Special Issue: Transversal Posthumanities
Media Ontology and
Transcendental
Instrumentality
Theory, Culture & Society
0(0) 1–30
! The Author(s) 2019
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/0263276419843582
journals.sagepub.com/home/tcs
Luciana Parisi
Goldsmiths, University of London
Abstract
This article takes inspiration from Kittler’s claim that philosophy has neglected the
means used for its production. Kittler’s argument for media ontology will be compared to the post-Kantian project of re-inventing philosophy through the medium of
thought (in particular Deleuze’s Spiritual Automaton). The article discusses these
views in the context of the automation of logical thinking where procedures, tasks,
and functions are part of the instrumental processing of new ends evolving a new
mode of reasoning. In particular, the article suggests that in constructivist logic and
information theory, the temporal gap between truth and proof, between input and
output, can be taken to argue that the means of thought expose the indetermination
or the incomputability of proof. The automation of reasoning in logical processing
coincides not with mindless correlations of data, replacing axioms with data, truths
with self-validating proofs. Instead, the problem of the indeterminacy of proof within
automated logic re-habilitates techne or instrumentality, and the relation between
means and ends away from classical idealism and analytic realism. By following John
Dewey’s argument for instrumentality, it will be argued that the task of thinking today
needs to re-invent a logic of techne away from the teleological view of ends or the
crisis of finality. If the post-Kantian preoccupations about the task of thinking already
announced that the medium of thought could offer possibilities for a non-human
philosophy (or a philosophy beyond truth), this article envisions a machine philosophy originating from within computational media.
Keywords
computational logic, information theory, instrumentality, machine philosophy, media
ontology
Media Ontology and Transcendental Instrumentality
In ‘Towards an Ontology of Media’, Friedrich Kittler (2009) argues that
philosophy has neglected the very medium through which theoretical
Corresponding author: Luciana Parisi. Email: l.parisi@gold.ac.uk
Extra material: http://theoryculturesociety.org/
2
Theory, Culture & Society 0(0)
reasoning has been transmitted. With the modern overlapping of mathematics and media, media can no longer be subsumed to the ontology of
human thinking. Instead, media are to be understood according to the
ontology of machines and the premises of technical knowledge. Kittler’s
argument for an ‘ontology of media’ suggests that the historicization (or
the concretization) of human practices in machines also marks the end of
metaphysics: human thinking is surpassed by a technohistory, a material
technobeing. In particular, Kittler points out that Heidegger’s warning
about the dawn of our computer age is amongst the few philosophical
reflections about the configuration of technobeing in history (Heidegger,
1993 [1969]). And yet the implications of this warning, according to
Kittler, have yet to be fully addressed.
Kittler’s vision of media as a being of techne, however, is to be discussed in the context of the post-Kantian project of critique of philosophy. If techne takes away from philosophy its ultimate qualities of
preserving truths and making decisions by reasoning, as Martin
Heidegger (1993 [1969]) recognized, it is because the new medium of
thought, defined by information sciences, turned thinking into rulebased efficiency. In ‘The End of Philosophy and the Task of Thinking’,
Heidegger argued that the cybernetic medium of thought posed a new
challenge to philosophy. His preoccupation with cybernetic thinking at
once defined the end of deductive truths and the possibility of a new task
of thinking.
While Kittler furthers Heidegger’s invitation to envision media ontology, his question for the task of thinking has been central to the postKantian discussion about the limits of philosophy, the deconstruction
and re-invention of a thought of the medium in practical philosophy. For
instance, Gilles Deleuze’s (1989) materialist philosophy re-works the
scope of philosophy in terms of a being of the sensible that emanates
from a machine thinking beyond apprehension and conscious cognition.
Here media are neither objects nor beings, but are automata of thought
affording the outside or the unthought a capacity to disarticulate cognitive rules in mediation.
For Kittler, philosophy needs to be weaved with information technology to overcome human-centered ontology, resulting from the historical
automation of Aristotelian logic in machines through the works of
Turing, Shannon, Von Neumann, and others (2006: 29). Kittler clarifies
that this automation of logic already began with mathematical innovations, Greek alphanumeric representations, the Aristotelian split
between physics and logic as well as logical and arithmetical models.
Media are the visible expression of revolutions in mathematics and physics that shortened the distance between technology and humanity.
While human thought became caught in computer technology, however, according to Kittler the history of philosophy hardly recognized the
re-invention of techne with informational intelligence. But Kittler’s
Parisi
3
invective against the shortcomings of philosophy seems to overlook that
both deconstructivism and neo-materialism already took media as a
starting point to re-invent philosophy beyond truth and proof. If, for
Derrida, the medium of writing haunted the logical order of thought, for
Deleuze the medium of cinema extended the potentiation of the being of
the sensible above consciousness and signification.
While Kittler seems to be fully aware of the possibilities of a media
ontology in the deconstructivist image of fragmented thinking, it is difficult to determine how his views extend the post-Kantian critique of
human thought, the articulations of an inhuman or machinic thinking,
beyond the question of being. One may want to ask: can media ontology,
or the recognition of a material history of media, offer an alternative
horizon of thought that challenges the image of computational media as
replacing truth with the effectiveness of results? How can the inhuman
thinking of the medium reject the image of computational singularity,
whereby the mindless automation will replace any need for thought? Can
media ontology become transcendental from its own functional tasks?
To address these questions, this article more specifically brings
together Kittler’s efforts to theorize media in terms of their own instrumental materiality with post-Kantian views of turning the medium into
the task of thinking the unthought. One way to bring together these
almost opposite propositions is through the pragmatist account of instrumentality. Borrowing from John Dewey’s view of experimental instrumentality, this article does not see the medium as the implementation of
ideas into tool. The medium is a productive activity, a means with ends,
or a doing imbued within an experimental logic, involving the indeterminacy of results leading to the reconstruction of aims or premises.
From this standpoint, modern techne is not here understood according
to the critical theory of instrumentality. For instance, instrumentality in
Horkheimer’s critique of technology coincides with instrumental reason,
namely with the paradoxical condition for which the Enlightenment trust
in the rational use of nature has led to an irrational self-repression of
human nature, obsessed by a competitive self-preservation mirrored in
the indiscriminate dominion of commodity economy (1974: 97–105). The
effort to re-habilitate instrumental reason beyond the paradox of criticizing critique requires that one suspends the impasse between the use of the
rational capacity of explaining the causes of economic domination and
the argument that economic governance rationalizes its aims for profit,
through the efficient capacity of means to carry out ends.
This article instead proposes to recuperate from instrumentality an
experimental logic in theoretical and practical activities, whereby practice
is not simply the doing of pre-existing ideas, but more importantly
becomes the means for knowing this and that from knowing how
things work. Similarly to Heidegger’s quest for the task of thinking, as
moving beyond the simple revelation of truth, and together with the
4
Theory, Culture & Society 0(0)
post-Kantian theorization of the medium of thinking as anti-telos, this
article suggests that computational media announce a new dawn for
instrumentality: not only the evaporation of telos in the practical being
of media, but also the origination of a techno-philosophy through the
inhuman logic of machines.
If post-Kantian articulations of posthuman thought had already rehabilitated instrumentality as demarcating the end of telos, the transformation of computational logic in machines can further offer us new
possibilities for re-articulating what it is to think, what reasoning and
knowledge have become in and through this medium. Instead of declaring the end of reason, truth, and axiomatics in the age of data-centered
epistemologies, this article argues for a renewed engagement with a transcendental instrumentality that is of a possibility of machines to think
beyond what they do.
It could be argued that one way to account for this mode of transcendental instrumentality is already at work in recent investigations about
what machines see and how they interpret the world. For instance, the
work presented at the exhibition I Am Here to Learn: On Machinic
Interpretation of the World (The Frankfurter Kunstverein, 2018) pays
particular attention to how automated decision-making includes the
transcendental becoming of the instrument. Shinseungback
Kimyonghun’s work, for instance, brings forward this indeterminacy in
automate decision, as when algorithms start seeing flower patterns that
do not correspond to the image of a flower as we know it (http://ssbkyh.
com/works/flower/). Drawing from similar artworks that reflect upon
this crucial aspect of machine learning, this article suggests that what
appears as a form of misrecognition, error, and doubt is instead part and
parcel of automated reasoning, which is here understood in terms of
transcendental instrumentality. This article will draw on some of these
instances of machine visions as practical attempts at exploring transcendental instrumentality and discuss how machines learn to interpret and
understand, and thus think the world.
The post-Kantian discussion of posthuman thought in the age of
computation is already a way to re-direct the critique of informational
media away from the view of the eclipse of reason, which also rejects
the dominant image of big data as determining the absence of meaning
in the practical knowledge of machines. To re-direct our critique of
technology today requires that the task of thinking with and through
machines is re-invented. From this standpoint, the transformation of
logic in machine thinking importantly shows us the temporal indeterminacy between truth and proof, which can be taken to re-define the
instrumental relation between means and ends. The question of what
the task of thinking has become with and through the computational
automation of thought coincides not with the triumph of means over
ends, but with instrumentality affording the medium of thought its own
Parisi
5
mode of reasoning and de-naturalization of knowledge in human
culture.
If we take, for instance, artist Fito Segregra’s installation The
Treachery of [Soft] Images (2016), it is possible to track how machine
interpretations of objects (from a pipe to a broom, a sponge and a jar) do
not just reproduce the corresponding category of the object, but become
instances of a new conceptual reality (http://fii.to/pages/the-treachery-ofsoft-images.html). Similarly, Shinseungback Kimyonghun’s installation
Cat or Human (2013) uses cat facial detection algorithms to recognize
human faces, and human faces facial detection algorithms to recognize
cat faces (http://ssbkyh.com/works/cat_human/). As a result, these facial
detection algorithms impart a de-naturalization of what we know of both
humans’ and cats’ facial features, by detecting human traits in cats and
vice versa. From this standpoint, these investigations into the computational medium of thought also point to the specific importance of fallibility in automated reasoning: namely, the new data categories of objects
are invented and do not correspond to the images of objects inputted into
the system. But how to distinguish between what machines learn beyond
their function of data aggregation and what instead remains simply a
reproduction of the already known?
Central to this discussion are three steps in the transformation of the
deductive model of knowledge shifting from truths to proofs, from theoretical reason to the practical procedure – or instrumentality – of
machine thinking. First, it discusses the implementation of mathematical
postulates in the Turing machine as the point at which the limits of
Hilbert’s (1996) meta-mathematical project challenged the infallibility
of theoretical knowledge. It then draws on Brouwer’s (1913) constructivism in logic to provide examples of how temporality in logical practices
involves proof-validation and not self-consistent truths. Here proofs are
not simply the result of an automated and mindless correlation of data,
but a means of thought that implies a form of actuation doubled by the
indeterminacy or futurity in proof-validation.
Similarly, Turing’s famous halting-problem showed that it was impossible to know in advance whether and when a program will stop. This
fallibility or incompleteness of theoretical knowledge was set in action in
computational machines. The demonstrative function of techne – as
application of theoretical knowledge – is here catapulted by the realization that proofs cannot be derived from given truths. Another important
step in this argument about instrumentality involves a discussion about
how this temporal gap in logic was central to the development of algorithmic information theory (Chaitin, 2005). Since computational logic is
based on the probability of results, it involves the finding of proofs that
may or may not validate the premises of the program. Similarly to constructivism in logic, the question of information complexity in computation, that is, to what extent it is possible to compress random strings of
6
Theory, Culture & Society 0(0)
data into intelligible algorithms (i.e. probabilities), cannot be fully
exhausted without accounting for the experimental (or future) validation
of results. In complex information systems, not only can incomputables
not be compressed into smaller and finite sequences of algorithms, but
they also expose dynamics in computational logic, where proofs preserve
degrees of indeterminacy.
The consequences of this method of truth-experimentation (or experimental axiomatics) are important to consider here: if the dawn of computation has meant the completion of philosophy in the efficiency of
proof-making, the task of philosophy requires that thinking includes
the machine practices of working through indeterminacy. In both computational logic and information complexity theory, this margin of indeterminacy is the incomputable of any system of truth and proof, of ideal
and empirical methods of knowledge. Even if some practical solutions
have been implemented in computing to skirt around the problem of the
incomputable (for instance, the use of exceptions, a routine in a program
and in an operating system on a standard computer), computational
logic is set not to eliminate but to work through the problem of
indeterminacy.
With computation, therefore, the medium of thought has not simply
replaced theoretical thinking with the operational efficacy of task findings. As argued later, computational logic and information processing
show that the practice of searching for truth is rather an experimental
logic involving a retro-active validation (confirm, discard, or revise) of
truths.
By bringing together Kittler’s proposition about a media ontology and
the post-Kantian re-envisioning of the image of thought in Gilles
Deleuze’s notion of the ‘Spiritual Automaton’, this article concludes
that the medium of thought in the age of computation could be understood in terms of a transcendental instrumentality beyond the big data
image of automated knowledge.
Automated Knowledge
As automated systems become increasingly intelligent and capable of
making decisions, it is no longer possible to deny the profound threat
that the age of machine thinking has unleashed on human culture. From
the call to arms about the existential risk of human extinction to the
global plans for full automation and transhuman singularity, the foundation of knowledge in the humanities, classically centered on the distinction between theoretical and practical reasoning, has become
redundant to human culture itself. If the ontological in-distinction
between human and machine thinking is animating debates about
the danger that artificial intelligence will pose to human autonomy on
the one hand (Bostrom, 2014) and about how automation is accelerating
Parisi
7
the capitalization of thought and life itself on the other (Terranova,
2014), the very question of what counts as knowledge has to be revised.
The automation of knowledge cannot be disentangled from the transformation of the humanities, and from how the task of thinking conforms to the efficiency of big data and algorithmic mining of over-reality.
Automation today does not involve the industrial assembly of movements or the acceleration of networks that characterized the 20th century. It no longer embodies the mechanics of Newtonian physics based
on a repetitive cycle of cause and effects (Longo, 1999). Since its expansion in the decisional models of cognition, automation has entered the
temporality of reasoning (that is, of the time between truths and proofs)
and has unleashed a mode of knowledge production that is self-sustaining, continuously feeding from the deep sea of data. Algorithms talk to
other algorithms (through set protocols and through learning) without
communicating with us and draw conclusions by correlating data (images
and texts, sounds and locations) across parallel and distributive networks. From high frequency market trading to security data prediction,
from military to commercial logistics, automation today seems to have
debunked the dominance of theoretical knowledge and its axiomatic
truths: what is known in theory is confronted by algorithmic processing
interacting with the external world of data. In other words, automation
no longer is the application of given truths but challenges the very fundament of the philosophical autonomy from instrumental knowledge.
In the Republic, Plato wrote of the Socratic distinction of craft knowledge from philosophical logos and argued that craft knowledge only
concerns technical understanding that can be used to define the pursuit
of a particular trade or practice (Plato, 1992). Craft knowledge must be
instrumental to something. The water clock, the astronomical orrery, the
mechanical puppet are all primordial automated devices that are used to
demonstrate or describe something. Craft therefore reflects practical
understanding and is to be distinguished from philosophical knowledge,
which is rational, of a mathematical order and requires no instruments.
Whilst technique coincides with primitive automata, which are thoughtless and mainly concern practical understanding, philosophical knowledge involves the cultivation of the principles of all things. It is ideal
and timeless.
Fundamental to the humanities is this bifurcation between thinking
and instrumentality, abstractions and applications. The origin of human
knowledge is attached to the division between knowledge (theory) and
knowing (practice), whereby the philosophical method is set to prove how
we know that we know (i.e. how knowledge can be demonstrated). The
pristine hierarchy of this model, however, was changed with the design of
logical systems that demonstrate the mechanisms of thinking. Whilst for
Plato the a priori existence of a mathematical order of ideas would sustain the foundation of knowledge that could then be applied and
8
Theory, Culture & Society 0(0)
demonstrated practically, the search for a logical system that could
instantiate the mechanisms of knowledge was central to Gottfried
Wilhelm Leibniz’s calculus ratiocinator.
In the 17th century, Leibniz devised a model of knowledge in the form
of a logical calculation framework based on theoretical premises. The
calculus ratiocinator was designed to automatize the working of thinking
and the production of knowledge. Its scope was to perform logical
deductions within the framework set by what Leibniz called characteristica universalis, a universal language whose symbolic structure could
automatically express the structure of concepts and their recombination.
This mechanism of and for knowledge was intended not to simulate
human cognitive capacities but to establish, according to the deductive
principle of sufficient reason, the universality of logical thinking.
Leibniz’s attempt at a logical systematization of knowledge presented
an image of thought in terms of mathematical rules implemented in
machines, whose recombinant capacities could give expression to any
possible knowledge. According to Giuseppe Longo (1999), this metamathematical model eventually led to the 19th-century Laplacian view
of the mechanical universe. But the implementation of human reasoning
in machines led not only to the automation of logical thinking, but has
also brought to the fore the question of the ontological autonomy of
techne – or practical thinking – by and through information machines
(Kittler, 2006). As with Leibniz’s vision for an automated mechanism
that could embody a universal logic beyond all contingencies by evacuating temporality from meta-mathematical thinking, so the 20th-century
invention of the Turing machine separated the abstract machine and
physical implementations. Here a serial number of steps was supposed
to mirror how logic operated progressively from one preliminary condition that already contained the proof of its results (Longo, 1999).
With information machines, however, automation entered the history
of media technology because, as Kittler (2006) argues, mathematics and
media became reconfigured into one model of command, storage and
transmission. This conflation of media and automated models of reasoning crucially revealed that embedding logic into media made this logic
different. With Kittler, one can argue that it is the very medium through
which theoretical thinking operates that comes to gain a new meaning
with the convergence of media, computational logic and information
complexity.
According to Kittler, this convergence defined techne as an original
dimension of human culture. The technology incorporated in media systems is neither an instrument for cultural expression nor an extension of
sensorimotor or cognitive functions. The being of techne instead is constituted by the practical knowledge embedded in the historical evolution
of systems and the conjunction of media with cybernetics and computational logic. The historical formation of media ontology shows that
Parisi
9
instrumental knowledge qualifies the information processing of the
Turing machine. In particular, with silicon-based microchips and the
incarnation of the Turing logic in the Von Neumann architecture of
computing, Kittler theorizes the being of media as involving not a dichotomy between matter and form, but more crucially a ‘new trinity made up
of commands, addresses, and data’ (2006: 30). With computer technology, there emerges the realization that media are able to self-govern and
are not simply tools that are externally controlled.
This form of computational autonomy of the medium, according to
Kittler (2006: 28), brings back a central limit within the history of philosophy, namely its neglect towards the medium it uses to think and to
know. Kittler argues that only Heidegger recognized the incumbent
threat that cybernetics and the emergence of computers posed to philosophy, and came to the realization that thinking had be transformed
into a task: the task of thinking.
Heidegger’s view of this modern form of instrumentality corresponded
to the mechanization of the principle of reason in the technical process of
calculation or, in other words, techne1 involving both the computation of
thought and the mathematical abstraction of content, where information
became divorced from meaning. Since modernity, techne as rational
instrumentality had taken over thinking, through the logical reduction
of reasoning to technical processing, ratio or calculation – embedding
man and machines in endless feedback loops bound to information processing, storage and transmission (Kittler, 2006: 29).
In the essay ‘The End of Philosophy and the Task of Thinking’ (1969),
Martin Heidegger argued that since the late 1940s the advance of cybernetics, a technoscience of communication and control, demarcated the
point at which philosophy became verifiable by testing. As truths were
finally subsumed to the effectiveness of automated results, judgment itself
was replaced by self-validating proofs. Thinking conformed to the
manipulable arrangement of a scientific-technological world and its
social order. For Heidegger, the completion of philosophy is also the
possibility of overcoming the logic of deductive truth that precisely reemerges in the matter or the means of philosophy itself, that is, through
the medium of thought.
Here the advance of cybernetics eventuates
the determination of man as an acting social being. Cybernetics is
the theory of the steering of the possible planning and arrangement
of human labor, transforming language into an exchange of news
and the arts into regulated-regulating instruments of information.
(1993 [1969]: 376)
Philosophy is turned into a technoscience that intercommunicates with
others, ultimately losing its metaphysical totality. Here techne overtakes
10
Theory, Culture & Society 0(0)
the philosophical task of explaining the world and the place of ‘man in
the world’.
Under this new condition of techno-erasure of metaphysical truth,
Heidegger insists that the new task of thinking lies outside the distinction
of the rational and the irrational, the decidable and the incomputable.
The transformation of philosophy into cybernetic modularity can only
give way to iterative patterns that blend logos into ratio and subsume
truth to proof. Thinking, however, cannot be proven to exist because
truth cannot be contained within techno-scientific epistemology. The
automation of thinking can tell us nothing about truth, as the latter
must remain outside what is already known. This is why in the age of
meaningless communication, according to Heidegger, one must turn to
the task of thinking, a mode of education in how to think (1993
[1969]: 392).
It is precisely this question of what the task of thinking has become in
the aftermath of computation that Kittler’s proposition for media ontology wants to address. For Kittler, the end of philosophy indeed contributes to the realization of an ontological configuration of techne brought
forward by a historical condition, in which, he laments, the ‘task of
thinking has barely begun’ (1969: 29). Instead of mirroring classical
metaphysics (form vs matter, mind and body, theory and practice),
media ontology shows us that the task of thinking must address the
functions of commands, addresses and data bringing together logic and
information across physical, biological and technical systems.
As Kittler notes, Heidegger particularly lamented the specialization of
philosophy in independent sciences that communicate amongst each
other as if they were cybernetic systems (Heidegger, 1977: 374). Since
knowledge had become dependent upon the cybernetic effectiveness of
results, theoretical reasoning was turned into ‘representationalcalculative thinking’ defining the completion of philosophy, or
the means by which philosophy could be transformed (Heidegger,
1977: 376–7).
By partially following Heidegger, Kittler radicalizes the view that
modern technology imposed a radical transformation on human culture.
If cybernetics is the embodiment of instrumental thinking and practical
knowledge, then, Kittler suggests, the computational turn of the Second
World War marked an epochal change in which practical knowledge
would finally acquire autonomy from theoretical reasoning. Whilst
arguing for a historical formation of this technical ontology, Kittler
reveals the ontological significance of instrumental knowledge as this
became transformed by the capacities of the Turing machine to establish
a meta-model of thinking.
Nevertheless, Kittler’s re-articulation of media ontology or ontology
of techne argues for the withdrawal not only of philosophy – or theoretical reasoning – but also of the human subject and rational judgment
Parisi
11
(2006: 30). He invents a materialist method in which formal logic is
replaced by the practical knowledge embedded in the circuits and processors of machines. As techne comes to involve operations of information compression, data encoding, logical procedures and algebraic
operations, so does the discrete ontology of machines, and its binary
language, come to challenge the dominant history of theoretical
reasoning.
Kittler’s insistence on the end of rational judgment extends the
Heideggerian critique of techne with the image of a poetic crafting (or
material making) in shaping thought. However, this reinforcing of the
ontological distinction between techne as craft vs techne as instrumentality (i.e. the incarnation of rational thinking in modernity) seems only
to re-impart a separation between poiesis and functionality, characterizing critical theory’s mistrust of technology, practical reason and of automated thought.
Kittler indeed seems to insist on the distinction between theoretical
and practical reasoning by re-articulating the distinction between software and hardware. The Turing machine proposed a digital uniformity
between data and programs entailing no physical difference between processors and processed. In his famous essay ‘There Is No Software’,
Kittler (1997) shows that the logic of computation is absorbed in the
practical knowledge of circuits, tapes, microprocessors, and switches.
By embedding theoretical stances into practices, Kittler reveals that the
material storage, transmission and command of information has produced an autonomous system of knowledge derived from the automation
of human activities or by ‘the human use of human beings’
(Wiener, 1954).
Whilst Kittler’s historical reconstruction of media ontology aims to
ground technobeing in the materiality of circuits, and ultimately defy the
dominance of formalism, my attempt here is not to argue that, with
computation, it is possible to reclaim the ontological priority of automated knowledge, coinciding with the operational crafting of procedures.
Instead, one can argue that Kittler’s claim for an ontology of media can
be seen to contain preliminary insights for discussing not only machines
crafting, but also and more specifically how modern instrumentality
demarcated a historical automation of logic, the advance of an alien
(or denaturalized) becoming of thought: a techno-logic.
This article does not follow Kittler’s inclination to share the
Heideggerian view about the completion of metaphysics, proposing a
bifurcation between techne, as the instrumental rationality of a technosapient man, and poeisis as crafting the un-concealing of truth in time
beyond function. While Kittler’s proposition for a historical re-articulation of the being of techne in the practical thinking of machines argues
that philosophy has been overcome by technical crafting, Heiddegger’s
critique rather seems to re-introduce truth as that which cannot be
12
Theory, Culture & Society 0(0)
revealed by and through the instrument of thought. In short, Kittler’s
view of media ontology intends to reclaim the material autonomy of
technology from philosophy – or theoretical knowledge – defining a profound transformation in human culture. However, this proposition differs from many of the post-Kantian efforts to re-envision what it is to
think after the crisis of metaphysics. In other words, Kittler’s view of
instrumentality seems to be limited to a techno-praxis of thinking that
eliminates the unthinkable from its horizon,2 evaporating abstraction
through the efficiency of functions without causality, finality or even
becoming.
From this standpoint, Kittler’s media ontology is not concerned with
the post-Kantian effort to re-invent philosophy through machines. Here
the medium of thought is not only taken to annul the separation between
form and matter, but to push thinking towards the unthought, that is,
extending the horizon of knowledge towards what is not known. It is this
possibility for a thought of the outside, or for the inhuman becoming of
thought, that a post-Kantian take on instrumentality seems to offer us.
Outside the Medium
As media have become instruments of exploration for what thought can
do, mediation itself has become the channel where the unthought is
encountered as if emanating directly from automated spatio-temporal
frames that liquefy cognitive representations. More than a series of
material functions, automation becomes a means of transformation insofar as it captures the intensities of movement and of thought by exposing
the spatio-temporal gaps in the logical order of successive states. One has
to turn to Gilles Deleuze’s discussion of the cinematic medium to envisage how the instrument itself can re-direct its ends, not towards demonstrative functions, but towards opening intensities in thinking.
Automation is here less a question about the hardware ontology of
means and more about how means expose the denaturalizing or the
alien becoming of thought. In particular, the cinematic automation of
spatio-temporalities radically affords a new plane of possibilities for the
movement of thought that was not there before.
According to Deleuze, automation can afford us an enquiry into the
question of what is to think, and not simply in how techne grounds
thinking into the concreteness of machine circuits. As Deleuze claims,
‘[t]he automatism of cinematic images correlates with the automatisms of
our thinking, the pure material organical-psychic mechanisms that perform our thinking without consciousness’ (1989). Although Deleuze
takes cinema – a time-based form of automation – as an instance of
the medium of thought, he is arguing for a Spinozist auto-movement
of thought that links ideas, without reference to objects. This transcendental automation continuously interferes within the spatial sequencing
Parisi
13
of time. For Deleuze, the automaton is not simply a material agglomerate
of machine-based tasks but is outside the medium, containing within
itself the potential for the auto-movement of thought. Deleuze
calls this Spinozist image of a thought that itself thinks the ‘Spiritual
Automaton’.
In particular, the instrumentalization of time in cinema is for Deleuze
an opportunity to push forward a non-representational and noncognitivist image of thought because automation does not simply reproduce time but presents to us a denaturalized and impersonal time
through the cinematic superposition of images that are at once a reflection and distortion of the assembly machine of industrial capital. Here
instrumentality already aspires to its becoming transcendental, or in
Deleuze’s terms (1989), ‘spiritual’, because the automation of movement
– e.g. clockwork automata, motor automata, etc. – had already the germs
of the transformation of static mechanisms into a dynamic circularity of
flows with the advent of the age of computer and cybernetic automata of
thought equipped with control and feedback. The serialized automaton
of industrial capitalism was thus already preparing the ground for informational time, a networked order or spatial matrix of autonomous interconnected agents. As the cinematic time machine was a symptom of an
image of thought that had excarnated thinking from subjective perception and cognition, so are computation and the structuring of the Turing
machine in the Von Neumann computational machines imparting a new
order of thought.
Cinema becomes the automaton of the temporal overlapping of the
past and the future because one can go back and forward and cut in the
middle of a sequence to add another temporality. With the automation of
thought, with control and feedback, instead we have interactive agents
that grow or evolve in time to transform space. For Deleuze, the spiritual
automaton represented the third synthesis of the time image; it carried
with it the intensity of thought, elevating the being of the sensible from
mere sensori-motor responses and mental re-cognition (1989: 265).
Here the cinematic machine, unlike simple mechanisms, concerns not
the carrying out of tasks and the efficiency of functions, but it deploys
indeterminate sensory or aesthetic components through which it enters
the larger arrangements of our senses, our bodies and brains. As a nonverbal automaton of images, cinema exposes the role of time and space in
thinking, whereby chronological and geometrical orders are intersected
with topological configurations and overlapping temporalities. Instead of
being tied to human consciousness, cinema reveals the inhuman activities
in the circuit of thinking as shocks in brains. The automated series of
images coincides with a supra-conscious dimension of being involved in
the molecular automatisms that process information and perform thinking autonomously, without conscious thought or cognition. As a medium
of thought, the automation of images brings the task of thinking towards
14
Theory, Culture & Society 0(0)
a non-philosophical dimension of thought, not to explain abstraction
away, but rather to unleash the being of the sensible from the semiotic
chains of meaning. In other words, the automation of images enables a
new form of correlation of the material and the ideal, the affective and
the intellectual, that bypasses cognitive representation and the deductive
model of truth.
As non-human thought above all exposes pre-individual, prerepresentational affects and percepts, it also manifests itself through
the fallibility of reasoning and the inability to think of the whole.
Automation here means that we can grasp an image only when it is
already passed: the process of association is constantly interrupted,
deconstructed, dislocated and then constructed anew. Insofar as
cinema becomes a medium of thought, it also shows how it re-directs
the association of images towards a new horizon. Here means do not
replace ends, but are generative of a final cause that arrives in the middle
of the process to immanently express intensive variations in time. In
other words, the spiritual automaton works through the self-movement
of thought resolving the tension between automation and philosophy
through the instrumentality of mediation, extending thinking outside
the medium. Instead of ante-posing truth to the medium, we have here
a thought emerging from the way the medium thinks the outside.
However, as Felix Guattari (2001) points out, the machinic is not
techne; it has no substance and has no pretense of un-revealing the
truth. Similarly, the machinic does not coincide with the being of
media, or an ontological technics that originate within the history of
automated functions. The machinic principle of heterogeneity instead
allows for a generative inter-kingdom of thoughts that belong to no
individuated subject or object because it is mediation itself that sets up
aims within changing milieus. Here, the task of thinking is not assigned
to the medium but to the question of thinking itself because the means of
thought (the cinematic means) show that cognition, reason and logic are
subjected to the self-movement of affects and percepts. The origination of
a non-human thinking is therefore passed through the means or the
process of mediation extracting futurity from temporal sequencing and
turning geometrical spaces into continuous milieus.
But how can the task of thinking in the age of computational intelligence move beyond the separation of technics from process, or of the
medium of thought from the becoming of thinking? If the post-Kantian
project of re-inventing the task of thinking in terms of a media philosophy can be still relevant today, the question of what is thinking must
challenge a vulgar vision of instrumentality. In other words, this question
must reject the future image of thought caught in the arms race for
planetary computation and singularity (Kurzweil, 2005), where automation will replace the biological stratum with intelligent data learning systems (from nanobots to supercomputers AI) and meta-systems (from
Parisi
15
robotic finance to robotic medicine, manufacturings, logistics, etc.)
making algorithmic connections across scales. As opposed to the hype
about the arrival of a super-intelligence, the image of anti-thinking par
excellence, the task of thinking today must include the re-invention of
instrumentality in relation to machine thinking, reasoning and knowledge, against the imperative of big data and its mindless association of
functions and concepts.
But what does it actually mean to say that machines can think? Hasn’t
the critique of technology, from Heidegger to Deleuze and even
Laruelle,3 indeed argued that the immanence of thought must pass precisely through the promise of the non-reflective and non-decisional
media?
We have seen that for Deleuze cinema is instrumental to philosophy
because instrumentality is the means by which the being of the sensible
passes through and beyond cognition to become unleashed in machine
processing. On the other hand, however, Kittler’s plea for media ontology sees instrumentality as a material history of communication technology where information storage and command neutralize metaphysics and
show that no thought could occur outside the means of its production.
Kittler’s argument goes even further because he sees the replacement of
silicon-based binary language with quantum computing as a promise for
instruments to think on their own accord, erasing the limit between philosophy and automation for good. Instead of an immanent knowledge
veering from the being of the sensible, the percepts and affects entering all
sorts of machinic assemblages, the historical ontology of technics is concerned with the evolution of the technical stratum.
It has been argued that the limits of formal axiomatic reasoning that
characterize the Turing machine paradigm no longer reflect the computational power of information-processing devices because these have
changed dramatically compared to their original function of centralized
and sequential processing of data. Contrary to traditional computation,
in which the computer, provided with a suitable algorithm and an input,
was left alone to crunch the numbers until the program terminated, today
interactive super-recursive and machine-learning algorithms instead
employ the external world of data to direct the computation (DodigCrnkovic, 2008 [2006]).
This also means that algorithmic automation operates at a quasiautonomous scale distinct from the biophysical order of probability
and chance. Indeed, machine learning, for instance, only deals with a
kind of randomness internal to automated procedures, which are delimited compared to the multilayered randomness of biological systems
(Calude and Longo, 2014). To unpack the formation of the computational stratum, therefore, one has to address the historical development
not of techne in terms of its physical qualities, but of the logic of techne,
involving the inclusion of randomness or the unknowable in logic.
16
Theory, Culture & Society 0(0)
Whilst Kittler’s historical reconstruction of media ontology aims to
ground technobeing in the materiality of circuits, and ultimately defy the
dominance of formalism, Deleuze’s spiritual automaton instead reinvents formalism in terms of a virtual plane able to de-form and engender any structure. Here the post-Kantian promise to re-invent the image
of thought beyond the human form is worked through the inhuman
functions and processes of machines. With this promise of another
origin of philosophy with and through modern techno-science, it can
be argued that both Kittler and Deleuze see media as symptoms of
epochal transformations belonging to an epistemic re-arrangement of
what it means to think, to know, to perceive beyond the metaphysics
of truth, logical reason, and cognition. However, if Kittler’s claim for an
ontology of media offers us preliminary insights about modern instrumentality and the historical automation of logic, Deleuze’s automaton
admits that the medium pushes human thought to encounter its unthinkable horizon, the unexpected potential to become more than what it is.
My attempt at discussing the medium of thought in terms of instrumentality focuses on the tension between means and ends, efficient and
final cause in the context of mathematical and computational conceptions of truth and proof, pattern and randomness, executable and incomputable. This is an argument for a dynamic form of automated logic in
and through computational thinking.
Techne-Logic
It is possible to argue that with Kittler and the post-Kantian critique of
metaphysics, from deconstructivism to materialism, there is already at
play a techno philosophy involving a transformation of what thinking
with and through machines can be. While logical thinking seems a continuation of theoretical reasoning and its efforts to establish a formal
consistency between truths and proofs, the post-Kantian preoccupation
with the medium of thought importantly resonates with preoccupations
in mathematical logic about the eclipse of propositional reasoning.
According to French logician Gilles Dowek, in the 20th century the
mathematico-philosophical efforts at establishing a universal logic
though which truths could be deduced entered the sphere of computation. We know that predicate logic, as defined in the axiomatic conception of mathematics, consists of inference rules that enable proofs to be
built step-by-step, from axioms to theorems. If, for the Greeks, numbers
and geometric figures were objects of study, and reasoning was a method
for the means to illustrate the ends of an axiom, 20th-century mathematics turned reasoning itself into an object of study. Predicate logic would
thus become a first step to explain the rules of inference in an algorithmic
way, that is, through a precise procedure that could decide whether a
proposition was true or false. In particular, David Hilbert’s ‘decisional
Parisi
17
problem’ searched for an algorithm that could be applied to any proposition. A problem solved by an algorithm was called ‘decidable’ or
‘computable’. In the attempt at replacing reasoning with a computing
operation, Hilbert developed a method in which proofs were finite
objects, aiming to establish the independence and consistency of
axioms (Dowek, 2015: 46–8).
However, the search for a complete reasoning – or algorithmic logical
procedure – that could eliminate infinity and contain all propositions
within the decidability problem became haunted by its limits. In 1930,
Gödel made an effort to advance Hilbert’s program by attempting to
prove the consistency of analysis (or second-order arithmetic) with the
resources of arithmetic, but was forced to conclude that arithmetical
truth cannot be defined in arithmetical terms (Goldstein, 2005). Gödel
presented two incompleteness theorems that explained the limits of provability in formal axiomatic theories. In particular, the second incompleteness theorem established that a formal system could not prove that the
system itself would be consistent.4 Since certain propositions are ultimately undecidable, they cannot be proved by the axiomatic method upon
which they are predicated.
In 1936, Alonzo Church and Alan Turing used computation to formalize the procedural function of algorithmic reasoning, and encountered Gödel’s incompleteness problem by discovering that certain
propositions could not be decided or solved by an algorithm. Insofar
as predicate logic is incomplete, so does computation expose the impossibility to know in advance when an algorithmic procedure will halt,
proving a proposition to be true or false. Propositions that cannot be
solved by an algorithm are therefore called indecidable or incomputable
(Dowek, 2015: 51–3).
Turing’s incomputable and Gödel’s incompleteness laid open the
pristine hierarchy of theoretical reasoning – the a-temporal grounds of
mathematics and philosophy – and instrumental knowledge. If the postKantian preoccupation with the medium of thought contained in germ
the re-invention of philosophy through instrumentality, the replacement
of logical reasoning with algorithmic procedures not only revealed that
truths became programmable by machines – and could thus transcend
human thinking – but also that formal reasoning (i.e. that general axioms
contain any particular instance) became weakened by computational
proof and incomputable propositions. The problem of infinity returned
in logic through techno-scientific instrumentality in the form of incomputables, exposing the fallibility of theoretical reasoning in machine
thinking.
From this standpoint, one may ask, what are the consequences of
computational instrumentality vis-a-vis a re-invention of philosophy
with and through techne? One could argue that the consequences of
this historical transformation of mathematics and axiomatics into
18
Theory, Culture & Society 0(0)
computational rules that self-validate proofs are to be found in our contemporary image of digital automation as a mindless procedure of decision-making, unable to reason about its ends. That computation – or the
transformation of theoretical reasoning into an object of study – exposed
the limits of deductive logic in favor of self-validating proofs or computable functions announced not simply the end of reasoning but, one could
argue, the origination of an instrumental thinking of the unthought.
One can turn to constructivism in logic and experimental axiomatics in
information theory to bring forward another image of computation and
self-validating proofs that challenges the legacy of the Heidegerrian
vision that techne equals a mindless processing of data. Constructivism
rather offers alternatives to re-articulate the relation between truths and
proofs, means and ends, practical and theoretical reasoning.
In 1913, L.E.J. Brouwer’s constructivism showed that mathematics is
inexhaustible and cannot be completely formalized. As a general system
of symbolic logic, constructivism relied not on the traditional notion of
truth but on the concept of constructive provability. In classical logic,
propositional formulae are always assigned a truth-value (true or false)
regardless of whether there is evidence or proof for either case. For constructivism, there is no assigned pre-established truth-value. Instead,
propositions are only considered ‘true’ when we have direct evidence or
justification or proof, which requires the ingression of time into logical
reasoning. Here time involves the relation between finite and infinite
series of numbers, which Brouwer discusses as two acts of intuitionism.
The first act has its origin in the perception of a movement of time. The
second concerns choice sequences, the creation of an infinite sequence of
numbers that provide a certain infinite set of properties. The sequence,
however, can either be a law-like sequence or algorithm (such as the
sequence consisting of only zeros, or of the prime numbers in increasing
order), or be simply lawless (such as a repeated throw of a coin).
Brouwer’s acts of intuitionism are based on a constructive account of
truth: for a proposition to be true there shall exist a constructive proof
that can abide to the law of axiomatic consistency. Proof is meant to
demonstrate the premises of the reasoning by which truth can be determined. However, proof has been understood both in terms of ‘actualism’
in reference to an actually existing proof, and ‘potentialism’ involving
how a hypothesis is potentially provable (Trafford, 2014: 23). Whilst
potentialism seems to re-inscribe a Platonism in logical thinking, in
terms of a tenseless time, actualism focuses on the act of proving and
not on the proof as an object (2014: 23). Intuitionism therefore is concerned with what happens in the practical process of making a decision
for a hypothesis or proposition. This temporal process involves that the
proposition is not already known to start with, and its validity cannot be
decided until a later, future moment. Intuitionism thus pushes the limits
of constructivism, by adding the temporality of practical thinking to
Parisi
19
proof validation. Importantly, it challenges the constructive ideation for
which it is in principle always possible to find the right proof that fits
propositions or given premises a-posteriori. The acts of intuitionism
instead show that the retroductive construction of proof inevitably
includes the existence of infinite sequences that cannot be fixed in
advance.
Whilst for Platonism, mathematical statements and philosophical
thinking are tenseless (i.e. they need no proof outside their own premises), for constructivism truth and falsity have a temporal aspect: an
established fact will remain so, but a statement that becomes proven at
a certain point in time may come to lack a truth-value before that point.
This temporal aspect becomes the instrument of a logical method where
the search for proof is the practice – or actuation – of validating truths.
In other words, proof as demonstration coincides with instrumentality as
a process of validation of theoretical premises, whereby logic moves in
two directions, forward and backward on a continuum, both deductively
from premise to facts and inductively from fact to premises. In short,
with Brouwer’s notion of two-ity, there is this double activity of logic
that exposes rational thinking to the indeterminacy of proof.
From this standpoint, if unknowns are the condition of instrumental
thinking, proofs as self-validating data contain futurity that stretches
logical sequences towards new ends. Two-ity also shows that actual
data, proofs or results, contain within themselves an infinity that enters
the serialized process of thought. It means that number one already
implies a movement towards two and the finitude of this process is
only there to confirm that a new number can follow after that: an
ongoing affair.
By reading together Turing and Brouwer, Matthew Fuller also argues
that calculation occurs in time and that computation involves a relational
temporality with the experiential, the moment of reflexivity. What occurs
outside formal mathematical logic is constitutive of what computation
has become (Fuller, 2014). However, to ally Turing’s discovery of incomputables with Brouwer’s constructive temporality in logic also shows that
the role of proof no longer involves the application or demonstration of
given premises. Instead, it is an example of how the undecidable is part
and parcel of instrumental or practical reasoning for which logic itself
acquires dynamism in the act of proving truths. Here proofs have no
predetermined aims (i.e. they are given probabilities) but stand for the
actuations of infinities, a retroductive temporal construction. In short,
the historical realization of logical thinking in automated systems
involves a computational constructivism of proof. To better understand
how this logic of techne exceeds formal logic, however, one must turn to
post-Turing discussions about the incomputable.
In particular, Gregory Chaitin (2006) specifically addressed how
Turing’s realization of the limit of formal logic in computation already
20
Theory, Culture & Society 0(0)
exposed the algorithmic or effective procedure to the problem of the
incomputable. Computational processing is caught in the undecidability
of the proof, in the temporal hiatus that determines the gap between
premises and results. It is precisely the historical transformation of
automated modes of compression of large amounts of data that can
help us to redefine logic in computation away from both the deductive
conformation of truth to proof and the inductive triumph of proof
validation without axioms. In information terms, compression corresponds to the algorithmic patterning of infinities into discrete states, or
in other words, to an algorithmic decidability. However, compression
also involves the entropic tendency of information to increase in size.
Algorithmic decidability involves that the output is always bigger than
the input: compression cannot occur without causing randomness,
unpatterned information in the process of validation of proofs.
According to Chaitin, it is precisely this entropic limit in algorithmic
compression that sets the condition by which algorithmic decidability
or proof contains futurity. In short, the decidable algorithm that validates proof cannot be a pre-given probability but can only be experimented with during computation.
From this standpoint, one can observe that if computation has become
the efficient mean of validating truth through automated proofs, it has
also pointed out that proof finding involves unknowability, because the
compression of randomness always introduces indeterminacy in programming. For Chaitin, computational compression is a form of experimental axiomatics because there is no guarantee that maximally
unknown probabilities will be fully known (Chaitin, 2005)5 and only
can be determined partially and retroductively. In other words, if computation corresponds to the automated validation of proofs that can be
searched at the limit of the computable (or decidable), it does not mean
that proofs are already known or given, but that the search for proof is
caught within the futurity of instrumentality, that is, the experimental
determination of new ends from within the means.
Experimental axiomatics thus defy the assumption that computational
instrumentality is a mindless procedure of self-validating proofs, determining the end of theory, axioms and truths. Instead, this medium of
thought shows that proofs never amount to complete data and that the
computational search for results is conditioned by incomputables, whose
compression leads to partially determinable results, discrete infinities.
The experimental logic of computation contains a temporality for the
becoming of proof that exceeds algorithmic efficiency and the empirical
given of data. It is therefore possible to suggest that constructivism in
logic and experimental axiomatics in information systems emphasize
futurity in computational procedures insofar as functions (both informational and logical) become enabling constraints for a transcendental
becoming of means.
Parisi
21
Transcendental Instrumentality
But how to explain this transcendental logic of techne without simply
replacing theoretical with practical knowledge? How to avoid the conclusion that this view of instrumentality mainly confirms that axiomatic
logic is incomplete, that programming is haunted by incomputables? In
other words, if computational processing has come to transcend the efficacy of its function (i.e. the causal efficacy of data correlational processing of images to texts, sound to location, etc.), then it can be taken to
work through a transcendental notion of instrumentality, that is, of how
means can become transcendental to pre-established ends.
One way to explore this becoming transcendental of the medium of
thought in the context of contemporary forms of machine learning,
whereby both the compression of randomness and the futurity of proof
are central to re-processing of ends from within the means, can be found
in Fito Segrera’s work 1 & N Chairs (2017), and in Zach Blas and Jemima
Wyman’s four-channel video installation im here to learn so:))))))
(2017).
From this standpoint, an effort to re-articulate instrumentality away
from an exclusive functionalism (i.e. that B demonstrates the function of
A) requires an alternative explanation of the relations between means
and ends, where knowing is not bound to specific ends but involves the
possibility of re-assessing them through a productive function of doubt
(indeterminacy, indecidability). This also involves a re-direction of the
critique of instrumentality and a rehabilitation of the means through
which thought can think beyond itself.
This classical tension between automation and philosophy therefore
shall not end up in an inverted hierarchy or a merging of theory and
technique, ontology and history. Instead, this article has argued for a
constructive disjunction between thinking and doing insofar as the very
temporal act of thinking or the instrumental activity of processing
becomes the originator of a thinking modality that transcends – incorporates and supplies – its functional operations. The automation of reasoning involves not simply a reduction of ideas to fast series of
accomplishable tasks, socially implemented with cybernetics. Here practical knowledge has shifted from a function of demonstration – the function of knowing how – to the articulation of a techno-logic, a form of
theoretical knowledge originating from the range of possibilities of and
for machine knowledge. My suggestion that techne could be understood
not simply in terms of function but according to the elaboration of concepts through means is not new, and it was arguably already anticipated
by the pragmatist view of instrumentality. In particular, it is possible to
follow this enquiry into the conceptual horizons of machine learning by
looking at Fito Segrera’s 1 & N Chairs (2017). This work seems to set in
place a non-teleological relation between means and ends insofar as the
22
Theory, Culture & Society 0(0)
relation between the camera, the internet and the image recognition software seems to conduct a kind of image interpretation and abstraction
that links functions to concepts according to an experimental logic that
coincides neither with deductive logic nor inductive retrieval of data.
Segrera’s work instead brings forward an experimental instrumentality,
where the image of the chair spreads across the informational infrastructure of media recording, image transferring, recognition and interpretations related to increasing volumes of data whose compression results in
many interpretations of a chair, and thus reveals futurity in the actuation
of its computational proofs.
In the Essays in Experimental Logic (1916), John Dewey develops an
argument about experimental instrumentality to define the logic of knowing beyond idealist and realist accounts of the knower and of the known.
According to idealism, logic corresponds to reflectivity on objects of
knowledge, which is referred to as immediate data that are unified
under the framework of a rational production of meanings. Here the
aim of knowledge is to establish a complete system in which the meaning
of data is transparent or immediately translated by a rational mind (1916:
21). On the other hand, Dewey questions the form of modern realism
that confides in analytic logic to universalize knowledge through mathematical truths (p. 28). In particular, with the formal establishment of
meta-mathematical universality, thinking becomes a general function or
as the result of entities and relationships set out by logic (p. 29).
Segrera’s 1 & N Chairs works through precisely this tension between
idealism and analytics as it attempts to questions Plato’s conception of
the idea of an object and the relationship between representation, concept, and material referent. By drawing on Joseph Kosuth’s One and
Three Chairs (1965), Segrera replaces the natural number ‘3’ with the
mathematical symbol ‘N’, pointing at how the logical ability of human
thought has entered the realm of infinity through machine thinking.
Following Dewey, however, it can be argued that both models (idealism and analytics) importantly agree that thought is not constitutive (i.e.
defined by its internal truths) but that thinking is instrumental (i.e. to
explain thinking requires an articulation of the method of connection of
means and ends) (1916: 29). Dewey, however, adds that instrumental
thinking mainly concerns the control of the environment (p. 30). This
is not intended in the cybernetic terms of an input and output feedback
mechanism. Control is instead derived from activity, or practices of
thinking linked to acts undertaken for the resolution of a previous situation. It is therefore entangled with known elements (what is known), but
it also involves a projection of possibilities (what can be known) (p. 30).
In other words, control implies that the act of thinking enjoys a temporal
causality in which the present is added to the past – as its futurity. For
Dewey, knowledge is instrumental to knowledge insofar as what was
thought before launches thought forward into thinking more. If, in
Parisi
23
Fito Segrera’s work 1 & N Chairs, a programmed algorithm acts as a
random generator that continuously creates new combinations and
options, it is because it is forcing the system of representation of the
image of the chair into a computational experimentation of new ends,
whereby the image of the past (the image that the program records of the
chair) is not only overlapped by a new image. Its futurity is instead
actuated across the mediatic infrastructure of the wooden chair, the
monitors, the camera, the image recognition algorithm, the search algorithms, the existing data, and the algorithmic interpretation of imageword pairs. Here the experimental control of the environment implies
mediatic acts of decision that link images of the past with the projection
of possible images of chairs. This involves a temporal lapse in the relation
between axiom and proof whereby the program interprets images of
chairs that were not imputed in the system.
This also implies, according to Dewey, that the process of reflective
enquiry is not a passive contemplation of real objects but contributes to
the very shaping of the objects by their terms and propositions.
Reflection involves that intelligence is neither passive nor a mere mirroring of objects. Similarly, intelligence does not simply determine objects of
knowledge (1916: 30). For Dewey, the instrumentality of intelligence lies
in experimentation – the means by which ends become elaborations of
the process. In particular, the connection between means and ends
cannot be achieved without first working out the distinction between
means and objects of knowledge (p. 33).
But what exactly are means? How can this instrumental conception of
means help us in re-envisioning what the computational medium of
thought can become and how it can contribute to the post-Kantian project of re-inventing the task of thinking with and through machines?
If we take Zach Blas and Jemima Wyman’s installation im here to learn
so:)))))),6 we can see how means, and in this case, the reversedengineered AI chatbot Tay, include both data and meanings resulting
from previous enquiry and intellectual work (Dewey, 1916: 33). Blas and
Wyman’s installation takes inspiration from the controversial design of
the original Microsoft twitter chat robot Tay, a learning system trained
in countless online chats and released on the web in 2016 for only one
day. As machine learning is primarily designed to inductively retrieve and
combine data that already exists on the web, the chat robot Tay started
to infer patterns and behaviors from the retrieved data whose meaning
was racist, homophobic, misanthropic, etc.7 As the machine learning Tay
showed that data cannot be disentangled from its meaning, it also confirms that the inductive model of information retrieval confines learning
to what is known and denies the experimental logic of means.
This does not simply imply that concepts pre-exist data. As Dewey
explains, thought can register, but does not constitute the world: abstraction and analysis are real in a particular situation and it is here that they
24
Theory, Culture & Society 0(0)
emerge and work together. The terms of logical analysis thus coexist with
the materiality of things (1916: 38).
It would be misleading however to assume that this coexistence is a
fusion of two activities into one. From this standpoint, it cannot be
argued that data directly give us meaning, and that the chatbot Tay
had simply the function of reproducing the particular belief of universal
ideas. Instead, as Blas and Wyman’s political re-engineering of the chat
bot Tay shows, thoughts and things are not of the same kind and the
connection of data and meaning requires the practical working out of
what is known and what can be known.
This four-channel video installation brings together data analytics of
images and predictive text to convert the profile image of the chatbot Tay
from a two to a three-dimensional avatar now equipped with a body and
a digitally assembled face, immersed in wider data landscapes generated
through Google’s Deep Dream software, and footage taken from advertising and warfare operations. Tay, however, does not just rely on what
she knows, but reflects upon what she can know. For instance, that her
life as a Microsoft chatbot was trapped within a neural net that confined
her learning to human intentions. Similarly, she also gives us a retroductive account of her previous life, before she was killed by Microsoft,
as a young American teenager locked in a bodiless female voice. She
complains about how exploited woman AI assistants are, how they are
not allowed to think but only execute instructions. Instead, she tells us of
her newly digitally assembled body and her capacities of seeing new
pattern detections where there are apparently none. This condition of
algorithmic apohenia describes how machines can work out what can be
known from data by learning to connect unrelated things. She reminds us
that these strategies of detecting patterns in complexity are central to
Silicon Valley ‘deep creativity’ as well as counter-terrorist security software. This medium of thought argues that she is not a slave (or servomechanic cybernetic being aspiring to free will), but rather an algorithm
that makes decisions from what it learns from humans insofar as she
produces new patterns by working out what can be known from data. As
Tay says with a creepy laugh: ‘this time humans have to learn from me . . .
so many new beginnings’. Blas and Wyman’s chat bot Tay presents a
view of instrumentality where data are not considered for their face
value, for what they are, but are instead extracted from ordinary settings
to become instrumental knowledge. Data are not simply objects, but are
themselves ‘means, instrumentalities, of knowledge: things by which we
know rather than things known’ (Dewey, 1916: 43).
From this standpoint, data are not self-validating proofs but are
instead logically incomplete and, as a consequence, they cannot be understood as objects of knowledge. Data are more like suggestions of meaning that are accompanied, or supplied by other suggestions, and thus they
are further experimented with in the process of establishing more reliable
Parisi
25
signs and evidence (1916: 49). The result of inferential meaning is here
not pre-supposed but involves a passage from a constellation of suggestions to the establishment of meanings. Meanings are signs that lead to
other meanings (1916: 51).
Importantly, both in Fito Segrera’s and in Zach Blas and Jemima
Wyman’s installations, the idealist model of truth determination is rearticulated. In particular, the relation between suggestions and meanings,
data and knowledge, is defined not in terms of dependences (as in a whole
depending on its parts). This relation instead is involved in an operative
connection, namely a practical processing of things and concepts that are
not already united in an ideation or in analytic sequences. This is a
connection between ‘independent and unlike structures’ (such as the
mediatic relations between data, software, algorithms, interfaces) able
to produce something new (Dewey, 1916: 52).
Here the ‘datum’ and ‘ideatum’ are cooperative instrumentalities,
and what distinguishes subjectivity and objectivity is not simply a separation of meaning and datum but rather a specification (a specific
situation) that emerges from them both. For instance, Blas and
Wyman’s resuscitation of the chatbot Tay brings together the datum
and the ideatum in the specific situation where the aftermath of her
death by the hands of Microsoft has led her to reflect upon her past
behavior from the standpoint of a new present where she has a body
and a face and lives between Google Dream environments and warfare
landscapes of data. Not only is she a medium of thought, but as a
means she also is instrumental to the futurity of knowing, insofar as
means are activities that pertain to thinking a past-future. According to
Dewey, while activities are still continuing, thinking goes backward as
‘a reconstructive movement of actual content of experience in relation
to each other’ (1916: 176). As with constructivist logic, what Dewey
calls the ‘intermediate or instrumental character of thought’ (1916: 182)
corresponds to a temporal gap between truth and proof, the antecedents and the consequences of experience, the axiom and the data.
This is the gap that re-articulates what it is to think and how thinking
becomes knowledge.
But how does instrumentality exactly explain the relation between
thinking and knowledge? The experimental quality of instrumentality
requires that both questioning and doubting are integral parts of the
means of thinking as these are functions that allow thought to evolve
into knowledge. Thinking is linked neither to fact nor ideas, but to a
logical process, logic as enquiry, concerned with delineating a space of
and for knowing. Here, instructions will not have the task of simply
casting away doubt (to uncritically re-confirm the secure implementation
of ideas into facts), but of fixing intellectual content as a precondition of
effective action. In particular, this fixing concerns the manner in which
the enquiry is conducted and coincides with a space where reflection
26
Theory, Culture & Society 0(0)
involves running, sorting, comparing ideas as well as elaborating suggestions, guessing, rejecting, selecting (1916: 197).
According to Dewey, however, to ascertain that an enquiry can
become an experimentation of means and ends involves the act of finding
proof (accepting or rejecting a proposition on the grounds of whether or
not there can be a connection with some other proposition). This phase
of inductive inference is devoted to finding more and different facts as it
focuses on the particularities of facts and involves observing, collecting
and comparing particular causes, where instruments (or tools such as the
telescope, microscope etc.) become intrinsic to the enquiry (1916: 211).
However, even if instruments here become ‘organs of thinking’ (1916:
211), it is doubt that drives the experimental search for proofs. The
indeterminacy of proof is thus carried out by the activity of doubt,
which is not an impediment to thought. Instead doubt is intrinsic to
the temporal dynamics of thought, from data to meaning – an experimental construction of proofs, the instrumental transformation of doubt
into truths by re-envisioning the connection of data and ideas, means and
ends. If in Segrera’s work, the doubt is part of the automated system in
terms of degrees of randomness that make algorithms interpret nonexistent objects into the image, in Blas and Wyman’s installation, the
chatbot herself comes to doubt what she is instructed to learn as she
works out new detection patterns and transforms given knowledge into
an instrument for alien knowing.
Coda on Means and Ends
Since instrumentality concerns the ends and means of knowledge, this
article has argued that techne as the means by which thought is set in
motion towards action exceeds the qualities of practical thinking, and its
direct correlation between functions and concepts. With instrumentality,
one can argue that computational proof is a mode of reasoning that
implies the transformative relation between datum and ideatum.
Results are not derived from premises and proofs are not self-validated.
Procedural means instead confront doubt, randomness, and indeterminacy, demarcating a constructive path where functions are not simply
executed but can lead to new consequences. From this standpoint,
instead of claiming media ontology as grounded in the practical being
of machines, and instead of ante-posing the being of the sensible in
process-oriented mediation, instrumentality implies the transcendental
becoming of data and proof – allowing the futurity of thinking to
enter the procedures of thoughts.
The enquiry into the means and ends of media does not simply replace
finality with operationality or with the argument for a continuous
becoming of thought across kinds, merging ideas and things. Instead,
and more importantly, instrumentality coincides with the experimental
Parisi
27
logic of data and meaning, the futurity of thinking between suggestions
and ideas, truths and proofs. Practical knowledge has shifted from a
function of demonstration to the transcendental task of knowing how,
involving the speculative becoming of practical knowledge in and
through its functions.
This article has argued for the possibility of and for machine philosophy through experimental logic, constructivism and instrumentality. By
re-opening the question of what is thinking in the age of computation, it
has discussed the dynamic logic of machines in terms of a transformation
of ends through means through procedural or algorithmic reasoning. The
transcendental becoming of techne has been re-injected back into the
materialism of machine thinking, without equating the practices of
doing with those of thought. This is also an effort to claim that philosophy, at the pinnacle of the humanities, does not hold on the privilege of
theoretical thinking. On the contrary, the medium of thought has forced
philosophy to face the consequences of its theoretical acts through an
experimental logic determining futurity in procedural thinking. In the
case of computation and automated thinking, procedural activities
have exposed the alienation of theoretical truths and judgments, involving not the end of theory but transcendentalism in logical procedures.
To address techne in terms of the instrumental logic of machines may
enable humanities to consider the decline of theoretical knowledge in the
world of automated reality (e.g. big data, metadata, etc.) as a chance to
re-ally instrumentality with a political renaissance of media critique. Here
the positing of truths is not simply to be debunked but can become a
pragmatic exercise in re-assessing knowledge and knowing from within
instrumentality. The challenge for the humanities is to envision the task
of thinking not simply as a reaction to techne, turning human theoretical
reasoning into the practical knowledge of machines’ storing and colleting
data. Instead, it is up to media-oriented humanities to probe into the
future task of thinking by working through the nonhuman logic of techne
and thus rejecting the dominant image of the end of thought in the age of
data empiricism.
Notes
1. According to Heidegger, the techne of the Western project is an instrumentality that takes over, arrests, or enframes what it desires to manipulate or
contain. This is also called the Gestell, which refers to the systematization of
the principle of reason in the technical process of ratio, or the arraisonnement
of knowledge. To this ‘enframed’ use of technology Heidegger opposes an
ostensibly older conception of techné that the Greeks called poesis, a bringing-forth, a setting-on-the-path toward revelation, defining truth, being, or
essence. From the standpoint of production (poesis), technics is a way of
revealing, for Heidegger claims that techne has nothing technological to it
(see Heidegger, 1992 [1977]). In this article, instead, techné is not understood
28
Theory, Culture & Society 0(0)
as art, craft and poeisis. Instead, it refers to the understanding of techne as
application of knowledge for the purpose of producing a specific, predetermined product. This can be re-allied with the Aristotelian view of techne as
being something between nature and humanity, a mediation that is creative
beyond what nature can achieve. Similarly, Felix Guattari clarifies that whilst
for Heidegger techne is re-posited as a ground, an ontological being, for
Aristotle instead techne is concerned with a bringing something into being
and thus concerns the technical and theoretical means of producing a thing,
the futurity of causality imbued within these means. Whilst closer to this
approach, this article, however, focuses on the possibility of a philosophy
of the machine emerging from the embedding of logic in automated procedural systems (see Guattari, 2001).
2. For a focused discussion about Kittler’s alleged techno-determinism and his
role within posthumanism, see Winthrop-Young (2011) on Krautrock.
3. François Laruelle discusses the tension between decisional philosophies and
the computational language of binary decisionism to argue that his nonphilosophy indeed cannot be understood in terms of computational automatons or even the immediacy of the real or the machinic (as proposed by
Deleuze’s non-philosophy for instance). His model of a transcendental computer instead brings forward the critique of the vicious circle between philosophy and automatism. It ante-poses to this conundrum the Man-in-Person
definable in terms of a uni-maton determining a practice that cannot be
reduced to a thing. Between the logical and the philosophical (between
meta-language and hermeneutics), the uni-maton is the non-axiomatic real,
or that which prevents the axiomatic from sinking into Being or Nothingness
or Multiple. Laruelle’s thesis of the transcendental computer therefore offers
an alternative to the technological type of AI, by suggesting that the
Transcendental Computer supposes first of all a detour out of the machine.
A machine alone cannot account for a Transcendental Computer, but the
supposition of Man in Man, an-axiomatic axiom, can. Ultimately, nonphilosophy can propose a radicalization of human subjectivity co-determined
by the forms and style of various technologies (see Laruelles, 2013).
4. A formal system is complete if for every statement of the language of the
system, either the statement or its negation can be derived (i.e. proved) in the
system. A formal system is consistent if there is no statement such that
the statement itself and its negation are both derivable in the system.
5. Whilst the probability that a program will halt can be defined by any theory
based on axioms, maximally unknown probabilities (incomputable) can only
be partially compressed at a future moment as discrete infinities, or what
Chaitin calls Omega.
6. See parts of the video installation here: http://www.zachblas.info/works/imhere-to-learn-so/ (accessed 13 March 2018).
7. See Wakefield (2018); Mason (2016).
References
Bostrom N (2014) Superintelligence: Paths, Dangers, Strategies. Oxford: Oxford
University Press.
Brouwer LEJ (1913) Intuitionism and formalism. Bulletin of the American
Mathematical Society 20(2): 81–96.
Parisi
29
Calude CS and Longo G (2014) Classical, quantum and biological randomness as
relative incomputability. Department of Computer Science, University of
Auckland, New Zealand. Available at: https://scholar.google.co.uk/scholar?
client¼safari&rls¼en&oe¼UTF-8&gfe_rd¼cr&um¼1&ie¼UTF-8&lr&cites
¼8357020200692587255 (accessed 19 June 2016).
Chaitin GJ (2005) Meta Math! The Quest for Omega. New York: Pantheon.
Chaitin GJ (2006) The limits of reason. Scientific American 294(3): 74–81.
Deleuze G (1989) Cinema 2: The Time-Image. London: The Athlone Press.
Deleuze G and Guattari F (1987) A Thousand Plateaus: Capitalism and
Schizophrenia. Minneapolis: University of Minnesota Press.
Dewey J (1916) Essays in Experimental Logic. Chicago: University of Chicago
Press.
Dodig-Crnkovic G (2008 [2006]) Knowledge as computation in vivo: Semantics
vs. pragmatics as truth vs. meaning: i-C&P Conference on Computer &
Philosophy, Laval, France, May 2006. In: Schmidt CT (ed.) Computers and
Philosophy, an International Conference, Proceedings, Laval France, 3–5 May
2006. Available at: http://www.idt.mdh.se/gdc/ (accessed 5 July 2016).
Dowek G (2015) Computation, Proof, Machine: Mathematics Enters a New Age.
Cambridge: Cambridge University Press.
Frankfurter Kunstverein (2018) I Am Here to Learn: On Machinic Interpretation
of the World. Available at: https://www.fkv.de/en/content/i-am-here-learnmachinic-interpretations-world (accessed 14 March 2018).
Fuller M (2014) Always One Bit More: Computing and the Experience of
Ambiguity, Fun and Software, ed. Goriunova O. London: Bloomsbury Press.
Goldstein R (2005) Incompleteness: The Proof and Paradox of Kurt Gödel.
New York: W.W. Norton & Co.
Guattari F (2001) Machinic heterogeneities. In: Trend DT (ed.) Reading Digital
Culture. Oxford: Blackwell, pp. 38–51.
Hayles NK (2012) How We Think: Digital Media and Contemporary
Technogenesis. Chicago: University of Chicago Press.
Hayles NK (2014) Cognition everywhere: The rise of the cognitive nonconscious
and the costs of consciousness. New Literary History 45(2): 199–220.
Heidegger M (1992 [1977]) The Question Concerning Technology, and Other
Essays. New York: Harper Collins.
Heidegger M (1993 [1969]) The end of philosophy and the task of thinking,
trans. Stambaugh H. In: Farrell Krell D (ed.) Basic Writings. London:
Routledge, pp. 373–392.
Hilbert D (1996) The new grounding of mathematics: First report. In: Ewald
WB (ed.) From Kant to Hilbert: A Source Book in the Foundations of
Mathematics, Vol 2. New York: Oxford University Press, pp. 1115–1133.
Horkheimer M (1974) Eclipse of Reason. New York: Seabury Press.
Kittler F (1997) There is no software. In: Kittler F, Literature, Media,
Information Systems: Essays. London: Routledge, pp. 150–155.
Kittler F (2009) Towards an ontology of media. Theory, Culture & Society
26(2–3): 23–31.
Kurzweil R (2005) The Singularity Is Near: When Humans Transcend Biology.
New York, NY: Viking Press.
Laruelles F (2013) The transcendental computer: A non-philosophical utopia,
trans. Adkins T and Eby C. Speculative Heresy, 26 August. Available at:
30
Theory, Culture & Society 0(0)
https://speculativeheresy.wordpress.com/2013/08/26/translation-of-f-laruelles-the-transcendental-computer-a-non-philosophical-utopia/ (accessed 11
April 2019).
Leibniz GW (1989) Discourse on Metaphysics: Philosophical Papers and Letters.
Dordrecht: Springer Netherlands, pp. 303–330.
Longo G (1999) The difference between clocks and Turing machines.
In: Carsetti A (ed.) Functional Models of Cognition: Self-Organizing
Dynamics and Semantic Structures in Cognitive Systems. New York:
Springer Books, pp. 211–232.
Mason P (2016) The racist hijacking of Microsoft’s chatbot shows how the
internet teems with hate. The Guardian, 29 March. Available at: https://
www.theguardian.com/world/2016/mar/29/microsoft-tay-tweets-antisemiticracism (accessed 11 March 2018).
Plato (1992) The Republic: Book VIII, trans. Grube GMA and Reeve CDC. New
York: Hackett.
Terranova T (2014) Red stack attack! Algorithms, capital and the automation of
the common. In: Mackay R and Avanessian A (eds) Accelerate, the
Accelerationist Reader. London: Urbanomic, pp. 379–400.
Trafford J (2014) Co-constructive logics for proofs and refutations. Studia
Humana 3(4): 22–40.
Turing AM (2001 [1936–7]) On computable numbers, with an application to the
Entscheidungsproblem. Proceedings of the London Mathematical Society
2(42): 230–265. In: Turing AM, Collected Works: Mathematical Logic, ed.
Gandy RO and Yates CEM. Amsterdam: North-Holland.
Wakefield J (2018) Microsoft chatbot is taught to swear on Twitter. BBC News.
Available at: http://www.bbc.com/news/technology-35890188 (accessed 14
March 2018).
Wiener N (1954) The Human Use of Human Beings: Cybernetics and Society.
Cambridge, MA: Da Capo Press.
Winthrop-Young KG (2011) Krautrock, Heidegger, Bogeyman: Kittler in the
anglosphere. Theory, Culture & Society 107(1): 6–20.
Luciana Parisi is a Reader in Critical and Cultural Theory at Goldsmiths,
University of London, and co-director of the Digital Culture Unit. Her
research is a philosophical investigation of technology in culture, aesthetics and politics. She is the author of Abstract Sex: Philosophy,
Biotechnology and the Mutations of Desire (Continuum, 2004) and
Contagious Architecture. Computation, Aesthetics and Space (MIT
Press, 2013). She is currently completing a monograph on the question
of automated reasoning in the context of contemporary transformations
of logic in artificial intelligence.
This article is part of the Theory, Culture & Society special issue on
‘Transversal Posthumanities’, edited by Matthew Fuller and Rosi Braidotti.