Kornblith wants to give an account of knowledge as a natural kind. But first, what is a natural kind? Often, a natural kind is defined as a category that is required both for the explanation of observed phenomena and to make predictions. It is possible then that knowledge is a natural kind if it is need to explain and predict the behavior of animals (and people). One problem with this account of knowledge as a natural kind is that knowledge will disappear given more intelligent scientists. Here I am specifically focusing on the requirement that knowledge be required for explanation. If we accept a materialist position then we accept that all of an animal’s behavior does ultimately have microphysical causes, and that in general an animal’s behavior could be determined from information about its genetics and environment. A super-intelligent reasoner could then deduce from an animal’s DNA what kind of brains would develop, and in which ways it is likely to act given its environment. The super-intelligent reasoner would have no need to invoke knowledge to explain behavior, since unlike us they have no need to resort to abstractions in order to deal with the complexity of the animal’s brain. One way to resolve this problem would be to broaden natural kinds to include anything that could be used to explain and predict, but to do so would be to allow such unnatural things as gravity ghosts to be classified as natural kinds (spirits which are responsible for gravitational phenomena).
Kornblith, however, does not define a natural kind this way. He argues that a natural kind is a collection of properties that is resistant to change. Unfortunately this definition is no better, because it would deny that unstable chemicals and elements are natural kinds. Although it is true that such things are rarely found in nature, we certainly don’t want to claim that they are unnatural or not a kind. Perhaps then a better definition of a natural kind is a classification based on o-physical properties, where an o-physical property is defined as any property that is required for a theoretical explanation of phenomena or a property that supervenes on such properties. This definition doesn’t eliminate the possibility of knowledge since, even if all behavior can be explained in terms of neural activity. knowledge can still be said to supervene on that activity. Gravity ghosts, however, are ruled out, since they have properties that do not supervene on the fundamental physical properties.
Now that we have defined what a natural kind is we can make the first step towards understanding knowledge as a natural kind; understanding belief as a natural kind. A belief, according to Kornblith, is the needs of an animal combined with a representation of the external world. This may be fine for animals, but it doesn’t seem to quite capture what we mean by belief, at least when we talk about people. For example, I might believe that I own a certain book, and although this does have to do with a particular representation of the world it certainly doesn’t involve my needs. Similarly, we may also have beliefs such as “if Goldbach’s conjecture is true then the only odd untouchable number is 5”, which don’t seem to involve any representation of the external world at all.
Later, when Kornblith is explaining why plants turn towards the sun for more light, an idea is proposed that could be seen as the basis for a better definition of belief, specifically that a belief is information used to inform multiple behaviors. This statement is basically equivalent to a more traditional definition of belief, that a belief is a disposition to act as though some statement was true (where the statement might be something about the world, such as “berries are good to eat”, or something more abstract). Kornblith’s “second” definition of belief picks out the same beliefs picked out by the more traditional definition because if an informational state affects multiple behaviors then what we have is a disposition to act as if the informational state were accurate, and it is easy to see that the content of the informational sate can be captured by some statement, and thus they are a disposition to act as if some statement was true, the other definition of belief.
Once we have an understanding of belief as a natural kind we can build upon that to describe knowledge as a natural kind that is a subset of belief. Knowledge, according to Kornblith, is a true belief generated by a reliable process. By reliable Kornblith does not mean reliable for a particular animal in a particular situation, but reliable on average for the members of its species in their natural environment. There are two problems, though, with this account of knowledge. The first is a problem that arises for many theories of truth and knowledge, specifically the question “is this theory itself knowledge?” I would have to say that, by Kornblith’s own criteria, it is not. I will admit that it is a belief, and it may even be true, but to be knowledge it would have to be generated by a reliable process. However, we know from the history of philosophy that the process by which members of our species produce definitions of knowledge is highly unreliable, since there are many definitions, and only one of them can be right. For similar reasons none of our theories in mathematics and science would be deemed to be knowledge, since there are always many failed theories that come before a successful one.
The second problem is slightly more foundational. It is a premise of Kornblith’s that cognitive ethologists can identify some natural kind of animal thinking as knowledge. I don’t doubt that they can indeed identify natural kinds of animal thinking, but I do doubt that they have special access to the concept of knowledge that allows them to know that one of the natural kinds they have identified is knowledge as we understand it. It seems possible that there may be in reality three natural kinds of thought: “belief”, “knowledge”, and “proto-knowledge”. What I am proposing is that it is possible that cognitive ethologists have correctly identified a natural kind, but incorrectly labeled it knowledge when it is really proto-knowledge. What needs to be done is to define what exactly each of these categories is, and then see which one the natural kind cognitive ethologists have discovered is, almost the reverse of Kornblith’s methodology, which seems to be to look for natural kinds and then call the most likely candidate knowlegde.
What both these objections hinge on is that our conception of knowledge is intrinsically normative, knowledge is something we want to have, and to say something is knowledge is to recommend it. Thus we expect a definition or explanation of knowledge to instruct us how to acquire knowledge and how to separate real knowledge from beliefs we only mistakenly think of as knowlegde. For example, traditional accounts of knowledge can provide a description of justification that can fill this role (but they fail for other reasons). Kornblith’s account, on the other hand, is essentially descriptive. Although it tells us that we should have reliable processes as the basis for our belief formation mechanisms, it doesn’t tell us how to identify reliable processes or to improve existing ones, because to do so would require pre-existing knowledge about the reliability of processes, leading to an infinite regress from which we would have to conclude that it was impossible to acquire knowledge. Perhaps then Kornblith, like Quine, would be happy to simple settle for such a descriptive account, but as Fred Dretske said, “If you can’t make one, you don’t know how it works.” Thus I would argue that if a definition of knowledge can’t describe how to acquire knowledge it hasn’t really explained what knowledge is.
1 “I want to claim that knowledge is, in fact, a natural kind,” Kornblith, Knowledge and its Place in Nature, page 61
2 “… I take natural kinds to be homeostatically clustered properties, properties that are mutually supporting and reinforcing in the face of external change. … Why does H2O count as a natural kind? … The chemical bond that joins these atoms provides the newly formed unit with a degree of stability that is not found in just any random collection of atoms.” Kornblith, Knowledge and its Place in Nature, page 61
3 “… a physical property is a property which either is the sort of property required by a complete account of the intrinsic nature of paradigmatic physical objects and their constituents or else is a property which metaphysically (or logically) supervenes on the sort of property required by a complete account of the intrinsic nature of paradigmatic physical objects and their constituents. … Let us say that any property which is physical by the lights of an object based conception is an o-physical property.” Daniel Stoljar, “Two Conceptions of the Physical” (2001)
4 “Once we recognize the existence of internally represented animal needs together with representations of features of the environment, we have the beginnings of a belief-desire psychology. … Just as we explain human action by attributing beliefs and desires as causes of action, we explain the behavior of a wide variety of animals as the causal consequence of their beliefs and desires.” Kornblith, Knowledge and its Place in Nature, page 38
5 “… the sole role that that information plays is in getting the plant to move in the direction of the light; the information about the sunlight is not available for other, more diverse, informational interactions. When you and I come to believe that sunlight is present in a particular direction, however, that information is available to interact with our other internal states so as to inform an extremely wide range of behavior. It is this fact about us that requires talk not merely of a representational system of information-bearing states, but talk of beliefs.” Kornblith, Knowledge and its Place in Nature, page 42
6 “The concept of knowledge which is of interest here thus requires reliably produced true belief.” Kornblith, Knowledge and its Place in Nature, page 58
7 “Knowledge on this view first enters our theoretical picture at the level of understanding of the species, rather than the individual.” Kornblith, Knowledge and its Place in Nature, page 57
8 There is also a Gettier-like case for Kornblith’s definition of knowledge as follows: a group of people is raised for many generations in an artificial environment where everything written in red is true. Thus these people come to have the reliable belief-formation mechanism of simply accepting everything they see that is written in red. One day one of these people comes across a magic 8 ball, and asks: “will I have soup for lunch”. The magic 8 ball displays “Yes”, written in red. And in fact the person does have soup for lunch. Was this true belief, formed by a reliable process, knowledge? However, I won’t actually propose this objection, since Kornblith would probably argue that our intuitions about knowledge were faulty in this case.
9 Specifically: since Kornblith’s account doesn’t inform us as to how to acquire knowledge it ends up being unable to recommend itself as knowledge (objection 1), and because knowledge is by its nature normative cognitive ethologists haven’t been working with the correct understanding of it; they see knowledge as simply the beliefs that we would judge to be “knowledge” given what the animal senses (it might seem like the animal should believe them, but since we consider only the true beliefs of the animal, and we know more about the truth than the animal does, of course it seems like the animal should believe them, after all they are true; this is not the position we are in though when it comes to knowledge, we don’t have special access to the truth except through knowledge), but knowledge also includes a criterion of what should be knowledge given what the animal senses, which they omit since they have no need of it (objection 2).
10 Fred Dretske, “A Recipe for Thought” (1994)