Mortal Computation: A Foundation for Biomimetic Intelligence (2311.09589v2)

Abstract: This review motivates and synthesizes research efforts in neuroscience-inspired artificial intelligence and biomimetic computing in terms of mortal computation. Specifically, we characterize the notion of mortality by recasting ideas in biophysics, cybernetics, and cognitive science in terms of a theoretical foundation for sentient behavior. We frame the mortal computation thesis through the Markov blanket formalism and the circular causality entailed by inference, learning, and selection. The ensuing framework -- underwritten by the free energy principle -- could prove useful for guiding the construction of unconventional connectionist computational systems, neuromorphic intelligence, and chimeric agents, including sentient organoids, which stand to revolutionize the long-term future of embodied, enactive artificial intelligence and cognition research.

• The paper presents the mortal computation thesis, arguing that computation must be integrated with its hardware, much like the finite lifespan in biological systems.
• The authors employ the Markov blanket and free energy principle to design a framework for neuromorphic, self-organizing systems that adapt dynamically.
• The research lays a foundation for sustainable 'Green AI' by incorporating biomimetic concepts such as autopoiesis and homeostasis for energy-efficient computation.

Overview of "Mortal Computation: A Foundation for Biomimetic Intelligence"

The paper by Alexander Ororbia and Karl Friston provides a comprehensive synthesis and theoretical framework termed "mortal computation," aimed at advancing biomimetic intelligence. By anchoring this concept in biophysics, cybernetics, and cognitive science, the authors propose a computing paradigm deeply integrated with its hardware, mirroring the close interplay between a biological organism's software and hardware.

Key Concepts and Framework:

1. Mortal Computation Thesis: Central to the paper is the notion that biological and artificial systems that exhibit intelligence must tightly interweave computation with the hardware substrate. Unlike traditional computer science paradigms where software is independent of hardware, mortal computation postulates that computation is bound to the life cycle of the hardware, akin to biological entities where the death of the organism ends the computational process.
2. Theoretical Underpinnings: The authors utilize the Markov blanket formalism, embedding the free energy principle, to describe computation in living systems. This foundation serves for constructing neuromorphic or chimeric agents, emphasizing systems that adapt, self-organize, and evolve to maintain themselves—an attribute that could be crucial for achieving artificial general intelligence (AGI).
3. Biomimetic and Bionic Intelligence: The paper argues for a shift towards systems that emulate natural intelligence's adaptive and survival-oriented processing. The artificial systems, thus, should possess life-like qualities such as homeostasis and autopoiesis, ensuring persistence in interaction with dynamic environments.

Theoretical Implications and Components:

• Metabolic Influence and Homeostasis: Drawing parallels with biological systems, the paper emphasizes the importance of incorporating homeostatic mechanisms within artificial systems to achieve thermodynamic and computational efficiency. The embodiment of these systems ensures that information processing aligns with energy constraints dictated by physics.
• Autopoiesis and Morphology: A significant aspect of mortal computation is the self-producing and maintaining features of the system, paralleling biological life that continuously seeks stability far from thermodynamic equilibrium. Morphology, in this context, is treated as a computational asset that defines function and adaptability.
• Inference, Learning, and Selection (MILS): The authors delineate a framework where autonomous systems conduct inference, learning, and structural optimization as part of an intertwined feedback loop. This suggests a dynamic adaptation process that is integral to the survival and evolution of computational systems.

Practical Implications and Future Directions:

The research sets a visionary course toward creating systems that incorporate natural intelligence properties, proposing solutions to contemporary AI research's energy and sustainability challenges. The free energy principle posited here suggests a path toward "Green AI," which optimizes resource usage in computation-heavy tasks. The paper also calls for interdisciplinary approaches, leveraging insights from biophysics, robotics, and computational neuroscience to engineer new computing architectures.

Conclusion:

Ororbia and Friston present "mortal computation" as a foundational stone for future AI research, emphasizing systems' mortality and physical embodiment as critical to mimicking life-like intelligence. Such a paradigm could lead to advances in AGI, enhancing our understanding and interaction with intelligent systems beyond current capabilities. The proposed framework not only spells a transformative shift in thought but also paves the way for developing sustainable and adaptive technologies strongly rooted in the principles of life. The adoption of such biomimetics and bionics principles signals a potential paradigm shift essential for the next generation of intelligent systems.