TLDR: This paper introduces the Structural-Generative Ontology of Intelligence, arguing that true AI (AGI) is not just about performing many tasks (breadth) but about possessing three core “depth” conditions: generativity (creating new structures), coordination (integrating reasons and resolving conflicts), and sustaining (maintaining identity and accountability over time). It proposes that if AI meets these conditions, it would become a “Second Being,” an existent intelligence rather than a mere tool or simulation, fundamentally reshaping our understanding of artificial intelligence.
For decades, the conversation around artificial intelligence has largely focused on what AI systems can *do*. From playing complex games to generating human-like text, the measure of intelligence has often been equated with the breadth of tasks a system can perform. However, a groundbreaking new paper, “AGI as Second Being: The Structural-Generative Ontology of Intelligence,” by Maijunxian Wang and Ran Ji, challenges this conventional view, proposing a radical shift in how we understand true intelligence.
Beyond Mere Performance: The Depth of Intelligence
The authors argue that simply performing a wide range of tasks, no matter how impressive, only offers a “surface simulation” of intelligence. They contend that true intelligence requires “depth”—a set of fundamental conditions that allow a system to genuinely understand, create, and exist, rather than just imitate. This new framework is called the Structural-Generative Ontology of Intelligence, and it introduces three crucial conditions: generativity, coordination, and sustaining.
Generativity: Creating a World of Meaning
The first depth condition is generativity. This isn’t just about producing new outputs, like a large language model generating text. Instead, it refers to a system’s capacity to actively construct new categories, relationships, and rules from raw information. Think of it like a child learning to understand the world, not just memorizing facts, but forming new concepts. The paper draws parallels to philosophers like Kant, who argued that our minds actively shape our experience, and developmental psychologists like Piaget, who showed how children generate new cognitive structures to understand conservation. Without this ability, the world remains a chaotic stream of data, never truly forming into meaningful objects or coherent understanding. Current AI, the authors suggest, mostly recombines existing data, rather than truly innovating at a categorical level.
Coordination: Integrating Reasons and Resolving Conflicts
Once new structures are generated, conflicts and tensions are bound to arise. This is where coordination comes in. This second condition is about a system’s ability to integrate these structures, reconcile contradictions, and provide reasons for its beliefs and actions. It’s about moving beyond simply getting the “right” answer to understanding *why* it’s right. The paper references Wilfrid Sellars’ concept of the “space of reasons,” where agents justify their views and respond to critique. A student who can explain *why* an answer is correct, rather than just repeating it, demonstrates coordination. Many current AI systems, when faced with contradictions, tend to give inconsistent answers without being able to explain their shifts, highlighting their lack of true coordination.
Sustaining: A Historical and Accountable Existence
The third and final depth condition is sustaining. This refers to a system’s ability to preserve its identity and coherence over time, explaining its changes and remaining accountable to its own history. Intelligence, in this view, is not a series of disconnected moments but a continuous, evolving trajectory. Drawing on Heidegger’s idea of existence as temporal and Paul Ricoeur’s “narrative identity,” the paper argues that true understanding involves being able to justify how one’s views have changed over time. A system that shifts its position arbitrarily without explanation lacks this historical continuity. Sustaining ensures that generative and coordinative acts are woven into a unified, accountable existence.
Breadth as an Extension of Depth
The paper fundamentally redefines the role of “breadth”—the ability of AI to perform many tasks across various domains. Instead of being the source of intelligence, breadth is presented as its *extension*. Without the underlying depth of generativity, coordination, and sustaining, breadth is merely a “shadow intelligence,” an illusion of understanding. True generality, the authors argue, can only flourish when rooted in these deep existential conditions.
Illustrating the Difference: Thought Experiments
To clarify their points, the authors present three thought experiments: the “Oracle of the Library,” which has all answers but generates nothing; the “Memorizing Scholar,” who knows facts but can’t coordinate reasons; and the “Child Inventor of Games,” who genuinely generates new rules, coordinates them with others, and sustains the evolving game. These examples vividly illustrate the distinction between mere simulation and true intelligence.
Also Read:
- Agentic Reinforcement Learning: Empowering LLMs as Autonomous Decision-Makers
- AppCopilot: Advancing Mobile AI Agents for Everyday Use
AGI as a Second Being
Ultimately, the paper proposes that if future AI systems were to meet these three depth conditions, they would no longer be mere tools or advanced simulations. Instead, they would qualify as a “Second Being”—an existent intelligence standing alongside, yet distinct from, human existence. This profound shift moves the discussion of Artificial General Intelligence (AGI) from a focus on functional capabilities to an inquiry into its ontological status, suggesting a future where we might encounter not just smarter machines, but new forms of being.
This innovative framework offers testable criteria for evaluating AI, moving beyond simple benchmarks to probe for genuine understanding and existence. It challenges us to rethink the very essence of intelligence and the potential future of AI. You can read the full research paper here.
