The Thinkers Mapping Human Intelligence in the Age of AI

4. The Broader Intellectual Landscape

Shannon Vallor — The Virtue Ethicist

Institution: University of Edinburgh, Centre for Technomoral Futures.
Key work: The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking (Oxford University Press, 2024).

Vallor is working in the tradition of virtue ethics and philosophy of technology, and her contribution is the most explicitly human-centered of this group. She invokes the Spanish philosopher Ortega y Gasset's argument that humans are fundamentally 'creatures of autofabrication' — future-oriented beings who must choose to make and remake themselves. AI, by contrast, is architecturally backward-facing: its responses are extrapolations from historical data, pointing always to where we have already been, never where we might go.

Her central metaphor — AI as mirror — argues that AI does not merely reflect human thought but recursively structures it. The mirror is distorting, not neutral. When we interact with systems trained on the aggregate of human output, we see ourselves reflected back, but the reflection is shaped by whoever controlled the training data, optimized for engagement rather than truth, and calibrated to the center of a distribution rather than to any individual's genuine situation.

Her most original contribution is the 'bootstrapping problem': the paradox of trying to cultivate moral and intellectual progress using AI systems that are inherently backward-looking, because AI learns from historical data and reinforces existing norms rather than fostering ethical growth. This creates a recursive trap — AI reflects back what we have already thought, and our capacity for genuinely new moral and intellectual vision atrophies.

In relation to Quattrociocchi and Floridi: Vallor is more explicitly normative and more concerned with virtue and character. Where Quattrociocchi asks what LLMs do to knowledge and Floridi asks what LLMs are, Vallor asks what LLMs do to us — to our moral agency, our capacity for genuine judgment, our ability to become different and better than we currently are. She would find both admirably rigorous and would argue their frameworks, while accurate, understate the character-level stakes for the humans who use them.

Gary Marcus — The Structural Critic

Institutional position: Cognitive scientist; author of Rebooting AI (with Ernest Davis, 2019); founder, Marcus on AI Substack.
Key claims: LLMs lack symbolic world models; no amount of scaling will produce genuine reasoning from pattern-matching alone.

Marcus occupies a unique position as the field's most persistent and technically credible internal critic. He began critiquing neural network limitations in 1992 and has been consistently right about the structural limits of pure statistical learning. His central thesis is architectural: LLMs lack the rule-governed, symbolic world models that underpin genuine reasoning, and the gap is not a matter of size or training data.

His chess example is clarifying: an LLM can repeat the rules of chess perfectly and answer questions about them. But then it will have a queen jump over a knight illegally when it actually plays. It has not built an internal model of chess — it has learned the linguistic patterns surrounding chess without building a representation of what chess actually is.

In relation to Quattrociocchi and Floridi: Marcus is more technically focused and more combative in style, but his core claim maps closely onto Quattrociocchi's causal fault line — the divergence between human causal reasoning and LLM reliance on correlation. He would agree with Floridi that the architectural problem will not be resolved by scaling. His prescription — hybrid neurosymbolic systems that combine statistical learning with symbolic reasoning — is more concrete than either Quattrociocchi or Floridi offers, though less developed as a philosophy of what AI should be in relation to human life.

Karl Friston — The Theoretical Neuroscientist

Institution: University College London.
Key framework: The free energy principle — all biological cognition minimizes surprise by maintaining a generative model of the world.

Friston is the most technically difficult thinker in this group and arguably the most theoretically original. His free energy principle argues that all biological cognition — from the simplest organism to the human brain — is fundamentally about minimizing the divergence between predictions and sensory input. He and Marcus agree that LLMs are 'just a mapping between content and content' with nothing in the middle — no understanding, no representation of cause and effect, no capacity to reason about the consequences of actions.

But Friston goes further. He argues that the critical missing element is the capacity to encode and quantify uncertainty — and that without the ability to represent its own uncertainty about the world, a system cannot evaluate the quality of its own model. This connects directly to Quattrociocchi's metacognitive fault line — but gives it a deeper theoretical grounding. For Friston, the absence of genuine uncertainty modeling is not a technical gap to be filled; it is a consequence of a fundamentally different objective function.

His vision for AI is of 'distributed ecosystems of intelligence from first principles' — systems that enhance rather than replace human agency, oriented toward jointly minimizing collective uncertainty through genuine understanding. This connects most directly to the vision of individual humans as irreplaceable knowledge nodes — a system where grounded human expertise is permanently in the loop, not as a quality-control step but as a primary architectural component.

In relation to Quattrociocchi and Floridi: Friston provides the deepest theoretical foundations for why the metacognitive fault line Quattrociocchi identifies is not a correctable flaw but an architectural consequence of how these systems are built. He also offers the most ambitious positive vision — though one that remains more theoretical than implemented.

Shoshana Zuboff — The Political Economist

Institution: Harvard Business School emerita.
Key work: The Age of Surveillance Capitalism (2019); ongoing commentary through 2025.

Zuboff is working at a different level from the others — less concerned with what LLMs are epistemologically and more concerned with the political economy that produces and deploys them. Her central concept, surveillance capitalism, describes a system in which human behavioral data is extracted, processed, and sold as a raw material for predicting and modifying human behavior at scale.

She has repeatedly emphasized throughout 2025 that the dominant narrative of an imminent superintelligent takeover serves a crucial ideological function: it diverts public and regulatory attention from the concrete concentration of economic and computational power already reshaping global society. The AI investment bubble — she has noted — is less a technology bubble than a surveillance bubble, with stratospheric valuations reflecting the anticipated value of exclusive access to behavioral data.

In relation to Quattrociocchi and Floridi: Zuboff is the political economist in a conversation dominated by philosophers and cognitive scientists. She would find Quattrociocchi's epistemological framing important but incomplete — the question is not only what epistemia does to individual judgment, but what economic interests profit from producing and sustaining it. She would find Floridi's governance work the most aligned with her concerns, and would push all of these thinkers to ask who benefits from the confusion between AI fluency and human understanding, and how that confusion is being monetized.