Biologically-inspired AI and Mortal Computation
Science & Technology
Introduction
In the race to build artificial intelligence (AI), there's a fundamental aspect that we've been missing: how to draw inspiration from real neurobiological systems and translate their functionalities into artificial frameworks. Professor Alex Oria, a researcher in computational neuroscience and cognitive science at the Rochester Institute of Technology, believes that it's possible to create AI that thinks more like humans. This idea stems from understanding why the human brain operates on just a few watts of energy while modern generative pre-trained transformers (GPT) consume energy equivalent to the carbon footprint of a large city.
Professor Oria emphasizes an embodied cognition view, suggesting a multi-faceted approach to creating bio-inspired intelligence that not only models neural processes but also examines the intertwined relationship between software and hardware. The process of generating bio-inspired intelligence presents a unique set of challenges, primarily due to the nascent nature of neuromorphic technology platforms. How can we overcome these obstacles to build a new generation of AI that is compatible with human emotions and cognition?
Professor Oria and his team have proposed a concept called "Mortal Computation," which refers to the idea that machine intelligence cannot be divorced from the substrates that implement it. Mortality in computation encapsulates the imperatives of self-preservation, closely linking it to thermodynamic principles that govern energy efficiency. Unlike traditional forms of computation that prioritize abstract models, mortal computation emphasizes the need for energy-efficient systems that mirror biological processes.
The paper presented by Oria and Professor Karl Friston discusses this framework's significance, particularly in understanding how natural systems preserve their organizational complexity. This idea connects back to existing fields like biomimetics and cybernetics, providing a foundational basis for building intelligent systems that embody the principles of biological learning.
In addressing the challenges of implementing these systems, Professor Oria highlights the critical role of Markov blankets, which serve as boundaries that allow entities to adaptively interact with their environment while preserving their internal states. Moreover, "Mills" (Mortal Inference Learning and Selection) provides a structured framework for these interactions across different timescales.
For bio-inspired AI to operate effectively, several strategies need to be evaluated. The work posited about neuroevolutionary approaches, like NEAT (Neuroevolution of Augmenting Topologies), suggests that we might unlock further potential in evolving neural networks that can adapt and self-organize their structures for specific tasks. Neuro-generative coding attempts to synthesize data in a way comparable to natural biological processes, showcasing the possibilities of creating robust learning systems that outperform traditional methods.
Ultimately, the conversation about biologically-inspired AI and mortal computation is vital, not just for understanding intelligence more deeply but for ethically guiding AI as it continues to evolve. Collaborations between disciplines, from neuroscience to machine learning, can pave the way for future advancements in artificial intelligence that are more aligned with human cognition and can coexist harmoniously with humanity.
Keywords
Biologically-inspired AI, Mortal Computation, embodied cognition, energy efficiency, neural networks, predictive coding, neuroevolution, Markov blankets, Mills framework.
FAQ
What is biologically-inspired AI? Biologically-inspired AI refers to the development of artificial intelligence systems that mimic the processes and structures of biological neural networks to improve efficiency and functionality.
What is Mortal Computation? Mortal Computation is a framework that suggests that machine intelligence is intertwined with the substrates that implement it, emphasizing self-preservation and energy efficiency based on thermodynamic principles.
How does predictive coding relate to biologically-inspired AI? Predictive coding is a model that explains how neural networks can make predictions and adapt their behaviors based on discrepancies between expected and actual inputs, making it a foundational idea for designing bio-inspired systems.
What challenges are associated with building these AI systems? Challenges include the complexity of neurobiological processes, the nascent state of neuromorphic technology, and the need for energy-efficient architectures that effectively leverage the principles of biological learning.
How do neuromorphic chips play a role in this field? Neuromorphic chips allow for parallel processing and energy-efficient computation, resembling biological neural dynamics, making them ideal for running biologically-inspired AI systems.