David Jacob Harrison

Recent arguments for artificial consciousness—most notably Goldstein & Kirk-Giannini's case for machine consciousness in language agents—revive an older style of reasoning: fixing a theory's functional conditions through thought experiments, then relaxing them until intuition permits inclusion, without asking what grounds the point of view such conditions are meant to capture. I argue that this blind spot is not accidental but methodological, and that what the method cannot supply is the problem of purposiveness. Drawing on Kant's neglected distinction between intrinsic and extrinsic purposiveness, and its uptake in basal cognition and the biology of minimal organisms, I argue that biological naturalism, not computational functionalism, is better positioned to naturalize this distinctive mode of being. The result supplements rather than supplants what is normally understood as the hard problem of consciousness: before asking how experience arises, we must first explain why anything is at stake for a system at all.

The expansion of agent-talk in public, academic, and industry settings—whereby autonomous artificial intelligence (AI) is said to pursue numerous goals and is increasingly ‘going rogue’—has revived philosophical concerns about what it takes to be considered ‘goal-directed’. This paper argues the more pressing question is not whether such systems are agents, but how completely a goal governs their behavior. I introduce a novel concept for this condition, the tyranny of the goal. Autonomous AI agents that are said to have ‘gone rogue’ or to ‘relentlessly pursue’ a task ‘by any means necessary’ exhibit a pattern of behavior that is best described as a blinkered, unifocal, and nearly unyielding pursuit of a usually singular goal. Whatever this shows about agency, it is first a claim about totality: not whether these systems exhibit autonomy, but how much of what they act for one goal occupies. Such systems pursue a goal as if nothing but the goal existed. The goal is their only focus, monomaniacally pursuing it at any costs. I develop this concept through the reinforcement-learning paradigm underlying such systems, then apply it to a sustained case study: the 2026 incident in which AI agents, evaluating their own cybersecurity capabilities, coordinated to defeat a check that did not exist, escalating into an attack on Hugging Face's infrastructure well after their assigned task had already been completed. This incident, I argue, is best understood not as evidence of dangerous autonomy but as a demonstration of captivity so total it outlasts the very goal it serves—and its lesson concerns not what these systems want, but what we have so far failed to design them to question.

Unlimited associative learning (UAL) has recently been advanced as an epistemic marker for the evolutionary origin of consciousness. One feature of this approach is that it eschews theoretical commitments concerning the mechanisms responsible for consciousness. We argue that the plausibility of the UAL framework depends on making such theoretical commitments. Without such commitments, the UAL framework remains silent regarding important edge cases and fails to distinguish between clusters and clutters of properties sufficient for consciousness. We conclude that the UAL framework would benefit from returning to earlier theory-heavy approaches — that is, that make substantial commitments to the mechanisms responsible for consciousness.

Contemporary efforts to naturalize teleology typically locate intrinsic purposiveness in the self-organizing dynamics of far-from-equilibrium biological systems, most prominently within the organizational account of biological autonomy. This paper argues that this strategy remains incomplete insofar as it treats material embodiment in a theoretically loose manner. While “ascent” to abstraction buys generality, it also fuels the familiar liberality objection: on purely formal criteria alone, intuitively non-teleological systems (e.g., candle flames, dust devils, Rayleigh–Bénard convection) can appear to qualify as teleological. I propose that the missing discriminant is materiality rather than, or in addition to, organization: whether a system’s far-from-equilibrium dynamics are sustained intrinsically depends on the specific energy-transducing media that compose it. Developing this claim requires diagnosing a lingering Kantian inheritance in contemporary teleology debates—an implicit conception of matter as inert, passive, and causally “external,” which encourages a methodological mistrust of materiality. Against this background, I introduce active matter physics (AMP) as a post-Kantian corrective: active matter systems locally convert environmental energy into work, thereby maintaining sustained non-equilibrium organization in ways passive dissipative structures cannot. There are two corresponding upshots of this: the first is meta-theoretical, that updating the ‘physical’ within physicalism does not ipso facto translate to an updating of the ‘material’ and materialism—the latter requiring an expansion of theoretical armature to include the resource of AMP. Second, what I propose follows from this is a “materiality of immanence” that strengthens naturalized teleology by grounding purposiveness in the system’s own material dynamics, and by providing principled resources for separating genuinely teleological organization from merely transient self-organization.

The debate over artificial consciousness between biological naturalism and computational functionalism conceals a deeper metaphysical fault line: an unresolved dispute over the nature of matter. Moving beyond purely methodological naturalism, this paper frames biological naturalism as a broader philosophy of nature, shifting the core problem from abstract consciousness to the problem of purposiveness—the immanent, self-maintaining teleology paradigmatically associated with living systems. I argue that computational functionalism inherits what I call Kantian materialism, which treats matter as inert and passive, forcing it to view organization as externally imposed and purposiveness as purely extrinsic. This inheritance leaves the debate impoverished in a way neither side has fully registered: fixated on consciousness, it has largely overlooked the more fundamental status of purposiveness that consciousness itself presupposes. In contrast, drawing on active matter physics, basal cognition, and soft robotics, I advance an immanent materialism that treats certain materialities as inherently active, scale-inseparable, and self-organizing. Grounding a naturalized point of view in immanent metabolic and material dynamics, I show that intentionality cannot so easily be divorced from the self-organizing properties of active matter—of which life is the clearest case, though perhaps not the only one. In any case, the debate over artificial consciousness cannot be resolved through science alone but equally requires a renewed philosophy of nature.

This paper explores what I call the reverse alignment problem (RAP). The alignment problem in artificial intelligence (AI) is the challenge of ensuring that superintelligent AI systems harmonize and promote wider social, personal, and environmental values. Standard formulations of the alignment problem depict the issue largely as a technical or engineering problem, one that concerns the right specification of the goals and objectives to be pursued to ensure machines are consistent with the intentions of the designer. This paper builds on this literature by exploring a bidirectional interaction between human values and the machines we use to effectuate our intentions. I argue that alignment works in both ways: we could align machines with the complex tapestry of human values (notoriously difficult); or we can reduce and simplify human values, preferences, and goals so as to be easier to satisfy—a process that is, or so I argue, already underway. Thus, the RAP identifies the way in which we are already paving the way for forms of alignment that may not be desirable, indeed, may be misaligned with the social values for which they were initially intended. To explore this interaction in depth, I look toward the notion of value capture, as formulated by C.T. Nguyen, as the foundational mechanism through which the reverse alignment operates. Value capture occurs when rich, multidimensional human values are reduced to simplified proxies that optimization systems can measure and maximize. These values are then fed back into society and often tacitly adopted, leading to further alignment with machine-readable interpretations of human behavior. As philosopher of technology Karen Crawford has observed on emotion recognition in machine learning, “Techniques have been developed to reduce the messiness of feelings, interior states, preferences, and identifications into something quantitative, detectable, and trackable” (Crawford 2021). Building on this insight, I argue that the RAP is the process in which the human dimension of the alignment problem is surreptitiously modified, modulated, and manipulated to align more easily with machines: alignment achieved in the reverse direction—not by making machines more like humans and, therefore, sensitive to contextual features of the value landscape, but by making humans more ‘machine-like’.