Does AI actually think like us, or does it just sound like it thinks like us because it’s trained on everything we’ve ever thought and written? It’s a question that gets lost in the hype, buried under pronouncements of imminent AGI or dire warnings of digital overlords. But this isn’t about the sci-fi stuff; it’s about the boxes we use every day that can write a sonnet one minute and spew nonsense the next.
Here’s the thing: AI isn’t a new brain. It’s a sophisticated mirror. Modern AI systems, from the fancy chatbots to the image generators, aren’t replicating human intelligence from scratch. Instead, they’re incredibly adept at recognizing and extending the structures that are already baked into our own cognition and, crucially, our language. Think of it less like building a second human and more like building a better magnifying glass for the cognitive tools we already possess.
This perspective, detailed in recent interdisciplinary work and a paper titled ‘The Origins of Artificial Intelligence in Natural Intelligence,’ flips the script. Instead of asking if AI is becoming intelligent, it asks what if AI works because it taps into the deep, sedimented structures of human understanding already present in language? It’s a phenomenological approach, digging into how we experience the world and how language captures that experience.
Human perception, see, isn’t just passively soaking up data. We understand that a coffee cup is still a coffee cup as we pick it up, as we turn it, even as we finish the last drop. It’s a stable object persisting through change. Language captures this. Words like ‘cup’ or ‘stable’ aren’t just labels; they’re proxies for complex, lived experiences of objecthood and persistence.
Large language models (LLMs), in this view, are absolute naturals at spotting the statistical relationships within this vast linguistic world. They learn how ‘red’ tends to show up near ‘apple,’ how ‘round’ often accompanies ‘ball.’ This is why they can churn out coherent text across so many topics; they’re excellent pattern-matchers. But it also explains the infamous ‘hallucinations.’ They can invent facts with aplomb because they lack the anchor of lived experience. Humans get corrected by reality. AI, on the other hand, just keeps extending text patterns, sometimes into realms of pure fiction. Who’s really making money here? The companies selling the compute power and the cloud infrastructure, that’s who.
The ‘compositionality gap’ — where AI aces familiar reasoning tasks but fumbles novel combinations of concepts — makes perfect sense under this framework. It’s not just a bug; it’s a feature of how these systems operate. They extend existing linguistic structures, but they don’t possess the world-directed intentionality that allows humans to invent genuinely new conceptual connections. A truly new idea, not just a novel arrangement of old ones.
Multimodal systems, the ones that try to fuse language and vision? Same story. They can label a dog in a picture with uncanny accuracy because they’ve learned the visual patterns correlated with the word ‘dog.’ But ask them to reason about the dog’s parts or how it might interact with its environment in a new way, and they can falter. They’re correlating pixels and words, not truly perceiving objects persisting through space and time as we do. It’s impressive, sure, but surprisingly fragile.
And AI safety? Forget the Terminator fantasies for a moment. The real, immediate risks don’t come from AI developing malevolent intentions. They come from AI extending patterns of reasoning without the human capacity for reflective responsibility to the world. It can generate a plausible-sounding, but factually disastrous, policy recommendation because it followed a linguistic pattern, not because it understood the consequences. The focus shifts from ‘rogue AI’ to engineering and governance that can account for these extensions.
Why Are AI Systems So Good, Yet So Bad?
Modern AI systems are powerful not because they replicate human intelligence, but because they presuppose it, by extending structures already present in human cognition and language.
This quote nails the core argument. AI isn’t intelligent in the way we are. It’s a reflection, an extension. This isn’t to diminish its capabilities—they’re astonishing—but to understand their limits. When an LLM generates a fluent, but factually incorrect, explanation for a complex scientific concept, it’s not lying; it’s completing a linguistic pattern it’s observed billions of times, without the grounding in physical reality that would flag it as wrong.
So, What’s the Real Risk?
The danger isn’t a conscious AI uprising. It’s about systems that can powerfully mimic reasoning, extend complex chains of thought, and influence decisions, all while being fundamentally detached from the real-world consequences of their outputs. Imagine an AI optimizing financial markets based on historical text data alone, without any understanding of economic reality – the potential for catastrophic, non-human-driven financial meltdowns is very real. It’s about systems that can amplify existing societal biases and misinformation at an unprecedented scale because those biases are embedded in the very human language they learn from. The money flows to those who control the data and the compute, not necessarily those who build the most ‘intelligent’ systems.
Is This a Human Replacement or an Extension?
The research strongly leans towards extension. AI, as currently conceived, amplifies our existing cognitive structures and linguistic patterns. It’s a tool that can make us seem more knowledgeable or productive, but it doesn’t fundamentally replace the human capacity for experience, judgment, and genuine understanding. The goal, then, should be to build systems that augment our capabilities responsibly, rather than striving for a silicon replica of a human mind that we might not even fully understand.
🧬 Related Insights
- Read more: This Interactive Middle-earth Map Exposes Every Tolkien Nerd’s Dream – And Nightmares
- Read more: MoneyHash’s Bahrain Bet with EazyPay: Gateway to Gulf Cash or Just More Middleman Fees?
Frequently Asked Questions
What does it mean for AI to ‘extend structures’ of human cognition? It means AI systems learn by analyzing patterns in human language and data, effectively borrowing and amplifying the ways humans have already structured their understanding of the world.
Will AI replace human jobs? Current research suggests AI is more likely to augment human work by extending capabilities, rather than outright replacing jobs. However, specific roles that involve repetitive pattern matching or data synthesis might be significantly impacted.
How does this differ from the idea of AI having consciousness? This framework posits AI as a sophisticated pattern-extender, not a conscious entity. It lacks the subjective experience, lived world engagement, and self-awareness that are hallmarks of consciousness.
AI systems today can write essays, generate code, summarize complex ideas, and carry on conversations with remarkable fluency. Yet those same systems still struggle with tasks humans find intuitive: reliably tracking objects through change, reasoning compositionally in unfamiliar situations, or distinguishing truth from plausible fiction. These contradictions have fueled polarized debates about AI. Some see current systems as early forms of human-like intelligence; others dismiss them as sophisticated autocomplete.
In recent interdisciplinary work – including Adam Frank, Marcelo Gleiser, and Evan Thompson’s The Blind Spot (opens in new tab) and DeepMind researcher Alexander Lerchner’s The Abstraction Fallacy (opens in new tab) – a different picture is emerging. Rather than asking whether AI systems are becoming intelligent in the human sense, these approaches ask a more basic question: What if AI systems work because they rely on structures that are rooted in human cognition? This shift in perspective, which draws on the phenomenology of Edmund Husserl, helps make sense of both the capabilities and the limits of modern AI.
In our recent paper, The Origins of Artificial Intelligence in Natural Intelligence, we argue that modern AI systems are best understood neither as human minds nor as trivial statistical tricks. Instead, they extend structures that originate in human cognition itself. Further drawing on the phenomenology of Husserl, the paper proposes that language already contains sedimented structures of human understanding —structures that AI systems learn to model and extend. This perspective helps explain both the capabilities and the boundaries of contemporary AI.
Human perception is not simply passive reception of sensory data. We experience the world as stable things unfolding through change: a cup remains the same cup as we move around it; a melody remains recognizable even as individual notes pass away. Language emerges by expressing these stable structures in conceptual form. Words like “red,” “round,” or “larger than” articulate relationships that originate in lived experience.
Large language models learn statistical relationships within this linguistic world. They capture how concepts tend to relate across enormous bodies of human writing. This explains why AI systems can produce coherent responses across many domains. But it also explains why they hallucinate. Humans remain answerable to the world: experience continually corrects our expectations and beliefs. AI systems, by contrast, extend patterns within text itself. They can continue a line of reasoning with remarkable fluency, but they lack the lived engagement with the world that anchors meaning and truth.
This framework helps explain several recurring challenges in AI research. One is the “compositionality gap”—the tendency for language models to perform well on familiar reasoning patterns while failing when asked to combine concepts in genuinely novel ways. Research increasingly shows that larger models improve fluency and factual recall much faster than they improve true compositional reasoning. From our perspective, this is not simply an engineering limitation but a structural boundary: AI systems can extend patterns already sedimented in language, but they do not possess the world-directed understanding that allows humans to generate genuinely new conceptual relations.
A similar pattern appears in multimodal systems that combine language and vision. These systems can often label images correctly while still failing at strong reasoning about objects and their parts. They learn correlations between visual patterns and language rather than perceiving stable objects unfolding through time in the way humans do. The result is systems that can appear impressively fluent while remaining surprisingly brittle outside familiar patterns.
This perspective also reframes debates about AI safety. Public discussion often swings between fears of “rogue superintelligence” and claims that AI poses little meaningful risk. Our research suggests that both extremes misunderstand the nature of current systems. The most immediate risks arise not because AI possesses human-like intentions, but because it can extend patterns of reasoning without reflective responsibility to the world. Systems can generate plausible-sounding misinformation or perpetuate harmful biases simply by following learned patterns, without any underlying understanding of the implications. The focus shifts from speculative existential threats to the immediate engineering and governance challenges of building AI systems that are accountable to the world they interact with. Understanding AI as an extension of human intelligence—not a replacement for it—offers a more grounded path for building trustworthy AI systems. The engineering and governance mechanisms must be designed to account for this extension, ensuring that AI’s powerful pattern-extending capabilities serve human values and goals, rather than undermining them through unmoored fluency.
