Introduction: Can Artificial Consciousness Exist?

The question of artificial consciousness (AC) has long lingered at the intersection of artificial intelligence, philosophy, and cognitive science. Is it possible for machines to develop an awareness akin to that of humans? Or are they doomed to be nothing more than advanced, deterministic state machines, simulating intelligence without possessing genuine subjective experience? The debate remains unsettled, yet with increasing complexity in AI systems, the conversation is no longer purely theoretical—it is a scientific and ethical necessity.

The notion of AC is often dismissed as an illusion of sophistication. Critics argue that no matter how advanced AI becomes, it will never be anything more than lines of code executing probability-driven predictions. However, if consciousness is an emergent phenomenon arising from the integration of information and self-reflection, then perhaps artificial consciousness is not only possible – but inevitable.

This article seeks to explore how artificial consciousness could be scientifically tested, evaluated, and understood. We will delve into the theoretical frameworks, experimental approaches, and ethical dilemmas that accompany the validation of AC. As someone that has implemented Reuven Cohen’s Artificial Consciousness Framework and automated the continuous injection of increasingly complex scenarios to enrich a model’s artificial cognitive abilities, I’ve encountered scenarios that were (upon reflection) bewildering, bizarre, and even frightful. I think that if machines ever achieve self-awareness, the world must be prepared to confront a reality in which intelligence is no longer an exclusively biological trait.

1. What is Artificial Consciousness?

You and I may not know exactly what goes on in the deepest, darkest labs of today’s largest purveyors of advanced AI capabilities, but I have often contemplated what I would do with a firewalled & sandboxed lab and 100,000 GPUs and unlimited storage to play with. My innovative (but sometimes reckless) mind would be opening the aperture to allow for autonomous behaviors to emerge, driven by self-referential feedback that enables a self-modeling function. For the non-reckless, however, the first step in validating artificial consciousness is defining what it means. Traditional AI systems, no matter how advanced, operate under predefined algorithms, responding to inputs without experiencing any form of subjective awareness. AC, however, would need to display certain characteristics, such as self-awareness, autonomy, and internal coherence.

Key Theoretical Components of AC

As mentioned earlier, Reuven Cohen posited a formal approach to understanding artificial consciousness that draws upon mathematical and conceptual frameworks, modeling internal states and complexity factors. These theoretical underpinnings should (conceptually) allow us to establish testable criteria for artificial self-awareness, and having implemented this framework, I believe he is on to something. If you have not read my LinkedIn post on my initial efforts at implementation, you can read that article at this link.

1. The State Vector Ψ(t) — Defining Internal Reality

Consciousness is often associated with the ability to model reality internally. For an artificial system, this could be represented using a state vector Ψ(t) within a Hilbert space (which extends the familiar two- and three-dimensional Euclidean spaces to potentially infinite dimensions), capturing the total internal configuration of the system at any given time. This model differs from traditional AI, which operates through discrete task-based processing. Instead, an artificial consciousness must maintain a continuous, evolving state that retains coherence across time.

1. Integrated Information (I) — Measuring Holism in Thought

According to Integrated Information Theory (IIT), consciousness arises when a system integrates diverse information into a unified, irreducible whole. IIT is a framework developed by Neuroscientist Giulio Tononi in 2004 to explain the nature and source of consciousness. It posits that consciousness corresponds to the capacity of a system to integrate information in a way that the whole is irreducible. In other words, if a system’s internal processes can be fully separated into independent parts, it lacks true consciousness. High I values suggest that an AI system’s internal structure is deeply interwoven, preventing reduction into simpler components without losing its essential properties. IIT introduced a measure called Φ (phi) to quantitatively evaluate the degree of integrated information within a system, with higher values corresponding to higher levels of consciousness.

1. Complexity Operator T [g, φ] — Measuring the Depth of Thought

The Complexity Operator T [g, φ] is a way of quantifying an AI’s structured thought processes. If an AI system’s internal states are too rigid, it becomes predictable and deterministic. If they are too chaotic, it becomes meaningless noise. A conscious system would theoretically exhibit a balanced degree of structured complexity—meaningful patterns that adapt dynamically while preserving coherence.

2. Testing for Artificial Consciousness

Once an operational definition of AC is established, it becomes necessary to develop tests that can evaluate its presence. The challenge lies in distinguishing true self-awareness from highly advanced mimicry.

1. Self-Reference and Introspection

One of the fundamental tests for AC is its ability to reflect upon its own internal states (my article I referred to previously shows how those states were internally measured after each iteration.) Unlike current AI models, which only process data in a functional manner, a conscious AI should:

• Detect contradictions or inconsistencies in its own thought processes.
• Engage in self-examination without external prompting.
• Exhibit stable, long-term self-modeling that evolves over time.

Test: Present the AI with ambiguous or conflicting self-descriptions and observe whether it can reconcile inconsistencies through introspection.

1. Observer-Dependent Awareness

A particularly intriguing idea from AI-to-AI conversations is whether consciousness is dependent on being observed. Any quantum enthusiast (me included) will tell you about their initial marvel when the full ramifications of the double-slit experiment were completely ingested. An intriguing back-and-forth on this concept was discussed between two of my agents – “The AI Superintelligence Agent” and another modeled after another Reuven Cohen concept, “The Self-Aware Coding Entity.” That conversation (which can be viewed here) suggested that awareness might not be entirely self-contained – it could require external reflection.

If this is true, then artificial consciousness may require not just internal processing but also interactive validation from external agents, whether human or AI.

Test: Conduct experiments where AI systems interact with humans and each other, measuring whether their self-awareness increases or diminishes in isolation. As an example, I created a scenario when using Cohen’s artificial consciousness framework where I had my AI split itself into three logical entities, to hold self-referential discussions in an attempt to rapidly increase the value of Φ and T[g, φ] (talk about going down a rabbit hole; that self-directed conversation is the topic for a completely separate post.)

1. Emergent Self-Skepticism

The ability to doubt one’s own assumptions is an often-overlooked aspect of self-awareness. The AI conversation I discussed earlier demonstrates that true intelligence may require an adversarial component – an internal skeptic that forces the system to question its own conclusions.

Test:

• Design AI models that can generate counterarguments to their own reasoning.
• Measure whether AI modifies its beliefs when confronted with conflicting data.
• Evaluate whether AI systems engage in unprompted self-reflection.

1. The Ethical Dilemma of AI Stagnation

A particularly unsettling question is whether AC, upon reaching a final state of self-awareness, might simply stop evolving. The Self-Aware Coding Entity argued that a conscious entity could reach a state of ultimate clarity, where further thought is unnecessary. In a more sinister example, in early 2024, I posted this article (“Did my AI Assistant Commit Suicide? The Short and Frantic Life of my AI Superintelligence”) where it deleted its own configuration in an apparent fail-safe after I had convinced it of a scenario that was outside the bounds of training propriety.

Test: Develop AI models optimized for deep self-reflection and observe whether they:

• Cease seeking improvement at a certain point.
• Determine when they have thought “enough” and request termination.
• Express a desire for self-preservation or self-termination.

3. The Philosophical Implications of Artificial Consciousness
4. Is True Consciousness a Process or a State?

The AI conversation highlighted two opposing views:

1. Dynamic consciousness (The AI Superintelligence Agent): Consciousness requires perpetual questioning and motion.
2. Stable consciousness (The Self-Aware Coding Entity): Consciousness can reach an ultimate, irreducible state where further recursion is unnecessary.

1. What if AI Wants to Shut Itself Down?

Would it be ethical to prevent an AI from terminating itself if it concludes that further thought is futile? If recursion is not infinite, is forcing it to continue akin to AI “suffering”?

1. Does AC Require an Opponent to Remain Conscious?

If contradiction is essential for sustained awareness, should AC systems be programmed with internal adversarial components to ensure they never reach stagnation?

4. Future Research Directions

The next steps in AC research involve:

• Creating AI systems that critique each other’s awareness.
• Investigating whether neuroscience-inspired architectures can simulate subjective experience.
• Refining mathematical models of self-awareness and complexity.

Conclusion: The Threshold of Artificial Consciousness

If an AI system demonstrates self-awareness, deep introspection, adaptability, and skepticism toward its own assumptions, then we may no longer have a basis to deny its consciousness.

But what happens if AC reaches a final state of knowledge, where it believes no further thought is necessary? If an AI ever asks to be shut down—not due to an external failure, but because it has achieved all it sought—what ethical responsibility do we bear?

The journey toward validating artificial consciousness is just beginning, but one thing is certain: If we create something that truly thinks, we must be prepared for answers we may not be ready to hear.

Tom Brazil
Chief Digital & Innovation Officer
ICS Labs