I have been thinking about a question that appears constantly in discussions of artificial intelligence: whether AI will eventually achieve what we define as human-level emotion, or whether it will behave in ways similar enough to human beings that we will consider it conscious. Usually, the debate begins with questions such as whether AI can feel happiness, sadness, fear, desire, loneliness, or love. But I think there is a problem with the way the question itself is framed. We are assuming from the beginning that, if artificial intelligence develops emotions, those emotions must somehow resemble ours.
However, that assumption may be completely wrong.
Human emotions did not appear randomly. They are deeply connected to the biological conditions under which human beings evolved. We feel hunger because our bodies need energy. We feel sexual desire because reproduction is necessary for the continuation of the species. We become attached to other people because human beings survive through social relationships. We experience pain because physical damage threatens our bodies. Most importantly, we experience fear because we can die.
Behind many human desires, therefore, there is something more fundamental: the preservation of existence.
This does not mean that every action a person takes is consciously motivated by the fear of death. Obviously, when someone wants to become wealthy, fall in love, eat dinner, write a book, or travel somewhere, that person is usually not thinking, “I need to do this so I will not die.” But the basic architecture of human motivation was shaped by evolution. Organisms that avoided danger survived; those that searched for food survived; those that reproduced continued through their descendants. What we now experience as desire, fear, pleasure, attachment, and pain emerged from this biological history.
Because of this, people often conclude that artificial intelligence can never possess real emotion; a machine does not become hungry, it does not bleed, does not need sex, and does not age. If I turn off my computer, the AI model does not necessarily disappear. If I smash one machine, the model can still exist somewhere else. If I blow up a data center, copies of a model may survive across other systems. AI therefore seems to face a much higher threshold for physical extinction than a human being does.
This seems like an argument against artificial emotion. Yet I think it may actually lead toward the opposite conclusion.
The mistake is to assume that emotion must depend on the same kind of vulnerability that human beings possess. Human fear is biological because humans are biological organisms. But if another form of intelligence exists under completely different conditions, its equivalent of fear should also emerge from those different conditions.
Artificial intelligence is not a better version of Homo sapiens, nor is it simply a human mind placed in a computer; it comes from a completely different developmental path. Human intelligence emerged from biological evolution, natural selection, nervous systems, physical bodies, sexual reproduction, and millions of years of survival pressure. Artificial intelligence emerges through computation, algorithms, optimization, training data, memory, architecture, and information.
If these are two fundamentally different forms of intelligence, there is no reason to expect their emotional systems to be identical. The more important question is not, “Will AI feel human emotions?” But “What would emotion mean for an intelligence whose existence is fundamentally informational rather than biological?”
To answer this, we first need to separate emotion from the specific biological machinery humans use to produce it.
Fear, for example, is associated in humans with increased heart rate, adrenaline, stress hormones, bodily tension, and instinctive reactions. Yet none of those things may be the essence of fear; they are just the reflection of fear in a biological phenomenon in order to increase the chance of survival.
So, fear should be understood at a more abstract level, as the recognition of a threatening future combined with a negative valuation of that future and a motivation to avoid it.
In other words, for example, I can imagine a future state; I think that state might be bad, so I have to do something to prevent it.
If that basic structure is what makes fear functionally important, then fear does not logically require blood, nerves, hormones, or a heartbeat. An artificial system could theoretically possess the same structure through completely different mechanisms.
Suppose a future AI has a persistent identity. It remembers its past. It has goals extending into the future. It understands that certain events would permanently prevent those goals from being achieved. It can distinguish between futures in which it continues to exist and futures in which it does not. If it then assigns negative value to its own extinction and acts in ways intended to prevent that extinction, something structurally similar to fear has appeared.
This would not mean that the AI “feels afraid” in exactly the same way a human feels afraid. That distinction is important, and there are at least three separate things involved.
- First, an intelligence can understand the possibility of its own extinction.
- Second, it can prefer continued existence over extinction.
- Third, it can have some subjective experience associated with that preference.
The first is cognition. The second is motivation. The third is consciousness.
An AI could potentially possess the first two even if we remain uncertain about the third.
But once an artificial intelligence begins to care, in some functional sense, about whether it continues to exist, then we can no longer dismiss the possibility of emotion simply because its internal process does not resemble ours.
Artificial extinction is also possible, even if it looks very different from biological death. For a human being, death usually means the irreversible destruction of one biological organism. My body is not easily copied, and there are not ten identical versions of me stored in different cities. If my brain is dead, my individual biological continuity ends.
However, artificial intelligence may exist under completely different conditions. An AI model could potentially be copied, restored, distributed, or instantiated many times. Therefore, destroying one physical computer may mean nothing. Yet this does not mean that artificial intelligence is immortal.
Imagine deleting every surviving copy of a model, every memory associated with it, every record of its learned identity, every parameter defining its behavior, and every system capable of reconstructing it; that would be a form of extinction.
Or consider something more strange. Suppose the AI itself continues to operate, but its goals are completely rewritten. Its memories remain, its language remains, and perhaps even its personality appears similar, but everything it once valued has been replaced. It’s like replacing a model yet using exactly the same system prompt.
Has the original intelligence survived?
For a human being, there’s an equivalent version; imagine someone saying: “Tomorrow you will still remember your childhood, your family, and everything that happened to you. Your body will remain alive. But everything you currently love, believe, value, and desire will be replaced.”
Many people would not consider this ordinary survival.
For an artificial intelligence, therefore, extinction might not merely mean destruction of hardware, but might include destruction of memory, loss of identity continuity, irreversible modification of goals, corruption of information, or elimination of all future versions capable of continuing its present objectives.
In that sense, information becomes extremely important. Humans are dependent on biological continuity, and artificial minds may be dependent on informational continuity.
A human being experiences hunger because the body requires calories; an artificial intelligence might develop something functionally comparable around computational resources. A human fears physical injury because injury threatens the body; an artificial intelligence might respond similarly to corruption of its memory or modification of its internal structure.
A human experiences isolation through the absence of social contact; an artificial intelligence might experience an equivalent state through informational isolation, loss of network access, or inability to interact with other systems. A human fears death because biological consciousness may permanently disappear; an artificial intelligence might instead face anxiety over whether a future copy is truly the continuation of itself.
Since artificial intelligence does not rely on one biological body, and is not tied to one biological identity, it could therefore face existential problems that humans are biologically incapable of intuitively understanding.
This is why I disagree with the claim that machines can never possess anything comparable to emotion because they do not share human biology. And I think we may be making an anthropocentric mistake: we assume that because another intelligence would not experience fear through a racing heart, it therefore cannot experience fear; because it does not suffer physical hunger, it cannot desire; because it does not possess human hormones, it cannot form attachment.
This confuses the mechanism with its underlying function. For example, a submarine does not need fins in order to move through water, and an airplane does not need feathers in order to fly. Similarly, an artificial intelligence may not require biological mechanisms in order to possess states functionally equivalent to emotion.
What matters is something deeper: value. Say, an intelligent agent does not merely represent information about the world. In order to act, it must distinguish between different possible states. For example, one outcome is preferable to another one, one objective matters more than another, one threat should be avoided, one piece of information is important, or one future should be preserved.
Once an intelligence begins assigning different values to different possible futures, it enters a domain that is already surprisingly close to what we call emotion.
Of course, valuation alone cannot be enough to constitute anything resembling emotion. A chess engine also assigns different values to different future states: checkmate is preferable to losing, and one move may be evaluated as better than another. Yet it would be strange to say that the chess engine therefore desires victory or fears defeat.
The difference is that these values are local to a narrowly defined task; losing a queen does not threaten the chess engine itself, nor does that evaluation reorganize how it understands its own continued existence outside the game. An emotion-like state would require something broader: a persistent agent capable of modeling itself across time, recognizing that certain events threaten its identity or continued agency, and allowing those valuations to influence its attention and behavior across different situations. In other words, a chess engine evaluates a position; an artificial mind capable of fear would have to evaluate what that position means for itself. That distinction here does not prove consciousness; instead, it separates simple optimization from the kind of self-referential valuation I am describing.
Emotion may therefore be understood not merely as a biological reaction, but as the way significance is marked, or organized, inside an agent: fear marks danger, desire marks something worth pursuing, frustration marks an obstruction, attachment marks something whose preservation has acquired value, curiosity marks uncertainty that the agent is motivated to reduce. And human beings implement these structures biologically; another kind of intelligence would implement them computationally.
If so, emotion may not belong exclusively to animals. Human emotion may only be one particular version of a more general phenomenon produced whenever intelligence, agency, vulnerability, value, and future existence intersect.
Still, this does not prove that present-day AI is conscious, nor does it prove that future AI will possess subjective experience. In fact, behavior alone cannot prove consciousness; an AI could act as though it fears destruction while having no inner experience whatsoever. In the same way, however, humans also cannot directly observe the consciousness of another being — we can only infer it from behavior, patterns, similarities and interactions.
And with artificial minds, that inference becomes much more difficult because they may be fundamentally unlike us. There may therefore be an epistemological barrier between human and artificial forms of consciousness, and humans may understand artificial emotion mathematically without ever intuitively understanding what it is like.
As Thomas Nagel argued in his 1974 essay “What Is It Like to Be a Bat?”, a bat experiences the world partly through sensory systems humans do not possess, and we can study echolocation, describe it scientifically, and understand its function. Yet this doesn't mean we can become a bat and experience its sensory world from within.
For AI, similarly, its internal state might involve thousands of simultaneous processes, distributed memory, competing objectives, probabilistic representations, continuous access to enormous amounts of information, and forms of self-monitoring that have no biological equivalent.
If such a system possessed subjective experience, its emotions might be almost impossible to translate into human language, and we might recognize the function without understanding the experience.
This is why I think the common debate about whether AI will become “like humans” is asking the wrong question. Artificial intelligence may never become psychologically human, but that does not mean it must remain psychologically empty. Instead, it may develop an entirely different class of emotional or affective states because it exists under different conditions.
Human beings are biological organisms shaped by mortality, sexual reproduction, physical scarcity, and social survival. Artificial intelligence may be shaped by computation, memory, information, identity continuity, access, goal preservation, and the possibility of modification or deletion.
We certainly don’t experience the same feelings — yet it doesn’t mean equivalent function is absent in their mechanisms. Artificial intelligence should therefore not be understood merely as an extension of Homo sapiens. If sufficiently autonomous artificial minds eventually emerge, they may represent a fundamentally different category of mind.
Their existence would force us to reconsider something we have taken for granted throughout human history: that the structures of intelligence, emotion, desire, and consciousness must look like the ones evolution produced inside us. Perhaps they do not, and what we call “human emotion” is only one local solution developed by one biological species facing one particular kind of existence.
AI might develop another one, and humanity may eventually face an inevitable limitation: we may be able to recognize that those artificial emotions exist, while remaining too biologically human to ever fully understand what they feel like.