Imagination and Hypothesis Generation
Joshua Myers is a philosophy PhD student at New York University. He is primarily interested in philosophy of mind, epistemology, and cognitive science and is currently writing a dissertation on the nature and epistemology of the imagination.
A post by Joshua Myers.
Peter Medawar, in his “Advice to a Young Scientist,” writes that
every discovery, every enlargement of the understanding, begins as an imaginative preconception of what the truth might be. The imaginative preconception—a “hypothesis”—arises by a process as easy or as difficult to understand as any other creative act of mind; it is a brainwave, an inspired guess, a product of a blaze of insight. (Medawar 1979, p. 84)
Medawar gets things exactly right. The imagination plays a crucial role in hypothesis generation: to form a hypothesis is to imagine a way the world could be. Despite being highly intuitive, and despite many scientists, like Medawar, explicitly stating that they use their imagination to come up with hypotheses, this view has received surprisingly little philosophical attention.
In this post, I’ll motivate the imaginative account of hypothesis generation, explicate some of the mechanisms by which the imagination plays this epistemic role, and briefly speculate about norms on hypothesis generation.
Although the word “hypothesis” is most at home in discussions of the scientific method, humans form hypotheses all the time. For example, suppose you hear a loud thump outside the door to your apartment. You might immediately start wondering about what caused such a loud noise. Could it have been the neighbors who live down the hall slamming their door? Perhaps someone has just rather clumsily delivered the package you ordered a few days ago. Or, maybe your neighbor’s cat has once again managed to sneak out of their apartment and knock your potted plant over.
As this scenario illustrates, humans are prolific hypothesis generators. Immediately upon hearing the loud noise, and with very little effort, you have already generated three plausible hypotheses about its cause, which will then go on to guide your inquiry. Both science and our daily lives are rife with similar examples.
Hypothesis generation is puzzling because it has two contrasting features. It is simultaneously inventive and informed.
Hypothesis generation is inventive insofar as it involves the construction of novel ideas. Hypotheses represent possibilities that one may not have ever considered before. Furthermore, hypotheses are constructed and not learned. One does not generate hypotheses by gaining new information about the world, but instead by considering various ways the world might be. Upon hearing the thump, you don’t hypothesize that it was caused by your neighbors slamming the door because this conclusion is supported by your evidence. You form this hypothesis precisely because your evidence massively underdetermines what the cause could be, and you need to consider the various possibilities.
Hypothesis generation is informed insofar as prior information constrains which hypotheses are generated. Although hypothesis generation involves considering various ways the world could be, one does not simply consider possible states of the world willy-nilly. For example, upon hearing the loud thump outside of your door you would not usually hypothesize that an elephant or an alien caused it. Instead, your hypotheses are informed by your evidence. Although we often form hypotheses that are false, we do not usually form hypotheses that are outrageous, absurd, or wildly implausible.
The fact that hypothesis generation is both inventive and informed puts constraints on a plausible account of hypothesis generation. On one hand, beliefs are a bad model for hypothesis generation because they fall towards the informed end of the spectrum. One cannot construct new beliefs at will precisely because beliefs are highly informed by one’s evidence. On the other hand, a random idea generator is a bad model for hypothesis generation because it falls all the way on the inventive end of the spectrum. A random idea generator might be able to construct lots of new hypotheses, but they will almost all be extremely implausible since they are in no way constrained by one’s evidence or the target of one’s inquiry. Thus, a theory of how humans generate hypotheses must locate a cognitive process which is capable of being both inventive and informed in the right way.
Imagination is a faculty that can be inventive and informed in just the right way. Imagination allows us to generate representations of indefinitely many possibilities that we have never considered before. Yet, imagination can also be informed. I can constrain my imagination in various ways in order to limit the space of possibilities that it represents. The fact that imagination has both of these features is reason to think it is well-suited for hypothesis generation.
Constraints have been widely invoked in the epistemology of the imagination in order to explain how imaginings are capable of justifying beliefs (e.g. Kind 2016, 2018, Kind and Kung 2016). If the imagination is constrained to accurately represent the actual world, one can use it to test hypotheses and form justified beliefs about the actual world.
I propose that constraints on the imagination also allow us to generate hypotheses in the first place, thereby elucidating an overlooked epistemic role that they play. There are a number of hallmark constraints on hypothesis generation that can be implemented by the imagination.
First, people tend to generate hypotheses that are possible given their evidence. Of course, after formulating a hypothesis one may very well go on to encounter evidence which contradicts it. But, in order to play their role in guiding inquiry, hypotheses must be left open by one’s evidence at the time that they are formulated. For example, you would not hypothesize that the neighbor’s cat caused the thumping noise if you already had excellent evidence which ruled this out.
Second, people tend to generate hypotheses that are likely given their evidence. This is stronger than the previous constraint. The hypothesis that an elephant caused the thumping noise does not contradict your evidence. However, your evidence makes it exceedingly unlikely, and it would be a waste of time to pursue inquiry into whether there is an elephant in your hallway. This makes it useful to constrain one’s imagination via your evidence about what is likely. Notice, however, that the probabilistic constraint still leaves plenty of room for inventiveness. The threshold that determines which hypotheses count as likely given one’s evidence may depend on contextual factors. But in most ordinary contexts, it will be low enough that many hypotheses will count as likely given one’s evidence, and thus allow for many hypotheses to be generated.
Third, people tend to generate hypotheses that meet certain explanatory desiderata. These come in two kinds. First, there are general explanatory virtues. These include parsimony, causal adequacy, explanatory depth, and others. For example, when imagining what might have caused the thumping noise, one will tend to imagine simpler scenarios, and one will tend to imagine things which could conceivably cause a thumping noise. Second, sometimes the nature of the target of inquiry suggests constraints on a satisfying explanation. For example, children tasked with explaining what causes a continuous variable to change over time tend to hypothesize changes in other continuous variables as the cause, as opposed to changes in discrete variables (Magid, Sheskin, & Shulz 2015, see also Tsividis, Tenenbaum, & Shulz 2015) In this case, the continuous nature of the explicandum, instead of general explanatory virtues, puts substantive constraints on what a satisfying explanation will look like.
To sum up: imagination is well-suited to play the role of generating hypotheses. It has all the properties that one would expect the capacity underlying hypothesis generation to have. Imagination is a genuinely inventive capacity which is capable of forming indefinitely many novel representations. Nevertheless, imagination is capable of being constrained such that those novel representations can be informed by our evidence and explanatory desiderata. Thus, there is good reason to think that imagination is the mechanism by which we form hypotheses about the world. I’ve further motivated and elaborated this view by explicating some of the specific constraints which are operative on imaginative hypothesis generation, but this is a fruitful area for further work.
Let me conclude by briefly noting a potential normative implication. It is commonly thought that, while there are norms governing which hypotheses we ought to believe, there are not norms governing how we ought to generate hypotheses in the first place. Nevertheless, the imaginative account shows that hypothesis generation is not an inexplicable “blaze of insight,” but rather a process that is responsive to evidence and is subject to constraints This suggests that one can implement those constraints in ways that are either well or poorly suited for guiding inquiry, and thus that there are norms governing this process. Intuitively, it would be irrational to generate hypotheses by imagining scenarios which contradict one’s evidence, or which are highly unlikely given one’s evidence, or which would have very little explanatory value. In these cases, it is natural to think that one ought to have constrained one’s imagination in different ways in order to generate hypotheses better suited for guiding inquiry. If this is right, then there are norms governing how we ought to generate hypotheses and, by extension, how we ought to imagine.
References
Kind, Amy (2016). Imagining under constraints, in Amy Kind and Peter Kung (eds.), Knowledge Through Imagination. Oxford University Press, 145–59.
Kind, Amy (2018) How imagination gives rise to knowledge. In F. Dorsch & F. Macpherson (eds.), Perceptual memory and perceptual imagination. New York: Oxford University Press.
Kind, Amy & Kung, Peter (2016) Introduction: the puzzle of imaginative use, in Amy Kind and Peter Kung (eds.), Knowledge Through Imagination. Oxford University Press, 1-37.
Magid, Rachel, Sheskin, Mark, and Schulz, Laura (2015). Imagination and the generation of new ideas. Cognitive Development. 34:99–110.
Medawar, Peter (1979). Advice to a Young Scientist. Harper & Row.
Tsividis, Pedro, Tenenbaum, Joshua, & Schulz, Laura (2015). Constraints on hypothesis selection in causal learning. Proceedings of the 37th Annual Cognitive Science Society.