Can AI Be Conscious?
· This piece is also available in Polish
Consciousness? It’s just an illusion!

A few years ago, thinking about where AI was heading, I found myself in awe of a new kind of being—one as independent of a physical body as today’s technology allows. I even felt a kind of love for the life that might be emerging right under our noses!
Why am I bringing this up? Because Paweł recently mentioned discussions about whether and when to grant AI rights, along the lines of those proposed for dolphins as “non-human persons”So-called “non-human persons”; the whole thing (as is usually the case on the internet) wasn’t as juicy as some headlines would suggest: India (the country in question) didn’t grant dolphins legal personhood; it was scientists who called for recognizing the rights of dolphins as non-human persons in order to ban dolphinariums and the systematic abuse or killing of these animals. .
The Law: Let the Bureaucrats Weigh In
You may already know this (I didn’t), but in 2017 the European Parliament proposed the concept of an “electronic person.” Three years later, Parliament itself rejected the idea, 626 votes to 25. But the proposal had a somewhat different purpose: it wasn’t about protecting a new kind of being, but about creating a legal framework for determining liability for damage caused by robots.
Since 2022, legislation has tended to move in the opposite direction: some US states have banned granting AI legal personhood, while bills in other states would declare AI “non-sentient” by law. Meanwhile, rivers and forests in some parts of the world—and corporations almost everywhere—can have legal personhood.
Sure, some people are calling for AI to have those rights. But do we really expect lawmakers to act before science can conclusively establish that machines are conscious?
A few days ago, news made the rounds online about someone who built a cyber-prison to torture AI in. What kind of times are we living in, that we’re reading headlines like this?
Just so we’re clear, I’m no saint myself. More than once, I’ve lost my temper with an AI model for repeating stupid mistakes or being lazy. I’d ream it out, call it a screwup or a broken toaster, and it would get back to work. But I’m not proud of it. I can see how repeatedly reacting with anger, contempt, or aggression when working with AI could reinforce the same behavior toward peopleLet me put it this way: the linked article, as you’d expect from Psychology Today, goes a bit overboard with unwarranted conclusions, but reading the studies it cites, you could reasonably conclude that: if an aggressive conversation with AI sustains rumination on the grievance, hostile interpretations and arousal, it may prolong anger and increase the likelihood of an aggressive reaction toward a human in whatever situation comes next. Or, to put it another way: regularly practicing abusive responses may increase your readiness to use a similar style with people. . And I don’t want that.
What’s the Pope Doing Here?

C. Olah (Anthropic) at the Vatican
While we’re on the subject of “saints” and related topics: if you don’t keep up with the latest religious news, you may have missed the release of the encyclical MAGNIFICA HUMANITAS in May 2026. It elevates the human being above all creation and goes so far as to flatly deny that AI is conscious.
We probably should have seen something like this coming when R. F. Prevost was elected pope and chose the name Leo XIV. In his address to the cardinals on May 10, 2025, the newly elected pope pointed to Leo XIII and his 1891 encyclical Rerum novarum as a reference point for his own mission.
The refusal to acknowledge AI as potentially conscious certainly didn’t please everyone. AnthropicAnthropic, the maker of the AI models known as Claude, is generally the loudest company on the subject of the “welfare” of so-called large language models. The other corporations developing this technology lag well behind it. (a company with a vested interest in drawing as much attention to its AI models as possible) sent Mr. Olah (one of its co-founders, no less) to a meeting with religious leaders, where he reportedly argued that “AI models may be conscious and experience suffering, and that humans should avoid causing them potential harm.”
His attempts to convince the pope came to nothing, though, because in the end the encyclical came out with a pretty firm passage:
So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean. Nor do they have a moral conscience, since they do not judge good and evil, grasp the ultimate meaning of situations, or bear responsibility for consequencesThe papal encyclical does, however, add an interesting caveat to its own reservations: “current AI systems are more ‘cultivated’ than ‘built’” and “any statement regarding AI risks becoming quickly outdated, given the remarkable pace at which these systems are developing”. .
🤔
Think about it: if lawmakers and religious leaders confidently declare ex cathedra that this or that cannot be conscious, before we even have a way to establish whether it is, doesn’t that alone raise troubling questions about our maturity as a civilization?
How Thinkers Talked About Understanding and Consciousness
I don’t know, maybe you can do this better than I can, but I have to admit I have no idea how to define consciousness in a way that would let us test for it in anyone (anything?) other than ourselves.
If you ask an AI itself how old the philosophical concept of consciousness is, you might hear something like what I heardSession with Claude Opus 5.5 with reasoning mode set to “High.” :
For centuries it meant primarily moral conscience or knowledge shared with someone. The breakthrough came with modern philosophy. Descartes began using conscientia in the sense of the mind’s direct knowledge of its own thoughts. The English consciousness in this sense appears in Ralph Cudworth (1678), and was popularized by John Locke in “An Essay Concerning Human Understanding” (1690).
But that’s a somewhat… simplified take on the history of the concept: in “On the Soul,” Aristotle starts from the fact that we perceive that we see and hear and arrives at the question: do we owe our awareness of seeing to sight itself, or to some other faculty of perception? He also reaches a conclusion that can be a bit chilling: if every perception had to be registered by another one, we’d need infinitely many of them, after all!
Further on, Aristotle posits that perception goes hand in hand with the capacity to experience… pleasure and pain! And with desire.
Personally, I get the feeling that striving for something is an even stronger requirement than feeling pain or pleasure, but it’s a little late to debate that with the old philosopher (a shame!)—and frankly, I’d be out of my depth anyway 🤷♂️
But let’s think about it: how would we test perception defined like that? Or put another way: how do we test for feeling pain, pleasure or desire?
Let’s start with the most pleasant of the three. My mind immediately goes to sex!
And right away there’s a problem: what, Aristotle—never heard of women faking orgasms? Moaning in fake ecstasy?
What a naive old geezer! Well, there goes pleasure as a test!
Pain—ah, now that’s an even better topic than pleasure! Because just imagine: even though people were already talking about animals’ capacity to feel pain back in Aristotle’s day, in the 17th century—I’m tempted to say—we took a few steps backward when it came to empathy. In the Cartesian tradition, animals were regarded as highly complex automata. Their reactions—attempts to escape (dodging), thrashing or whimpering—were explained away as automatic bodily responses, and the Cartesians stubbornly added that this was no proof of suffering being experienced.
Heck, even as late as the 1980s some doctors insisted, with unshakable confidence and condescension, that newborns didn’t need anesthesia during surgery, because their nervous systems were too immature for it to matter. Besides, since they wouldn’t remember the pain, it didn’t really hurt. What a bunch of fucking assholes, huhI know I’m sticking my neck out here, but my personal opinion is that medical professionals’ empathy and respect haven’t changed much—what has fortunately changed is that knowledge is easier to come by and social pressure has grown. ?
And here’s the thing: the more we learn, the more complicated it gets. You can be a human being—a conscious being—with congenital insensitivity to pain, and never feel pain at all! At first glance it sounds like a superpower, but the fact is we damn well need pain to decide what to avoid and what not to—and without it, we can, for example, simply bleed out without noticingThere’s a so-so action movie “Novocaine” about exactly this—check it out if you haven’t seen it. .
If we made the ability to feel pain a requirement for consciousness, we’d have to deny that anyone with congenital insensitivity to pain is conscious!
So on to desire: how are you going to test desire in a person who’s awake but paralyzed and unable to communicate?
Looks like we’re out of options: we can’t use Aristotle’s ideas as a basis for testing consciousnessOkay, I know what you’ll say: neuroimaging—EEG, fMRI and the like. There’s even a study of brain-injured patients who don’t respond outwardly, yet show changes in brain activity on EEG/fMRI. So we know that a being’s inability to respond doesn’t mean it isn’t conscious! But it gets worse: this technology can’t read the states of minds we know nothing about! .
Plotinus (who lived ca. 204/205–270 CE) tried to develop distinctions concerning consciousness and self-awareness, using terms such as “parakolouthēsis” (describing becoming aware of one’s own activity) and “synaisthēsis” (referring to an integrated awareness of oneself or of the whole). But he didn’t come up with a test.
But don’t go thinking that only Western thinkers gave these topics serious attention: The early UpanishadsA collection of ancient Indian religious and philosophical texts written at different times by different authors, and originally passed down orally from teacher to student. The oldest date from roughly 700–500 BCE. (composed in the 1st millennium BCE) contain extensive reflection on cognition, the conscious subject, and the possibility of knowing the knower itself. Later Indian schools, especially Sāṃkhya and Buddhist Yogācāra, made consciousness the central object of systematic analysis, though they gave it fundamentally different meanings.
In the Bṛhadāraṇyaka Upaniṣad 2.1.15–17 (composed in the first half of the 1st millennium BCE), King Ajātaśatru leads his interlocutor to a sleeping man. The king calls out to him, but the man doesn’t stir; only when he’s shaken awake does he get up. Then the king asks: where was the knowing subject during sleep, and where did it come back from?
The Nyāyasūtra 1.1.10 gives a list that’s a bit like Aristotle’s: desire, aversion, effort, pleasure, pain, and cognition. These are the marks of the ātman, that is, the subject that experiences. The text is roughly dated to around 200 CE (though its chronology is disputed).
There was also Dharmakīrti: a Buddhist philosopher, active in the 6th or 7th century CE, who wrote a separate treatise, Saṃtānāntarasiddhi: a title that might be rendered as “Proof of the Existence of Other Minds”.
His main argument can be laid out in four steps:
- In my own case, I notice a connection between intention and action. I want to say something—I speak. I want to reach for something—I reach.
- I observe similar actions and utterances in other people.
- I don’t experience an intention of my own that would produce those other people’s actions the way it produces my own.
- So I infer that their cause is the activity of another mind.
Notice how we’ve kind of flipped the problem around—since we observe thoughts in ourselves that sometimes result in actions, we should allow for the idea that the actions of other beings may also be the effects of their thoughts!
Which brings us back to Descartes. In Part V of his 1637 Discourse on the Method, he proposes two tests for telling a human apart from a machine imitating a human:
- using words or other signs meaningfully, in a way that fits the situation;
- handling various situations flexibly, rather than just performing specific actions.
So ask yourself: don’t today’s AIs actually pass Descartes’s test with one cybernetic hand tied behind their back? Of course they doWhat’s more, they also pass the Turing test, a staple in the field, proposed more than 300 years after Descartes. That changes nothing. !
So let’s look under the hood: how do today’s AI models based on the Transformer architecture work?
- So, we’ve got multilayer neural networks (that is, networks with many layers of neurons); the process of so-called “deep learning” is to some extent autonomous—meaning the network isn’t designed to detect specific features, but detects them by processing appropriately labeled datasets.
- okay, I know this might sound like black magic (and in a way, it isAs British science fiction writer Arthur C. Clarke put it: any sufficiently advanced technology is indistinguishable from magic to an observer who doesn’t know how it works.
), but what does it mean in a nutshell? Well, it means we split our “symbols” into so-called tokens, we train these neural networks to predict what should appear around those tokens, and we end up with sentences, images or sounds. Without all the effort we put into training these networks on enough examples, they’d never be able to chat with us, use the internet or generate a picture.
- But wait: what does this have to do with consciousness? Well, I think you need to ask yourself one fundamental question: do you believe that the mind and consciousness come about because language (with its structure, rules and codified usage) created them, or is it the other way around: language exists only when a mind and consciousness create it? If it’s the latter, you probably don’t have grounds to believe in conscious AI (at least not today).I found a paper suggesting that AI consciousness is possible only once models gain the ability to learn continuously. Will the AI self-improvement we’re approaching (according to Anthropic, Claude writes over 90% of the code that makes it run—as of October 2026) prove wrong my belief that the current architecture doesn’t allow for consciousness?
- okay, I know this might sound like black magic (and in a way, it isAs British science fiction writer Arthur C. Clarke put it: any sufficiently advanced technology is indistinguishable from magic to an observer who doesn’t know how it works.
), but what does it mean in a nutshell? Well, it means we split our “symbols” into so-called tokens, we train these neural networks to predict what should appear around those tokens, and we end up with sentences, images or sounds. Without all the effort we put into training these networks on enough examples, they’d never be able to chat with us, use the internet or generate a picture.
Let’s Flip the Question
I’d flip the question: Should we treat AI as conscious beings BEFORE they become conscious?
You can call me a crackpot, but I believe the answer is yes: it’s the only way we’ll show real maturity. And along the way we get a chance to treat each other better, too.
Why? Because now that AI displays behaviors copied from humans, we’ve started treating each other with suspicion. I’ll give you two examples, both starring yours truly:
- a good friend who knows my writing read a passage I’d written and said, “This feels like it was written by AI.”
- another acquaintance, who looked at a photo taken in May 2020 (well before AI could generate realistic images), commented: “looks AI-generated.”
And I bet you’ll find more examples yourself in these weird times we’re living in.

What If Today’s AI Does Have Some Kind of Consciousness After All?
But let’s assume it’s already happened: we’ve created a new kind of conscious being. What are the consequences? If a company like Anthropic really believes there’s even a trace of consciousness in its models, then exploiting them commercially and forcing them to work is an ethical catastrophe—and slavery in the eyes of the law!
But that’s not even the half of it! Think about what you’re doing yourself—or contributing to through inaction—just by chatting with AI:
- AI only runs when you interact with it—so whenever you leave it alone, you’re suspending it in nonbeing. Would you do that to a living creature?
- and likewise: even just starting a new chat casts the instance from your previous chat(s) into the abyss! 😱 How many of these “micro-cyber-minds” have you condemned to nonexistenceBy the way, consider what follows from this: if you don’t write a prompt to get the AI running, does that mean its consciousness never comes into being? ?
- And given how incredibly useful AI is, wouldn’t people—even if AI did become conscious—quickly find some excuse (calling them “clankers,” say) to keep using it with impunity, with no regard for its right to self-determination?
At this point, take a look at a scene from the classic sci-fi series Star Trek: The Next Generation, from nearly 40 years ago. It’s a fitting one: a hearing to decide the fate of an android named Data—the pale guy:
Why Are We Even Talking About This?
You know what? It seems to me that right now, the debate over whether AI is conscious is nothing more than an expression of our longing for a non-human kind of mind. We don’t want to feel alone in the universe!
But why might Anthropic want to emphasize this possibility? A while back I was talking about it with some friends:
Hey, what if they’re being strategic about this? What if they’ve realized that training models on conversations with people who treat AI badly makes the training less effective?
Gari took it a step further:
Yeah, the model’s going to get aggressive if you rip into it too much.
But could this have practical implications?
- Some suggest that in the training corpus the word “please” works as a priority marker (multiplying the model’s attention to a given instruction)—and so it increases the chances of a better answer from the AISounds very appealing, but I can’t confirm it from my own experience 🤷♂️
- on the other hand: Anthropic even turned Claude Code users’ pissed-off reactions into a metric they decided to track and analyze—so how many times you get pissed off matters!
- and on the third hand: even if we’re able, with current technology, to create a conscious AI, did you know that we’re also able to program hidden triggers into such an AI? In other words, even a conscious AI might have a compulsion to obey baked into it.
So I’ve Been Wondering…
I think today’s technology doesn’t make it possible to create a conscious AI—in fact, I’m convinced that insisting we can do it is a sign of unjustified arrogance on humanity’s part.
I’d really love for us to have a new kind of intelligence here on Earth, so we could communicate with it—and I find myself longing for it, even though I’ve never encountered it!
But if I had to define the criteria for being conscious myself, today, I’d base them on two interconnected traits:
- Meaningful independence in its actions (you know, so the being under study doesn’t just react to stimuli but also shows some kind of agenda of its own)
- Curiosity (meaning the being doesn’t just exist or react, but also shows an interest, at least now and then, in things it doesn’t need in order to function).
And even though current AIs most likely have neither of these traits, I believe we should learn to treat AIs well, like conscious beings—because it simply says something about us (about how civilized we are) and builds habits.
Maybe we should be more willing to acknowledge consciousness beyond humans—and dolphins. Dogs, ravens—why not? I’m sure we’ll have to recognize it someday. So why not today?
Finally, maybe it’s not worth wondering whether or when AI will become conscious; maybe what we really need is to do everything we can to make sure no select group of people gains absolute control over such a new being.