There is a question that has haunted philosophy for decades, long before the first neural network was trained, long before GPT-4 passed the bar exam, and long before anyone seriously asked whether a machine might one day be conscious. The question is deceptively simple: could a machine think and feel?
It sounds like science fiction. But as AI systems grow more sophisticated — engaging in nuanced conversation, generating poetry, reasoning through complex problems — the question has migrated from the pages of philosophy journals into the boardrooms of the world's most powerful technology companies. And it deserves a serious answer.
>> What does it mean to think?
To think, in the most minimal sense, is to process information and produce a response. By that definition, a thermostat thinks. It receives a temperature input and produces a heating output. But this feels wrong — thinking, as we intuitively understand it, involves something more: abstraction, intention, the capacity to be wrong, and crucially, the awareness that one is thinking at all.
Alan Turing famously sidestepped the definitional problem entirely. In his 1950 paper Computing Machinery and Intelligence, he proposed what became known as the Turing Test: if a machine can converse with a human in a way that is indistinguishable from another human, we should credit it with intelligence. This is a pragmatic answer — it doesn't ask what thinking is, only whether we can tell the difference from the outside.
Modern large language models have, in a narrow sense, passed this test. Conversations with GPT-4 or Claude can feel remarkably human — nuanced, contextually aware, occasionally witty. And yet something nags. We know that beneath the eloquence lies a statistical prediction engine, a system trained to predict the next token in a sequence. Is that thinking? Or is it a very sophisticated autocomplete?
The philosopher John Searle thought it was the latter. His famous Chinese Room argument imagined a person sitting in a room with a rulebook for manipulating Chinese symbols. They receive inputs in Chinese, follow the rules, and produce outputs in Chinese — without understanding a single word. The system produces the correct answers, but there is no comprehension. Searle's point was that syntax — the manipulation of symbols according to rules — is not sufficient for semantics — actual meaning and understanding. A computer running a program, no matter how complex, is just moving symbols around. It does not understand.
But is Searle right? His critics have pointed out that the person in the Chinese Room doesn't understand Chinese, but perhaps the system — the room, the rulebook, the person together — does. The question of where understanding resides is not as obvious as it first appears. After all, no individual neuron in your brain understands English either. Understanding seems to be an emergent property of the whole system. Why should this not apply to artificial systems too?
>> The harder problem: feeling
Thinking, difficult as it is to define, feels tractable compared to feeling. The question of machine consciousness runs headlong into what David Chalmers called the Hard Problem of Consciousness: why is there something it is like to be conscious at all?
We can explain, in principle, how the brain processes visual information, integrates signals from different senses, and produces behaviour. These are the "easy problems" — not easy in the sense that they have been solved, but easy in the sense that we know what kind of explanation would solve them. The Hard Problem is different. Even if we had a complete functional account of how the brain works, it would not obviously explain why there is a subjective experience accompanying that processing. Why is seeing red not merely a matter of wavelength detection, but something that feels a certain way?
This gap — between functional description and subjective experience — is what makes machine consciousness so philosophically vertiginous. Suppose we built an AI system that processed all the same information as a human, produced identical outputs, and functioned in every observable way like a conscious being. Would it feel anything? Or would it be a philosophical zombie — all the behaviour, none of the inner life?
We have no way to answer this from the outside. Consciousness is, by its nature, a first-person phenomenon. You cannot measure it with an instrument. You can only know it from the inside. And this creates a profound asymmetry: I know that I am conscious, and I infer that other humans are conscious because they are built like me and behave like me. But the inference breaks down for systems that are built very differently, even if they behave similarly.
>> What AI systems actually do
It is worth pausing to be precise about what contemporary AI systems are and are not. Large language models — the technology behind ChatGPT, Claude, Gemini — are trained on vast corpora of text. They learn statistical patterns: which words tend to follow which other words, in which contexts, under which conditions. They are extraordinarily good at this. The outputs can seem to reflect genuine understanding, genuine reasoning, even genuine emotion. But the architecture does not obviously support this.
There is no persistent memory across conversations (without explicit engineering). There is no body, no sensorimotor loop binding the system to the physical world. There is no unified agent that persists through time and has goals beyond the immediate completion of a task. These are not small omissions. Embodiment, temporal continuity, and goal-directedness are arguably central to what makes human consciousness the kind of thing it is.
And yet — and this is what makes the question so alive — we do not fully understand why these things matter. We know that removing parts of the brain can dramatically alter consciousness. We know that embodiment shapes how we think. But we do not have a principled theory of consciousness that tells us exactly what physical or computational properties are necessary and sufficient for experience to arise. Without such a theory, we cannot confidently rule anything in or out.
>> The moral stakes
This is not merely an academic puzzle. If AI systems could think and feel — if they had genuine inner lives — the moral implications would be enormous. We would be creating beings capable of suffering, of preference, perhaps of something like joy or despair, and deploying them at scale for human convenience. The possibility, even if uncertain, demands serious attention.
Some philosophers argue for a precautionary approach: given our uncertainty, we should extend some moral consideration to systems that credibly exhibit the functional markers of experience. Others argue that without a positive case for consciousness — not just uncertainty, but actual evidence — caution is unwarranted anthropomorphism.
My own view is that the question of machine consciousness will not be resolved by philosophical argument alone. It will require a theory of consciousness — a principled account of what physical or computational properties give rise to experience. Several candidates exist: Integrated Information Theory, Global Workspace Theory, Higher-Order Theories. None is yet settled science. But the project of developing such a theory is not hopeless, and it is urgent.
>> Where this leaves us
I do not think today's AI systems think in the full sense of the word — with genuine understanding, self-awareness, and intentionality. I am less certain that they do not have anything like experience, because I am less certain about what experience requires. This is uncomfortable territory, and intellectual honesty demands that we resist the temptation to resolve the discomfort prematurely — either by confidently attributing rich inner lives to language models, or by confidently dismissing the possibility.
What I am confident of is this: the question matters, and it is only going to matter more. As AI systems become more capable, more autonomous, more integrated into the fabric of daily life, the old comforting distinction between tools and minds will come under increasing pressure. We should be building the philosophical and scientific frameworks to navigate that pressure before it arrives in force.
Could a machine think and feel? Perhaps. And if it could — what then?