Brown, J. R., & Fehige, Y. (2014). Thought Experiments. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Fall 2014). Retrieved from http://plato.stanford.edu/archives/fall2014/entries/thought-experiment/

Summary

This entry in the Stanford Encyclopedia of Philosophy gives an outline of what thought experiments are, the history of thought experiments, and the main schools of thought regarding whether thought experiments are useful and how they work.

First, while the entry does not give a specific definition of thought experiments, it does describe some common features of them: visualizing a situation in our imagination, “running” it, seeing what happens, and drawing a conclusion. Thought experiments do not necessarily result in some objective truth (e.g., following the thought experiment about throwing a spear through the edge of the universe, we now know of topologies where the space could in fact be simultaneously finite and unbounded, like a circle). Thought experiments frequently cannot be run as real experiments “for physical, technological, ethical, or financial reasons… but this needn’t be a defining condition of thought experiments”.

One possible taxonomy for thought experiments categorizes them firstly as either constructive or destructive. Destructive thought experiments may illustrate a contradiction in a theory (e.g. Galileo’s falling objects), show that the theory is in contradiction with other beliefs (unrelated to the theory) that we hold (e.g. Schroedinger’s cat), undermine a premise of the though experiment itself (e.g. Thomson’s violinist), or slightly modify the original version of the thought experiment in order to produce an outcome which calls the conclusions from the original thought experiment into question. Constructive thought experiments provide positive support for a theory by illustrating the implications of a theory’s claims (e.g. Newton’s cannonball).

The first stage of philosophical investigation of thought experiments began in the 18th and 19th centuries (Novalis, Hans-Christian Orsted). It was again revived in the beginning of the 20th century, marking the 2nd stage (Duhem, Mach, Meinong), and then again in the first part of the second half of the 20th century (Koyre, Kuhn, Popper), marking the 3rd stage. The current investigation is the 4th stage (Brown, Norton).

There are a number of prominent views regarding thought experiments:

Skeptical objection (e.g. Duhem, Wilkes) — a denial that thought experiments are useful, and that they are no substitute for a real experiment. Most skeptical objections are specific to particular fields, rather than to thought experiments as a whole.
Intuition (e.g. Brown) — one version of the intuition account is Platonic intuition, in which the claim is that what is determined in a thought experiment is “a priori (though still fallible) knowledge of nature, since there are no new data involved, nor is the conclusion derived from old data”. The other version is naturalistic intuition, though it’s not entirely clear to me what this is.
Argument (e.g. Norton) — thought experiments are really just inductive or deductive arguments.
Conceptual constructivism (e.g. Van Dyck, Gendler, Camilleri, Kuhn) — thought experiments enable conceptual change; i.e. they “[help] us to re-conceptualize the world in a new way”.
Experimentalism (e.g. Mach, Sorenson, Buzzoni) — thought experiments are just experiments which generate “uncontrollable images of facts acquired in past experiences with the world”.
Mental model — thought experiments are the manipulation of a non-propositional mental model, rather than a physical model. The mental model account ties into the idea of “literary fiction as thought experiments”. The idea is that fiction brings us to construct a hypothetical scenario and use our imagination to let it play out.

One more recent development in the understanding of thought experiments is the question of whether computer simulations can be thought experiments.

Takeaways

There is something interesting about thought experiments—and mental simulation in general—in that we can imagine scenarios which are physically impossible and which may violate many natural laws. At the same time, as some of Schwartz’s experiments show, we can’t always use imagery in ways that violate natural laws. So why is it that we can in some cases, but cannot in others? There is something interesting going on between the interplay of actual perceptual or motor imagery (which presumably must approximately conform to experience with the world as it is goverened by our sensory modalities rather than higher level cognition) and the capacity for abstract of symbolic thought. I can conceptualize the idea of gravity going in the opposite direction, and even run thought experiments regarding what the consequences would be of that. But I can’t do things like visualize a glass with water in it rotating upside down without the water spilling out. So it seems to me that certain types of thought experiments—particularly ones which violate physical laws—are not using information from low level sensorimotor processes. Or at least not using the full information the same way. So where does that information come from, then? Do we construct more abstract models about how the world works initially based on low-level information, but which are conceptual/abstract enough to be manipulated during the course of a thought experiment?