Computational Metaphors Oversimplify Embodied Biological Reality

The AI "Plato Problem": Why Idealized Models Risk Obscuring Reality

This conversation with Professor Mazviita Chirimuuta reveals a fundamental tension in AI and neuroscience: the allure of elegant, idealized models versus the messy, complex reality of biological systems. The hidden consequence of this pursuit of "real patterns" is the potential for entire fields to be led astray by oversimplification, mirroring historical scientific missteps. Anyone invested in understanding the true nature of intelligence, whether human or artificial, will gain an advantage by recognizing how our cognitive limitations and philosophical biases shape our scientific endeavors. This analysis highlights why a focus on "haptic realism"--knowledge gained through active engagement--offers a more robust, albeit more challenging, path than abstract, disembodied computation.

The Kaleidoscope of Idealization: When "Real Patterns" Lead Us Astray

The pursuit of understanding, particularly in fields as complex as neuroscience and artificial intelligence, often relies on abstraction and idealization. As Professor Mazviita Chirimuuta explains, abstraction involves ignoring details, much like frictionless surfaces in introductory physics. Idealization, however, goes further, attributing properties to a system that are known to be false, such as assuming infinite populations in genetic models. While these tools are essential for creating tractable theories, Chirimuuta warns of a profound pitfall: the "Plato Problem" or the "kaleidoscope effect." This philosophical stance, prevalent in AI research, posits that the universe is fundamentally a neat, mathematical code, and our task is to decompose reality into these underlying rules. The danger lies in assuming these idealized models are the underlying reality, rather than useful, albeit simplified, representations.

"The idea that the mathematical representation is getting you more to the truth the underlying truth of how things are as opposed to what i call the def like the down to earth view of what abstraction is and mathematical representation is that it's something that we do because of our complex um cognitive limitations."

-- Mazviita Chirimuuta

This perspective, Chirimuuta argues, can blind us to the inherent limitations of our own finite cognitive capacities. Instead of recognizing abstraction as a necessary tool for finite knowers, it's elevated to a method of uncovering a pre-existing, perfect form. This can lead to a disconnect from the concrete, biological systems we aim to understand. The history of science offers a cautionary tale: reflex theory, a dominant paradigm in the late 19th and early 20th centuries, treated simple reflex arcs as explanatory universals. Despite prominent physiologists like Charles Sherrington acknowledging these as idealizations, the theory was pursued relentlessly. This pursuit, as Chirimuuta notes, ultimately failed to explain the full spectrum of observed data, demonstrating how an elegant but ultimately false idealization can derail scientific progress. The computational theory of mind, with its focus on brains as computers, can fall into a similar trap, prioritizing computational equivalence over biological reality.

The Spectator's Fallacy: Why Disembodied Knowledge Falls Short

The prevailing computational view of the brain, which likens cognitive processes to machine operations, carries significant implications for how we understand intelligence and consciousness. Chirimuuta critiques this perspective by drawing on John Dewey's "spectator theory of knowledge." This theory, which posits that knowledge is acquired through passive observation, is contrasted with a more active, "haptic" approach. Haptic realism emphasizes that knowledge is constructed through direct engagement and interaction with the world. We don't just "see" reality; we "touch," manipulate, and alter it to understand it.

"The contrast here with is with an ideal of knowledge which is based on this idea that we can know things in a disengaged way if you think of vision as the archetype of knowledge... but if you think that scientific knowledge in particular is more kind of touch like you can't ignore the fact that we um sort of run into things we have to pick things up engage with them ultimately change them in order for us to acquire knowledge of them."

-- Mazviita Chirimuuta

This distinction is crucial when considering artificial intelligence. If intelligence is merely a matter of computational equivalence, then any sufficiently complex system, including a rock, could theoretically be programmed to exhibit cognitive properties. This challenges the very notion of what makes brains special. Chirimuuta suggests that the mechanisms of the brain are deeply intertwined with its biological nature--its living tissue, its biochemical signaling, and its metabolic processes. These are not mere implementation details but fundamental aspects of cognition. The "lottery ticket hypothesis" in neural networks, where most of the network can be pruned after training without loss of function, might seem to support the idea that biological specifics are vestigial. However, Chirimuuta counters that the extreme energy efficiency of biological cognition suggests a deep economy, implying that these "biological details" are essential for survival and function, not extraneous baggage.

The Agency of Embodiment: Beyond Input-Output Mapping

The debate around artificial intelligence and understanding often circles back to the question of agency and embodiment. Chirimuuta argues that true understanding, as humans experience it, is not reducible to input-output mappings, as behaviorism might suggest. Instead, human cognition is deeply integrated with sensory-motor engagement and our biological existence. This means that abstracting language models, detached from a physical body and the capacity to interact with the world, are unlikely to achieve genuine understanding.

"I think certainly there's more to human understanding than that. I think that a thing about human cognition and animal cognition in general is that my view is that it's not a set of discrete modules that work separately from one another. I think language is bound up with sensory motor engagement..."

-- Mazviita Chirimuuta

Furthermore, Chirimuuta highlights the concept of "distal causes" in relation to agency. Human agents are distinguished by their consistent beliefs and ideas, allowing them to respond to stimuli that are distant in time and space. This contrasts with non-living physical systems, whose actions are largely determined by proximal causes. For instance, past experiences and future aspirations significantly influence human behavior, a level of temporal and spatial responsiveness not typically found in machines. This sensitivity to the distal, she posits, is a key differentiator for cognitive systems. The computational theory of mind, by focusing on abstract computations, risks overlooking these embodied and temporally extended aspects of cognition, potentially leading to a flawed understanding of intelligence and consciousness. This is a critical point for AI development, suggesting that merely replicating computational functions may not replicate genuine understanding.

The Heideggerian Warning: Technology, Finitude, and the Loss of Self

The conversation takes a philosophical turn with a discussion of Martin Heidegger's critique of technology. Heidegger viewed technology, including what we now call AI, as the culmination of a metaphysical tradition that encourages a desire to transcend human finitude--our inherent limitations as bounded knowers. This tradition, he argued, fosters an aspiration for a boundless, universal realm of knowledge, a perspective that can be seen in the AI dream of a disembodied, all-knowing entity.

Chirimuuta connects this to the contemporary perception of technology, particularly the "cloud," which is often presented as immaterial and disconnected from real-world constraints like resource consumption and energy usage. This perceived immateriality, she suggests, masks the material realities of technological infrastructure. Moreover, the increasing mediation of our lives through digital interfaces--from online transactions to social media--can lead to a sense of disconnection from our immediate, physical environment. While humans are imaginative beings who co-create digital worlds, the ethical implications of this shift are profound. The potential for children to have less direct social interaction due to screen time raises concerns about their future social development and well-being. This ongoing "experiment" on the next generation underscores the need to critically examine our relationship with technology, recognizing that our finitude and embodiment are not limitations to be overcome, but fundamental aspects of what it means to be a knower.

Key Action Items: Navigating the Labyrinth of Abstraction

• Embrace Haptic Realism in AI Development: Prioritize building AI systems that engage with the physical world through sensory-motor interaction, rather than solely relying on abstract computational models. (Longer-term investment)
• Critically Evaluate "Real Patterns": When analyzing data or designing models, actively question whether identified patterns are inherent to reality or imposed by our own cognitive biases and limitations. (Immediate action)
• Study Historical Scientific Oversimplifications: Regularly examine case studies like reflex theory to understand how seemingly elegant abstractions can lead entire fields astray. (Ongoing learning)
• Resist the "Brain as Computer" Metaphor's Ontological Claims: Acknowledge the utility of computational models but refrain from asserting that brains are computers. Focus on the biological basis of cognition. (Immediate action)
• Integrate Embodiment into AI Understanding: Recognize that genuine understanding is likely tied to embodiment and sensory-motor experience, which current disembodied LLMs lack. (Strategic consideration)
• Consider the "Distal Cause" Principle for Agency: When assessing AI capabilities, look beyond proximal input-output responses to evaluate sensitivity to temporally and spatially distant information. (Analytical framework)
• Acknowledge Human Finitude in Technological Aspirations: Be wary of AI goals that aim to transcend human limitations of embodiment and bounded knowledge, as this may reflect a misunderstanding of cognition itself. (Philosophical grounding for R&D)