Extending Human Intelligence Through AI

Extending Human Intelligence Through AI

At a glance

• Modern AI systems are powerful not because they replicate human intelligence, but because they build upon it, extending structures already present in human cognition and language.
• This perspective helps explain both AI's remarkable capabilities and its recurring limitations, including hallucinations and reasoning breakdowns.
• This research argues that AI safety is a system-level challenge, shifting focus from "rogue AI" narratives to leveraging engineering and governance.
• Viewing AI as an extension of human intelligence—rather than a replacement—offers a more grounded approach to developing trustworthy AI systems.

Today's AI systems can write essays, generate code, summarize complex ideas, and engage in conversations with remarkable fluency. However, they still struggle with tasks that humans find intuitive, such as reliably tracking objects through changes, reasoning compositionally in unfamiliar situations, or distinguishing truth from plausible fiction. These contradictions have fueled polarized debates about AI. Some view current systems as early forms of human-like intelligence, while others dismiss them as advanced autocomplete tools.

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), presents a different perspective. Instead of questioning whether AI systems are becoming intelligent in a human sense, these approaches pose a more fundamental question: What if AI systems function _because_ they rely on structures rooted in human cognition? This shift in perspective, drawing on the phenomenology of Edmund Husserl, helps clarify both the capabilities and the limitations 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 originating in human cognition itself. Building further on Husserl's phenomenology, the paper suggests that language already contains embedded 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 merely the passive reception of sensory data. We experience the world as stable entities evolving through change: a cup remains the same cup as we move around it; a melody remains recognizable despite individual notes changing. Language emerges by expressing these stable structures conceptually. Words like "red," "round," or "larger than" articulate relationships rooted in lived experience.

Large language models learn statistical relationships within this linguistic framework. They capture how concepts tend to relate across vast bodies of human writing. This explains why AI systems can produce coherent responses across various domains. However, it also explains why they sometimes hallucinate. Humans are accountable to the world: experience continually corrects our expectations and beliefs. In contrast, AI systems extend patterns within text itself. They can continue a line of reasoning with remarkable fluency but lack the lived engagement with the world that anchors meaning and truth.

AI Extends Human Cognition

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 but fail when combining concepts in genuinely novel ways. Research increasingly shows that larger models improve fluency and factual recall much faster than true compositional reasoning. From our perspective, this is not just an engineering limitation but a structural boundary: AI systems can extend patterns already embedded in language but lack the world-directed understanding that enables humans to generate genuinely new conceptual relations.

A similar pattern appears in multimodal systems that integrate language and vision. These systems can often label images correctly while still failing at robust reasoning about objects and their parts. They learn correlations between visual patterns and language rather than perceiving stable objects evolving over time as humans do. The result is systems that can seem impressively fluent while remaining surprisingly fragile 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 persuasive but ungrounded outputs, automate flawed decisions at scale, or execute harmful actions if embedded in poorly governed environments.

This helps explain why AI safety is increasingly shifting from model safety to system safety. In practice, organizations already rely on layered safeguards—what the industry increasingly calls “harnesses”—to constrain, validate, and monitor AI behavior. Rather than temporary patches, our paper argues that these mechanisms reflect something fundamental about AI architecture itself: trustworthy behavior emerges from the work of builders of AI systems responsible for their behavior, a responsibility that cannot be delegated to or shared with models.

This interpretation aligns closely with how enterprises increasingly approach trustworthy AI deployment. Organizations need systems that can extend human intelligence while remaining governable, auditable, and aligned with human oversight. Understanding AI as a derived form of intelligence clarifies why layered governance, evaluation, and operational controls matter so deeply.

PODCAST SERIES

AI Testing and Evaluation: Learnings from Science and Industry

Discover how Microsoft is learning from other domains to advance evaluation and testing as a pillar of AI governance.

Looking ahead, we believe phenomenology offers more than a critique of AI—it offers a framework for understanding its promise. AI systems reveal something profound about human cognition itself: that meaning can be formalized, extended, and scaled in powerful new ways. The central societal risk of AI thus turns out to be kicking away the ladder of its origins in human experience and cognition—misinterpreting AI as a rival intelligence that diminishes our humanity and thus, in turn, diminishing the true promise of AI itself.

The question, then, is not whether AI will replace human intelligence. It is how we can responsibly build systems that extend human understanding while remaining grounded in the world from which that understanding arises. If we mistake AI systems for autonomous minds, we risk over-trusting them. If we dismiss them as trivial tricks, we risk overlooking one of the most important technological developments of our time. A more grounded interpretation recognizes both truths at once: AI is a genuine extension of human intelligence—and precisely because of that, humans remain responsible for how it is understood, governed, and used.