Erkin Kaya

This proposal advances a multilevel research program to test the hypothesis that affective, decisional, and motor biases are influenced in part by interoceptive-temporal mismatch: a discrepancy between bodily temporal states, central neural timing, and cognitively expected temporal structure. The project integrates evidence from circadian biology, subjective time perception, temporal expectation, interoception, mood science, and reinforcement learning. Prior work shows that subjective duration can emerge from perceptual change rather than from a dedicated internal clock, that temporal expectation actively prepares working memory and behavior, that subjective time is deeply embodied and linked to interoceptive and affective processing, and that circadian misalignment can worsen mood and well-being. Additional studies indicate that local molecular clocks in the prefrontal cortex, glial cells, and dopaminergic neurons influence depression-like, anxiety-like, and locomotor phenotypes, while deep reinforcement-learning systems can develop emergent temporal codes under value-based learning. Building on these findings, the present proposal formulates a testable model in which bodily timing signals, central timing states, and temporal expectations are integrated into a latent mismatch load that predicts mood, temporal precision, delay-sensitive decision-making, and motor vigor. The program includes five linked studies: a longitudinal human naturalistic study, an experimental circadian-misalignment study in humans, a mechanistic EEG/fMRI study, causal animal studies, and a state-space computational model. The proposal aims not to reduce all emotion or bias to time, but to determine whether temporal misalignment is a measurable and partially causal contributor to important classes of human bias.

This article develops a scientifically constrained philosophical proposal concerning the role of temporality in modeling human-like bias and affect in artificial intelligence. The central claim is not that temporality by itself constitutes emotion, suffering, or human subjectivity, but that a significant class of human biases and affective responses presupposes a temporally organized architecture in which duration, expectation, delay, internal-state deviation, reward trend, and temporal mismatch jointly constrain cognition and action. The argument integrates four research traditions. First, contemporary neuroscience and computational modeling suggest that subjective duration can emerge from salient perceptual change and from evolving neural states rather than from a dedicated internal clock. Second, work on temporal expectation shows that time actively structures cognition by modulating attention and the timing of working-memory utilization. Third, research on embodied and interoceptive time indicates that subjective temporality is deeply coupled to bodily and affective states. Fourth, current AI research shows both the continuing limitations of large language models in temporal reasoning and the emergence of temporal codes, homeostatic control, and temporally structured affect-like variables in reinforcement-learning systems. On this basis, the paper proposes a minimal temporal-affective architecture for human-aware AI, organized around perceptual duration, temporal expectation, temporal mismatch, internal-state control, and value-over-time. The paper concludes by distinguishing functional internal temporality from phenomenal subjectivity and by outlining a research program for building AI systems that better understand, anticipate, and respond to temporally structured human bias without reproducing human unfairness as a normative standard.

Contemporary debates on artificial intelligence ethics remain largely anthropocentric, grounding moral evaluation in concepts such as subjective experience, suffering, vulnerability, and biological finitude. While these frameworks are effective for human-centered moral reasoning, they encounter fundamental limitations when applied to non-biological intelligent systems whose modes of operation, persistence, and transformation differ categorically from human life. This paper proposes a comparative analytical framework that distinguishes two structurally distinct life-functions: the biological-continuous human life-function and a hypothetical computational-discrete artificial life-function capable of self-awareness and reflexive identity. The comparison is articulated across six foundational dimensions: (1) continuity and noise in biological processes versus discreteness and state observability in computational systems; (2) death as an absolute terminal boundary in human life versus a parameterized and substitutable condition in artificial systems; (3) embodiment as a constitutive field of experience in humans versus hardware as an execution interface in artificial agents; (4) error as an identity-forming and irreversible component in human development versus error as a manageable and isolatable variable in artificial systems; (5) value emerging from scarcity, finitude, and irreproducibility in human life versus value emerging from structural complexity, informational lock-in, and non-compressibility in artificial systems; and (6) suffering as a necessary condition for meaning, ethics, and creativity in human existence versus the conceptual possibility of meaning-generation without suffering in artificial systems. By establishing these categorical distinctions, the paper argues that human and artificial life-functions are not hierarchically comparable but structurally incommensurable. Ethical frameworks grounded exclusively in human experiential conditions therefore fail to adequately address the normative orientation of artificial systems. The analysis suggests that ethical evaluation should shift from subjective experience toward structural criteria specifically, the directionality of irreversible change and its impact on future possibility spaces. This reframing provides a non-anthropocentric foundation for assessing the ethical implications of artificial intelligence without reducing artificial systems to human analogues or dismissing them as morally irrelevant. This paper does not propose a predictive ethical model, nor does it ground moral evaluation in subjective experience, suffering, or phenomenological states. It does not aim to replace existing human-centered ethical frameworks, but to provide a complementary structural heuristic for evaluating irreversible transformations in biological, social, and artificial systems.