What Are Thought Experiments?

The basic mechanics

Every thought experiment has roughly the same shape: a stipulation, a question, and a check on whether the answer feels right.

The trolley problem shows how this works. A runaway trolley is about to kill five people. You can divert it onto a side track where it will kill one person instead. Should you? The scenario strips away almost everything that makes real decisions complicated. Relationships, uncertainty, context: all gone. What’s left is a pure structure. One action, two outcomes, a decision. That artificial simplicity is the point.

Philosophers call thought experiments “intuition pumps,” a phrase Daniel Dennett coined. A well-constructed one surfaces something that was already there, a reaction you had but hadn’t articulated, by putting it in a situation where it has to become visible.

What you do next is the philosophical work. If your intuition says one thing and the argument says another, one of them has to give. Either you revise the intuition (maybe it’s a bias, a heuristic, something cultural) or you revise the argument (maybe a premise is wrong, or the conclusion doesn’t actually follow). The thought experiment is what forces the choice.

Where they came from

The practice is ancient. Socrates used hypothetical scenarios constantly, imagining ideal cities, rings that grant invisibility, prisoners in caves, to test whether moral and political intuitions held up under pressure. Plato’s Republic is built around the thought experiment of constructing a just city from scratch, not because anyone was going to build one, but because working through the requirements illuminates what justice actually is.

The German word Gedankenexperiment (thought experiment) was popularized in the 19th century, but the practice goes back as far as recorded philosophy. Aristotle used it in physics. The Stoics used it in ethics. The 11th-century Persian philosopher Avicenna described a “floating man” scenario: imagine you were created in mid-air, blindfolded, without sensory input, unable to feel your own limbs. Would you still be aware of yourself as existing? He argued yes, and used it to argue that the soul is prior to the body.

Galileo demolished Aristotelian physics without building a single piece of apparatus. He imagined tying a heavy stone to a light stone: if Aristotle is right that heavier objects fall faster, the combined object should fall at an intermediate speed, since the light stone slows the heavy one. But it’s also heavier than either stone alone, so it should fall faster. The contradiction disproves the theory. No lab required.

Einstein worked out the foundations of relativity by imagining himself riding alongside a beam of light. Maxwell imagined a tiny demon sorting fast and slow molecules to reverse entropy, an exercise that eventually connected thermodynamics to information theory. Schrödinger put a cat in a box. Not idle curiosities. Some of the most productive instruments in the history of science.

What makes a thought experiment good

A bad thought experiment proves too much. If the stipulation is incoherent, if the scenario can’t even be imagined consistently, any conclusion follows and nothing has been established. If it stays too close to the real world, ordinary considerations flood back in and the scenario hasn’t isolated anything.

A good one is tight. It changes exactly one thing (or a specific cluster of things) and holds everything else fixed. It produces an intuition strong enough to have evidential weight. And it connects back to something that matters outside the scenario, so that however you answer it, you’ve learned something about your actual commitments.

The best ones also last. The trolley problem was introduced by Philippa Foot in 1967. It has generated several hundred academic papers and is more actively debated today than when it was published. The Chinese Room, from John Searle in 1980, looked like an argument about AI at the time. Forty years later, with large language models in the picture, it reads like a live question.

Six categories

Thought experiments cluster around the questions they were designed to probe. This library has 153 of them, organized into six categories.

Why they work for conversation

Most interesting questions about values, justice, and what kind of person someone is can’t be discussed directly. “What do you value most?” produces a rehearsed answer. “Are you a good person?” produces a defensive one. Abstract debates about politics or ethics produce entrenchment because everyone already knows where they’re supposed to land.

Thought experiments approach the same questions sideways. A runaway trolley, a room with Chinese symbols, a box with a cat: the hypothetical gives people enough distance to think without the usual defensiveness. The real question gets through anyway.

You also can’t answer “it depends” and leave it there. The trolley is moving. Someone has to decide. That pressure is what pulls out a real answer rather than a considered one, something closer to what a person actually believes rather than what they’ve decided to believe.

Good conversations tend to happen when someone changes their answer midway through a variant, or notices that two of their answers don’t add up. Having to explain the inconsistency out loud is where the philosophy becomes personal.

How to use one

The scenario is the starting point, not the whole thing. Start by getting agreement on the stipulated facts. The point is to respond to the hypothetical as given, not argue about whether it’s realistic. Then ask for the answer. Then ask why.

The “why” is where the philosophy is. Once someone commits to an answer, the follow-up questions come naturally: what principle is that based on? Does it hold in a slightly different version of the case? Could you universalize it? You don’t need to know the philosophical literature. You just need to keep asking what would have to be true for the answer to be correct, and whether those things actually are.

Most thought experiments come with variants. The trolley has a footbridge version. Teletransportation has a double-copy failure mode. The Chinese Room has the systems reply. These aren’t distractions. They isolate which feature of the original case was doing the work in your intuition. If your answer changes between the original and the variant, the difference between them is what you actually care about.

There’s no requirement to reach a conclusion. The goal isn’t consensus. The thinking is the point.