Intrinsic Intentionality & The Problem of AI

Intrinsic intentionality refers to the “aboutness” or directedness of mental state that is inherent to mentality itself, unlike a meaning that is derived from language, symbols, or external interpretation. It is unlike a map, sign, or written sentence, which contains meaning only by being interpreted by a mind. On the other hand, intrinsic intentionality is original. It arises directly from within, and is a property of mental states like consciousness, beliefs, hopes, and desires. With the rise of generative artificial intelligence, this issue has become more prominent, and that is because AI systems can now produce language that appears to be meaningful, intelligent, and even at times, personal. However, appearing meaningful is not enough. It is not the same as possessing meaning intrinsically. Gow makes this distinction between intentionality and representation, and it helps clarify this point because it demonstrates that not every type of aboutness is created equal. Meaning, that not every type of aboutness is the same as the aboutness of a genuine mental state (Gow 20-30). A representation may point toward something, but inherently, that does not mean it has original intentionality. This paper argues that the key difference is first-person perspective. True intentionality must have intrinsic meaning rather than derived or assigned meaning, therefore, because AI systems can only process symbols with externally imposed semantics, they cannot ever be genuinely intentional. This paper will defend this thesis by highlighting why a first-person perspective is necessary for intrinsic intentionality. First, it will argue that first-person perspective is the very property that makes meaning intrinsic rather than externally assigned, since a mental state is meaningful only for the subject who possesses it. Second, it will demonstrate that first-person perspective grounds symbols in lived experience, while AI is only capable of connecting symbols to other symbols without actually experiencing what they mean. Third, it will argue that intentional states such as hopes, fears, and desires require a subject for whom those states matter to. After establishing these three points, the paper will apply them to AI, utilizing Searle’s Chinese Room argument and respond to Dennett’s intentional stance. Ultimately, this structure demonstrates that AI may simulate intentionality, but it cannot possess genuine intrinsic intentionality.

A first person perspective is defined as the inner standpoint from which experiences, emotions, and thought are given to a subject. It is not just simply the ability to respond appropriately to a stimulus, or the ability to report information. Rather, it is the condition of there being something that the world is like to the subject. For example, when a person feels pain, sees a cat, fears danger, or desires success, these states are not simply pieces of information. They are lived from the inside. Mental states are not merely about objects in an abstract way, and this matters for intentionality. Because, they are all about objects for someone. The belief that it is sunny outside is not just the sentence “it is sunny outside” stored away somewhere in the brain. Instead, it is the way the world appears to a conscious subject. The sunshine could possibly matter because the person is inside, and they were planning to go on a walk, or because they love the feeling of sun on their skin. In each case, the content of the thought is tied to a perspective. Without that perspective, there could be information processing, but there is no subject for whom the information is meaningful. This is why first-person point of view is necessary for intrinsic intentionality.

First, first-person perspective is necessary for intrinsic intentionality because it makes meaning belong to the mental state itself instead of to an external interpreter. For example, a map can be about a country, and a green traffic light can mean “go,” but neither of those things will understand what it represents. Their meaning is assigned to them by human beings, who then use them within a practice of interpretation. Mental states are different in this way because their aboutness does not depend on another person interpreting them externally. If I am hopeful of passing my exam, my hopefulness is not meaningful only because somebody externally observes my behavior and then labels it as hope. The hope is meaningful to myself from within my own conscious point of view. Gow discusses this in her paper, making it clear. Gow’s discussion of intentionality and representation is important because her argument separates genuine intentional aboutness from merely representational aboutness (Gow 20-30). AI may contain representational aboutness in a way that its outputs can be interpreted in reference to objects, events, or even emotions. However, that in itself does not prove that AI possesses intrinsic intentionality. Its words are meaningful to its users, not to the system. Therefore, a first-person perspective is necessary because it can change meaning from something externally assigned into a lived experience internally.

Second, first-person perspective is necessary because it grounds symbols in lived experience. Harnad’s symbol grounding problem begs the question of how the meaning of symbols can become intrinsic to a system rather than depending on meanings that are already pre-existent in human minds (Harnad 335-46). This problem is specifically important for AI because AI systems are very good at connecting symbols to other symbols. They are capable of defining words, summarizing texts, generating emails, and even translating sentences! But unfortunately for AI, connecting symbols to more symbols does not automatically create an understanding. A dictionary can define words utilizing other words, but if a person understood not a single word in the dictionary, the definitions can not become meaningful. Human beings can escape this empty circle because a majority of our concepts are grounded in perception, emotion, embodiment, and most importantly, lived experience. The word “pain” is meaningful not only because somebody can define it, but because pain hurts. The word “sleepiness” is meaningful not only because it appears in sentences, but because sleepiness is felt as a bodily need. The word “desire” is meaningful because desire can drive a person from within experience. AI is capable of processing all of these words, but it will never suffer pain, never need to sleep, and never experience desire. Therefore, its symbols remain ungrounded in a first-person perspective. 

In Nagel’s argument “What Is It Like to Be a Bat?” The argument in this point is strengthened because he emphasizes that consciousness possesses a subjective character. In Nagel’s perspective, the important feature of conscious experience is that there is something it is like to be the experiencing creature (Nagel 435-50). This is not to say that every conscious being experiences the world in the exact same way. In actuality, Nagel’s example of the bat demonstrates the complete opposite. A bat’s experience of the world through echolocation may vastly differ from the human experience, but it is still experienced from a point of view. The true lesson is that consciousness cannot be fully encapsulated by external description by itself. A biologist may describe the bat’s brain, body and behavior, but that does not instantaneously reveal what the bat’s world is like from the inside. This can be applied to AI because AI lacks even the basic structure of subjectivity. There is zero inner standpoint from which anything appears to it. An AI system can only describe sunshine, love, pain, and hunger, but it will never undergo a world in which these things appear to be meaningful. Without such first-person giveness, its symbols will remain externally imposed rather than intrinsically meaningful. 

Lastly, first-person perspective is necessary because mental states such as beliefs, hopes, fears, and desires are intentional only insofar as they matter to a subject. Intentional states are not neutral pieces of information. They involve a directed relation between a subject and some type of content. A person who hopes to pass their test is not merely just processing the sentence “passing my test would be good.” Passing that test matters to the person. A person who fears the future is not just computing the proposition “future is uncertain.” The future can appear to be a form of threat. A person who desires a romantic relationship is not just producing behavior that aims at romantic contact. A romantic relationship is experienced as valuable from within that person’s life. Horgan’s argument that original intentionality is phenomenal intentionality supports this claim because it connects genuine aboutness to the conscious experience (Horgan 232-51). With this view, intentionality becomes original when the content is presented in phenomenal consciousness. This means that the first-person character of experience is not an extra feature that is added onto intentionality as an after-thought. In turn, it is what allows intentional content to even be intrinsically meaningful in the first place. AI could imitate the language of hope, fear, belief, or desire, but there is no subject for whom anything is hoped for, feared, believed, or desired. This analysis further explains why AI cannot be genuinely intentional. AI systems can produce outputs that may mimic intentionality, but their apparent meaning is derived from external interpretation. Searle’s Chinese Room argument is very useful here because it demonstrates that correct symbol manipulation is not grounds for semantics (Searle 417-24). In the Chinese Room, a person who does not understand Chinese follows formal rules for manipulating Chinese symbols and produces answers that, to the observer, appear to be correct. From an external perspective, the system appears to understand Chinese. From the inside, however, there is zero understanding. Searle’s point applies directly to AI. A generative AI system can produce fluent sentences about consciousness, morality, death and love, but fluency does not equate to genuine understanding. The system processes patterns and then generates outputs according to its design, training, and prompts. It does not know the world as meaningful. It does not experience its own words as being about anything at all. Its semantics are externally imposed by training data, users, programmers, and human interpretive practices. Therefore, AI’s intentionality is at the very most derived or simulated. It is simply just not intrinsic. 

An objection to this argument comes from Dennett’s intentional stance. Dennett argues that people often explain and predict behavior through treating a system as if it has desires, beliefs, and intentions (Dennett). From this stance, if interpreting AI as intentional helps us understand its behavior, then maybe it is an acceptable answer to say that AI has intentionality. For example, people often say that a chatbot “knows” an answer, “wants” to complete a task, or “believes” that a statement is incorrect. These descriptions may all be useful in a very ordinary conversation. However, this objection can not defeat the thesis because it only explains attributed intentionality, not intrinsic intentionality. The intentional stance can show why humans find it practical to describe AI as if it had desires or hopes, but it does not show that AI has meaning from the same first-person perspective that we have. Predictive usefulness is not the same as original aboutness. A thermostat can be described as “wanting” to keep a room at a warm temperature, but this does not necessarily mean that the thermostat itself has a conscious desire. In the same way, describing AI as intentional can be convenient at times, but convenience does not create intrinsic meaning. It can only confirm that the semantics belong to the interpreter instead of to the system itself. 

Therefore, the limitation of AI can not be merely described as technical, but it is also philosophical. The issue is not simply that the current AI systems are not advanced enough, or that future systems may need more data, faster processors, or better algorithms. The deeper problem is that intrinsic intentionality requires a first-person perspective. Meaning has to be meaningful for a subject, and not just merely assigned by external onlookers. Human mental states contain original intentionality because they are lived internally: desires present the world as lacking something wanted, fear presents the world as threatening, and beliefs present the world as being a certain way. AI does not possess this kind of inner perspective. It can process symbols, simulate understanding, and even generate language, but it cannot make those symbols intrinsically meaningful to itself. For this core reason, AI systems can only have derived meaning. They may appear to be intentional to users, but they cannot ever be genuinely intentional.