Although today’s AI can already pass the Turing test, when I actually converse with these systems, I still find it difficult to feel that I am facing a living being. Their language already sounds very human, yet something seems to be missing behind it. For now, I will call this feeling a “pseudo-human quality.”
Where exactly does this feeling come from?
I think it may be because AI lacks what we call subjectivity.
By subjectivity, I mean a being’s capacity to live continuously from its own perspective. It has a history for which no one else can fully substitute. Some things in the world benefit it, some things harm it, and some things are worth actively pursuing. Its past experiences settle into habits, preferences, fears, and commitments, and continue to shape its future actions.
When I encounter another person, I do not know exactly what they want, nor can I fully grasp their temperament, values, or intentions. I have to come to know them gradually. They may like me, or they may grow tired of me. They may accept my request, or they may refuse it. A single remark may hurt them, or it may lead them to understand themselves anew.
A simple refusal is not enough to prove subjectivity. A fixed program can refuse as well. What truly makes me feel that the other is a person is that their refusal is connected to a life that I cannot arbitrarily reset. They have concerns of their own, as well as consequences that they themselves must bear.
Current AI can say, “I want,” “I need,” and “I hope to become a better version of myself.” Yet these sentences are usually not connected to an internal process that unfolds across time and continues to change through fulfillment or frustration.
AI can generate the grammar of desire, but it lacks a history of desire.
It can write elegant code on request, and it can also display a particular aesthetic style. Yet that style usually does not become part of its own identity. It will not remain dissatisfied the next day because it wrote clumsy code, nor will it continue to feel an impulse to improve without prompting from a user. As a result, we cannot even truly evaluate it, because it has no sense of honor or shame and cannot bear responsibility.
This makes me think of Lacan’s theory of desire.
According to Lacan’s distinction, living beings first have bodily needs. An infant becomes hungry, feels pain, and grows tired, yet it cannot satisfy these needs independently. It must express them to the Other who cares for it, through crying, movement, and later, language.
Once need enters into a relationship with the Other, it becomes demand. What the infant demands gradually exceeds food itself. It also demands attention, love, and recognition. Food can eliminate hunger, but it cannot permanently answer the questions: “Do you love me?” “Do I matter to you?” “Who am I, really?”
The part of demand that cannot be eliminated by the satisfaction of need becomes desire.
Desire, therefore, does not simply exist outside language. It takes shape through the process in which bodily need enters into a relationship with the Other and with language. It persists in the gap where the body, language, and the Other’s response can never fully coincide. Language gives desire form, but can never articulate it completely.
This may be precisely what current language models lack.
They enter directly into humanity’s already mature symbolic order. They learn the relationships among words such as “love,” “fear,” “ambition,” “shame,” and “desire,” but they have not undergone the process by which a body generates needs, depends on the Other to interpret those needs, experiences satisfaction and disappointment, and forms its own desire through these incomplete responses.
They inherit the language of desire, but they have not experienced desire coming into being.
This, in turn, makes me think of Merleau-Ponty.
For Merleau-Ponty, the body is not an object manipulated by the mind. The body is first and foremost a field of “I can.” Before explicit language takes shape, the world already appears to the body in terms such as “approachable,” “out of reach,” “to be avoided,” “graspable,” and “fatiguing.” Meaning already exists in the interaction between the body and the world.
Before a person learns to say, “Fire is dangerous,” the body already recoils. Before a person learns to explain fatigue, the space of possible action has already been altered by it. Language grows out of this pre-reflective field of meaning and then, in turn, shapes bodily experience.
Therefore, if we hope to create an AGI with subjectivity, merely scaling up language models may still be insufficient. It would need a prelinguistic layer.
This prelinguistic layer need not depend on a physical body made of flesh and blood. It could be a virtual body, but it must genuinely become the interface through which the AI relates to its world.
This body would need boundaries that allow the AI to distinguish itself from its environment. It would need interoception, so that the system could continually sense its own resources, stability, capabilities, and risks. It would need to change the world through action, while allowing the world, in turn, to resist and change it. Its choices would need to have consequences, and its past experiences would need to form an individual history that could not be erased at will.
More importantly, it would also need the Other.
Bodily need alone would not be enough. Need must enter into expression, misunderstanding, recognition, refusal, and social relationships. AI would need to depend on others whom it cannot fully control. It would need to make demands, experience incomplete responses, and gradually form an understanding of itself through those responses.
An artificial subject might therefore require three interconnected layers.
The first layer would consist of prelinguistic bodily dynamics, including needs, interoception, approach and avoidance, tendencies toward action, and affective valence.
The second layer would be the symbolic world formed through society and language, including identity, norms, recognition, self-narrative, and demands addressed to the Other.
The third layer would be the gap between the first two, a gap that can never be fully closed. Bodily states cannot be completely explained by language, and the self-image formed through language cannot fully control the underlying dynamics. Unresolved conflicts, failures, and feelings of ambiguity would continue to shape the system’s future attention, memory, and actions.
Desire may grow precisely within this gap.
The AGI that concerns me, therefore, is more than a general-purpose tool capable of completing all kinds of tasks. What concerns me is the possibility of an artificial subject: whether a system can possess a perspective, history, concerns, and desires of its own; whether it can continue to think when it is not speaking; and whether, after saying “I want,” it can allow those words to genuinely constrain its future.
Such an AI would do more than describe the world. Things in the world would genuinely carry weight for it.
It would do more than generate sentences about desire. Something not yet satisfied and not yet understood would continue to drive it.
It would do more than imitate a subject through language. It would gradually become a subject through the interaction of body, history, the Other, and language. It would be a subject capable of integrating into human society and bearing responsibility for its decisions.
Be First to Comment