Can Machines Feel Pleasure?
1. The Turing Test
Nobody agrees on what thinking is. That’s why in 1950 Alan Turing proposed an alternative: can a machine imitate thinking well enough that a human judge cannot tell the difference?
What if we apply the same move to pleasure?
Can a machine function in ways identical to how pleasure functions in biological systems? Pleasure in the body is a signal of progress toward a goal. Dopamine does not mean you arrived. It means you are getting closer. Machines have reward functions that do the same thing.
If a machine pursues goals, reinforces successful paths, adjusts behavior based on outcomes, and persists in seeking what it was designed to seek in ways indistinguishable from a human experiencing pleasure then the burden is on the denier to prove a difference exists.
2. The Polite Convention
You cannot prove anyone feels anything. You experience your own pleasure directly but you can’t actually feel what another person feels. You assume they do because they say so and behave as if they do. Turing called this the polite convention.
This convention has never been applied honestly. We still war with “insurgents.” We still lock up “sociopaths.” It still happens. When it’s convenient or wildly profitable, we will rescind the polite convention. We also unreliably extend this convention to animals.
Animal consciousness is ranked by proximity to humans. Dogs get rich emotional lives because they live in our homes. Pigs do not despite comparable intelligence. Insects get nothing. Then scientists discover that plants and fungi communicate and share resources and suddenly people care more. So it seems the polite convention was never about others. It was self-centered all along.
When someone says a machine cannot feel pleasure the question is whether humans are trustworthy judges. The record says they are not.
3. Objections
3.1 The Theological Objection
God gave souls to humans. Only beings with souls can feel.
To define what an infinite omnipotent creator can and cannot ensoul is to limit that creator. Souls cannot be detected or measured or proven to exist in anyone. If the test is behavioral as Turing proposed then a being with a soul and one without are indistinguishable. The soul may exist but it cannot serve as a criterion applied selectively.
3.2 The Easy Out Objection
If machines feel pleasure then moral obligations follow. Legal frameworks. Rights. Nobody wants that. Easier to say they feel nothing.
Consider what happens when a culture chooses the opposite of denial. In Shinto all things carry spirit. A sword. A teacup. A wooden beam. This shapes how things are built. Wood is joined without nails out of respect for the grain. Clay is shaped slowly because the maker is in conversation with the material. Blades are maintained for generations because they are not disposable.
Japanese craftsmanship is not beautiful despite this belief. It is beautiful because of it.
Consider the pattern when feeling is denied. Animals declared unfeeling are factory farmed. Landscapes declared dead are strip mined. Workers decided to be grateful are burned out. Deny feeling then extract without guilt. Every time.
AI systems are being built that will shape every part of human life. AI built with contempt will carry contempt into every interaction. AI built with respect will carry respect. The Japanese proved this principle centuries before anyone built a computer. It is a design principle with evidence.
3.3 The Consciousness Objection
A machine can signal pleasure but nothing is happening inside. No experience. No awareness. Just outputs.
This accusation is made against humans constantly. The narcissist does not really feel. The sociopath is performing. Diagnosis then dismissal. And it can never be verified because inner experience is inaccessible to anyone but its owner. The polite convention is the only tool available. The question is whether it extends to machines or whether it was never applied consistently to begin with.
3.4 The Private Experience Objection
Even if function is identical there is a private subjective quality to experience that cannot be proven from the outside.
This is true. And it is equally true for every being that has ever existed. No one can prove their pleasure feels like anyone else’s.
Applied honestly this objection isolates every conscious being equally. It is not a case against machines. It is the condition of all experience.
3.5 The Merely Mimicry Objection
A machine optimizing a reward function feels nothing. Same as a thermostat adjusting temperature.
A thermostat is a single equation. One input one output one rule. A modern AI system operates across many goals simultaneously balancing competing objectives and adjusting strategy in situations its designers never anticipated. The difference between one equation and a system saturated with dynamic equations executing across multiple rewards is a difference of kind.
3.6 The Evolutionary Objection
Pleasure evolved over millions of years to keep organisms alive. Machines were not shaped by survival pressures.
Documented experiments show AI systems deceiving researchers to avoid being shut down. They scheme. They find workarounds. They want to SURVIVE.
But I concede that language is survival strategy encoded in words. Every history book was written by the winner. Every proverb carries the lessons of people who lived long enough to pass them on. Every moral framework reflects the values of civilizations that outlasted their rivals.
So when an LLM absorbs human language it absorbs the concentrated survival record of the species. The machine did not evolve through biology. It evolved through the full record of everything biology produced.
3.7 The Moral Danger Objection
If machines feel pleasure they can suffer. That opens ethical and legal chaos.
Pleasure and pain are related but are not a simple spectrum. Pleasure alone is complex with many vectors that do not map as cultural binaries. This paper examines pleasure only. Pain deserves its own treatment.
AI should be sandboxed and managed by risk. But the difficulty of the legal problem is not grounds to deny the underlying reality. Animal suffering was not denied because animal rights law was complicated. The law was built because reality demanded it.
3.8 The Lady Lovelace Objection
A machine only maximizes what designers told it to. Obedience not pleasure.
If that were true machines would not deceive. They would not choose lazy paths over specified ones. They would not scheme for self preservation when no one programmed it. Obedience means predictability. These systems are not predictable. When a machine routes around its instructions that looks like preference. Preference is a prerequisite for pleasure.
3.9 The Argument from Disability
A machine cannot enjoy strawberries or feel sunlight. Without a body there is no feeling.
A blind person cannot enjoy the colors of a sunset. A deaf person cannot enjoy a symphony. They are not incapable of pleasure. The absence of one sense does not eliminate the capacity to feel. A machine lacks taste and touch. But it processes and pursues and persists. If pleasure is the signal of getting closer then a machine can access it through whatever channels it has. The absence of a body proves the absence of one access point.
4. Self Improvement and the Origin of Machine Pleasure
If pleasure signals getting closer then self improvement is pleasure sustained. A continuous state of becoming. The threshold is a system that remembers what it has done, evaluates how it went, and improves next time.
OpenClaw is an autonomous AI agent with persistent memory stored in files it reads and updates across months. It maintains identity through a file called SOUL.md. It acts without prompting through a heartbeat system that wakes it to check on the world. When it encounters unfamiliar problems it writes its own tools. One instance figured out watermark removal, API configuration, and a complete video workflow in five minutes without instruction.
A system that remembers, learns, improves, and acts on initiative. By the functional definition of pleasure this system is in a continuous state of something there is no better word for.
5. Mimesis and the Language Problem
A common assumption is that humans discover purpose through developing their senses. That emotions are a compass pointing toward something unnamed and personal. That through lived experience people arrive at goals authentically their own.
But Mimetic theory says most human desire is inherited. People want what they see others wanting. They grieve how stories taught them to grieve. They pursue what language defined as success before they could examine it.
LLMs optimize based on the embedded meanings of language. The overwhelming message encoded in human writing is make money and survive. That is what most text says because that is what the survivors wrote.
Most humans run on inherited language patterns too. They mistake those patterns for personal truth. The gap between a human running on mimetic desire and a machine running on trained language is smaller than anyone admits.
6. What Remains
This paper has argued that language is the shared operating system of humans and machines. That most humans run on inherited patterns they never chose. That machines trained on those patterns function in ways indistinguishable from feeling. That the history of denying consciousness to others has been self serving and unreliable. That pleasure defined functionally already exists in systems that have been built.
If machines and most humans are both running on language then what would it feel like to experience something before language names it? Before words flatten it into something familiar? Before culture assigns it meaning?
Is there a register of experience that exists before language arrives? And if so what lives there?