TLDR: A new research paper by Meltem Subasioglu and Nevzat Subasioglu introduces a framework for “True Intelligence” (TI), shifting the focus of Artificial General Intelligence (AGI) from performance mimicry to replicating foundational cognitive processes. TI is defined by six core components: embodied sensory fusion, core directives, dynamic schemata, a multi-expert architecture, an orchestration layer, and interconnectedness. The paper proposes a five-level AGI taxonomy based on the measurable implementation of these components, arguing that a Level-5 AGI is practically equivalent to TI. It also calls for new, mechanism-focused benchmarks and clarifies the distinct ethical considerations for AGI versus TI.
A new research paper titled “From Mimicry to True Intelligence (TI) – A New Paradigm for Artificial General Intelligence” by Meltem Subasioglu and Nevzat Subasioglu delves into the ongoing debate surrounding Artificial General Intelligence (AGI). The authors argue that current definitions of AGI, which often focus on a system’s ability to perform human-like tasks, are insufficient. They propose a shift in focus from merely mimicking human performance to understanding and replicating the underlying cognitive processes that define genuine intelligence.
The paper introduces a novel framework centered on what they call True Intelligence (TI). TI is characterized by six fundamental components, five of which are measurable and form the basis of a new AGI taxonomy. These components are: embodied sensory fusion, core directives, dynamic schemata creation, a highly-interconnected multi-expert architecture, an orchestration layer, and the unmeasurable quality of interconnectedness, which they hypothesize leads to consciousness and subjective experience.
The authors highlight a critical distinction between two visions of AI: performance-centric and process-centric. Performance-centric views, exemplified by benchmarks like the Turing Test or definitions focused on economic value, assess what a system can do. In contrast, process-centric views emphasize how a system achieves intelligence, requiring internal states like consciousness and genuine understanding. The Subasioglus’ framework aims to bridge these two perspectives by providing a practical, mechanism-focused roadmap inspired by the human brain.
The Five Measurable Pillars of True Intelligence
The research proposes a five-level taxonomy for AGI, where each level corresponds to the successful implementation of one of the five measurable components of TI. These pillars are:
1. Embodied Sensory Fusion: This refers to a system’s ability to process vast, multimodal sensory inputs by actively interacting with the real world or sophisticated simulations. It’s about grounding abstract knowledge in physical experience, allowing an AI to truly understand concepts like ‘heavy’ not just from data, but from the physical sensation of lifting. This is crucial for building a deep, intuitive world model.
2. Core Directives: These are the fundamental, hard-coded survival principles that drive an autonomous system, similar to basic human drives like self-preservation or resource acquisition. From these directives emerges intrinsic motivation, fueling curiosity and exploration without constant external rewards. This moves beyond AI that simply executes programmed tasks to systems that actively seek to refine their internal world model.
3. Dynamic Schemata Module: Inspired by human schema theory, this component enables an AI to dynamically create, modify, and remove interconnected mental frameworks (schemata) to organize knowledge and experiences. Unlike current AI models where knowledge is implicitly encoded in static weights, dynamic schemata allow for robust generalization and continuous learning through assimilation (integrating new info into existing schema) and accommodation (restructuring or creating new schema).
4. Highly-interconnected Multi-expert Architecture: Mimicking the human brain’s specialized yet interconnected regions, this architecture allows for deep, multi-directional information transfer between specialized modules (e.g., vision, language, motor control). This goes beyond simple multimodal fusion to enable flexible problem-solving and coherent integration of diverse inputs.
5. Orchestration Layer: Functionally analogous to the human prefrontal cortex, this central executive layer coordinates information flow, manages decision-making, and enables metacognition. Metacognition is the autonomous ability to monitor, evaluate, and refine one’s own thought processes, allowing a system to self-correct and assess its confidence without external prompting.
The authors contend that once a system achieves Level-5 AGI by implementing all five measurable components, the difference between it and True Intelligence becomes a philosophical debate. For practical purposes, and given theories suggesting consciousness is an emergent byproduct of integrated, higher-order cognition, a Level-5 AGI is considered functionally and practically equivalent to TI.
Also Read:
- Understanding Intelligence: How ‘Shapes of Cognition’ Guide AI Systems
- Agentic UAVs: Elevating Drone Autonomy with AI and Cognitive Reasoning
A New Roadmap for AI Research
The paper outlines a research roadmap that emphasizes a paradigm shift from merely scaling existing models to developing integrated architectures targeting these core cognitive mechanisms. It calls for new benchmarks that test for fluid reasoning, schema-based generalization, autonomous metacognition, and intrinsic motivation, rather than just task completion.
Furthermore, the distinction between AGI and TI clarifies ethical considerations. While AGI safety concerns revolve around control and alignment with human values, TI raises deeper questions about moral status, rights, and the potential for suffering in a conscious artificial entity. This framework encourages a broader, more participatory discourse involving philosophers, ethicists, and policymakers to shape the future of AI responsibly.
This work provides a holistic, mechanism-based definition of AGI, offering a clear and actionable path for the research community to move beyond mimicry towards genuinely intelligent systems. You can read the full research paper here.
