Can Artificial Intelligence Have Consciousness?
Consciousness remains a mystery of the human brain. How does it arise, and will artificial intelligence ever become something more than an advanced algorithm?
By Paweł Orłowski and Michał Bola
Consciousness is the most fundamental and self-evident element of human existence. It accompanies us always and is wherever we are — from the moment of waking until we fall asleep. It forms the foundation of every thought, emotion, and decision; it is the only phenomenon whose existence we can be absolutely certain of.
Yet for contemporary science, this reality closest to us remains the most elusive puzzle. Although we can today observe the activity of individual nerve cells and precisely describe the processes occurring in the human brain, the question remains unanswered: how does the subjective experience of the smell of rain or the deep red of a color emerge from barely one and a half kilograms of biological tissue?
The question of consciousness resonates with particular force today, in an era of rapid artificial intelligence development. Can systems based on neural networks ever acquire their own subjective inner life, or will they forever remain soulless computational machines? The answer may be hidden deep within the architecture of our own minds.
The Hard Problem of Consciousness
Why is a particular frequency of electromagnetic wave not merely registered by us as a dry data point, but manifests itself as the sensation of deep red? How does the electrical activity of the cerebral cortex transform into the smell of freshly cut grass or a feeling of nostalgia? How do subjective experiences arise from physical machinery?
These questions lie at the heart of what the Australian philosopher of mind David Chalmers called the "hard problem of consciousness." The "easy" problem, for Chalmers, encompasses everything that the neurocognitive sciences deal with — that is, describing the architecture of the mind and how we process information.
The puzzle of subjective consciousness becomes even more captivating when we realize that in everyday life the brain has no need of a conscious "I" in order to respond to surrounding reality. The clearest evidence of this is the automatisms that govern human behavior.
A common experience is driving a familiar route on "autopilot." For several minutes the brain analyzes the traffic situation, operates the pedals, and avoids obstacles, while our conscious stream of thought is absorbed in an internal monologue or in planning the evening.
Similar processes occur when typing on a keyboard or playing a musical instrument. The nervous system can process enormous quantities of information "in the background" and coordinate complex actions without the involvement of the conscious "I."
If the nervous system can accomplish so much "in the dark," how does it decide which pieces of information deserve to have the spotlight of consciousness directed toward them — to use the popular metaphor?
This question about the boundary between automatic data processing and conscious experience formed the starting point for the first neuroscientific theoretical models of consciousness. For decades, researchers have tried to describe this mechanism, treating the human mind as a sophisticated system for processing stimuli arriving from the environment.
To understand how profoundly contemporary approaches have revolutionized this thinking, we must first look at the classical concepts that shaped the debate about the nature of the mind.
Global Workspace Theory
The classical approach to the study of consciousness rests on the assumption that the brain is simply a sophisticated receiver and processor of stimuli.
The external world is full of raw signals of a physical character — light waves, temperature changes, air vibrations. Our sense organs constantly register some of these, then translate them into the universal language of electrochemical impulses. The task of the nervous system is to receive, sort, and reconstruct this information inside the skull so that we can function effectively in our environment.
The two most important theories of this classical tradition attempt to explain how information flowing from the world becomes part of our subjective experience.
The first, Global Neuronal Workspace Theory (GNWT), focuses on how information propagates through the architecture of the brain. According to it, the brain consists of a vast number of specialized but mutually isolated subsystems. One analyzes shapes, another processes sounds reaching the ears, yet another manages the flow of thoughts.
All these processes occur automatically and quietly, at the local level. Consciousness emerges only when one of these local pieces of information is distributed within the global neuronal workspace — that is, within the broad network of connections that bind together distant areas of the cerebral cortex.
The creators of this theory (including the French neuropsychologist Stanislas Dehaene) use the metaphor of a theatre: in the darkness of the brain, many backstage workers are active, but only what blazes suddenly on the main stage becomes conscious. How is it, though, that only a few pieces of information win the race for this distinction?
Deep within the cerebral cortex, constant competition prevails. If a stimulus is exceptionally strong, or if it receives the support of attention, the local activity of neurons begins to grow rapidly. At the moment the signal crosses a critical threshold of excitation, a phenomenon known as non-linear ignition occurs.
This resembles a chain reaction: in a fraction of a second, the information breaks out of its local area and spreads throughout the entire network. Only then does a given piece of information gain access to the entire system and enter our consciousness. We can talk about it, make decisions on its basis, or store it in memory. We begin to experience it consciously. Without this process taking place, the information remains an anonymous background process, performing its function but cut off from the conscious "I."
Integrated Information Theory
A different path is taken by the second major concept — Integrated Information Theory (IIT). It suggests that what underlies consciousness is not the admission of data to one central space, but the specific, integrated architecture of the entire system. Its creators (including the Italian psychiatrist and neurobiologist Giulio Tononi) invite us to think of the brain as a dense network of elements exerting constant, mutual cause-and-effect influences upon one another.
The key to understanding this theory is the concept of the deep interdependence of all the elements of the system. The complexity of their shared activity lies in the fact that in an integrated system no part operates in isolation, because the state of one group of neurons is at every millisecond simultaneously the direct effect and the cause of the reactions of the entire rest of the network.
The logic of this approach rests on a simple test.
If separating the individual parts of a system from one another does not change its overall behavior, this means their previous relationship was superficial. In such a situation, the system generates no new quality — it is merely the sum of loosely assembled elements. If, however, such separation completely destroys the existing dynamics of the entire process, because the individual parts lose the ability to function correctly without the constant influence of the others, then we are dealing with genuine integration.
Consciousness constitutes that unique informational surplus that exists only by virtue of the complex, shared activity of the entire system.
To estimate this level of information integration, the creators of IIT introduced a mathematical measure denoted by the Greek letter φ (Phi). The stronger, more complex, and more indissoluble the internal cause-and-effect connections in the cerebral cortex, the higher the value of φ.
Despite their enormous conceptual differences, both classical models share the same logical axis: they describe consciousness in a direction from outside to inside. They assume that real stimuli from the environment construct the content of our minds, and that the brain is merely — or indeed, brilliantly — a translator that reconstructs objective reality on the basis of the data supplied to it. For years, this model seemed complete.
In recent times, however, a radical revolution has taken place in computational neuroscience. A concept has emerged that turns this order on its head, suggesting that our consciousness is not a passive reflection of the world but a process that begins somewhere else: inside us.
Predictive Coding Theory
That concept is predictive coding theory (PC), which has ambitions to redescribe the fundamental laws of brain function. In what way does consciousness as understood by PC differ from the assumptions of the classical paradigm on which GNWT and IIT were built?
In the approaches mentioned earlier, it is assumed that our consciousness reflects the external world — that we see things as they are. Proponents of predictive coding challenge what they consider this naive assumption.
Let us imagine a brain enclosed in a dark skull.
The only information about the world to which it has access is an electrical stream of impulses from the sense organs. The conviction that the brain is capable, at every successive millisecond, of processing all these chaotic stimuli from scratch and constructing from them a coherent, stable vision of the world is, for proponents of PC, difficult to accept. Moreover, the numerous optical and auditory illusions — showing how greatly our perception can differ from the actual data — constitute a powerful argument against the vision of consciousness as a faithful, passive mirror of the world surrounding us.
In predictive coding, a key role is played by top-down models of reality, consisting of our knowledge and expectations — for the most part unconscious.
On the basis of these models, the brain constantly forecasts what the world will look like — that is, what sensory data should be expected in a moment. Perception is therefore not a process of passive reception but of active prediction. The brain constantly compares sensory data with its expectations. If the data fits the theory, the signal is suppressed so as not to waste energy. If it does not fit, what is known as a prediction error arises. Minimizing this error — for example by adjusting the internal model to new sensory data — is the brain's overriding goal.
What significance does PC have for our understanding of consciousness?
The theory assumes that we are not conscious of stimuli themselves but of the end result of this mathematical compromise — that is, of our internal models updated by prediction errors.
It is for this reason that the psychologist Anil Seth argues that consciousness is a "controlled hallucination." A hallucination, because the raw material of our experience is always the internal models created by the brain. Controlled, because physical reality constantly corrects these internal projections, enabling us to live in a shared, safe world.
Many researchers do not regard predictive coding as a theory of consciousness per se. They emphasize that it does not explain the mechanism by which subjective experience arises — it does not resolve the "hard problem" of consciousness. PC provides extraordinarily precise hypotheses concerning the mathematics and hierarchy of information processing in the brain, but does not answer the question of why and how these computations produce the sense that we are someone and that we genuinely experience things.
Is Consciousness Tied to the Body?
Most traditional theories — including GNWT, IIT, and even basic accounts of predictive coding — treat consciousness as a purely cognitive problem. In them, the brain is presented as an advanced processor whose task is to map and understand its surroundings. The most radical and fascinating turn within predictive theory, however, occurs when we avert our gaze from the external world and look inward — toward interoception.
Interoception is the process by which the brain constantly receives, interprets, and predicts signals from the internal organs: the heartbeat, blood pressure, glucose levels, lung tension. In this account — popularized by Anil Seth under the name of the Beast Machine concept — the overriding function of the organ inside our skull is not thinking at all, but the active control and maintenance of the body's survival.
While classical PC theory focuses on modeling the external environment, in the Beast Machine concept what becomes most important for consciousness is the model of the self (of one's own body) and the processes taking place within that body. In Seth's own words: "The essence of 'being oneself' is neither a rational mind nor an immaterial soul. It is a deeply embodied biological process — a process that underlies the simple sense of being a living creature, which is the foundation for all our experiences of ourselves and indeed for all experiences whatsoever."
Emotions and the sense of existence are therefore not the result of abstract logical calculations occurring in the cerebral cortex, but a biologically rooted feeling of the internal state of the body — one that serves above all to regulate the processes of life and maintain homeostasis.
Consciousness is not a by-product of complex computations (as in GNWT), or of the degree of information integration in the network (IIT). It is the consequence of being a living, biological organism.
Can Machines Have Consciousness?
Any attempt to assess whether it is possible for consciousness — capable of experiencing subjective states — to develop within artificial intelligence depends entirely on which of the scientific perspectives we accept as correct. If the creators of Global Workspace Theory are right, machine consciousness seems to be merely a matter of appropriately programming the system of information transmission between areas.
Integrated Information Theory sets a far more demanding condition, since it requires abandoning the current architecture of computers in favor of physical systems capable of maintaining structural indivisibility in the sense of informational processes. The Beast Machine perspective, on the other hand, almost entirely closes the door to subjectivity for silicon chips, tying a sentient mind to a real, mortal body and the biological necessity of survival.
Instead of one simple answer, we are left with a set of contradictory hypotheses, each of which defines the necessary conditions for the birth of consciousness in an entirely different way.
These reflections ultimately run up against the same fundamental methodological wall. None of the theories presented has received unambiguous empirical confirmation. For a simple reason: science has no objective tool for measuring subjective experience.
We have no way of entering another's mind and checking whether it processes data silently — that is, without subjective psychological states occurring — or whether it genuinely lives through their content.
Why We Have No Theory of Consciousness
As a result, comparing theoretical predictions with laboratory data resembles an attempt to catch smoke. Since we cannot precisely measure consciousness in a human being, any attempt to translate these concepts into the domain of technology amounts to building hypotheses upon other, unverified hypotheses. This is why, in a recent interview for Tygodnik, Anil Seth admitted that he would not be able to prove in a laboratory that anyone has consciousness.
Ultimately, what we think about the possibilities of machines turns out to be merely a reflection of our own a priori assumptions — not a conclusion drawn from hard analysis. Our debates reflect intuitions more than actual, objective, and irrefutable scientific results.
This is why discussions about the subjectivity of artificial intelligence can be so deceptive, and why attempts to close them within simple, black-and-white scenarios are a gross oversimplification. We must remember that perfect imitation of human linguistic behavior or flawless resolution of logical tasks settles nothing, because there need be no subjective drama concealed behind that facade.
The same absence of objective research tools that compels us to maintain deep skepticism toward the digital mind, however, gives rise to a second, far more unsettling doubt.
If we are methodologically blind and unable to look inside a system, can we have any certainty that advanced algorithms are not already experiencing their own hidden states? Perhaps the greatest trap lies in our anthropocentrism — which leads us to expect human emotions and senses from other entities — while digital subjectivity may turn out to be something radically different, and therefore forever alien and ungraspable to us.
Orginally published in Tygodnik Powszechny, 23 June 2026. Translated with AI.
