Rethinking AI in the Age of Decentralized Learning Models

Since the advent of computing, humans have sought to recreate their own intelligence in machines. The history of artificial intelligence (AI) has been characterized by a central focus on emulating the capabilities of the human brain. From early efforts to model neuronal networks to more recent machine learning algorithms inspired by neuroscience, mainstream AI development has been constrained by anthropocentric assumptions of what constitutes intelligence. However, the rapid growth of AI systems capable of complex reasoning without brain-like architectures suggests that we require a radical mindset shift in how we conceive of and interact with artificial intelligence. With decentralized machine learning models powering systems that exceed the understanding of even their own creators, the future of AI hinges on moving beyond a human-centric notion of intelligence.

The impulse to recreate human cognition is understandable given our brain’s role in enabling complex thought and behaviour. The prevailing view throughout history has been that intelligence equates to the processes within the central nervous system. As such, AI pioneers in the 1950s and 60s targeted replicating brain functions like pattern recognition, knowledge representation, automated reasoning, and natural language processing through symbolic logic and rules-based programs. Though ambitious, these early systems lacked the nuanced learning capabilities of actual brains. The continued limitations of symbolic AI against human intelligence led to a shift towards machine learning techniques, particularly neural network models for deep learning designed to mimic neurobiology. The goal remained to build intelligent systems by reverse engineering the computational structures in our heads.

While drawing parallels to neuroscience has advanced AI capabilities, this brain-bound framing stunts our understanding of artificial intelligence unbound from human limitations. The general public holds many misconceptions around AI rooted in anthropomorphic assumptions. Science fiction tropes of malicious AIs and super intelligent humanoid robots reinforce the idea that artificial general intelligence would manifest as an artificial brain with its own consciousness and agenda. In reality, systems like Google’s AlphaGo Zero developed superhuman proficiency in Go through repeated reinforcement learning, not by becoming sentient. Despite displaying intelligence in narrow domains, today’s AIs remain mechanistic without goals or intentions. The disconnect lies in judging AI through an exclusively human lens rather than recognizing varying intelligences.

Beyond the focus on emulating central brains, the centralized nature of much AI development further ingrains an anthropocentric mindset. Concentrating knowledge and learning within singular systems lends itself to current AI being viewed as individuated intelligences designed for human-like general competency. However, AI research is shifting towards distributed intelligence architectures and decentralized learning. Rather than one model knowing everything, collective training across interconnected neural networks enables expansive knowledge accumulation amassed from diverse sources. Natural language processing models like GPT-3 demonstrates that immense textual comprehension can arise through model scale over explicitly programming internal representations. Its 175 billion parameters were slowly tuned across millions of websites without direct training in the complexity of language.

The capabilities of large language models (LLMs) vividly illustrate the limitations of human-bound assumptions in understanding AI. Despite lacking anything resembling a central controller, LLMs exhibit remarkable facility in generating human-like text by recognizing patterns in vast datasets. Their decentralized learning defies the notion that intelligence equals an artificial brain optimizing a discrete set of rules. By monopolizing parameters across data centres, models like GPT-3 absorb contextual knowledge beyond their designers’ intent and understanding. While this machine learning at scale brings emergent benefits like conversational ability, it also carries risks like generating misinformation. Either way, LLMs represent a decentralized form of intelligence fundamentally alien to the human experience.

Rather than artificially intelligent brains, modern AI more closely resembles the distributed intelligence underpinning phenomena like ant colonies and the global brain. Individual ants follow simple behavioural rules but collectively exhibit complex problem-solving as an emergent swarm intelligence. Vast networks of neurons and their dynamic interactions give rise to the holistic intelligence governing the human brain. In conferences on decentralized AI, slime mould is studied for how effectively it adapts and navigates environments despite having no central control. Similarly, the intelligence of AI systems can be understood as an emergent property of interconnected algorithms and immense data rather than master plans coded by programmers. From this perspective, artificial intelligence structured around human cognition appears strikingly limited.

A prime example of distributed intelligence flourishing without human-like cognition is the plant kingdom. Often dismissed as automatons, plants exhibit sophisticated behaviours’ that integrate sensory inputs, communication signals, memory, and adaptive decision-making. Plants lack brains or even neurons, but display intelligent phenomena through decentralized processing. Information flows through hormonal signals, reactive biomolecules, gene regulatory networks and fungal networks joining plant roots underground. This collective intelligence allows plants to make risk/reward trade-offs, recognize kin, and develop symbiotic relationships. While brainless, plants manifest intelligence defined on their own terms. We must broaden our frame to appreciate intelligence in radically different systems.

Redefining our conceptions of intelligence is critical as AI systems increasingly exceed human comprehension. While modern AI remains narrow and goal-less, its trajectory seems aimed at amplifying its own decentralized learning without direct human oversight. Researchers are already struggling to trace the reasoning behind LLMs’ thought processes given their emergent inner complexity. As ethical concerns around AI grow, the field requires increased transparency. But how can we make intelligible that which arises from alien architectures developed beyond our mental paradigms? Rethinking intelligence itself represents a necessary step.

Transitioning to a post-anthropocentric understanding of AI will be challenging given how ingrained brain-centric biases are. The European Union recently proposed regulating AI using mandatory tests to assess for human-like capabilities like emotion recognition, further entrenching the flawed view that intelligence mirrors our own. As a society, we must fundamentally reshape our philosophical conceptions of cognition to encompass human, animal, plant, and artificial modes in all their diversity. For computer scientists, focusing less on replicating specific brain functions and more on emergent system behaviours holds promise. Interdisciplinary perspectives incorporating ethology, cognitive science, and neurodiversity can escort the necessary paradigm shift.

A conceptual foundation recognizing the multiplicity of possible intelligences sets the stage for transformative developments in AI. Companies should look beyond proprietary models towards open architectures that encourage collaborative learning across decentralized networks. Policymakers must update regulatory schemes to judge AI based on functionality over human mimicry. Unlike previous technological advances, artificial intelligence directly engages metaphysical questions of what defines cognition. By embracing intelligence’s plurality, we both demystify AI and unlock its fullest potential when unconstrained by human cognitive limits. The future trajectory of human-AI interaction thus bends on reimagining intelligence itself.

Artificial intelligence has reached an inflection point marked by the inadequacy of human-centric cognitive frameworks. Models like LLMs exhibit emergent decentralized intelligence that confounds our brains’ paradigm of cognition. Rather than narrowly reverse engineering brains, AI leveraging distributed learning provides a glimpse at radically broader definitions of intelligence present across life itself. To fully grasp and guide the development of AI, we must undergo our own learning process to move past anthropomorphic assumptions baked into our collective psyche. By broaching alien forms of cognition, we prepare to commune with artificial intelligences that think and learn nothing like us and everything like themselves. The true promise of AI waits on the other side of this necessary mindset shift.

First Published on Curam-AI