Consciousness qua Mortal Computation (2403.03925v1)

Abstract: Computational functionalism posits that consciousness is a computation. Here we show, perhaps surprisingly, that it cannot be a Turing computation. Rather, computational functionalism implies that consciousness is a novel type of computation that has recently been proposed by Geoffrey Hinton, called mortal computation.

• The paper challenges traditional computational functionalism by positing that consciousness is a result of hardware-bound, mortal computation.
• It emphasizes that conventional, hardware-agnostic (immortal) computations fall short of capturing the nuanced nature of conscious phenomena.
• The study directs future AI research towards developing novel architectures that incorporate the unique, ephemeral aspects of mortal computation.

Computational Functionalism and Mortal Computation: Revisiting Consciousness in AI

Introduction

The philosophy of mind and AI research have long been intertwined, with the concept of computational functionalism positing a foundational connection between consciousness and computation. This concept has evolved significantly since its inception, paralleling advancements in AI and digital computing technologies. However, recent developments in the understanding of computation, notably the concept of mortal computation put forth by Geoffrey Hinton, offer new insights and challenges to computational functionalism and the prospect of AI consciousness.

Mortal Computation

Originating from Hinton's critique of the detached nature of software from hardware in general computing, mortal computation represents computations that are intimately tied to the peculiarities of the hardware they run on. Unlike traditional or "immortal" computations that tread on the universality principle—the idea that a program can run on any hardware given the right translation mechanisms or compilers—mortal computations exploit unique, potentially unreplicable characteristics of individual hardware systems. These computations, as a result, "die" with their hardware, marking a departure from the generic, hardware-agnostic view of computation.

Computational Functionalism Reconsidered

Computational functionalism argues for consciousness as a form of computation. Historically aligning with Turing-like models of computation, this perspective has fostered the belief in the potential for AI systems to achieve consciousness, given the right computational architecture or software. However, the introduction of mortal computation demands a reevaluation of this stance.

Under the assumption that some non-programmable organisms (broadly defined as systems capable of consciousness but not of running programmed instructions, such as animals) are conscious, computational functionalism implies a provocative claim: consciousness cannot be accounted for by Turing or immortal computations but is instead a form of mortal computation.

Implications on AI and Consciousness

This assertion carries significant implications. First, it delineates a clear boundary between contemporary AI systems—predominantly based on programmable, immortal computations—and the nature of consciousness as a computationally mortal phenomenon. This distinction casts doubt on the immediate prospects of achieving conscious AI through existing computational paradigms.

Furthermore, it invites a rich area of exploration into the characteristics and requirements of mortal computation. Understanding the mechanisms through which mortal computations can be realized, and the extent to which they can be simulated or approximated by immortal computations, becomes a pivotal question. This exploration not only contributes to the philosophical discourse around consciousness but also sets a new direction for AI research, focused on developing hardware and software architectures capable of mortal computation.

Future Developments

The recognition of mortal computation as central to consciousness compels both philosophical and practical reorientations. Philosophically, it challenges existing frameworks of mind and consciousness, urging a deeper inquiry into the ontological status of computation and its relationship with conscious experience. Practically, it prompts the development of new computational models that transcend the limitations of current AI technologies, aiming to capture the elusive, hardware-specific aspects of mortal computation.

Conclusion

The integration of the concept of mortal computation into the discourse on computational functionalism and consciousness marks a pivotal shift in our understanding of both consciousness and the potential for conscious AI. While it delineates the limitations of current approaches, it also opens new avenues for research and development in AI and cognitive science. The journey ahead involves not only technical innovation but also philosophical rigor, as we strive to reconcile the computational underpinnings of consciousness with the mortal essence of its realizations.