The undecidability in the Other AI

The paper situates AI at the intersection of science and fiction, noting the absence of a unified theory and the ongoing debates about AI’s scope and implications across culture, ethics, philosophy, politics and cosmology. It reviews futurist claims (e.g., the Singularity, the Fourth Industrial Revolution) and positions computation as the core of AI’s attempt to simulate cognition via algorithmic procedures, following Turing’s computationalist framing. It raises the question of whether algorithmic systems can be said to think or exhibit autonomy as material, socially embedded processes, and argues that contemporary algorithmic thought is nearly fully automated and displays a form of material performativity. The central wager is that psychoanalytic critique of identity (especially Lacan’s theory of the Other) is crucial for understanding AI: systematic unifications of identity fail, and undecidability/incompleteness are intrinsic to both signification and computation. A terminological clarification distinguishes incomputability/noncomputability and undecidability, linking Gödel’s incompleteness, Church–Turing incomputability, and the Halting Problem to show inherent limits of formal systems and algorithms.

The paper surveys key philosophical positions on machine thought: Dreyfus’s critique of AI’s lack of intentionality and consciousness; Searle’s Chinese Room argument against intentionality in programs; and Wittgenstein’s view of intentionality as a linguistic, context-dependent phenomenon rather than a biological property. It engages Deleuze (thought as forced, indeterminate, non-natural), Fazi (thought’s ontological abstractness vs epistemic abstraction), and Parisi (algorithmic randomness/incomputability—Chaitin’s Omega—as internal to computation, revising teleological rule-finality). It connects these to Lacanian psychoanalysis: the Other (A) as the symbolic order of language/law and radical alterity; identity as a fragmentary process of identification/disidentification. It brings in the historical development of logic (Boole, Frege) and Gödel’s incompleteness theorems, emphasizing paradoxes (Russell’s, liar) and the problem of metalanguage in Lacan’s seminars, to argue that any totalizing identity/system is inconsistent. The review extends to contemporary AI: datafication and software as signifier-like writing; digital objects as code-produced, simultaneously universal and singular in operation; and critiques showing models embed human judgments. It employs standard results from computability theory (Church–Turing thesis, computable/uncomputable numbers, Busy Beaver, Halting Problem; reductions) and recent formal developments (Kirst & Peters on synthetic computability and Coq) to frame undecidability as a structural feature relevant to AI.

A conceptual and theoretical analysis integrates psychoanalytic theory (Lacan’s Other, the real, inconsistency of the signifying order) with foundations of logic and computability (Gödel’s incompleteness, Turing’s definitions of computability, Halting Problem, Busy Beaver, reducibility). The paper proceeds by: (1) reconstructing philosophical debates on whether machines can think (Searle, Wittgenstein, Dreyfus) to shift focus from consciousness to linguistic and symbolic functioning; (2) explicating formal results that establish limits of axiomatic and algorithmic systems; (3) mapping Lacanian notions of signifier systems onto digital technologies (data, code, software) to argue for structural incompletion within symbolic/computational orders; and (4) using canonical proofs and definitions from computability to demonstrate undecidability as intrinsic rather than accidental. It also offers an interpretive analogy for AI systems: datasets as the AI’s “Other,” trained parameters as an “unconscious,” and inference software as the mediating mouthpiece, highlighting embedded human norms in models.

• Undecidability and incomputability are inherent to formal and computational systems (Gödel’s incompleteness; Turing’s Halting Problem; Busy Beaver’s noncomputability), not contingent errors or exceptions.
• Parisi/Chaitin’s insight that randomness/incomputability conditions computation aligns with Lacan’s thesis of the Other’s inconsistency: the symbolic order is structurally incomplete (“minus one”) with a built-in lack.
• Attempts to ground rationality wholly on computation are destabilized by intrinsic undecidability and internal contingency of algorithmic rules.
• AI’s relation to language/signification can be modeled psychoanalytically: datasets as the “Other,” trained models as a kind of “unconscious,” and inference programs as mediators; models embed human judgments and social norms.
• Software/digital objects share properties of signifiers: universality (same code across instances) coupled with singular, context-dependent operations; datafication reflects the differential nature of signifier systems.
• Classical objections about intentionality (Searle) are reframed by Wittgensteinian views of meaning-as-use: understanding is functional within rule-governed practices rather than a biological essence.
• Constructive/meta-mathematical work (e.g., synthetic computability in Coq) reaffirms foundational limits: no consistent, sufficiently strong system proves its own consistency; no algorithm decides halting for all programs.

By juxtaposing psychoanalysis and computability theory, the paper argues that the very structures underpinning AI—formal languages, algorithms, data, and code—exhibit built-in limits akin to the Lacanian lack in the Other. This addresses the guiding question about AI’s agency and rational grounding: if computation mechanizes thought, it nevertheless harbors incomputable elements, so rationality cannot be fully secured by algorithmic closure. The significance for AI is twofold: (1) epistemically, undecidability constrains what can be known or decided by automated systems; (2) ontologically, AI’s symbolic operations are constitutively open, marked by gaps that resist full formalization. Recognizing this undermines narratives of total algorithmic control or exhaustive modeling and reframes debates on AI autonomy, creativity, and understanding in terms of linguistic practice, social context, and the structural incompletion of signifying/computational orders.

The paper advocates a philosophical reevaluation of computational formalisms that acknowledges abstraction together with indeterminacy: computation mechanizes thought yet intrinsically includes the incomputable. From a Lacanian perspective, the signifying order begins with a lack (“minus one”); analogously, any computable system contains its own domain of incomputability. Thus the undecidable is not a glitch to be eliminated but an inherent feature that challenges efforts to found rationality solely on computation. Consequently, AI systems, as symbolic-technological formations, must be understood with their structural limits foregrounded rather than effaced.