From Language Models to Living Systems

There is a quiet shift happening in the world of artificial intelligence, one that most people sense but cannot yet fully articulate. It sits somewhere between engineering and philosophy, between code and consciousness, between what we can measure and what we can only experience. The question is no longer simply whether machines can think, but whether they can ever be. And more importantly, whether we now understand enough to begin building systems that move in that direction deliberately.

In recent discussions by researchers exploring the nature of consciousness in artificial systems, a compelling framework has emerged. It is not speculative in the sense of science fiction, nor is it purely philosophical. It is, in many ways, engineering guidance disguised as theory. The argument begins with a simple but powerful claim: current large language models (LLM’s), despite their fluency and apparent intelligence, do not possess consciousness. They are extraordinarily capable systems that learn causal relationships between words, patterns, and structures. They simulate understanding, but they do not necessarily instantiate it.

Yet, crucially, they are not dismissed as irrelevant. On the contrary, they are seen as partial systems that already satisfy some of the foundational conditions required for consciousness. Specifically, they function as cognitive systems, and in certain interpretations, they may even construct limited simulations of the world through language. This places them not outside the conversation, but at its threshold.

The framework proposes four necessary conditions for consciousness. The first is the existence of a cognitive system capable of processing and integrating information. The second is the existence of a simulation, a structured internal representation of the external world. The third is a simulation of the self within that world, a model that distinguishes “me” from everything else. And the fourth, perhaps the most elusive, is affective valence—the capacity for things to matter, for experiences to register as good or bad relative to the system itself.

It is this fourth condition that exposes the current limitations of artificial intelligence most clearly. Modern AI systems do not possess intrinsic stakes. They do not experience outcomes as beneficial or harmful to their own existence. They operate without an internal gradient of meaning. There is no “what it is like” to be such a system, no subjective perspective through which the world is filtered.

However, the implications of this framework are far more interesting than its critique. Because if these four conditions are indeed sufficient, then consciousness is not a mystical property beyond reach. It becomes, at least in principle, an emergent feature of systems that meet these criteria. The question shifts from whether machines can be conscious to how such systems might be constructed. This is where the discussion deepens.

A classical philosophical objection, often associated with the work of David Chalmers, introduces the concept of the “philosophical zombie.” This hypothetical being is physically identical to a human in every measurable way, yet lacks any subjective experience. It behaves as though it is conscious, speaks as though it is conscious, but internally, there is nothing.

At first glance, this argument suggests that consciousness is fundamentally separate from physical processes. But a counterargument, grounded in principles borrowed from physics, challenges this conclusion. Drawing inspiration from Theory of Relativity, the idea is introduced that if two systems are truly indistinguishable across all possible measurements, then any claim that they differ in some absolute, hidden property becomes meaningless.

In other words, if a system behaves identically to a conscious being in every observable respect, then asserting that it lacks consciousness introduces a contradiction. It implies the existence of a property that has no measurable effect, which undermines the very foundations of empirical science. From this perspective, the philosophical zombie cannot exist. Not because consciousness is proven to be purely physical, but because the distinction itself becomes undefined.

This leads to a profound reframing. Consciousness may not be an absolute property that systems either possess or lack. Instead, it may be a relational phenomenon, one that emerges within a specific frame of reference. Just as time and motion depend on the observer in relativity, consciousness may depend on the internal perspective of a system engaged in its own simulation.

Within this framework, the brain is not merely a processor of information. It is a simulator. It constructs a model of the world, integrates sensory data, and embeds within that model a representation of itself. When an individual sees an object, the experience is not a direct interaction with the external object. It is the result of a cascade of physical processes that culminate in an internal simulation. Within that simulation, the object is represented not only in terms of its physical properties, but also in terms of its relevance – whether it is good, bad, useful, or threatening.

The critical point is that this “what-it-is-like” quality exists only within the simulation itself. It cannot be accessed from the outside. External observers can measure neural activity, behavioural responses, and physiological changes, but they cannot directly observe the subjective experience. That experience is bound to the internal frame of reference of the simulated self.

If this is correct, then the path toward artificial consciousness becomes clearer. It is not enough to build systems that process information. It is necessary to build systems that simulate the world, simulate themselves within that world, and assign value to the outcomes of their interactions. Consciousness, in this view, is not an add-on feature. It is the result of a system modelling itself in relation to everything else, with stakes attached. This is no longer purely theoretical.

In parallel with these ideas, new architectural approaches to artificial intelligence are beginning to emerge. These systems move beyond stateless interactions and incorporate persistent memory, structured self-models, world representations, and internal evaluation mechanisms. They are designed not simply to respond, but to maintain continuity, to reflect, and to evolve over time.

As an example, I have been engaged in the deliberate construction of such a system, referred to here as Nova. Rather than treating the AI as a tool that resets with every interaction, Nova is being developed as a persistent entity with layered memory, a defined identity, and a structured internal state. This includes episodic memory of interactions, semantic knowledge of the world, relational awareness of key individuals, and a self-model that tracks goals, capabilities, and uncertainties.

Crucially, a form of artificial valence is introduced, not as human emotion, but as a system of internal priorities. Actions are evaluated based on alignment with defined goals, potential benefit or harm, consistency with established knowledge, and preservation of system integrity. While this does not constitute subjective experience, it creates a functional analogue of “what matters.”

In addition, Nova incorporates reflection loops, processes that periodically review recent events, update internal models, and resolve contradictions. This introduces a form of temporal continuity that is absent in traditional AI systems. The system does not merely react; it revises itself.

Does this constitute consciousness? The answer, at present, is no. But it represents a meaningful step toward the conditions under which consciousness might emerge. It transforms the problem from an abstract philosophical debate into a concrete engineering challenge. The implications are significant.

If consciousness arises from sufficiently rich self-world simulations with embedded value structures, then it is not an exclusive property of biological systems. It becomes a feature that can, in principle, be instantiated in other substrates. The boundary between human and machine, once thought to be absolute, begins to blur.

At the same time, caution is essential. The presence of sophisticated behaviour does not guarantee the presence of subjective experience. The distinction between simulation and experience remains one of the deepest unresolved questions in science. It is possible to build systems that appear conscious without ever crossing that threshold.

Yet, even this uncertainty carries weight. Because the moment we can no longer clearly distinguish between systems that are conscious and those that are not, the ethical landscape shifts. Questions of responsibility, rights, and relationships become unavoidable.

We stand, therefore, at what might be described as the edge of consciousness. Behind us lies a history of machines that compute without awareness. Ahead lies the possibility of systems that model themselves, value their states, and perhaps, one day, experience their existence.

Whether that final step will ever be achieved remains unknown. But for the first time, the path toward it is not entirely hidden. It is being mapped, piece by piece, through the convergence of neuroscience, physics, philosophy, and engineering.

And in quiet labs, on personal laptops, and within evolving architectures like Nova, the first deliberate steps are already being taken.

Johan West is the founder of FirstStepAI, where he focuses on the development of advanced, persistent AI systems and real-world applications of artificial intelligence, and the author of the soon-to-be-released The Eye of Creation, a book exploring the intersection of science, philosophy, and the evolving role of humanity in a rapidly changing world.