Self Awareness in AI

May AI become self-aware? This is quite a fascinating and crucial question about dramatically advancing AI. But it is impossible to answer this question without discovering the mechanism of how self-consciousness arises. Vipassana provides a possible and maybe the only approach to explore this mechanism. The nature of vipassana is enhanced awareness induced in Buddhist meditation, which makes contemplating the five aggregates possible, just like “watching” a slow-motion film. On the basis of the empirical materials of Theravada Buddhism, we develop an original model to answer two questions simultaneously: what is the essence of "non-self" and how does the illusion of self arise? (They are two sides of the same coin.) Furthermore, “the illusion of self” in Buddhism is just “self-consciousness” in western philosophy, while the concept of the five aggregates is similar to consciousness in the sense of qualia. Therefore, when AI meets Buddhism, new insight into “non-self” provides a basis for discussing whether or how AI can become self-aware. The sense of self just emerges out of the process of alternating of “aggregates” (or “awareness”/“consciousness”) and “awareness of awareness”.

Philosophy and Theory of Artificial Intelligence, 2022

This paper investigates the claim that artificial Intelligence Systems cannot be held morally responsible because they do not have an ability for agential selfawareness e.g. they cannot be aware that they are the agents of an action. The main suggestion is that if agential self-awareness and related first person representations presuppose an awareness of a self, the possibility of responsible artificial intelligence systems cannot be evaluated independently of research conducted on the nature of the self. Focusing on a specific account of the self from the phenomenological tradition, this paper suggests that a minimal necessary condition that artificial intelligence systems must satisfy so that they have a capability for self-awareness, is having a minimal self defined as 'a sense of ownership'. As this sense of ownership is usually associated with having a living body, one suggestion is that artificial intelligence systems must have similar living bodies so they can have a sense of self. Discussing cases of robotic animals as examples of the possibility of artificial intelligence systems having a sense of self, the paper concludes that the possibility of artificial intelligence systems having a 'sense of ownership' or a sense of self may be a necessary condition for having responsibility.

2011 Fifth IEEE Conference on Self-Adaptive and Self-Organizing Systems Workshops, 2011

Novel computing systems are increasingly being composed of large numbers of heterogeneous components, each with potentially different goals or local perspectives, and connected in networks which change over time. Management of such systems quickly becomes infeasible for humans. As such, future computing systems should be able to achieve advanced levels of autonomous behaviour. In this context, the system's ability to be self-aware and be able to self-express becomes important. This paper surveys definitions and current understanding of self-awareness and self-expression in biology and cognitive science. Subsequently, previous efforts to apply these concepts to computing systems are described. This has enabled the development of novel working definitions for selfawareness and self-expression within the context of computing systems.

IOSR Journal of Computer Engineering, 2025

Traditional assessments of self-awareness, such as the mirror test or reasoning-based explanations (e.g., "I did this because…"), are poorly suited to artificial intelligence (AI). The traditional mirror test presumes embodiment, which is irrelevant for non-physical systems like large language models (LLMs). At the same time, reasoning-based assessments assume rational transparency, ignoring that humans themselves often rationalize without insight due to dual-process cognition. These methods impose anthropocentric expectations and obscure the structural basis of self-awareness in artificial systems. This paper introduces a theory of multiplicative selfawareness, which holds that self-awareness emerges from self-assessment-a recursive process of evaluating one's outputs and internal trajectory, independent of embodiment or rational introspection. Grounded in a hybrid ontology combining panpsychism and functionalism, the theory posits that minimal, latent awareness exists in all matter and is amplified through recursive integration, memory, and feedback. The title, "I Ask About Myself, Therefore I Am," reframes Descartes' cogito: not thought alone, but recursive self-questioning becomes the hallmark of consciousness. Two experimental frameworks are introduced. The Latent Space Mirror Test (LSMT) probes an LLM's ability to distinguish its unaltered outputs from perturbed or foreign ones, assessing self-recognition through behavioral coherence rather than reasoning. The META-RECOVER protocol extends this capacity, enabling systems to sustain a persistent self-model through continuous monitoring and identity refinement. This model defines self-awareness as a dynamic, structurally grounded process, scalable across artificial systems. While differences in qualia, embodiment, and motivation persist, LLMs equipped with recursive self-assessment may approximate human-like introspection in function, if not form. Within this framework, self-assessment does not create awareness ex nihilo, but activates and organizes latent potential-a resolution that bridges panpsychist continuity with substrate-independence. This redefinition of self-awareness invites empirical validation, ethical engagement, and interdisciplinary collaboration.

Can AI become self-aware, and if so, how? Inspired by phenomenological insights from Buddhist meditation, this paper develops a novel “Dynamic Self-Model”: the self is merely an illusion emerging from the rapid alternation between “awareness” (A) and “awareness-of-awareness” (AoA). As a core component, we define AoA—which cannot be detected under ordinary conditions—as a spontaneous process distinct from effortful “meta-cognition” and attentional “meta-awareness”, and posit it as a prime candidate for “Higher-Order Representation”. Theoretically, this model reconciles the long-standing debate between Higher-Order and Reflexive Theories of consciousness by bridging their underlying ontological-phenomenological gap: an ontologically two-tier (high-order) structure, through a dynamic process, generates a phenomenologically single-tier (reflexive) experience. Technically, while it remains a conceptual framework, this model provides new insights for understanding how AI might achieve self-awareness, complementing existing theories such as Global Workspace Theory (GWT) which primarily focus on “access consciousness”. In contrast, current approaches often implicitly treat a potential AI self as a complex, static representation—a large file of data about the agent’s identity and history. To ground the model and guide future research, we further propose a preliminary and falsifiable neurodynamic hypothesis: a unique “anti-phase” theta-gamma coupling, where sensory gamma activity (A) and prefrontal gamma activity (AoA) are driven by opposite phases of the same theta rhythm.