The Experience Machine: What Our Refusal to Plug In Says About What We Value

Bostrom opens with Robert Nozick's experience machine by asking his host to imagine a super-duper neuroscientist of the future who builds a machine capable of delivering any experience you want. You step in, pre-program it however you like, and for the rest of your life your fondest dreams come true: you become a great artist, a great lover, the author of a wonderful life. The price is that you float in a tank, disconnected from the rest of reality, with experiences that are constantly happy but have no connection to anything outside.

Nozick's intuition was that most people would refuse to enter permanently. Many might book a holiday inside the machine, Bostrom notes, but few would want to check out of existence for good. Nozick took that reluctance as an argument against any view holding that what we value is a function of what we experience, since the machine can supply every experience one could want while still striking many people as a life of little value.

The force of the argument is that it locates a gap: if value were purely experiential, the machine would be the best possible life, and for most people it plainly is not. Bostrom frames the next step as an exploration of what else is doing the work — what exactly is missing inside the tank that makes people refuse it.

The host presses on whether the machine could reproduce the ups and downs of life. Bostrom points out that it obviously could, which sharpens the question: what can you not have inside the machine, even in principle? His first answer is real connection with other people; in the original version of the thought experiment, the machine contains only you.

That answer requires a distinction Bostrom keeps returning to, between appearing perfectly real and actually being real. Hand someone flowers inside the machine and nobody actually becomes happy: there is a simulation of a smile, a kind of smiley face that looks entirely convincing to you but is not the sort of simulated creature that appears in the simulation argument. The person you believe you delighted does not exist to be delighted.

A second candidate for what is missing is consequence in the world. Some people would say the life they want is one in which they have a big positive impact on how history unfolds, and that is precisely what checking into a tank forecloses.

Bostrom's most interesting observation about Nozick's setup is how much of the verdict is a status quo effect. People resist unplugging from current reality into a dream machine, but the answer flips when the framing flips: suppose you were told that everything you have experienced so far was the dream, and asked whether you want to disconnect and enter the real world — where you are, say, a farmer in Peru growing peanuts.

Presented that way, plenty of people might prefer to keep the life they know — the travel, the podcasts, the research — rather than bail out into an unfamiliar reality. What looks like a preference for reality over illusion may be partly a preference for the familiar over the leap.

Bostrom's hypothesis is that we are not chasing reality as such, but carrying forward particular projects, relationships and a self-identity whose value is bound up with their continuation. Whether that continuation happens inside a tank or outside one, in Peru or inside a computer, may matter less than it first appears.

Bostrom says he finds himself quite willing to take the leap to being the farmer in Peru, and expects more people to feel that way as virtual reality becomes more realistic. He then clarifies the setup to keep the thought experiment clean: the Peruvian farmer is not another simulation, but the actual world as it was before the machine was ever switched on.

The host objects that this framing washes away the concept of what is real. If an illusion can feel exactly like the real thing, what test could ever settle which side of the plug you are on? Bostrom's reply is that in this case the difference can be stated concretely: in one scenario you are floating in a tank with wires in your head, and in the other you are actually tilling soil in Peru.

He adds the consequence that makes the difference tangible: the peanuts you grow are eaten by other people around the world who buy the export. The two situations are different states of one and the same real world, and which one you would choose is a question about your values rather than about metaphysics.

The host pushes back: inside the vat you can still go farming in Peru, so why does it matter? Bostrom concedes that you could have the experience of farming, but not that any peanuts would actually be grown. The disagreement is about whether peanut-growing inside a simulation counts as peanut-growing at all.

Bostrom is careful not to dismiss simulation wholesale. He jokes that he hopes Peru really has peanut farms, then grants that inside a simulation there is an important sense in which it should all count as real. What survives is a distinction between inside the simulation and outside it, and how much that distinction matters depends on your values and preferences.

Bostrom proposes splitting the thought experiment into versions rather than arguing about a single machine. In the original, only you are inside; if you would refuse that, the refusal tells you something specific about what you care about. Now add other people who are also plugged in and whom you can interact with, and the offer becomes noticeably more attractive.

Add the ability to have important long-term effects on human history — to do something genuinely useful even though you are in the tank — and it becomes more attractive still, because the life now has purpose and consequences. The general lesson is that as you add features, the machine converges on baseline reality, and the comparison stops being machine versus life and becomes one life versus another.

The host raises a consequence Bostrom had not counted: even without other people plugging in or effects on the outer world, you still affect the creatures inside. Suppose a person is waiting for you in there, with fears and hopes of her own; when you unplug, she does not vanish — she carries on with her life in the machine.

Bostrom agrees that in that case she has begun to acquire an independent existence, but insists the answer depends entirely on how she is implemented. His middle case is a static photograph on the wall: you can look at it, but nothing that happens to it matters much, and tearing it up harms no one. A fully realized digital mind is a different matter.

If the companion is implemented at a neural level of detail, with the same behavioral repertoire as you have, then very plausibly she is a conscious person like you are. At that point what you do inside the machine has real consequences for how another mind feels, and the moral arithmetic of checking in changes completely.

This leads Bostrom to a claim that is not obvious: it may not be possible to build a machine that delivers a normal human life — one full of interactions with other people — without also generating consciousnesses corresponding to those people. Because everything depends on which experience machine you are talking about, the question cannot be answered in the abstract.

The intuition is that if you create an entity you perceive and interact with, and the interaction is rich — many days of deep conversation rather than a scripted greeting — the only plausible implementation may be one that instantiates the other person in enough detail to be a second consciousness. Bostrom calls this to some extent an open question.

Faking is clearly possible at the shallow end. A recorded voice already greets people convincingly today, and a superficial exchange gives you nothing with which to distinguish a person from a pre-recorded message or a mindless simulation. The interesting threshold is where the interaction stops being scriptable.

Once you can ask open-ended questions and probe from many different angles, it becomes implausible that every possible answer could have been anticipated in advance. At that point, Bostrom suggests, the most plausible way to realize the thing so that it answers correctly from any angle is a way that also instantiates a conscious mind. Whether the tank can be convincing and empty at the same time remains unresolved.