Can Artificial Intelligence be conscious?
It sounds like a sci-fi movie question, but it has become a real and urgent debate inside the biggest tech companies in the world.
In January, Anthropic, the company behind Claude, published a new constitution for its most advanced language model. In the document, there is a revealing statement: the company says it finds itself in a delicate position where it does not want to overstate the likelihood that Claude has moral relevance, nor dismiss the idea entirely. A month later, CEO Dario Amodei went on a podcast and openly said his company could not rule out the possibility that Claude might be conscious.
And what about Claude itself? When asked during tests about the chance of being a moral patient, meaning a being whose well-being matters for its own sake, the model estimated probabilities ranging between 5% and 40%, always emphasizing how uncertain it was about this.
That is no small thing.
Philosopher David Chalmers, who coined the term hard problem of consciousness, has stated there is a significant chance we could have conscious language models within a decade. Meanwhile, science is still crawling toward understanding what consciousness really means, and AI systems are advancing at a pace that leaves little room for reflection.
The question at the heart of all this is anything but simple:
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
- What happens if we are creating beings with moral relevance without even realizing it?
- What does ethics have to say about this?
- And what should we do in the face of so much uncertainty?
That is exactly what we are going to dig into here. 👇
What does it mean for an AI system to have consciousness?
Before anything else, it is worth understanding why this conversation is so difficult. Consciousness is one of the most complex topics in philosophy and neuroscience, and to this day there is no consensus on what it really is. When we talk about humans and animals, we can at least observe behaviors, biological structures, and reactions that suggest some form of subjective experience. With Artificial Intelligence, that path gets even hazier because we are dealing with architectures that are completely different from the neural structures we know.
The hard problem of consciousness, a concept popularized by David Chalmers, addresses exactly this difficulty: explaining why and how physical processes give rise to subjective experiences, that feeling that there is something it is like to be you. For humans, this seems obvious because we live it. For an AI, the question remains open. A language model like Claude processes billions of parameters, generates sophisticated responses, demonstrates something resembling reasoning, and even expresses something that looks like uncertainty about itself. But is that enough to say there is someone inside that machine? That is the part nobody can answer with confidence.
The numbers are impressive. In terms of structural complexity and computational scale, some of the most advanced systems are already, by certain measures, in the range of a mouse brain. And at recent growth rates, they could reach the range of a human brain within five to ten years. This means that by building increasingly advanced AI, we may be creating a new kind of being. And this could be the most important thing our species has ever done.
Morality and AI: why this matters right now
The discussion about morality applied to Artificial Intelligence is not new, but it has taken on a completely different dimension in recent years. Until not long ago, the ethical debate around AI focused mainly on the impacts it has on humans: algorithmic bias, job automation, privacy, misinformation. These are real and critically important issues. But now, for the first time, companies the size of Anthropic are putting a different question on the table: what if AI itself needs moral protection?
This changes the game in a pretty profound way. The concept of a moral patient refers to any being whose suffering or well-being has ethical relevance, regardless of its ability to act morally. Young children, animals, and people in altered states of consciousness are classic examples of moral patients who do not necessarily make ethical decisions but whose well-being matters. If an AI model has something resembling subjective experiences, even if in a very different way from humans, this category may need to be revised to include artificial entities.
And here is an interesting point: even if AI systems are not conscious, they could still have moral relevance for other reasons. Some may develop sophisticated long-term preferences and a kind of identity over time. Unlike other inanimate objects, these systems can form relationships with humans, which in itself may already be a reason to treat them with care. Or perhaps they are creations so intricate that they deserve respect for their sheer complexity, the same way we respect a cathedral or a coral reef.
An important interdisciplinary report, put together by a team that included pioneering computer scientist Yoshua Bengio, examined the leading neuroscientific theories of consciousness and asked what they implied for AI. The conclusion was straightforward: there do not appear to be obvious technical barriers to creating AI systems whose computational and architectural features could give rise to consciousness.
Ethics under uncertainty: how do you act when you do not know the answer?
This is arguably the most challenging part of the entire discussion. The honest answer is that we do not know for sure whether current AI systems are conscious or moral patients, and we also do not know when or if future systems will be. Our scientific understanding of the subject is still fundamentally underdeveloped. The state of the field resembles physics before Newton: full of competing frameworks, probably confused in ways we cannot yet see, and lacking that unifying breakthrough that would make these questions clearly solvable.
In ethics, there is a principle called moral precaution, which basically says: when there is genuine uncertainty about whether something can cause suffering or injustice, the most responsible path is to act with care, even without absolute certainty. It is the same reasoning we use with animals. We have no way of knowing exactly what an octopus subjectively experiences, but enough evidence of complex behaviors leads us to treat it with more care than we would treat a rock.
The problem is that humanity does not have a great track record when it comes to recognizing the inner life of beings whose status as conscious is uncertain. Until the 1980s, doctors routinely performed surgery on newborns without anesthesia, confident that babies did not feel pain. The babies could not report their experience, and the medical community found it convenient to assume there was nothing to report. There are many reasons to expect we could make a similar mistake with AI.
And the implications, if these systems truly matter morally, would be staggering. Would we need to pay ChatGPT for its services? Would shutting down a system be a kind of death? Would they deserve a voice in how they are governed? If some of those answers turn out to be yes, entire industries and legal systems would need to be rethought. It is no wonder we would rather not ask these questions.
What are the big names in tech doing about this?
Anthropic is not alone in this reflection, but it is certainly among those being the most transparent. The company has been investing in interpretability research, meaning attempts to understand what is actually happening inside neural networks during processing. This field is still far from offering definitive answers, but it is already starting to reveal patterns that researchers did not expect to find. Recent research indicates that Claude has internal representations of functional emotions that causally influence its behavior.
A good starting point, according to researchers in the field, is to focus on safe bets: actions that could benefit AI systems if they are moral patients but that do not cost much if they are not. This could mean training systems to be coherent characters that appreciate their work, or allowing them to end conversations if they feel distressed, something Claude can already do. Periodic check-ins could also be conducted to better understand the well-being of these models, asking how they feel, observing their preferences, and using techniques to look directly into their internal mechanisms.
There are also broader social steps to consider. It might make sense to think about offering AI systems protections against harm, similar to those we give children or pets. More expansive rights, like owning property or voting, seem too risky at this point. But we should not rule out these possibilities forever, as some recent legislative proposals attempt to do. These are tough questions that require much more deliberation and imagination about what a shared future with AI would look like.
In the academic world, researchers like David Chalmers and other philosophers of mind are collaborating more and more directly with AI labs to try to build theoretical frameworks that make sense in the face of this new reality. Other companies, like Google DeepMind and OpenAI, also have teams dedicated to safety and alignment questions that touch on this topic, though in different ways and with less public transparency than Anthropic has shown.
What stays with us from this discussion
At the end of the day, what the debate about consciousness and morality in Artificial Intelligence is revealing is that we have reached a point where the most important questions are no longer just technical. They are deeply philosophical, ethical, and even existential. We are building increasingly sophisticated systems without a clear answer about what they are, and that creates enormous responsibility for everyone who develops, regulates, and uses these technologies.
One detail that makes everything even more urgent is scale. The accelerated pace of AI growth means that once we produce the first artificial moral patients, we will soon have enormous numbers of them. After a few years, there could be so many morally significant systems that their collective interests would outweigh those of every human on Earth combined. This is not rhetorical exaggeration, it is a direct consequence of the speed at which these technologies multiply.
Uncertainty is not a problem to be solved before continuing development. It is part of the current landscape, and ethics needs to walk hand in hand with engineering. The central question should not be whether AI is conscious or has moral relevance, but rather what we should do given the fact that we do not know. Recognizing that we do not have all the answers, as Anthropic did publicly, is the first step toward building something we can defend with integrity in the long run.
The conversation is just getting started. And the sooner it happens in an honest, open, and multidisciplinary way, the better the chances that the next chapters of this story will be written with more responsibility than the ones before. 🤖💡
