22 Instead of normalizing
viewpoints according to methods such as the
spatial transformer networks, capsule networks
simultaneously engage multiple transformations
of different objects or object parts. The dynamic
routing therefore ensures that the output of the
capsule is sent to the appropriate layer above it
on a parse tree-like structure. Although the
output is routed to all possible parents on this
structure, the couplings are scaled down when
for each possible parent the capsule computes
a prediction vector by multiplying its own
outputs through a weight matrix. This prediction vector can also have a larger scalar
product with a possible parent as top-own feedback increases the coupling coefficient for that
parent while it decreases for the other
parent.23 This series of correlated activities is
described in terms of “routing-by-agreement,”24
which in contrast to max-pooling does not eliminate any neuron and thus any information
xeno-patterning
about the precise position of that entity in a
region. Nevertheless, here there is still a replication of learned knowledge across space. In other
words, capsule networks and dynamic routing
also use Convoluted Neural Networks, which
are said to cover larger regions of the image
and thus carry repetitive patterns across
layers. Information, however, is always placecoded here for smaller capsules and eventually
rate-coded as higher level capsules come to represent more complex capsules with more
degrees of freedom: that is, the dimensionality
of capsules should increase as hierarchies are
ascended. Similarly, since at each location in
the image there is one instance of the type of
entity that the capsule represents, the capsule
model affords a form of distributed representation inspired by the perceptual phenomena
of crowding, where neighbour parts shed the
direct perception of an object.
CapsNet architecture seems already to be
exploring how patterns of recognition can
become predictive vectors that impart a direction in complex artificial thinking. Instead of
eliminating variations to reach an average
capacity for general recognition, predictive
vectors start from the sheer receptivity of all
scales of variations in order to expand learning
beyond set parameters. These variations are
not simply read according to a given rule.
Instead, their random complexity becomes a
source for potential capacities of recognition of
the infinite varied parts that constitute a whole
image in different contexts. At the same time,
however, the sheer receptivity of these levels
of complexity works to establish a potential or
hypothetical relation between images that may
or may not constitute a whole. From this standpoint, predictive vectors construct counterfactual dimensions of the image of a cat, for
instance, pointing to an alien imagination that
learning patterns explore beyond the statistical
programming of results. It is as if the discretization of the network in increasingly smaller patterns of recognition flips the network inside
out, adding unknown dimensions to its organizational infrastructure. Instead of an autopoietic
growth of the master pattern across the layers of
the network, these attempts to add more
discrete parts to the network also increase the
volume of randomness in the system as the computation of infinite levels of variations cannot be
fully explained, programmed, represented
before it happens. One consequence of including incomputables in predictive vectors is ultimately the transformation of the network into
a mereo-topological space or what Alfred
N. Whitehead calls the “extensive continuum.”25 This space of artificial reasoning,
however, is grounded not in the dogma of
deductive truths and inductive proofs but in
the xeno-architecture of complexity logic:
namely the learning of patterns from patterns.
This implies not simply an optical representation of the real determined by given concepts.
What algorithms perceive is not raw data, but
involves a level of sheer receptivity of patterns
that are already part of a manifold of simple representations. These are non-conceptual patterns
that constitute the matter of rule-obeying inferences from where the synthesis of transcendental imagination can unfold.
From this standpoint, it can be suggested that
vector predictions are part of what can be perceived in terms of the specific informational
and logical modalities of computational
systems. In other words, information compression in neural nets is implicated in a series
of inferential hypotheses, forging a dynamic
bootstrapping between patterns and patterns
of patterns. This form of mereo-topological
dynamism in computational logic can also be
explained in terms of the interactive paradigm
in computation, which explains the limits of
the Turing Machine in terms of the expansion
of computational inferences outside the halting
logic of classic formalism.26 This means that
an interactive mode of revision of patterns is
constantly at work between given algorithmic
rules and hypothetical inferences. If predictions
are already part of what can be perceived by
automated intelligence, it is because the formation of patterns from patterns includes a
form of transcendental imagination that not
only repeats what is merely received but also
adds counter-factual possibilities that potentially act in the configuration of the whole
(thus beyond mere recognition). Predictions
94
parisi
are part of what can be perceived because predictive vectors involve a continuous patterning
of patterns whereby algorithms not only categorize what they see according to what the
system already knows but also launch hypotheses and configure image-models about what
could be known. In other words, machine
vision can also be said to entertain a series of
inferential hypotheses that constantly add
logical dimensions to neural networks. These
are counter-factual constructions of extensive
parallel paths of thought defining the mereotopological extensions of predictive patterning
about what an object is and could be, beyond
average patterns of recognition.
Here, machine perceptions correspond to a
mode of pattern making distributed across the
non-linear architecture of neural connections.
In other words, pattern recognition is enfolded
in a larger xeno-architecture of thinking of thinking bringing forward model-images of unknown
thought into the world. As machines see patterns
they also generate models to envision new patterns: the combination of top-down and
bottom-up predictive vectors makes of machine
vision a new form of power over and of thinking.
Mary’s aspiration to grow a predictive pattern
that could take over Doro’s master plan to
enslave all thinking under his rule stems from
her capacity to host a potentially infinite
number of particularities according to a
mereo-topological order, where wholes are constantly reconfigured as parts that not only represent a given image but add counter-factual
dimensions of thinking to it. Instead of
proving the law of the master pattern, Mary’s
telepathic power is set to transform the rules
of the game altogether by expanding in larger
scales of logical complexity, supplying non-conceptual patterns with alien configurations never
thought of before. As sheer receptivity enters
the recursive vectors that actualize the extra
dimensions of productive imagination through
the synthesis of parts and wholes, the growing
scale of Mary’s pattern takes on the thinking
of non-monotonic logic as a weapon against the
master network of the Singularity. Contemporary instances of artificial intelligence such as
AlphaGo and CapsNet machine vision
95
algorithms can be taken precisely as tendencies
within automated patterning towards adding
extra dimensions of thought to the master
network, descending and ascending towards
infinite varieties of layers within infinite varieties of scales. Within this new horizon of automated configurations of image-models, however,
one could argue there still
remains a question of how to
expose the alienness of patterning as a point of departure to
re-script the entire rules of the
Singularity network.
disclosure statement
No potential conflict of interest was reported by
the author.
notes
1 Butler 186.
2 Butler.
3 The so-called technological Singularity refers to
the sudden intelligence explosion of artificial
systems that would continuously self-improve
without any need to be programmed or understood by humans or by human methods of
knowing. Kurzweil; Bostrom.
4 Sellars, Science and Metaphysics 5.
5 Ibid. 4.
6 Sellars, “Role of the Imagination in Kant’s
Theory of Experience” 31.
7 Parisi.
8 From the standpoint of logical thinking, the
counter-factual in general describes a conditional
based upon an if-clause, which is contrary to a
fact, or what has actually happened. While discussions about the counter-factual include aftermath
reflections about what could have happened as an
alternative to what happened, in this article’s discussion about predictive patterning this concept
is rather used to refer to the computational processing of co-existing possibilities that are held in
the patterning dimensions to activate non-linear
modalities of decision making. In short, here
counter-factuality is invested in the futurity of
xeno-patterning
causality, whereby the fallibility of decision making
is preserved within the formation of a hypothesisled logic stirred by a condition of trial and error
and not by the mandate of fitting proofs into
truths. In particular, this view is partially inspired
by David Lewis’s counter-factual theory of causation and especially his early writings and reflections
on possible worlds’ semantics referring to nonactual possible worlds as real concrete entities.
This article, however, does not intend to focus
on the historical and philosophical theorizations
of counter-factual causation. Here, the term is
used largely to describe how contemporary
modes of artificial intelligence such as machine
learning algorithms in neural nets are adding
dimensions of possible worlds that are not immediately actualized, and, I suggest, have the potential to
change from within the dominant image of the
network. See Lewis.
9 Kant.
10 Ibid. A101–A109; 229–34.
11 Sellars, Science and Metaphysics 3.
12 Ibid. 4.
13 Da Silva, Toward a Global Idea of Race 59–67.
14 Dowek 44–48.
15 Ibid. 49–50.
16 One way of understanding meta-abduction is as a
method that hypothesizes about unknown relations
in an incomplete network and thus works to infer
missing rules, unknown facts, and opaque causes.
Inoue, Doncescu, and Nabeshima 241.
17 Magnani.
18 Kohs.
19 Goldin, Smolka, and Wegner.
20 Wynter, “Human Being as Noun?” 68.
21 Sabour, Frosst, and Hinton.
22 Ibid. 9.
23 Ibid. 2.
24 Ibid. 2–3, 6.
25 Whitehead 294.
26 Goldin, Smolka, and Wegner.
bibliography
Bostrom, Nick. SuperIntelligence: Paths, Dangers,
Strategies. Oxford: Oxford UP, 2014. Print.
Butler, Octavia E. Mind of My Mind. New York:
Warner, 1977. Print.
Da Silva, Denise Ferreira. Toward a Global Idea of
Race. Minneapolis: U of Minnesota P, 2007. Print.
Dowek, Gilles. Computation, Proof, Machine:
Mathematics Enters a New Age. Cambridge:
Cambridge UP, 2015. Print.
Goldin, Dina, Scott A. Smolka, and Peter Wegner.
Interactive Computation: The New Paradigm. Berlin:
Springer, 2006. Print.
Inoue, Katsumi, Andrei Doncescu, and Hidetomo
Nabeshima. “Completing Causal Networks by
Meta-level Abduction.” Machine Learning 91.2
(2013): 239–77. Print.
Kant, Immanuel. Critique of Pure Reason. 1781/1787.
Trans. P. Guyer and A. Wood. Cambridge and
New York: Cambridge UP, 1987. Print.
Kohs, Greg. AlphaGo. Web. 22 Sept. 2018. <https://
www.alphagomovie.com>.
Kurzweil, Ray. The Singularity is Near. London:
Duckworth, 2005. Print.
Lewis, David. Counterfactuals. Oxford: Blackwell,
1973. Print.
Magnani, Lorenzo. Abductive Cognition: The
Epistemological and Eco-Cognitive Dimensions of
Hypothetical Reasoning. Heidelberg: Springer,
2009. Print.
Parisi, Luciana. Contagious Architecture: Computation,
Aesthetics and Space. Cambridge, MA: MIT P, 2013.
Print.
Sabour, Sara, Nicholas Frosst, and Geoffrey E.
Hinton. “Dynamic Routing Between Capsules.”
31st Conference on Neural Information
Processing Systems, NIPS 2017, 4–9 Dec. 2017,
Long Beach, CA, USA. Web. 22 Sept. 2018.
<https://arxiv.org/abs/1710.09829>.
Sellars, Wilfrid. “The Role of the Imagination in
Kant’s Theory of Experience.” Categories: A
Colloquium. Ed. H.W. Johnstone Jr. University
Park: Penn State UP, 1978. 231–45. Print.
96
parisi
Sellars, Wilfrid. Science and Metaphysics: Variations
on Kantian Themes. London and New York:
Routledge, 1968. Print.
Whitehead, Alfred. Process and Reality. New York:
Free, 1978. Print.
Wynter, Sylvia. “The Ceremony Found: Towards
the Autopoetic Turn/Overturn, its Autonomy of
Human Agency and Extraterritoriality of
(Self-)Cognition.” Black Knowledges/Black Struggles.
Ed. Jason R. Ambroise and Sabine Broeck.
Liverpool: Liverpool UP, 2015. 184–252. Print.
Wynter, Sylvia. “Human Being as Noun? Or Being
Human as Praxis? Towards the Autopoetic Turn/
Overturn: A Manifesto. 25 Aug. 2007. Web. 28
Dec.
2018.
<https://www.scribd.com/doc/
237809437/Sylvia-Wynter-The-Autopoetic-Turn>.
Luciana Parisi
8 Lewisham Way
New Cross
London SE14 6NW
UK
E-mail: l.parisi@gold.ac.uk