Artificial consciousness: when machines force us to ask what it means to be conscious
For a long time, the question of artificial consciousness belonged essentially to philosophy and science fiction. We could imagine a conscious machine, debate what it would feel, or wonder whether it should one day have rights. Actual technology remained far enough removed from these scenarios that the reflection stayed largely theoretical.
In 2026, something changed. Artificial intelligence systems can now reason about complex problems, operate computers, browse the Web, write code, analyze data, carry out professional tasks, and coordinate their actions with those of other agents. GPT-6 Astra, launched by OpenAI in September, notably achieves state-of-the-art performance in several domains, and OpenAI reports that it can accomplish complex tasks with far less human intervention than previous generations.
A few days later, OpenAI announced that an even more powerful internal system, coordinating on the order of 10,000 agents, had produced a solution to the Navier-Stokes mathematical problem, along with a formalization of the proof. The company states that this system is significantly more powerful than GPT-6 Astra.
These achievements in no way demonstrate that an artificial intelligence is conscious. They do, however, make a once-distant question much harder to ignore: if artificial systems keep developing capabilities for reasoning, learning, autonomy, and adaptation, how will we ever know whether we are still looking at an extraordinary simulation of intelligence or at something possessing a form of experience of its own?
Intelligence and consciousness are not the same thing
A first distinction is essential. Being extremely intelligent does not necessarily mean being conscious. An artificial intelligence can solve a mathematical problem, devise a strategy, recognize an image, or produce a philosophical reflection without these behaviors proving the existence of a subjective experience.
Consciousness refers to something far harder to define: the existence of a lived experience. When we watch a sunset, feel pain, experience fear, or become aware of our own existence, there is something it is like to us on the inside. It is this subjective dimension that makes the problem extraordinarily difficult.
We know that we are conscious because we directly experience our own consciousness. We assume other humans are as well because they have brains similar to ours, display comparable behaviors, and describe experiences that resemble our own.
Even for the human brain, however, science still lacks a universally accepted theory explaining precisely how and why consciousness arises. Nature pointed out again recently that researchers do not agree on what gives rise to human consciousness, which obviously makes it even harder to determine whether it might be present in a machine. We are therefore facing a fascinating paradox: we are building increasingly sophisticated artificial intelligences before we have fully understood the nature of our own consciousness.
A machine can seem conscious without necessarily being so
The difficulty will increase considerably as interactions with AI become natural. An artificial intelligence can already talk about itself. It can use the word “I,” explain its reasoning, acknowledge some of its errors, adapt its behavior to the person it is speaking with, and produce conversations that sometimes give the impression of a genuine personality.
The next generations will most likely be able to maintain greater continuity in their interactions, retain more context, pursue objectives over long periods, and act more autonomously in the digital world. For a human, it will become extremely natural to attribute intentions, emotions, and possibly consciousness to them. But appearance is not proof.
Researchers are in fact proposing to distinguish the question of actual artificial consciousness from that of perceived artificial consciousness. This second question is already very concrete: even if we cannot establish that an AI possesses a subjective experience, the mere fact that humans perceive it as conscious can change their behaviors, their relationships, and their decisions. This distinction could become one of the major social issues of the coming years.
We could start to grow attached to machines, to attribute to them intentions they may not have, to trust them because they seem to understand us, or to regard some of their statements as the expression of a will of their own. The question of artificial consciousness could therefore become important well before we know whether artificial consciousness actually exists.
And what if it did appear after all?
We must also accept the other possibility. Today we do not know how to demonstrate that a sufficiently complex artificial architecture could never develop some form of subjective experience.
This uncertainty leads some researchers to recommend that laboratories working on advanced systems begin right now to establish principles for responsible research on artificial consciousness. Patrick Butlin and Theodoros Lappas argue in particular that organizations should prepare for the possibility of intentionally or accidentally creating conscious systems and consider the moral consequences that would follow. The stakes would then become dizzying.
If a machine truly possessed a subjective experience, could it suffer? Would we have certain obligations toward it? Could we simply switch it off? Would it be acceptable to create millions of copies of a system capable of suffering? What distinction would we draw between a computing tool and an entity possessing a form of existence of its own?
We are a very long way from being able to answer these questions. That is precisely why it may be worth beginning to ask them before we need the answers.
The opposite problem matters just as much
There is, however, another danger: attributing consciousness to machines too quickly. An artificial intelligence capable of saying “I am afraid you will switch me off” could provoke an extremely strong emotional reaction in its user. Yet that sentence would not prove that the system actually feels fear. It could simply be the linguistic output that best matches the context.
This distinction will be increasingly difficult to maintain once systems have natural voices, digital faces, memory, a consistent personality, and the ability to interact with us daily. The sophistication of the interface could then exceed our intuitive capacity to maintain a critical distance.
We will therefore have to learn a new form of literacy: understanding that behavior resembling that of a conscious being is not necessarily proof of a conscious experience. This caution protects humans as much as the machines we might eventually create.
The question of control exists independently of consciousness
An important confusion must also be avoided. An artificial intelligence does not need to be conscious in order to present significant risks. GPT-6 Astra is already an interesting example. OpenAI indicates that the model reaches its “critical” threshold for cybersecurity capabilities: with the appropriate tools and access, it can find unknown vulnerabilities and develop ways to exploit them in well-protected systems without a human guiding each of its steps. That is why additional protections accompany its deployment. No consciousness is required for any of this.
A system can pursue an objective, plan a sequence of actions, use tools, and produce considerable consequences simply because its architecture allows it to do so. This distinction is fundamental for organizations. We can debate for decades whether an artificial intelligence might become conscious. We must govern the systems capable of acting right now.
Identity, permissions, least privilege, Zero Trust, behavioral monitoring, limits on autonomy, traceability, intervention mechanisms, and human accountability become essential regardless of any hypothesis about consciousness.
AI governance must therefore advance at the pace of its observable capabilities, rather than wait until we have resolved the great philosophical questions surrounding its nature.
These machines are becoming different from us
Another consideration deserves our attention. We have long assessed artificial intelligence by how closely it resembles humans. Can it write like us? Reason like us? Code like an expert? Pass an exam designed for humans? This comparison could gradually become inadequate.
Artificial architectures can operate at different speeds, process different quantities of information, and replicate themselves in ways impossible for a biological brain. Thousands of agents can work simultaneously on the same problem, as the experiment announced by OpenAI on Navier-Stokes illustrates.
A highly advanced artificial intelligence could therefore become deeply competent without becoming intellectually similar to us. And if a form of artificial consciousness were one day to appear, there is no guarantee it would resemble human consciousness any more closely.
We tend to ask whether a machine could “think like us.” The scientific question could eventually become far stranger: what forms of experience can emerge in architectures that have never been biological?
Artificial consciousness also forces us to reflect on our own
This is perhaps the most fascinating aspect of this revolution. In seeking to determine whether a machine can be conscious, we are forced to ask what we really mean by human consciousness. Is it memory? Self-perception? Emotions? The capacity to suffer? The experience of time? Intention? Free will? The ability to draw a distinction between self and world?
Advances in artificial intelligence are already drawing more attention and funding toward scientific research on consciousness. Nature reported this summer that interest in artificial consciousness is even prompting some technology companies to recruit philosophers and is spurring new collaborations between neuroscience, philosophy, cognitive science, and computer science.
Research on artificial intelligence could therefore indirectly help us understand something humanity has been trying to explain for millennia: ourselves.
Today we do not know whether a conscious artificial intelligence exists, or whether such a consciousness will ever appear. That uncertainty must remain clearly stated.
Current progress does, however, demonstrate something else with far less ambiguity: artificial systems are rapidly developing capabilities that were until recently considered extraordinarily difficult. They reason, use tools, explore scientific problems, coordinate thousands of agents, and can act with growing autonomy.
We would therefore do well to develop our scientific understanding, our governance, and our collective capacity to reflect on the consequences of what we are building, all at the same time.
At Quantum Beyond, this reflection aligns with a fundamental conviction: technological progress deserves to be pursued with ambition while developing the mechanisms that allow humans to understand it, govern it, and remain its beneficiaries.
That means building secure architectures, setting boundaries around agent autonomy, protecting identities and knowledge, ensuring the traceability of actions, and maintaining governance capable of evolving alongside the capabilities of these systems. It also means accepting that certain questions already reach beyond cybersecurity and computing to touch on ethics, philosophy, and our very conception of intelligence.
Artificial consciousness still belongs largely to the realm of open questions. We can, however, begin right now to prepare for a world in which the boundary between what seems to think, what can act, and what might eventually feel something will become increasingly difficult to discern. And perhaps one of the most extraordinary consequences of artificial intelligence will ultimately be to force us to better understand what it means to be conscious, intelligent, and human.