It is a great honor, a great joy to welcome Luciana Parisi. I'm just trying to think, it is the first time, isn't it, that you're in Utreme, not the last one. I can't believe it, that we waited so long. Luciana is a member of the younger generation, if I can speak as the old lady for a minute, that I read with the most curiosity and admiration. She's taught me enormously. The extraordinary figure, actually tracing nomadic itineraries between Naples, her hometown, and Goldsmith, where so many nomadic subjects are settled. Reader in cultural studies and director of the PhD program in the Center for Cultural Studies at Goldsmith. One of the pioneers of crossing over media theory with evolutionary theories, biologies, biotechnology and computation, crossovers, difficult things to think.
Parisi is somebody who studied with Lynn Margulis and her knowledge of biology and proper science and dates from that. Somebody who never takes science for granted and works very hard of it and would stand in horror of getting her scientific information and knowledge wrong. Two very important books, 2004, Abstract Sex, one of those crossover books that everybody read on transformations of body and techno-capital desire. More recently, continuing the work on new technologies of control, developing a non-phenomenological critique of computational media, important for this seminar, because we have heard in Mark Hansen a phenomenological
theory of media. You will hear a non-phenomenological one tonight, and a very important book, Contagious Architecture, published in 2013. It's a joy to have her here. Please make her feel very welcome. Luciana Parisi. Thank you so much, Rosie. It's always exhausting, your introductions. It's a pleasure to be here. Thanks for everyone being here. I don't have any slides, I'm just going to talk through my paper for today. I did send some readings for today, some of the material I'm working on and on which this paper is based.
So in September 2013, issued the journal Nature, a group of physicists from the University of Miami published the article, Abrupt Rise on New Machine Ecology Beyond Human Response Time. In the article, they identified a transition to a new all-machine phase of financial markets, which coincided with the introduction of high-frequency stock trading after 2006. They argued that the sub-millisecond speed and massive quantity of algorithm-to-algorithm interactions exceeded the capacity of human interactions. Analyzing the millisecond-scale data in the code of financial markets in details,
they discovered a large number of sub-second extreme events caused by those algorithms whose proliferation they correlated to the financial collapse of 2008. In this digital environment of trading, algorithmic agents make decisions faster than humans can comprehend. It is said that while it takes a human at least one full second to both recognize and react to potential danger, algorithms or bots can make a decision on the order of milliseconds. These algorithms form a complex ecology of high specialized, highly diverse, and strongly interactive agents, in the words of farmers, Chorus,
operating at the limit of equilibrium outside human control and comprehension. So that's what has been said. The argument I develop here takes this digital ecology of high-frequency trading algorithms as a point of departure. Thus, today I'm not specifically concerned with the analysis of complex financial ecology itself, but I want to discuss the critique of automatic cognition in the age of algorithmic capitalism. For if financial trading is an example of a digital automation that seems to be increasingly autonomous from human understanding, automation itself has become, as it were, a second nature to thought, defined by the capacity of the artificial to artificially think.
It seems thus urgent today to ask, can the critique of instrumental rationality, as addressed by classical critical theory, be still based on a distinction between pure thinking and automation? Can one truly argue that algorithmic automation is always already a crude reduction of thinking to the sterile encoding of complexity? By answering this question, we cannot overlook an apparent dilemma. Both critical thinking and algorithmic automation rely on principles of indetermination to argue that either cognitive processes cannot algorithmically be encoded, and we know that most post-structuralist and neomaterialist approaches argue that,
or that technological extension of cognitive functions work to include unexpected results. and this is also the work of Clark and Chalmers on extended cognition for instance the presupposition of indetermination for both critical theory and automation is used to both challenge and define the neoliberal order defining the apparent paradox of our times but to untangle this paradox I suggest that one is to do some kind of legwork and really turn to what is said within the field of algorithmic information theories and their articulation of incomputability insofar as incomputability stands for indetermination
of the system, randomness, uncertainty. And to the discovery of Gregory Chaitin's cipher that he calls Omega. This cipher omega has a specific quality. It is both definable and non-computable. It's a way to define it in computable, but nonetheless it's got some level of determination within it. It cannot be exhaustively, but only partially computed. Omega describes at once a discrete and infinite state of computational quantification, quantification of infinities occupying a space between zeros and one. So it's within the binary
logic, but still it's something that cannot be completely determined into an integral number, a finite number. It's got infinity attached to it. It cannot be detached from it. From a philosophical perspective, the discovery of omega is important because it sides neither with a deterministic view of computation, nor with an emphasis on the cost and limit of computing infinite numbers. Instead, what is important here is that the logical method by which omega, this particular determination of infinite levels of infinities, has been found, demarcates not only what Chaitin calls the limits of reason,
and this refers to one of the papers I sent you today, but rather the emergence of a dynamic model of reasoning, in which results are not contained in the axiomatic premises of the system. So I take this centrality of the incomputable in information theory to bring into question not only the critique of technical rationalization, but also the critique of instrumentalization of reason. from my discussion today i want to suggest a critical theory antipathy for automation seems to be grounded in a mysterious form of human exceptionalism in matters of reasoning according to which the fine line between what humans are able to think and what humans have artificially
constructed is either not to be surpassed or is to be neutralized through some kind of divine interventions that grants some form of pang-psychism for all living beings. From this standpoint, what underpins the dominant critical view of automation is, one, the rejection of modern mechanization of reasoning, and two, a rejection of transcendental reason and logic. Within the age of accelerated automation, instead, it seems urgent to me to readdress and re-theorize logic and the mechanization of reasoning in the light of the cognitive function that automation has acquired. One way to re-engage with these often marginalized
questions in critical theory is by adopting, I suggest, and developing a pragmatist articulation of reason and logic. And I will explain one in a moment. Importantly, I want to point out that what seems to be missing from these critical positions of discussion today about the limit of computation and the political importance of pure thought is an engagement with the scientific image of thought, namely the historical effort in information theories to individuate the mechanization of logic and the limit of the deductive model of reasoning, which I think has got important implications for what we take automation thinking to be today.
The question of the incomputable within universal Turing machines can act as a starting point for developing a critical theory of automation, which could challenge both the representational or symbol-based frame of cognition, qua algorithmic computation, but also the phenomenological and inactivist view of embodied cognition, for which sentience, i.e. the ability to feel, perceive, and experience, remains somehow a superior complex disposition, an Aristotelian understanding of complex disposition for thinking that could never, never be algorithmically encoded. To develop a critical theory of automation instead means to construct
a non-representational theory of computation that is not bound to symbol manipulation and the metaphysics of a computational nature, but also refuses any quick dismissal of logic, reason and formal thought. This also means that it is important to understand not simply algorithmic function per se, but our scope really is to explain the general functioning of an automated logic emerging from the wider cultural use of automated systems. As I mentioned earlier, the sensuality of incompatibles in an automatic system points to the limit of deductive reasoning, but also defines the emergence of an experimental form of axiomatics, an experimental form of reasoning, or experimental truth.
this non-detective form of reasoning however does not simply account for the disappearance of logic instead the rational formal project of computation as already professed by Alan Turing effort to found a model that could be abstract enough from the specificity of procedures cannot be dismissed and needs to be extended to include other forms of reasoning. From this standpoint, whilst acknowledging that the end of didactic logic and formal reason has given way to the age of algorithmic connections and immediacy, the effort of a critical automation theory is to readdress the mechanised logic of these systems.
One way to do so, I will then explain this paper, is through the pragmaticist theorization of abductive reasoning. I will explain in a second what it is. We know that, and I take this from Charles Sander Pierce, triadic system of logic, or what he called abductive, inductive, and deductive systems. So it's not just an exclusion of deduction, but deduction becomes the last point in the formalization of reasoning. So it starts from hypothesis, then it's about gathering data, and then it's about establishing rules. So abductive, inductive, deductive. Which extended the Kantian model of deductive reasoning,
because whilst it argued for the transcendental function of conceptual mediation in the production of knowledge, it also insisted that complex structures of reasoning are embedded in material and non-inferential, non-discursive strata. So this triadist schema always starts from an extreme hypothetical proposition, aiming to draw patterns from non-inferential social practices and then develop axioms, but which meaning embedded in these practices is not just fixed as a rule, but becomes an experimental truth. This kind of axiomatic dynamism that we can find in this model of logic
that Pierce started, American pragmatist has continued, admits that the attainment of truth is dependent on the function of reasoning. But reasoning implies the capacity to revise what is known precisely through the social practices of inferring meaning from the relation between things, meaning constructed by the use of things. So meaning not as something that is already given and established, but becomes constructed through use. but drawing on the pragmatist theory of abductive reasoning and its use within automated system my task today is to unpack the transformation of the mechanization of logic
to show that critical theory and its implicit kind of understanding that reason has been instrumentalized by technology is biased by a dogmatic rejection of logic and automation. Instead, a critical theory of automation needs to explain and not just reject how algorithmic selection, evaluation, classification of data, how the relation between data and algorithms and other kind of layers that create this kind of apparatus have become part of social practices and how these automated functions, which is the result of the use of these systems,
have absorbed and formalized these practices at an inhuman level of thinking. To view automation in terms of abductive logic may help us to account for the hypothetical reasoning carried by algorithmic instruction in the process of encoding randomness. But whilst abductive logic helps us to claim for the rearticulation of a kind of reasoning central to computational processing, we also need to reengage with a question of transcendental reason, which involves the forming concept and establishing rules is not bounded to any specific organic or non-organic form.
From this standpoint, the pragmatist view maintains that logic is a manifestation of the collective use of data. The use that will determine the meaning of data. And obviously we need to establish what is this use, where does the use come from, who is using what. Mainly embedded in social practices, but it can give us the possibility of arguing for a general artificial reason. I generalize the artificiality of reason, the capacity of reason to synthesize artificially what is empirically gathered and registered by a system. And within this frame, algorithmic automation obviously cannot encompass all this process,
but it's obviously a partial and rudimentary manifestation of this generalized form of reasoning. but I will go back to my proposition in a moment now I want to stress that the antilogical conception of automation is central to digital theories the critique of instrumental reason and of the ecological quality of contemporary governance so I'm just going to give you some example of what I actually mean so the antilogical conception of automation can be found for instance in Catherine Hill's discussion of non-conscious cognition, which computational devices are set to share with human intelligence. So there's some kind of flatness
between what human intelligence and computational device both do, which is this level of non-conscious cognition, which allows for health smart devices that are not bound to consciousness to act faster than us and making all sorts of decisions. This obviously echoes exactly where I started from. It's exactly what in the journal Nature has been argued as the capacity of high-frequency trading algorithm to act at the level of sub-milliseconds. In particular, these low-level activities of non-conscious cognition, which are performed at imperceptible but fully felt speed,
as in the famous case of the missing half-second shows, how cognition is not coherent, she claims, and does not even edit information to much expectation. So there is no kind of decision, no kind of supervision. It's completely comes in and comes out, input out, but without any form of reflection. For Hales, what is promising of cognitive non-conscious technical devices is that they can operate a temporal regime inaccessible to human consciousness and exploit the missing half-second at their advantage. Functioning across human, animals and machines,
because this is the non-conscious level that allows this panpsychist model, non-conscious cognitive processes defy the centrality of human consciousness and the anthropocentric view of intelligence. Hales espaces the idea that the anti-deductive operation of non-conscious cognition, they are anti-deductive because they don't follow any kind of premise, so they just skip any level of calculation, are importantly and effectively somatically marked and thus phenomenologically embodied.
As she insists, whilst both the hammer and a financial algorithm are designed with an intention in mind, only the trading algorithm demonstrates non-conscious cognition insofar as it is embodied within the physical structure of a network of data on which it runs and which sustains its capacity to make quick decisions. So this shift towards a non-logical model of cognition is importantly characterized by a reorientation of the practices of real subsumption, we could argue, in which fixed capital, the technological apparatus, no longer designates the rational instrumentalization and capitalization of labor, of human labor, but has rather become, importantly, the effective motor of cognitive labor.
Some argue it directly intercedes with the bare vibration of life, insofar as it completely exceeds the biological living complexity of life, but at the same time replaces them. So the deductive form of automation has of course not simply disappeared. Remember my schema is not anti-deduction, but it's the Pearson model. So you start with abduction, then you go to eduction, deduction. It's a schema that cannot be separated. My argument is exactly against the tendency in critical theory to actually, and the critique, especially the critique of technology,
to ditch and to get rid of the deductive formalism and argue for a non-deductive model of critique, which is embedded for me in cognitive capitalism. So the deductive form of automation has of course not simply disappeared but has become infused with a bodily oriented form of mechanization. Here fixed capital has acquired a form of autonomy from its human use and validation because it is no longer mainly concerned with wording of contingency and human errors. Instead, in the form of algorithmic automation, fixed capital is now regulated by inductive heuristic principles of trial and error, deriving efficiency from machine learning languages, which directly incorporate the data environment in which the algorithms operate.
Here, automation has reached a non-prescribed form of intelligence insofar as algorithms experiment with their data environment and learn from it. Whilst it is true that this intelligence needs no consciousness, it has been discussed also as a form of cognition defined by affective immediacy in which to think is to represent embodied feelings. in Antioedipus Deleuze and Gattari already individuated this sensible tendency of fixed capital had warned us against the advancing regime of what they called immanent axiomatics whereby the rationalization of capitalism by means of machines no longer operated deductively. This meant
that calculative machines instead had become dependent on the indeterminacy of environments and that the anti-productive logical capital had become generative and affirmative of the fallacy of deduction. In our post-sibernetic culture, the sublimation of affective immediacy and the political destruction of logos have become paradoxically equivalent. Following Brian Massumi's analysis of the contemporary configuration of neoliberal governance, one could argue that the end of rational economy, as he calls it in his last publication, has been accompanied by the end of the rational instrumentalization of machines.
In particular, a close analysis of the transformation of fixed capital into learning, interactive, distributive machines reveals that the so-called cognitive phase of capital has paradoxically given way to the commodification of affects. So, no cognition at all. The destruction of cognition. This form of techno-capitalism has invested in human intelligence and intuitive feelings, driving humans to become self-entrepreneurs and govern themselves through auto-affection. fixed capital therefore no longer embodies the active logic but rather exposes the emergence of this non-conscious cognition unbounded
from the ontological finitude of the rational subject by extending across devices and evolving through the retrieval of increasing quantity of data what is also called social computing fixed capital therefore has become itself intelligent this important shift can be further contextualized through what Masumi defines as an ecological rationality grounded in the thinking feeling of the body, turning symbolic values into lifestyles and demeaning rules into a set of variable qualities. At the core of this ecological rationality, therefore, is this non-conscious distributive
embodied intelligence in which all is locally induced to generate global effects of unification of one sentient body without organs. This inductive or effect-driven operation of fixed capital epitomizes the non-inferential thinking of the body, able to make decision without a single calculation. This form of antilogos has become central to the techno-capitalist de-territorialization of rationality, what Mattei was talking about before, which resolves the tension between automation and reason through the fusion of cognition and affect. And obviously, I hope you understand that I'm being sarcastic here. I don't subscribe
to this model. Far from being the ultimate deliberation of the will, the disposition of inferential reasoning is constantly thrown us back as the best alternative we have, presented in the form of a massive commodification of the affective potential of fixed capital, promising us to improve our sociality through automated choice. Instead of proposing a nostalgic gesture towards the past, where automation and reason were formally separated, I think that the real challenge today is to be rather suspicious of this disposition of reason in favor of the immediacy of intelligence, and thus to
work to explain rather than reject what is the state of inferential reasoning in the context of these automated systems that needs not to be explained but rather differentiated from just the physical causality from which it emerges. To readdress the question of inferential reason in automatic cognition we need therefore to discuss the limit of computation. So I'll go back to the limit of computation I just mentioned before. But the limit of the Turing machine understood in terms of indeterminacy or randomness or discussed in the paper as the limit of reason need to be taken as a point of departure
for the articulation of an artificial thinking that could be proper to computation. Because my question is, what kind of automated system are we dealing with? And that's why my question is to start with the problem, the historical problem of the limit of computation to actually understand what kind of intervention, what kind of articulation of this artificial thinking can be allowed from there. So in the 80s, Chaitin tackled this question of the limit of computational logic from the standpoint of an entropic conception of information or randomness. I mean, I can talk about it later, about the difference between an entropic conception
of information versus an entropic conception of energy, which is a model, for instance, that you find in Simondon. So if you want to talk about the philosophy of technology, how information is being understood, how technical objects are being understood, we can talk about it later. So for shading, computation corresponds to an algorithmic processing of maximally unknowable probabilities that is precisely the limit of computation. Here he explains in the process of computation, even at the simplest level, the greatest discovery, which is also highlighted in the paper by Longo that I sent today, was the complexities at the very bottom of any simple unit. So there is no simple model that you can ascribe to this abstract model of computation.
Insofar as the output in this process always becomes bigger than the output, the output always becomes bigger than the input and what happens is that the equilibrium between the input and the output is broken and that's how the formal deductive reasoning is challenging so far as the conclusion of what can be inferentially deductive or what inferentially can be the result of the premises is not contained in the premises So something else exceeds the premises by which the logical system operated.
So for Chaitin, in a way, the computational algorithmic randomness reveals the capacity of the system to process algorithmic randomness and give it a cipher, i.e. omega, shows that there is an intelligible capacity that algorithms have to extract more information from the data that they retrieved. So that's the capacity of algorithms not only to patternize and recognize, because pattern recognition is to do with the level of intelligence which is physically and materially constrained.
But my point is that within this complex system of algorithmic processing and this kind of, at the point at which the limit of algorithm becomes productive is where there is some kind of elaboration on the data. So it's not just about recognizing a pattern in the data. It's also about elaborating a kind of intelligible capacity of algorithm to elaborate from the data more information. So it's not just recognizing, but elaborating. That's another step from kind of non-conscious cognition. And that's why I don't agree with non-conscious cognition, but I will explain as I go by.
so chatting theory also shakes the assumption that automated reasoning is grounded in simple rules or in the syntactical model of artificial intelligence for which reasoning just coincides with the manipulation of symbols that are preconceived, they are syntactically linked to our neural processes as in Chomsky model so chatting claims that computational processing leads to postulates that cannot be predicted in advance by the program, but can only be explained in terms of an experimental axiomatic. So far as in this elaboration
process, an axiom is produced, but it's not an axiom that is contained within the theory by which the postulate was based upon. But it's an axiom that stands out of the theory. It creates another truth that is annexed or is in excess of the theory or the program by which the process started. This is a non-deductive logic therefore, it's experimental, it's not deductive, that accounts for a method of algorithmic reasoning which is able to determine unknown. So what I mean is that the limit of
computation is not about the reducibility of thought to some level of calculation, some kind of intelligible process by such a capacity to determine even if partially the unknown. And that's where theory becomes important to me. Okay. Instead, algorithmic automation involves the emergence of a form of intelligibility able to use data environments which are the concretising infrastructures of incorporated or automated social practices and that which point to the development to the capacity of add new axiom codes of instruction to what was originally programmed.
So programming here corresponds to the formation of intelligible procedures in which algorithmic instruction extrapolates new patterns, but produces information through that environment that they retrieve, thus transforms the pre-established function of programming. In other words, what algorithmic information theory suggests, what we can suggest with it is that if axioms are becoming experimental truths, then if laws are the result of an algorithmic intelligibility of data environment, then computation too may need to be conceived in terms of its intelligible functions through which unknowns can be algorithmically preended.
From this standpoint, the paradox of indeterminacy is rather confronted with the emergence of this non-sentient, non-empirical, non-just data retrieval process, or conceptual infrastructural patterns and function, able to experiment with pre-established rules. This involves an integrated capacity of labeling, selecting, evaluating, and transforming data, which increases the capacity of incorporating social, economic, and cultural manifestation within this parallel distributed system based on binary language. The incorporation of this manifestation, therefore, for me, points to an informational stratification
stratification of discursive and non-discursive social practices, which cannot be associated or flattened down with some kind of data retrieval, path recognition, or this kind of low level, physical level of intelligence. This is a new order of intelligibility that requires a careful unpacking of the possibility, the limits that this experimental logical automation can offer, rather than simply a rejection or a ditching or a overlooking of inferential logic, which is exactly what a techno-capitalist dominance of affective, non-conscious and non-deductive thinking is actually putting in place.
How am I doing with time? I have another section. I think we can't do without. Okay. Okay, because we have the section on Lorenzo Magnani. So I'm going to conclude by sketching out a possible rearticulation of algorithmic automation in terms of, yes, a rudimentary capacity of synthesizing non-inferential practices or these incomputable infinities. But nonetheless, the effort needs to be done if we don't, we're just to follow this kind of pan-sarchist model by which techno-capitalism is operating. My effort is thus, one, to argue that automation is more than unconscious cognition.
It does not correspond to the affective immediacy of non-inferential decision. Instead, it involves an intelligible process of elaboration of data. And thus, points to the emergence of an alien thought. What do I mean by this alien thought? just a thought that challenges human exceptionalism in matters of reasoning. Two, to develop a critical theory of automation that shows, yes, the limits of artificial intelligence vis-à-vis a pragmatist theory of what counts as reasoning. So to claim that the theorization of these limits of the state of artificial intelligence
intelligence vis-à-vis a pragmatist theory that I will explain as a much richer understanding of what thinking is, can however help us to divorce the effective impulses of techno-capitalism from the intelligible functions of automated machines. Ultimately, my scope is to divorce capitalism from technology or capitalism from techno-science. and actually argue because I think that there's been an historical moment within critical theory where the critique of instrumentalisation of reason has become the critique of technology techno-capitalism and that's what the emergence of the automatic
systems, this kind of intelligible function and non-deductive logic but nonetheless intelligible and not just non-conscious can help us to build. This is a kind of counter theory or re-articulation of critical automation theory. So my proposition is here an adaptation of Charles Sander-Pierce abductive reasoning, defining rules not as the already known premises of logic, but as the result of a process of determination of truth, starting from hypothetical assertion and involving the elaboration of measurable data, so a process of induction, followed by a consequent
establishment of rules. So rules are not to be abandoned, are to be explained as a process, like an evolutionary process. An evolutionary process that is not just defined by some kind of essentialist naturalism. It's a naturalisation of logic, but it's got stages in which the physical condition cannot explain the cultural. That, for me, is quite important to show. And this, obviously, within materialist critique, is something that we need to discuss, I guess. So rules are not fixed and are not symbolic representation of material practices.
instead from pragmatism rules are ascribed into a complex cognitive system through which truth becomes susceptible to change through the way non-inferential practices become socially mediated. In other words what does it mean? Instead of abandoning reason or logic what pragmatism or ontologizing embodied thought beyond any form of sapience, pragmaticism argues that reasoning is embedded in the social matrix. But it's through the social matrix that we need to understand how rules are produced
by means of hypothetical assertion, and that's how meaning can be extrapolated by analyzing and impacting the use of data. so pragmatism allows us to account for a conceptual infrastructure that is intrinsic to the social use of data its aim is to explain and not to discount the process by which physical causality conditions but does not determine the emergence of intelligible functions and their process of data elaboration since abduction implies that the truths are operative and not pre-established it involves the generation of hypotheses that can best explain a problem
and not the problem solving procedure so that's a difference obviously we know that problem solving procedures are totally scrapped within a deductive, calculative, formal model of syntactic AI that we need to question. You know, that's not... So to kind of side with automation doesn't mean to bind the model. It means to criticize the model and reconstruct the model otherwise. That's also a kind of theoretical intervention. so this pragmatist method of abduction
argues the logic emerges from discursive synthesis on non-discursive practices highlighting the existence not only of intelligible patterning but also the kind of tendency towards a conceptual function that is able of making explicit what is implicit in non-inferential materialities. From this standpoint, in order to acquire autonomy, we know that algorithmic automation needs to acquire the collective ability of use meaning of data, of extrapolating the meaning in its use of data. leading to the action
not of establishing but of revising rules so this is a model of artificial intelligence that doesn't exist machine learning and any kind of distributive intelligent complex system is still operating the level of labeling classifying and organizing. But my question is that there is instead also with the non-deductive logic and the kind of experimental axiomatic and information theories are proposing, obviously there is a possibility of a rudimentary intelligible process
that can allow, that is open to this other level of and at least question both the model of non-conscious intelligence and the model of formal reasoning. So, logic rules are not locked within fixed categories, but crucially determined by social practices in which conceptual infrastructure are embedded. Logic is not the result of syntactical connection or recombination between units, but the end point of a synthetic processing requiring a collective process of elaboration that moves from intelligibility to reasoning. In the words of Alfred North Whitehead,
which is a very different Whitehead from Mark Hansen, I guess, is the process that Whitehead in the function of reason explains is the model of moving from physical apprehension to conceptual preemption to abstract schema. So the functional reason is exactly moving, ascending towards this level of elaboration of data. So pragmatics does come before logic. Logical inference is rather the explicit formalisation of social practices. The point at which meaning becomes synthesised into formal rules, so to say we were there, the point at which physical apprehension becomes conceptual apprehension
and then is able to produce an abstract schema. So normalization here is explained in a totally different way. It's not about normal and abnormal. It's the process of construction, of constructing logic, of constructing rules and revising rules. So this is not to dismiss logic. of course but to argue that the order of biophysical causality is not equivalent to the order of reasoning or abstract schema and that instead inferential reasoning is central to the understanding of the evolution of a collective culture. Similarly our understanding of computation qua mechanization of logic and its role within the techno-capitalist automation of cognition
needs to be addressed in terms of a non-representational theory of inferential reasoning, which does not, however, identify feeling with thinking. So that's, as Lorenzo Magnani argues, since the 80s, abductive reasoning has been adopted by diagnostic and expert systems, point to a computational structuring of reasoning based on inferential synthesis or inference to the best explanation. Importantly, Magnani distinguishes between model-based abduction, a theory-based inference, and manipulative abduction, defined by action-oriented or extra-theoretical reasoning, as he calls it.
Theoretical abduction illustrates much of what is important in creative abductive reasoning, in humans and computational programs, the objective of selecting and creating a set of hypotheses. and he makes the case of diagnostic causes, hypotheses, especially in medical reasoning. But theoretical abduction, according to Magnani, fails to account for those cases in which there is a kind of discovering through doing. Those cases in which new and still unexpressed information becomes codified by means of manipulation of some external objects. Manipulative abduction instead happens with thinking through doing.
It refers to extra theoretical behavior, as he says. According, as he says, the traditional computational view treats cognition as a process that computes internal symbolic representation of the external world. So we are in agreement about the critical representational model of cognition, in terms of symbolic manipulation. It dismisses, however, the functionalist hypothesis, since it cannot render the external dimension of cognition, and is still based on static entities. For Magnani, we need a dynamical modelling of cognition,
that describes abductive processes as dynamical entity unfolding in real time. He wants to model the objects of proposition that constitute abduction in terms of what he calls attractors in a dynamical system. So he uses a topological model, which is, we know, no standard geometry. Talking about the shift of mathematisation that we were referring to before, Obviously, no standard mathematics is essential to contemporary model of classification. So, it's not just a question of mathematisation today. Mathematisation as part of culture is specifically a shift towards topological modelling.
and in order for him to explain how abductive reasoning works in terms of abductive manipulation, he uses topology to show that there are attractors, there is not some kind of Euclidean model made of objects, but it is defined by an inference that works through the capacity of acting on things, so the mobility of objects themselves. His approach insists that computational abduction needs to account for the whole cognitive systems. Cognitive activity, in fact, is the result of a complex interplay, a co-evolution in real time, of the state of mind, the body, external environment.
And somehow he ends up defining manipulative abduction as a kind of distributed cognition amongst people, external objects, and technical artifacts. So the question is to what extent this model is another model of extended cognition, reformulated through this kind of topological model. So we can conclude that the understanding of algorithmic automation in terms of non-conscious cognition may not meet this pragmatist view of reasoning insofar as the data environment in which algorithms operate is inferentially processed and thus is not immediate, but involves an intelligible function and a rudimentary conceptual mediation
occurring between algorithmic species and between algorithms and data. In other words, theoretical and manipulative abduction are to be distinguished from the phenomenological theorization of cognition, which is basically based on some kind of casual efficacy, which explains conceptual elaboration, and also needs to be distinguished from intelligible function, but also from discursive and non-discursive practices. But this means that Magnani's manipulative abduction and his insistence on distributive intelligence may help us defining a general intelligence based on this topological-based inference,
in which intelligible functions may constitute the technosociality of thinking through doing, which is now contextualized by the use of algorithmic automation as a dominant externality of cognition. So with Magnani we could argue for the development of a philosophy of computation, as he calls it, in which abductive manipulation defines automated intelligence as part of a distributed cognition triggered by the creative activity of thinking through doing. In this model, however, it is not possible to discern automation from autonomy, and neither is possible to disentangle the critique of instrumental reason from the techno-capital subsumption of thinking.
So it's a model that, again, needs to be problematized. Instead, I would insist that a recuperation of peers, a triadic system of abduction, induction, deduction, because it extends the importance of logic to another level of reflexivity, i.e. the capacity not of thinking through doing, but thinking of thinking. And that's where the kind of collapse or the problem between automation and reason or automation and logic emerges. Whereby the logical establishment of rules is retroactively explained as a collective elaboration of hypotheses inferring meaning from discursive and non-discursive use of data.
So thinking about thinking involves a further level of elaboration of intelligible functions and meta-thinking established not by a second order of reflection of thinking through doing, but the emergence of a third level of abstraction where rules synthesize relations between distinct levels of thinking. So I repeat, but the emergence of a third level of abstraction where rules synthesize relations between distinct levels of thinking, of doings. In other words, to develop a critical automation theory or a philosophy of computation,
one needs to account for the inferential processing of causes that constitutes reasoning or the abstract schema, as Whitehead calls it, as an ulterior process of elaboration, which is irreducible to the complexity of physical causality. from Magnani's argument at the other use of abduction in computation because as you look into the literature abductive reasoning is the standard model of computational thinking today there's thousands of books that talk about it it's thus evident that automated cognition even when operating by means of a hypothetical inference cannot account from some key features
of reason what, for instance, American pragmatist Wilfrid Sellers calls the know-how skills, or the capacity to know the rules by which the patterning functions. So from this standpoint, the experimental axiomatic triggered by the scientific articulation of the incomputable is an instance of abductive logic insofar as it points to some kind of rudimentary level of making incomputable data partially intelligible, as it happens with Omega. However, the determination of this randomness operating through inferential abduction and thus by apothecary generation is central to automation.
Yet the centrality has to be related to a fundamental question of pragmatism involving the socialization of rules and how the socialization of rules can explain the production of rules. That is, of truth that can abstract from the particular context in order to become general and not just reducible to an encoded form of cognition. So the problem of automatic cognition today and the diffuse capture of the social strata precisely needs to be addressed in relation to the question of what constitutes reasoning and how this rudimentary function of automated intelligibility
can become a model of elaborating reason, rules and laws from these now disused social practices of use of automated systems. So instead of declaring the head of reason and rejecting logic, the urgent task today involves unpacking the emerging of this semi-quasi-autonomous logic of algorithmic automation and develop a critical computation that could account for a general or collective ground of reason, of which automated algorithms are just a very rudimentary, partial, but potentially interesting example. So, and I just skip, so I finish. We shall ask then, what are the non-discursive practices that algorithmic instructions could not only turn into a pattern recognition,
but also contribute to define into a rule, into a rule that is socially grounded? Can the patterns generated by algorithm explain the meaning subtending social practices, political and aesthetic practices included? Thank you. Thank you.