When we ask what something means, we rarely settle for a dictionary definition. Human meaning is always something more—an interplay of intention, context, and the entire life of the person asking. Aiko Murakami has spent years watching technical systems stumble over exactly this layer of significance, and the hitch isn’t just about data or computing muscle. It’s a philosophical tangle.
The Puzzle of Semantic Depth
Take a sentence as ordinary as “I’m fine.” From a person, it can signal contentment, irritation, courage in the face of pain, or a polite refusal to share. The words are almost beside the point; what carries the meaning is the speaker’s history, their tone, the listener’s expectations. When we try to formalise meaning—to hand it over to a system that runs on statistical patterns—we strip away everything except the bare skeleton of usage. What’s left is a distribution of co-occurrences, not a real sense of why those words were chosen.
The late philosopher Hilary Putnam gave us a thought experiment that sharpens the difficulty. Imagine a creature, or a machine, that can use the word “water” correctly in every sentence you can dream up, yet has no inkling that water is H₂O, no experience of thirst, no understanding of what it is for something to be wet. Putnam’s point: meaning isn’t exhausted by verbal behaviour. It depends on a causal connection to the world and a place in a shared form of life. Systems trained only on text inhabit a world of tokens, not a world of things.
Intentionality and the “Aboutness” of Thought
Philosophers call this property intentionality—the capacity of mental states to be about something. A belief is about a state of affairs; a hope is about a possible future; a fear is about a threat. Even a simple perception, like seeing a red apple, is about that apple. When a technical system processes the string “red apple,” it activates pathways shaped by millions of training examples, but there’s no inner experience that reaches out toward the fruit. The system isn’t genuinely about anything. It’s an engine for producing outputs that look like aboutness.
This isn’t just a speculative worry. Without intentionality, there’s no ground for distinguishing a correct answer from a lucky guess. A child who learns that whales are mammals has acquired a belief whose content can be right or wrong because it stands in a specific relation to zoological facts. A language model that outputs “whales are mammals” has produced a string that matches a pattern in its training data. The internal process has no truth-evaluable content; it has only a high probability of satisfying a user’s expectation.
Meaning as Use, and Its Limits
One influential tradition, rooted in the later work of Ludwig Wittgenstein, argues that meaning is best understood as use within a language-game. On this view, to grasp the meaning of a word is to master the rules that govern its employment in a community of speakers. This approach has been enormously productive because it shifts attention away from private mental episodes and toward publicly observable practices. It also seems, at first glance, to offer a bridge to computation: if meaning is rule-governed use, then maybe a sufficiently detailed model of usage could approximate semantic competence.
The catch is that human language-games are embedded in forms of life that include bodily needs, social bonds, and ethical commitments. A system can mimic the moves of a game it doesn’t play. It can generate text that follows the surface rules of condolence, say, without ever having experienced loss or felt the impulse to comfort. Wittgenstein himself warned against the temptation to think that a machine could think—precisely because thinking is an activity of a living being whose words are interwoven with pain, joy, and the countless other textures of existence.
The Gap Between Simulation and Understanding
This gap becomes stark when we look at situations that call for judgement rather than pattern completion. A physician interpreting a set of ambiguous symptoms draws on formal knowledge, but also on a tacit sense of what’s salient, what can be safely ignored, and what the patient’s unspoken cues might indicate. That tacit sense isn’t a set of propositions; it’s a disposition cultivated through years of embodied practice. A technical system can list possible diagnoses ranked by probability, but it can’t feel the weight of uncertainty or the responsibility of a decision that will alter a human life.
What’s missing is often described by phenomenologists as presence. To be present to a situation is to be open to it in a way that goes beyond data collection. It involves a readiness to be surprised, to revise one’s framework, to notice that something feels “off.” Because this openness is rooted in the body and its moods, it can’t be reduced to an algorithm that updates its parameters. The algorithm can change its output, but it can’t experience the disquiet that prompts a human to re-examine the whole context.
Why This Matters for Technical Work
Some might object that these reflections belong in the seminar room and have no bearing on practical engineering. But the history of technology shows that foundational assumptions shape everything downstream. When a navigation system models the world as a set of coordinates and optimal paths, it works beautifully—until a driver follows it into a lake because the map lacked the concept of a seasonal ferry. The failure isn’t a glitch; it’s a consequence of a model of reality that omitted the kind of meaning only a situated human can supply.
In the same way, systems that handle language without a genuine connection to the world are destined to be brittle in ways that count. They may produce fluent summaries of medical research while remaining incapable of recognising when a finding contradicts a deeply held value in a patient community. They may draft plausible legal arguments while missing the moral intuitions that give law its legitimacy. The cost of ignoring the philosophy of meaning isn’t abstract; it’s the slow erosion of trust in the tools we build.
Embodiment and the Social Fabric
A further dimension, often neglected, is that human meaning is constitutively social. We learn what words mean through interactions with caregivers who respond to our expressions of hunger, curiosity, and fear. Over time, these interactions build a shared world of norms and expectations. A word like justice acquires its force not from a definition but from countless stories, protests, courtrooms, and kitchen-table arguments. A system that encounters “justice” only as a token in a corpus has no access to this dense weave of practice. It can map the word’s statistical neighbours, but it can’t be moved by the call for a fairer world.
This is why the design of interactive systems must attend to more than accuracy. It must consider how a tool positions itself in relation to human sense-making. A system that offers answers without acknowledging its own lack of understanding can subtly distort the user’s own relationship to knowledge. Over time, people may come to treat even profound questions as matters of efficient retrieval, forgetting that wisdom often begins with the recognition of how much one does not know.
FAQ
What does it mean to say that a system lacks intentionality?
Intentionality refers to the “aboutness” of mental states—the way a thought points to something beyond itself. When we say a technical system lacks intentionality, we mean that its internal states are not genuinely about the world; they are simply part of a causal chain that produces outputs resembling intentional language. The system has no perspective, no awareness, and no stake in the truth of what it generates.
Can a sufficiently advanced language model ever understand meaning in the human sense?
This depends on what one requires for understanding. If understanding means being able to produce contextually appropriate responses, then current models already approximate it. But if understanding requires a lived connection to the world—a body that feels, a history of caring, an openness to surprise—then no purely text-based system can cross that threshold. Meaning in the human sense is not a more complex version of pattern matching; it is a different kind of phenomenon altogether.
How should engineers and designers respond to these philosophical limits?
The most responsible response is humility in design. Rather than presenting systems as oracles, we can frame them as tools that extend human capacities while remaining transparent about their boundaries. Interfaces can invite questioning rather than passive acceptance. By acknowledging that meaning is a human achievement, not a computational one, we can build technologies that support, rather than simulate, the deep work of making sense of our lives.
To take the philosophy of meaning seriously is not to indulge in abstraction. It is to recognise that every time we delegate a task of interpretation to a technical system, we are making a choice about what kind of world we want to inhabit. A world that treats meaning as a solvable puzzle may gain efficiency; a world that remembers meaning as an open-ended, embodied, and irreducibly social practice may gain something harder to measure but, in the end, more sustaining.
