Issue #13 May 2018

The Problems of Science: What We Do to Things When We Study Them

Over the last two centuries, science has become more modest in its epistemological presumptions. After the failure of logical positivism, cracks appearing in the analytic project, and relatively recent discoveries such as the theory-ladenness of human perception, its original aim of mapping reality as it is independent of human cognizers seems to be as elusive as ever.

For example, Van Fraassen has argued that scientific theories no longer aspire for (capital ‘T’) Truth, but seek to be merely empirically adequate. Moreover, our theory choices are increasingly more a matter of Kuhnian underdetermination rather than Popperian falsification. And, on top of that, figuring out the ‘ultimate character of reality’ (as perhaps even someone as recent as Einstein might have put it) has been swapped for carving nature at its joints and identifying homeostatic property clusters.

One pillar of scientific realism that is still standing, it seems, is the solid boundary between classified entities and classifications. That is, on the one hand there are kinds we discover in nature, and on the other there are entities which we categorize as belonging to this or that kind. These entities, in turn, are not influenced by our classifying them as such. When we identify some universal and subsequently categorize its particulars, we are sorting the mind-independent world (identifying homeostatic property clusters), increasing our knowledge of the universe. In carving out these categories, we are thereby not changing the entities that we are cataloging.

Or are we?

· · ·

Ian Hacking and interactive kinds

The work of the Canadian philosopher Ian Hacking (most notably his The Social Construction of What?) sheds revealing light on this question. Traditionally, the assumption has been that reality does not particularly care for our attempts at dissecting it. Probably, it indeed doesn’t care, but the presupposition that reality is a ‘fixed target’ for us to aim at, might be more controversial than you would have thought.

The reason for this is the existence of interactive kinds.

Interactive kinds are kinds such that the act of formulating predicates corresponding to them is conducive to bringing them into being or to modifying them in substantial fashion, which in turn affects our concepts of them. Due to this feedback effect, which Ian Hacking calls the ‘looping effect’, such kinds are not indifferent to our classificatory practices by which we identify them, categorize them and act on these categorizations. This makes them ‘moving targets’: because our investigations may significantly change them, we might, in turn, be forced to revise our theoretical beliefs about them. Interactive kinds are therefore not “definite classes identified by definite properties” (Hacking 2006, 23).

Are only classifications of humans possibly interactive? This is debated. According to Hacking humans kinds are the only interactive kinds, because awareness of and identification with the classification are necessary conditions for the feedback effect to occur in the required way and only humans are able to be aware and identify with classifications. Others have denied these necessary conditions for the occurrence of the feedback effect in the required way. Cooper (2004) gives some counterexamples: because marijuana is classified as illegal the plants are grown in attics and wardrobes altering their physical appearance. As another example, the characteristics of domestic livestock change over time because particular animals are classified as being the ‘Best in Show’ and are used in selective breeding–sheep and pigs would now look very different if it weren’t for our classificatory practices.

However, such extra-epistemic classifications are not what Hacking had in mind. Legality and ‘best in show’ are, arguably, not properties we ascribe to entities when we seek to figure out what reality is like. Rather, those characteristics are predicated for non-epistemic reasons. Marijuana is classified as illegal because, by doing so, we want people to act in certain ways; it’s not a classification that flows out from our attempts to ‘carve nature at its joints’.

Hacking (1986, 78–79) gives examples which are more faithful to the original spirit of his theory:

“Take four categories: horse, planet, glove, and multiple personality. It would be preposterous to suggest that the only thing horses have in common is that we call them horses. We may draw boundaries to admit or to exclude Shetland ponies, but the similarities and differences are real enough. Arguably the heavens looked different after we grouped earth with the other planets and excluded the moon and sun, but I am sure that acute thinkers had discovered a real difference. Gloves are something else: we manufacture them. I know not which came first, the thought or the mitten, but they have evolved hand in hand. That the concept ‘glove’ fits gloves so well is no surprise; we made them that way. My claim about making up people is that [categories of people] are more like gloves than like horses. The category and the people in it emerged hand in hand.”

Hacking’s insight is that when we classify people as having this or that condition (multiple personality disorder, ADHD, depression, et cetera), we are not only discovering a kind of person (a human being with this or that property or such-and-such brain condition has ADHD), we are creating one as well.

Acting under a description

The reason for this is that humans react to being labeled. Except when we interfere, what things are doing, and indeed what horses are doing, does not depend on how we describe them. But some of the things that we ourselves do are intimately connected to our descriptions. Many philosophers follow Elizabeth Anscombe — who is famous for her theory in Intention (1957) that an agent can only act intentionally in a particular way when she is able to describe her action in that way — and say that intentional human actions must be actions under a description. This is not mere linguistics, for descriptions are embedded in our practices and lives. But if a description is not there, then intentional actions under that description cannot be there either.

What horses, mountains, and microbes are doing does not depend on our words. What happens to tuberculosis bacilli depends on whether or not we poison them with BCG vaccine, but it does not depend upon how we describe them. Of course, we poison them with a certain vaccine in part because we describe them in certain ways, but it is the vaccine that kills, not our words. Human action is more closely linked to human description than bacterial action is. What is happening to the microbes and the patient (or more precise: the biological strata patient’s body) is entirely independent of my correct or incorrect description, even though it is not independent of the medication prescribed. The microbes’ possibilities are delimited by nature, not by words. What is curious about human action is that by and large what an agent is deliberately doing depends on the possibilities of description. Hence if new modes of description come into being, new possibilities for action come into being in consequence.

When we ‘discover’ depression, it becomes possible to be a depressed person and act depressed. Arguably, before this ‘discovery’, that was not possible. That is, before our classification, even if we grant the biological basis of depression, there were undoubtedly myriad of ways of ‘being depressed’. Once the category of depression gets added to a cultural repertoire, as a single condition beneath its myriad expressions, it becomes possible for us to all be depressed in the same way. The expressions of depression then begin to unify. The behavior of people (which is the diagnostic basis for attributing the category of depression at all) changes. This means that we are not classifying a static reality.


This raises an interesting question: when the kind we identify is an interactive kind, are we discovering it, or inventing it? Does ‘the depressed Homo sapiens’ really exist? After all, if it wasn’t for our classificatory practices, things would have been different. The million-dollar question, then, is whether interactive kinds are real. Hacking thinks they’re not, that we “make them up”.

In the rest of this essay, I’ll argue that this is in some sense correct, but question whether non-realness follows from mind-dependence.

But first, it must be noted that the existence of interactive kinds constitutes a challenge to the traditional scientific realist picture according to which the kinds we identify are indifferent to our labeling (Ali Khalidi 2006, 336). One philosophical significance of Hacking’s identification of interactive kinds is that it shows that at least some kinds are not impervious to being conceptualized, named and acted upon (ibid., 340). If this is correct, it gives us reason to think that realism cannot be the correct account for all of our classificatory practices.

Even so, as Ali Khalidi (ibid., 342) points out:

It may be thought that their interactive character is precisely what makes [interactive] kinds non-real, and indeed that this is partly what underpins the distinction between real and non-real kinds. The fact that the realist picture does not apply to such kinds may be taken to be no threat to realism, but rather an indication that [interactive] kinds are not real.”

That is what Ian Hacking seems to think. He self-identifies as a dynamic nominalist, denying that interactive kinds are “recognized” (1986, 78), holding instead that they “come into being” (ibid., 78). He rejects realism about interactive kinds (ibid., 76–78) and repeatedly approvingly cites case studies which indicate that (his examples of) interactive kinds are not “real entities” awaiting “discovery” (ibid., 74). He goes on: interactive kinds are “not determined by the nature of things” (1999, 6–7): we “make them up” (1986, 74), they are “consequences of ways in which we represent the world” (1999, 33). According to Hacking, interactive kinds are not real, but their reality is created (1986, 84). It seems that for Hacking, the fact that interactive kinds are subject to the looping effect vindicates nominalism about them.

However, the lack of reality that is implied by Hacking’s terminology might be too strong, because the non-reality of interactive kinds does not follow from what Hacking says about them. That a kind is interactive — subject to the looping effect — shows that it is mind-dependent: it does not thereby vindicate claims to the extent that we make it up and create its reality and its existence depends on people having thoughts about it.

In other words, mind-dependency of a kind does not entail irrealism about this kind, because not all forms of mind-dependence are subjectivism. Once we acknowledge this distinction, we can see that mind-dependency of a kind is not an argument for non-reality of said kind, because there is a difference between saying that a kind is interactive and would not have existed without our classifications (mind-dependence) and saying that it is interactive and is present solely in our minds and discourse (subjectivism). Although interactive kinds are not defined by definite properties, there is no lack of correspondence between the correctness of the classification of some entity as belonging to some interactive kind and this entity having certain properties.

What is knowledge?

Let’s sum up our assessment of Hacking’s argument.

Interactive kinds pose a challenge for scientific realism because they show that there are kinds which are not indifferent to our classificatory decisions. Nonetheless, I depart from Hacking in thinking that establishing interactivity of a kind does not suffice to establish irrealism about it. By overlooking the distinction between mind-dependence and subjectivism, that inference mistakenly casts the mind-dependence of interactive kinds as an argument for non-realism about them. This is a mistake because interactivity of a kind does not mean that it is not really there: that a kind is subject to the looping effect does not mean that it is made up because this feedback mechanism is not disconnected from properties in the world. The observation that interactive kinds would not have not existed without classificatory practices does not mean that they are merely subjective: they are not unresponsive to human beliefs and actions, but neither are they relative to human minds.

Rather than calling interactive kinds out as unreal, such considerations indicate that the correct response to Hacking’s identification of them might be to rethink the aptness of mind-dependence as a criterion by which to adjudicate the real from the non-real.

Hacking’s work indicates that the ‘spectator model’ of human cognition is a too idealistic picture of science. According to this model, we stand by the sidelines of reality and investigate it without interfering in it. As a spectator, we merely observe what happens. However, this passive role for the knower is probably an inaccurate element of the model. As we increase our self-understanding, it is more and more revealed that knowing is active.

First, knowers have an essential role in determining the objects of their knowledge. The way in which we differentiate objects of knowledge is not set in stone but dependent on the one who seeks knowledge.

Second, more strongly and more contentious, perhaps knowers also inevitably influence the objects of their knowledge and there cannot be any ‘objective’ knowing in the first place. When philosophers seek, in the words of Thomas Nagel, “a view from nowhere”, or as Bernard Williams put it, “an absolute conception of the world”, they are striving after an illusion, because human knowledge cannot be assessed from such an Archimedian point of view.

Maintaining that kinds that are interactive are thereby unreal, is an exaggeration. That said, Hacking’s philosophy provides reasons for thinking that acquiring knowledge is not really something we do from nowhere. Hacking reminds us that reality is not a static object that we investigate. Rather, it is for a large part of moving target because it changes as a consequence of our inquiries.

Maarten van Doorn writes essays about why we believe what we do, how societies come to a public understanding about truth, and how we might do better.

Works Cited

Ali Khalidi, Muhamed. 2010. “Interactive Kinds.” The British Journal for the Philosophy of Science, 61: 335–360.

Anscombe, Gertrude. 1957. Intention. Cambridge, MA: Harvard University Press.

Cooper, Rachel. 2004. “Why Hacking is Wrong about Human Kinds.” The British Journal for the Philosophy of Science, 55: 73–85.

Hacking, Ian. 1986. “Making Up People.” In Heller, T., Sosna, M., and Wellbery, D. (eds.), Reconstructing Individualism: Autonomy, Individuality and the Self in Western Thought. Stanford, CA: Stanford University Press, 222–236.

Hacking, Ian. 1999. The Social Construction of What?. Cambridge, MA: Harvard University Press.

Hacking, Ian. 2006. “Making Up People: Clinical Classifications.” London Review of Books, 28: 23–26.


May 2018


Crooks, Elitists, and the Progress of Philosophy

Daniel Rhodes in conversation with Julian Baggini

The Problems of Science: What We Do to Things When We Study Them

by Maarten van Doorn

Defending the Classical Languages from the Charges of Racism

by Carl O’Brien

Practicing ‘Literariness’: a reminder for philosophers and philosophasters

by Cary Campbell

The Harmonic Void: Descartes’ Extended Substance

by John C. Brady