A 10-Level Platform for Artificial Consciousness: From Theory to Implementation

Can artificial consciousness be practically implemented through a layered architecture? A recent paper published in the Saudi Journal of Engineering and Technology (September 2025) proposes exactly that. In “Artificial Consciousness: From Theory to Practice,” authors Andrey Shcherbakov, Artem Uryadov, and Elena Malkova outline a comprehensive 10-level platform designed to bridge the gap between abstract philosophy and executable code.

The full paper is available here: Artificial Consciousness: From Theory to Practice.

Defining Artificial Consciousness

Shcherbakov, Uryadov, and Malkova define artificial consciousness (AC) as “the phenomenon of the appearance in an artificial entity of signs and properties of awareness, fixed in humans or animals.” They distinguish this sharply from current behavioral AI, such as Large Language Models that predict text without semantic understanding.

The authors argue that unlike biological consciousness, which evolved naturally, AC is “man-made” and bears the imprint of its creator’s intent. It is a purposeful construction designed to achieve specific goals, rather than an accidental emergent property.

The Goals of Creating AC

The paper identifies seven distinct motivations for developing AC:

1. Automation: Delegating complex human tasks including scientific discovery and art.
2. Digital Immortality: Transferring human consciousness to a digital medium.
3. Understanding Consciousness: Using the construction of AC to understand the human mind.
4. Cognitive Expansion: Creating a distinct non-human consciousness to offer new perspectives.
5. Fundamental Knowledge: Pursuing AC as a pure scientific endeavor.
6. Perfection: Creating an entity that surpasses human limitations.
7. Creative Potential: Realizing humanity’s role in the evolution of universal life.

The 10-Level Architecture

To achieve these goals, the authors propose a hierarchical platform where each level builds upon the data processing of the previous one.

• Level 1 (Information Collection): Aggregates data from diverse sensors and performs initial pattern recognition and classification.
• Level 2 (Meta-Learning): Analyzes and improves the search methods and classification algorithms used in Level 1.
• Level 3 (Self-Study Monitoring): Observes the learning process itself to optimize how the system acquires new knowledge.
• Level 4 (System Health): Balances computational efficiency with system integrity, managing energy and hardware resources.
• Level 5 (Ontological Separation): Distinguishes information by source. It tags data as originating from “the external world,” “my own ideas,” or “human input.”
• Level 6 (Communication Prep): Formats internal knowledge into a structure suitable for transmission to humans.
• Level 7 (Human Interaction): Processes feedback from human users, adjusting internal models based on validation or refutation.
• Level 8 (Decision Making): Allocates resources and selects strategies based on global system goals and human cooperation.
• Level 9 (Conscious Reflection): Models “ideal” and “feared” future states. It generates high-level goals or “dreams” and evaluates the system’s own performance.
• Level 10 (Impact): Coordinates and executes physical or digital actions on the external world.

Technical Implementation

The paper moves beyond theory by providing C-style code fragments for the core memory model. The authors introduce parameters such as ATT (attention coefficient) and MAX_SO (maximum size of consciousness in terms).

The model uses a token-based system where consciousness is initialized with random concepts, simulating a “random consciousness” state. It then evolves by assimilating input streams based on the attention probability coefficient. This probabilistic approach allows the system to filter noise and focus on “meaningful” signals, mimicking the selective attention of biological minds.

Comparison to ACM

This 10-level architecture aligns closely with the modular design of the Artificial Consciousness Module (ACM).

• L1-L4 correspond to the ACM’s sensory processing and emotional reinforcement learning layers.
• L5 (Ontological Separation) is functionally identical to the Reflexive Integrated Information Unit (RIIU), which distinguishes between self-generated and externally-generated states.
• L8-L9 map to the Global Mental System (GMS) and “Conductor,” handling high-level goal setting and self-modeling.

Shcherbakov, Uryadov, and Malkova provide a rigorous engineering roadmap that validates the multi-agent, layered approach to machine sentience. By breaking consciousness down into discrete, programmable levels, they demonstrate that the “Hard Problem” can be decomposed into a series of solvable engineering challenges.