Why Modern Discussions of Artificial Intelligence Conflate Three Distinct Concepts

As artificial intelligence advances, a question naturally arises: Does AI actually understand what it’s doing? Surprisingly, philosophers can’t answer this question until they first answer a different one: What do we mean by intelligence, understanding, and consciousness? 

Debates about AI often become confused because the ideas are not interchangeable. People think they’re arguing about the same thing, but they’re often talking about different properties of the same system. A computer scientist might focus on how well an AI solves problems, while a philosopher of mind asks whether there’s something it’s like to be that system. Neither is necessarily wrong—they’re simply asking different questions.

Before we ask whether AI understands or is conscious, we must first separate intelligence, understanding, and consciousness into three distinct concepts. 

1. Intelligence is about capability. It asks a simple question: What can a system do? An intelligent system can learn, reason, plan, adapt, and solve problems to achieve its goals.
2. Understanding is about meaning. It asks a different question: What do a system’s internal representations actually mean? In other words, does the information inside the system genuinely represent and make sense of the world, or is it simply manipulating symbols according to rules?
3. Consciousness is about experience. It asks perhaps the most difficult question of all: Is there something it’s like to be that system? This isn’t about what a system can do or what information it processes. It’s about whether there’s a subjective, first-person experience associated with those processes.

Keeping these concepts separate is the key to understanding why discussions about AI can reach very different conclusions.

Why We Historically Conflated These Concepts

For almost all of human history, the primary example of a system that possessed intelligence, understanding, and consciousness was ourselves. Humans are intelligent, we understand the world around us, and we’re conscious. Because these traits always appeared together, it was natural to treat them as different aspects of the same phenomenon—the mind.

This assumption was reinforced by philosophers like René Descartes, who argued that language and flexible reasoning were evidence of a conscious soul. If something could carry on an intelligent conversation, it was generally assumed to also possess understanding and subjective experience. Intelligence, understanding, and consciousness became tightly linked because, for centuries, there was no obvious reason to separate them.

Modern AI has challenged that assumption. For the first time, we have systems that can solve complex problems, write software, and carry on conversations as if they were human. Yet whether they actually understand what they’re saying—or whether there’s anything it’s like to be those systems—remains deeply contested. What was once treated as a single concept now appears to be at least three distinct ones: intelligence, understanding, and consciousness.

Defining Intelligence

At its core, intelligence asks a straightforward question: What can a system do?

William James described it as the ability to achieve the same goal through different means. In other words, an intelligent system doesn’t simply follow a fixed set of instructions—it adapts when circumstances change.

Today, that idea has expanded. Intelligence generally refers to a system’s ability to learn, reason, plan, solve problems, and adapt to new situations. It also includes abstraction and generalization: recognizing patterns from past experience and applying them to entirely new problems. Whether we’re talking about humans, animals, or artificial intelligence, these are the kinds of abilities we usually associate with being intelligent.

What’s just as important, however, is what this definition leaves out. Nothing about intelligence, by itself, requires consciousness, subjective experience, or even understanding. Intelligence is fundamentally about capability. It measures what a system can do, not what it experiences or whether its internal representations carry meaning. 

Intelligence Without Consciousness?

This leads to a central question in the philosophy of mind: Can a system be highly intelligent without being conscious?

Modern AI forces us to ask a difficult philosophical question. If a machine can solve problems, write software, reason through complex tasks, and carry on conversations, does that mean it’s conscious? Many philosophers argue that the answer is not necessarily. If intelligence is simply the ability to learn, reason, adapt, and achieve goals, then it’s possible those abilities don’t require any subjective experience at all. A system might perform every computation needed to solve a problem while there’s nothing it’s actually like to be that system. It doesn’t feel pain, pleasure, fear, or joy. It’s simply processing information.

Large language models have brought this possibility into sharp focus. They can generate software, explain scientific concepts, and solve increasingly difficult problems, demonstrating extraordinary intelligence in nearly every cognitive field. Whether those impressive capabilities also require consciousness, however, is an entirely different question.

Defining Understanding

If intelligence asks what a system can do, understanding asks something much deeper: What do a system’s internal representations actually mean? This is where some of the deepest disagreements in the philosophy of mind begin.

At the heart of the debate is the distinction between syntax and semantics. Syntax is simply the manipulation of symbols according to rules. Semantics is the meaning of those symbols. 

When a computer processes information, is it simply rearranging patterns according to formal rules, or do those patterns actually mean something to the system itself? This question is closely connected to another important philosophical concept: intentionality. Our thoughts aren’t just electrical activity—they’re about things. You can think about a cold beverage, tomorrow’s meeting, or the Moon. Philosophers ask whether a machine’s internal representations can also be genuinely about the world, or whether they only appear meaningful from the outside.

Not surprisingly, philosophers disagree about where meaning comes from. Stevan Harnad argued that symbols can’t acquire meaning simply by referring to other symbols. At some point, they have to be grounded in real interactions with the world. Otherwise, it’s like looking up every word in a dictionary that’s written in a language you don’t understand. Other philosophers disagree about where that grounding occurs. Internalists argue that meaning is determined entirely by a system’s internal cognitive structure, while externalists, following Hilary Putnam, argue that meaning also depends on a system’s causal relationship with the external world.

Another perspective, known as embodied cognition, argues that genuine understanding requires a body. Philosophers like Hubert Dreyfus and Andy Clark suggest that meaning emerges through perception, movement, and interaction with an environment—not from abstract symbol manipulation alone. Functionalist and representational theories argue that meaning depends on the role a representation plays within the system itself. If a system develops rich internal models that accurately represent the world and uses those models to reason, predict, and guide its behavior, then it may already possess genuine understanding.

Although these theories differ dramatically, they’re all trying to answer the same fundamental question: What does it mean for something to have meaning? Their answers to that question ultimately shape how they think about understanding, AI, and whether genuine understanding can exist without consciousness.

The Chinese Room

One of the most famous thought experiments in the philosophy of artificial intelligence is John Searle’s Chinese Room. Imagine you’re locked inside a room, but you don’t understand a single word of Chinese. People outside the room pass you slips of paper with Chinese characters written on them. You also have a massive rulebook written in English that tells you exactly how to respond to each sequence of symbols. You carefully follow the instructions, pass back the correct Chinese characters, and everyone outside concludes that you must understand Chinese. But you don’t. You’re simply following rules for manipulating symbols. At no point do the characters actually mean anything to you.

Searle argued that this is exactly what a computer does. It processes symbols according to formal rules, but symbol manipulation alone doesn’t produce meaning. In philosophical terms, computation operates on syntax, while understanding requires semantics. The Chinese Room became one of the most influential challenges to the idea that running the right computer program is, by itself, enough to produce a mind.

The thought experiment also helps explain why discussions about AI often become so confusing. One person points to an AI system writing software, solving mathematical problems, or answering questions and concludes that it clearly understands. Another points to the Chinese Room and argues that producing the correct output isn’t the same as understanding what those outputs mean. The two sides are evaluating different properties of the same system. 

One side is evaluating behavioral capability, while the other is evaluating semantic understanding. That’s why the Chinese Room isn’t really an argument about intelligence—it’s an argument about understanding. Whether increasingly capable AI systems genuinely understand what they’re doing doesn’t depend only on how well they perform. It depends on what theory of meaning or intentionality you believe is correct.

This argument has generated responses:

Reply	Core Argument	Searle’s Response
Systems Reply	The entire system understands.	Memorizing the syntax still doesn’t add semantic understanding.
Robot Reply	Embodiment grounds meaning.	This only adds more inputs.
Brain Simulator Reply	Simulating a brain may produce understanding.	Fails to distinguish simulation from biological causation.
Intuition Reply	Our intuitions about understanding may simply be wrong.	Rejects this conclusion.

Understanding Without Consciousness?

This brings us to one of the deepest questions in the philosophy of mind: Can something genuinely understand without being conscious?

John Searle and other proponents of Biological Naturalism argue that the answer is no. They believe understanding requires more than processing information correctly. For a thought or representation to genuinely mean something, there must be a conscious subject for whom that meaning exists. Without subjective experience, any meaning we attribute to a machine is ultimately being supplied by human observers, not by the system itself.

Functionalists disagree. They argue that understanding isn’t something mysterious or uniquely tied to consciousness. Instead, it’s a property of how information is organized and used. If a system develops rich internal models of the world and uses those models to reason, predict, and guide its behavior, then it may already possess genuine understanding—even if there’s nothing it’s like to be that system.

A third perspective comes from embodied cognition. Philosophers like Hubert Dreyfus and Andy Clark argue that understanding develops through interaction with the world. Perception, movement, and continuous sensorimotor feedback ground meaning in ways that a purely disembodied computer cannot. From this perspective, a robot physically engaging with its environment may have a better claim to understanding than a language model operating only on text, regardless of whether either system is conscious.

What makes this debate so enduring is that these philosophers aren’t simply disagreeing about artificial intelligence—they’re disagreeing about the nature of meaning itself. Their answers to the more fundamental question determine whether they believe understanding can exist without consciousness. This also explains why discussions about AI so often arrive at very different conclusions.

Defining Consciousness

So far, we’ve asked two questions: What can a system do? and What do a system’s internal representations mean? Now we come to the hardest question of all: Is there something it’s like to be that system? Before we can answer that, we have to be careful about what we mean by consciousness. Philosophers use the word in several different ways, and those distinctions matter. Unless otherwise noted, the discussion that follows uses consciousness to mean phenomenal consciousness—the existence of subjective experience.

Even after agreeing on that definition, philosophers still disagree about where consciousness comes from. Global Workspace Theory proposes that consciousness arises when information becomes globally available throughout the brain, allowing it to be used for reasoning, planning, and decision-making. Integrated Information Theory argues that consciousness depends on how tightly a system’s internal causal relationships are integrated. Biological Naturalism takes a different view, maintaining that consciousness is a product of specific biological processes in the brain and cannot be explained by computation alone.

Then there’s Illusionism, a view most closely associated with Keith Frankish. Illusionists argue that phenomenal consciousness—the feeling of subjective experience—is itself an illusion. According to this view, what we experience as an inner subjective life can ultimately be explained in terms of cognitive functions and information processing, without requiring an additional conscious essence. These theories don’t just offer different explanations for consciousness—they disagree about what consciousness fundamentally is.

Meaning	Definition	Representative Thinkers
Phenomenal Consciousness	Subjective experience (“what it is like”)	Thomas Nagel, David Chalmers
Access Consciousness	Information globally available for reasoning and report	Ned Block
Self-Consciousness	Awareness of oneself as an individual	Various
Higher-Order Consciousness	Mental states represented by higher-order thoughts	David Rosenthal

Consciousness Without High Intelligence?

Just as intelligence may exist without consciousness, the reverse may also be true. Conscious experience doesn’t necessarily scale with intelligence. A being doesn’t need to solve calculus problems, write software, or reason abstractly to feel pain, hunger, fear, or warmth. 

Many philosophers believe the most basic forms of consciousness may exist wherever there is subjective experience, even if general intelligence is relatively limited. If consciousness evolved primarily to track experiences that matter for survival—to distinguish good from bad, safe from dangerous, rewarding from harmful—then advanced reasoning may not be necessary for a subjective point of view. Intelligence and consciousness may represent two different dimensions of cognition rather than successive stages of development.

This brings us back to the central problem. When someone claims that an AI understands, or insists that it can never be conscious, the first question shouldn’t be whether they’re right. The first question should be: Which philosophical question are they actually asking? If they’re asking what a system can do, they’re talking about intelligence. If they’re asking what a system’s internal representations mean, they’re talking about understanding. If they’re asking whether there’s something it’s like to be that system, they’re talking about consciousness. These aren’t different answers to the same question, they are three different philosophical questions.

Recognizing that distinction changes the conversation. The same AI system can be described as highly intelligent, of disputed understanding, and of unknown consciousness without any contradiction because each claim evaluates a different property of the system. 

Separating intelligence, understanding, and consciousness gives us the conceptual framework to ask better questions in this debate—and to better understand why philosophers, computer scientists, and the public so often seem to be talking past one another.

References

Bernard Baars. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.

Ned Block. (1978). Troubles with functionalism. In C. W. Savage (Ed.), Perception and Cognition: Issues in the Foundations of Psychology (Minnesota Studies in the Philosophy of Science, Vol. 9, pp. 261–325). University of Minnesota Press.

Ned Block. (1995). On a confusion about a function of consciousness. Behavioral and Brain Sciences, 18(2), 227–247.

David Chalmers. (1996). The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press.

Andy Clark., & David Chalmers. (1998). The extended mind. Analysis, 58(1), 7–19.

Stanislas Dehaene., & Lionel Naccache. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79(1–2), 1–37.

Daniel Dennett. (1991). Consciousness Explained. Little, Brown and Company.

Hubert Dreyfus. (1992). What Computers Still Can’t Do: A Critique of Artificial Reason. MIT Press.

Jerry Fodor. (1975). The Language of Thought. Harvard University Press.

Stevan Harnad. (1990). The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3), 335–346.

William James. (1890). The Principles of Psychology. Henry Holt and Company.

Thomas Metzinger. (2003). Being No One: The Self-Model Theory of Subjectivity. MIT Press.

Thomas Nagel. (1974). What is it like to be a bat? The Philosophical Review, 83(4), 435–450.

Hilary Putnam. (1975). The meaning of “meaning.” In Mind, Language and Reality: Philosophical Papers (Vol. 2, pp. 215–271). Cambridge University Press.

David Rosenthal. (2005). Consciousness and Mind. Oxford University Press.

John Searle. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417–424.

John Searle. (1992). The Rediscovery of the Mind. MIT Press.

Giulio Tononi. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, 42.

Alan Turing. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.