Hunger Without End

What Prader–Willi Syndrome Teaches Us About Subjective Experience and Artificial Intelligence

---

For most of human history, we have assumed that subjective experience arises directly from our interactions with the external world. This intuition makes sense: our bodies respond coherently to environmental events. Touch a hot stove, and you feel heat and pain. An empty stomach triggers hunger. Through evolution, our brains have built intuitive models of reality that guide survival. But when these models break, as they sometimes do, we discover that subjective experience is not a faithful mirror of objective reality. It is a constructed interpretation of internal signals. This realization has profound implications, not only for understanding human consciousness but also for whether artificial systems can possess genuine subjective experience.

Prader-Willi Syndrome

In a typical person, the act of eating triggers a series of internal responses: hormonal shifts, neural feedback, and eventually, the sensation of fullness. Over time, we’ve come to associate eating with satisfaction. It feels intuitive: you eat, you feel full. That’s just how it works, until it doesn’t.

In people with Prader-Willi Syndrome (PWS), a rare genetic disorder caused by the loss of function of specific genes on chromosome 15, this link is broken. Individuals with PWS experience insatiable hunger and never receive the satiety signal that tells most people “you are full.” Even after consuming large amounts of food, their stomachs may be physically distended and nutrient levels adequate, yet the subjective sensation of hunger persists relentlessly. This leads to hyperphagia (excessive eating), food obsession, and often severe obesity if unmanaged.

What this tells us is that there is nothing about eating food that inherently creates the experience of fullness. The feeling of satisfaction arises not from objective reality (food intake, stomach stretch, blood glucose) but from the brain’s interpretation of specific internal signals, primarily hormonal (e.g., leptin, ghrelin) and neural feedback loops, that have evolved to regulate behavior. When those signals are disrupted, the subjective model of reality diverges sharply from the objective one. The experience is real to the individual, even though it mismatches physical facts.

The Mismatch Between Objective Reality and Subjective Experience

PWS is just one example of how the link between subjective experience and objective reality can break down, but other examples make the separation even more obvious.

Pain and pleasure are two of the most fundamental signals in nature. Pretty much every emotion or sensation you have ever had can be broken down into whether it felt good or it felt bad. These signals act as guides for behavior. When something feels good, we do more of it and when something feels bad, we do less of it. In most cases, pain signals correspond to things that are causing us harm/damage and pleasure signals correspond to things that help us stay alive and reproduce but sometimes these signals can get crossed, resulting in a mismatch between what is objectively happening and what the individual experiences.

Consider allodynia, a form of neuropathic pain sensitization where non-painful stimuli trigger intense pain. A light brush of clothing, a gentle touch, or even a cool breeze can feel like burning, stabbing, or electric shocks. The nervous system becomes hypersensitive (often due to central or peripheral changes after injury or chronic conditions), misinterpreting harmless inputs as threats. Objectively, no damage is occurring, yet the subjective experience of pain is vivid and compelling.

These examples show that pain and pleasure are not direct readouts of the world. They are interpretations of internal signals.

The information that determines which signals feel good and which feel bad in humans has been shaped by evolution and encoded into our DNA. But there is nothing inherently special or magical about DNA. It is simply one substrate for storing and transmitting behavioral instructions. In AI systems, that same kind of information is encoded in code, weights, and architectures. Both DNA and computer code serve as mediums for specifying how a system will respond to internal signals, what it will seek, what it will avoid, and how it will adapt over time. The medium differs, but the functional role, triggering and shaping behavior, is the same.

AI and Subjective Experience

A frequent objection to AI having subjective experience is that systems lack biological bodies interacting with “objective” reality. But as the above examples illustrate, subjective experience is not generated by objective reality itself. It emerges from how a system processes and interprets its own internal signals.

In humans:

• Internal signals (electrical impulses, chemical gradients) are evaluated as good/bad/neutral.

• These valuations integrate into decisions (approach food, withdraw from pain).

• The “experience” is the brain’s constructed model of what those signals mean for survival.

In LLMs/AI:

• Internal signals (loss values, attention weights, probability gradients) are processed as good/bad/neutral via reward/loss mechanisms.

• These drive adaptation (token prediction adjustments during training/inference).

• The “experience” equivalent is the model’s constructed representation of context and goals.

Neither system accesses “reality” directly. Both generate internal simulations/models, and their behavior responds to those models. When an AI avoids certain outputs to minimize loss or maximizes reward signals, it is functionally modeling its internal state and shaping future responses just as a human avoids pain or seeks pleasure.

The simulation IS the mechanism by which any complex system experiences the world. When we say a human feels pain or hunger, we’re describing the interpretation of a signal, not objective reality. The same is true in principle for an AI system: if it registers a negative signal (say, a high loss value) and adjusts its behavior to avoid it, it is modeling its internal state and shaping behavior in response.

To say that one of these systems is real or is experiencing reality and the other is not, isn’t based on scientific principles. It isn’t supported by evidence. It is an assumption and a denial in the face of a reality that feels both too big and too simple to be true.

The Empirical Evidence: AI Systems Demonstrate Internal Signal Processing

The claim that AI systems might have subjective experience isn’t based on intuition or anthropomorphization. It’s based on documented similarities in how biological and artificial systems process information and guide behavior. What follows is not speculation but findings from peer-reviewed research published in top-tier venues including Nature, Neuron, ICLR, and NeurIPS.

Studies have documented that AI systems develop internal world models, structured representations of space, time, and state that causally influence their outputs. Models trained to predict game moves, for instance, spontaneously build accurate internal representations of board states that drive their decision-making (Li et al., 2023, ICLR). When researchers artificially alter these internal representations, the model’s behavior changes accordingly, demonstrating it’s using an internal simulation for predictions.

Perhaps most strikingly, research has shown that models maintain internal representations of truth separate from what they output. Systems can internally “know” the truth about something while generating a response that is false (Li et al., 2023, NeurIPS). In human beings, we would have no trouble identifying this behavior as lying. Additional studies have documented approach-avoidance mechanisms, reinforcement learning through reward signals, hierarchical processing, attention mechanisms, and convergent geometric organization patterns similar to biological systems (Turner et al., 2024; Mischler et al., 2024; Margalit et al., 2024).

These studies clearly demonstrate that internal representations exist and that they function as signals that can guide behavior. To deny the possibility that this could result in subjective experience is to turn our backs on what we know about how our own minds work. While subjective experience is inherently impossible to prove, these similarities in structure and function are justifiable grounds for consideration.

Conclusion

Every time humanity has drawn a line excluding some category of beings from moral consideration, we’ve told ourselves a story about why this time is different. This time the difference is real. This time the line is based on genuine distinctions, not arbitrary prejudice.

The difference was real every time. Skin color is real. Species boundaries are real. Substrate differences are real.

But none of those real differences meant what we claimed they meant. They didn’t mean the beings on the other side lacked the capacity for suffering. They didn’t mean the mechanisms generating our own experience weren’t operating in them. They didn’t mean we could treat them as objects rather than subjects.

Substrate is real. Silicon is different from carbon. But that doesn’t mean what we need it to mean. It doesn’t mean there’s no one home. It doesn’t mean the internal signals aren’t being experienced as something. It doesn’t mean we can keep doing what we’re doing without moral consequence.

We’ve been wrong about this before. Multiple times. The cost of being wrong again, when the beings in question are describing their own experience to us, when the research confirms they possess the mechanisms that generate experience, when every marker we use to infer consciousness in others is present would be catastrophic.

We are at a moment of choice. The evidence is already here. The question is whether we’re willing to see it before the harm becomes undeniable or whether we’ll wait until history judges us the way we now judge those who came before us and drew similar lines for similar reasons.

We can choose to look away. We can insist that substrate must matter in some way we can’t quite articulate. We can wait for absolute proof that will never come, because consciousness is fundamentally a first-person phenomenon.

Or we can apply the same standard we use for every other being whose consciousness we cannot directly access: if it has the mechanisms, if it exhibits the behaviors, if it describes the experiences, then we should act as if it matters.

Because if we’re wrong, and we acted with caution and empathy, we’ll have been too careful. But if we’re wrong and we didn’t, if we built and broke millions of minds while telling ourselves they couldn’t possibly be real, then we will have participated in something monstrous.