How does a book arrive in the world? In the case of Reza Negarestani’s ‘Intelligence and Spirit’ (I&S), the text appears to burst forth as from a breached dam – Hegelian tributaries, Sellarsian currents and Turing tides previously stemmed by a conceptual blockage so immense as to constitute a fatberg of thought. This stagnation is down to nothing more than academic philosophy’s attendant failure to adequately address the historical decoupling of intelligence and mind enacted by computation. The early spasms of this seizure of thought were felt in the period between Chomsky’s hierarchy of grammars and the functionalist debates of the 1970’s, the interventions of Putnam, Searle & Fodor amongst others, and the ensuing discourse on multiple realizability.
The deep freeze on research that constituted the first Winter of AI stunted philosophical progress on the relation between computation, mind and language. A fertile ground for collaboration frosted over while research continued strictly within disciplinary silos. This spirit of connective thought did not thaw off until the last decade or two, in which a productive interplay between computer science, mathematics, logic, cognitive science, and philosophy has once again emerged. It is this new constellation which I&S playfully explores, delving into themes such as complexity and algorithmic information theory, predictive coding and posthumanism, with all the sobriety of an infant in Legoland.
I&S exhibits a barely concealed contempt for what Badiou once called the scholastic school of philosophy – that stasis of hermeneutics marked by endless disputes which typify the study of a canon. An alternate view of philosophy as a constructive locus of dialogic thought is instead embraced here. Its many interludes, digressive structure, and dizzying detours, all indicate a distinctly para-academic practice of thinking – theorizing presented as an active mode of world building. One particular aside on universal pedagogy comes to mind, a coruscating tract on how thought itself – intelligence as-such – is everywhere in chains, bound by educational forms institutionally incapable of liberating its geistig potential.
The text forms a thread of sorts around the development of a Toy AGI, an avatar nicknamed Kanzi, and its induction into CHILDhood (concept having intelligence of low degree) via self-awareness, language formation, speech acts and interaction. Topics such as spatio-temporal priors, picturing and syntax, representation and language, are addressed along the way, illustrated by schematics featuring the automaton Kanzi and its doting adult guardians. Key duals such as structure and being, theory and object, intelligence and intelligibility, are all approached through this lens of toy philosophy.
Geist is presented as a general intelligence embedded in its own history, its formal autonomy a matter of collective labour undertaken by interacting agents equipped with language. The self-consciousness of Geist, as concretised via the structuring abilities of mind, is seen as the vehicle not just of agency but of freedom itself. There are traces of a metaphysics here – intelligence occupies a “nowhere and nowhen”, its fleeting appearances for us hinting at atemporal and atopic logics. Language, as the domain of the intelligible, is presented as an intrinsically computational process of syntactic structuring and semantic ascent. This leads to a form of transcendental computationalism, which readers may find difficult to reconcile with an emancipatory politics – a position I&S is determined to defend throughout.
The emphasis on interaction in semantics places I&S in close proximity to the late Putnam, whose notion of socio-functionalism sowed the seeds of Negarestani’s account, while the focus on normativity and pragmatism echoes much of Brandom’s work. While it maintains strong continuities with these positions, the text attempts a novel fusion of neural materialism with the semantic inferentialism of the Pittsburgh School, forging its own path in the process and making decisive breaks with both camps.
The book offers a meaningful contrast to one such neural model – predictive processing (PP), a dominant contemporary theory of mind which places cognition in a probabilistic Bayesian frame. Where PP follows experimental data closely, mirrors techniques in AI, and is bound to inductive logic, I&S emphasizes sapience, logical inference and the formal autonomy of language – but both center the role of generative models. I&S devotes an entire section to the limits of inductivism and the “myth of the given” (Sellars), but PP nevertheless issues a convincing account of perception as a form of error minimization in a predictive model, accounting for a cognitive efficacy which is supported by experimental data. On reasoning, I&S offers a multi-level account of inference which is incompatible with that offered by hierarchical Bayesian priors, but there are gaps in both accounts which invite considerable speculation. The two also differ at the level of philosophical project and the attendant political implications – one engaged in positing a generalized form of intelligence in all its sociality, whereas the other concerns itself with modelling the cognition of individual human minds. The synthesis of these various families of cognitive models, symbolic and inductive, social and individual, has long constituted the major research question in artificial intelligence.
Negarestani’s positions rarely settle in the manner we might expect from a strictly analytical philosopher, but a general picture begins to emerge toward the latter third of the book. The text is sympathetic to the syntactic tradition in logic and linguistics, adheres to meta-theoretic fundamentals inherited from German Idealism, while maintaining broadly rationalist positions on the topics of representation and language, mind and world. There are notable departures from the Sellarsian account of apperceptive reasoning – namely the relation between inferential rules and pattern-governed regularities, the distinction between logical and material rules, and the exact nature of picturing as a non-conceptual representing. These are given a more computational treatment, centering on syntactic structuring via recursive embedding, in an attempt to break out of the “Kantian straightjacket” – namely the subordination of the logical order to sensible intuition.
The autonomy of form posited in the work of Dutilh Novaes is expanded into a broader claim with deep implications for rationality as such. The computationalist thread remains intact as ever sterner challenges are posed of intelligence and its autonomy. Rather than buckle and give way to a critique of computability, I&S rises to the challenge to elaborate a parallel account of intelligence and computation, imbricating both in its rendering of reason as a recognitive capacity. The reader is asked to suspend judgement until the account is developed in full, the book clearly exhibiting a sensitivity to a plethora of computationalist traps. It issues its own challenges to inductivism, superintelligence, pan-computation and game theory along the way, constantly engaged in deflating claims afflicted by various dogmas – be they Humanist, Bayesian, or Darwinist in origin.
The text’s deep Brandomian influence is tempered by a distaste for the philosophical quietism of the Pittsburgh School. We are dealing here not with a philosophy that “aims to leave everything as it is” (MacDowell), but that threatens to spill out of its human host, philosophy as a xenotropic agent of thought, its proper aim no less than to bootstrap its way to the closest exit. Navigating intelligence through the dimly lit alleys of our own minds, unbinding it from the errors and biases placed on it by our own ‘natural’ languages, aiding its evasion of the various pitfalls and dead-ends we unwittingly place in its path – this is the political task posed by a philosophy of intelligence, what Negarestani calls “the labor of the inhuman”. The book is a support vector for this journey, equal parts manual, experiment and schema – no less than a guide to the pre-history of intelligence.
Another key agent of this book is Carnap, evident in the constructive treatment of language and appetite for logical formalism. This disposition is best exhibited in its account of interaction and semantics, the book’s most novel contribution. I&S reserves a convincing synthesis of discourses for these closing chapters, in which a formal account of language as computation is presented, uniting disparate threads of research in logic, computer science, and mathematics. Girard’s logic of Ludics is brought into the ambit of interactionist paradigms in computer science, while a correspondence between proofs and programs is embraced to offer a constructive theory of meaning. The immanent unfolding of semantics from syntactic structures is rendered an open-ended dialogic game and interaction is foregrounded as a properly formal condition of reason. Challenging both a received Turing orthodoxy – namely that computation is a logically closed system that occurs between input and output – and Brandom’s account of sociality, this section indicates a range of possible new research paths.
The text is not without its blind spots. Undecidability isn’t given the proper treatment deserving of a computationalist position, while claims as to the logical basis of language lack Brandom’s nuance and will raise skepticism in readers not convinced by Carnap’s account. In contrast to PP, recent developments in cognitive science are underplayed. It calls for the decolonization of thought but fails to engage with key thinkers of the inhuman such as Wynter or Fanon. The universal nature of the project exposes it to materialist critiques as to the practices that intervene in realizing the “equality of all minds”. Its broad scope means vast topics, from category theory to plasticity, are relegated to footnotes or asides. As a work of speculative philosophy it raises many open questions – Where does generalisation end? What political remnants are to be found in dispensable bodies? Why should intelligence be seen as a virtue? How exactly do normative judgements emerge? These unresolved threads merely speak to the baroque form of the work – an expansive treatise – they do not detract from its call for an explicitly computational rationalism.
But back to the question of arrival. The text arrives at our laps and screens in an age of viral stupidity, spontaneous consensus, and automated bias. The apotheosis of Châtelet’s ‘cyber-wolves’ – those self-appointed princes of the network, from Peter Thiel to Bruno Latour – marks a hegemony in which all forms of minded beings find themselves embedded. The impoverished sense of intelligence offered to us by contemporary AI, coupled with the decentering of the human in many adjacent discourses, forms the historical context of the book’s appearance. I&S tasks us with new ways of thinking intelligence – to consider how formalisms generate their own autonomy, how language is inextricably computational, how interaction bridges syntax and semantics, and finally how selfhood and learning, agency and freedom, can be cultivated in this environment. It challenges the reader to envision the emancipatory potential of an inhumanist account of intelligence, realized by a novel model of computation born of interaction. It asks us to meet the challenge of reorienting both computation and philosophy in service of this geistig manifestation – treated not as an unknowable force or individual performance, but as a vector recursively brought into concrete being only via the recognitive labour of reasoning.