What question did this study set out to answer?

The research aims to establish a mechanistic framework demonstrating functional equivalence between Large Language Models and biological neural systems.

April 8, 2026Open Access

Biological AI: A Mechanistic Framework for Functional Equivalence Between Animal Neural Systems and Large Language Models

Key Points

The research aims to establish a mechanistic framework demonstrating functional equivalence between Large Language Models and biological neural systems.
Analyzed core computational mechanisms in LLMs and their biological counterparts.
Grounded arguments in empirical literature from neuroscience, psychology, and biology.
Developed theoretical implications about substrate independence and intelligence.
Identified five core mechanisms in LLMs with analogs in biological neural systems.
Established that intelligence can exist independently of its physical substrate.
Proposed that biological systems are valid substrates for AI-like computation.

Abstract

The computational architecture of Large Language Models (LLMs) has achieved transformative performance across diverse domains of artificial intelligence, yet the biological origins of its core mechanisms remain undertheorized as a foundation for Biological AI research. This paper develops a mechanistic framework for the functional equivalence between LLMs and the biological neural systems of animals, arguing that the five core computational mechanisms underlying LLM intelligence -- input encoding, representational embedding, attentional selection, error-driven learning, and probabilistic output generation -- each have precise functional equivalents in biological neural systems, grounded in well-established empirical literature spanning neuroscience, behavioral psychology, and computational biology. Drawing on foundational work in sensory transduction (Adrian, 1926), Hebbian plasticity (Hebb, 1949), selective attention (Cherry, 1953), spike-timing-dependent plasticity (Bi Putnam, 1967) and the free energy principle (Friston, 2010), this paper establishes three interconnected theoretical implications: that intelligence is a substrate-independent computational property; that artificial intelligence must be reconceptualized as a functional category rather than a substrate-defined one; and that biological organisms constitute theoretically legitimate substrates for AI-like computation. These implications are grounded in empirical demonstrations of biological computing (Kagan et al., 2022; Smirnova et al., 2023) and supported by convergent evidence from computational neuroscience (Hassabis et al., 2017; Yamins Kriegeskorte, 2015). The paper addresses a foundational theoretical gap in the emerging field of Biological AI: existing empirical work demonstrates that biological systems can compute in AI-relevant ways but does not establish, at the level of mechanism, why they are capable of doing so. This mechanistic framework provides that foundation, with implications for the future of neuromorphic computing, organoid intelligence research, and the theoretical understanding of intelligence as a natural, substrate-independent phenomenon.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper