The epistemology of thought experiments: A non-eliminativist, non-platonic account


Several major breakthroughs in the history of physics have been prompted not by new empirical data but by thought experiments. James Robert Brown and John Norton have developed accounts of how thought experiments can yield such advances. Brown argues that knowledge gained via thought experiments demands a Platonic explanation; thought experiments for Brown are a window into the Platonic realm of the laws of nature. Norton argues that thought experiments are just cleverly disguised inductive or deductive arguments, so no new account of their epistemology is needed. In this paper, I argue that although we do not need to invoke any Platonic insight to explain thought experimentation, Norton’s eliminativist account fails to capture the unique epistemological importance of thought experiments qua thought experiments. I then present my own account, according to which thought experiments are a particular type of inductive inference that is uniquely suited to generate new breakthroughs.

This is a preview of subscription content, access via your institution.

Fig. 1


  1. 1.

    By “should” I do not mean that fulfilling these desiderata is a necessary condition for a good account of how we can gain knowledge from a belief-forming process; for some processes, it may be impossible to fulfill all three to our satisfaction. For instance, there is tension between the Modeling and Parsimony Criteria that might require a trade-off. Rather, I mean that an account that fulfills these criteria enjoys an explanatory advantage such that it is prima facie more plausible than an alternative account that does not.

  2. 2.

    Brown’s argument, if sound, would show that universal or otherwise robust versions of empiricism about scientific knowledge are false. Presumably, it would not impugn more limited versions of empiricism.

  3. 3.

    Galileo presents this thought experiment in the “Second Day” of Galilei (1962, original 1632).

  4. 4.

    Thank you to Malcolm Forster for making me aware of the similarity of these two thought experiments. For an illustration of Einstein’s elevator experiment, see Brown (1992, p. 272). For a more detailed discussion, see Norton (1991: 137–138).

  5. 5.

    Perhaps the most famous thought experiment in quantum physics is the Einstein-Podolsky-Rosen Paradox. I doubt that this would count as “Platonic” in Brown’s sense since it does not explain the phenomena that it points out, and indeed, there has been no definitive explanation of paradox yet. See Brown (1991) for his discussion of the paradox, and Norton (1993: 37) for a response.

  6. 6.

    A helpful analogy here was suggested by an anonymous referee. In the duck/rabbit illusion, we can switch between perceiving a duck and a rabbit in the same picture, and this perceptual shift is achieved by focusing attention on different parts of the picture.

  7. 7.

    I will consider this argument in more detail in Section 6 when I discuss my alternative account and will argue that my account fares better than Norton’s at withstanding Brown’s criticism here.

  8. 8.

    Gendler (1998) levels a similar claim against Norton’s view. She argues that for some reasoners, Galileo’s falling bodies thought experiment can lead to knowledge that its argumentative reconstruction would not.

  9. 9.

    Norton’s (2004) primary line of attack against Brown’s Platonic view is that it can only account for the reliability of thought experiments by transforming them into arguments and analyzing their logical or inductive soundness. Norton takes this to be evidence that thought experiments are simply disguised arguments (1146).

  10. 10.

    When judging theories on the basis of multiple criteria, some of which might conflict with each other, it is often impossible to decisively demonstrate the superiority of one theory over another. I owe the useful talk of “plausibility points” to David Enoch (2011).

  11. 11.

    It is unclear whether Brown takes this argument to establish a conclusion this strong or merely to provide an explanation of how intuitive knowledge of nature might be possible; that is, it is unclear whether Premise 6 states a genuine logical entailment or merely a plausible analogy. The second interpretation is supported by Brown’s statement that “laws are abstract entities, so they could be perceivable, too”. However, in the next breath, he makes the stronger claim, “Laws and numbers are both outside of space and time. If we can see one, then we should be able to see the other” (Brown 2004: 1131, italics mine). Given that Brown does accept a strongly Platonic explanation of thought experiments, I learn toward the stronger interpretation of his argument. Thank you to an anonymous referee for helpful clarification on this point.

  12. 12.

    That is, they are important in the context of justification and not merely important in the context of discovery. Norton does acknowledge that thought experiments can be useful heuristically, providing a way for people to grasp otherwise difficult scientific concepts. He may agree that thought experiments are important in the context of discovery but insist that they do not have any importance in the context of justification over and above that of the argument forms that they conceal. On my view, thought experiments, qua thought experiments, have importance in the context of justification by being particularly reliable instances of inductive inferences.

  13. 13.

    Brown also emphasizes the importance of isolating essential properties in the running of a thought experiment, characterizing the process as a type of natural kind inference (1991, 45–46). While I do not think it is necessary to invoke natural kind properties to explain thought experiments, nothing in my account precludes it. My account seeks to explain how reasoners isolate projectible features of the experiences that they call upon in thought experimentation, and this is where most of my disagreement with Brown’s theory lies.

  14. 14.

    Perhaps “simulates” is too controversial for my purposes here since I wish to remain somewhat agnostic about the actual psychological process of agents performing thought experiments. I think that my account would not change substantially if one were to describe these steps as involving description, mental modeling, or other processes as opposed to simulation.

  15. 15.

    The Simulation and Idealization steps of my account share a close resemblance with Tamar Gendler’s theory of thought experiments. Gendler (1998, 2004) emphasizes the “quasi-sensory intuitions” that thought experiments raise, arguing that “their contemplation may bring us to new beliefs about contingent features of the natural world that are produced not inferentially, but quasi-observationally” (2004, 1154). I agree that the Simulation and Idealization steps allow us to arrive at beliefs about our experiences that we would not have otherwise, and that this component of the thought experiment may be non-inferential. However, my view differs from Gendler’s in its emphasis on the inferential step that applies this knowledge about experienced cases to new, unexperienced cases.

  16. 16.

    I am grateful to an anonymous reviewer for helpful clarification on this point.

  17. 17.

    There might be other processes by which she could arrive at the knowledge that is ultimately yielded by the thought experiment. For an extreme example, she could directly observe that the earth moves around the sun and that cannonballs launched from ships on the earth land back at their source. Conceivably, there are other inductive arguments that might also be effective, though these arguments will not be epistemologically identical to the particular thought experiment in question.

  18. 18.

    On my view, then, there are several ways for a thought experiment to go awry. If an individual lacks the requisite experience, the thought experiment will not get off the ground. If she inaccurately simulates and idealizes from known cases, then she will not have isolated the appropriate feature of her experience. This is where reasoners go wrong in the rod and slotted plate case described by Norton (2004: 1140). They fail to include in their simulations the proper kinematics, including the Lorentz transformation that eliminates the supposed contradiction in relativistic motion. I conjecture that this failure stems from our inability to observe this kinematic feature in our everyday experience. Lastly, the thought experiment could fail if a correctly isolated observed case is not of the same kind as the unobserved case to which knowledge about the former is extended.

  19. 19.

    I tentatively conjecture that my framework can be fruitfully applied to some thought experiments in philosophy, particularly ethics. Take, for example, Alistair Norcross’s (2004) Fred thought experiment. Individuals are asked to imagine themselves witnessing the horrific treatment of puppies by Fred and to reflect on their intuitive judgments of the rightness or wrongness of Fred’s actions. Then, it is argued that Fred’s behavior is identical in all of its morally relevant features to another case–the commercial farming of pigs–and therefore, judgments about the former should also apply to the latter. Trolley cases, Judith Jarvis Thomson’s violinist case, and Robert Nozick’s experience machine experiment follow a similar structure. While the first steps of these thought experiments may not have been actually experienced, they are in a sense closer to what individuals have experienced or can imagine experiencing. I am less confident that my account applies to thought experiments such as Gettier cases or Goodman’s grue case that appeal to theoretical judgments more than actual experience. I am grateful to an anonymous referee for inquiring about this point.

  20. 20.

    See Part 1 of Einstein (1920) for a discussion of the light clock and other issues regarding relativistic time.

  21. 21.

    Brown might deny that this thought experiment demands a Platonic explanation since it can be derived from the two postulates of special relativity: the principle of relativity and constancy of the speed of light. On this interpretation, the thought experiment does not prompt any new insight into the laws of nature. This may be the case for someone who is already well-versed in special relativity, but I think my account sheds light on how the thought experiment leads someone without this prior scientific knowledge to arrive at new knowledge about the nature of time. I am grateful to an anonymous reviewer for pointing out this alternative conception of the light clock experiment.

  22. 22.

    It only predicts the time dilation phenomenon when paired with auxiliary assumptions about the speed of light, how time is measured, etc.


  1. Brown, J. R. (1991). The laboratory of the mind: Thought experiments in the natural sciences. London: Routledge.

    Google Scholar 

  2. Brown, J. R. (1992). Why empiricism won’t work. PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association, 2, 271–279.

    Google Scholar 

  3. Brown, J. R. (2004). Peeking into Plato’s heaven. Philosophy of Science, 71, 1126–1138.

    Article  Google Scholar 

  4. Einstein, A. (1920). In R. W. Lawson (Ed.), Relativity: The special and general theory, trans. London: Methuen & Co Ltd. Project Gutenberg.

    Google Scholar 

  5. Enoch, D. (2011). Taking morality seriously. New York: Oxford University Press.

    Google Scholar 

  6. Galilei, G. (1962). Dialogue concerning the two chief world systems, Ptolemaic and Copernican. Second revised ed. Berkeley: University of California Press.

    Google Scholar 

  7. Gendler, T. S. (1998). Galileo and the indispensability of scientific thought experiment. The British Journal for the Philosophy of Science, 49, 397–424.

    Article  Google Scholar 

  8. Gendler, T. S. (2004). Thought experiments rethought—and reperceived. Philosophy of Science, 71, 1152–1163.

    Article  Google Scholar 

  9. Goodman, N. (1979). Fact, fiction, and forecast. Cambridge: Harvard Press.

    Google Scholar 

  10. Godfrey-Smith, P. (2011). Induction, samples, and kinds. In J. K. Campbell, M. O’Rourke, & M. H. Slater (Eds.), Carving nature at its joints: Topics in contemporary philosophy, V. 8 (pp. 33–53). Cambridge: MIT Press.

    Google Scholar 

  11. Norcross, A. (2004). Puppies, pigs, and people: Eating meat and marginal cases. Philosophical Perspectives, 18, 229–245.

    Article  Google Scholar 

  12. Norton, J. D. (1991). Thought experiments in Einstein’s work. In T. Horowitz & G. J. Massey (Eds.), Thought experiments in science and philosophy (pp. 129–144). Savage, MD: Rowman and Littlefield.

    Google Scholar 

  13. Norton, J. D. (1993). Seeing the laws of nature (review of James R. Brown, The Laboratory of the mind: Thought experiments in the natural sciences). Metascience, 3, 33–38.

    Google Scholar 

  14. Norton, J. D. (1996). Are thought experiments just what you thought? Canadian Journal of Philosophy, 26, 333–366.

    Google Scholar 

  15. Norton, J. D. (2004). On thought experiments: Is there more to the argument? Philosophy of Science, 71, 1139–1151.

    Article  Google Scholar 

Download references


I am grateful to two anonymous referees, Malcolm Forster, Elliott Sober, Larry Shapiro, Reuben Stern, Michael Goldsby, Naftali Weinberger, and an audience at the Boulder Conference on the History and Philosophy of Science for valued comments on earlier drafts.

Author information



Corresponding author

Correspondence to Hayley Clatterbuck.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Clatterbuck, H. The epistemology of thought experiments: A non-eliminativist, non-platonic account. Euro Jnl Phil Sci 3, 309–329 (2013).

Download citation


  • Thought experiments
  • Galileo’s ship
  • Platonism
  • Empiricism