†Truth and Serving the Biological Purpose
Since most human biological traits are selected by evolution, the semantic
representation relation may also be selected by evolution, but obviously, false
beliefs with survival value can arise in biologically normal situations. Therefore,
there is a genuine puzzle. I first argue that teleosemantic answers to the puzzle
offered by Papineau and Millikan are inadequate. Then, I offer an answer based
on a structural theory of content proposed in a previous paper. It suggests that
evolution selects only the content of the most primitive components of inner
representations. Moreover, false beliefs with survival value for the entire bio-
system can be biologically normal, because semantic content and truth is
determined by a subsystem of the entire bio-system.
1. Papineau and Millikan on False Beliefs with Survival Value
Teleosemantics (Papineau, 1993; Millikan 1993, 2004; Neander 2004) tries to characterize the semantic representation relation between inner representations and
† The research for this paper is supported by Chinese National Social Science Foundation (grant number 05BZX049). I would like to thank Princeton University and my advisors John P. Burgess and Paul Benacerraf for the graduate fellowship and all other helps they offered during my graduate studies at Princeton many years ago. Without them, my researches would not have started. * Department of Philosophy, Peking University, Beijing 100871, China.
external entities or states of affairs by referring to the biological functions of inner representations in serving the ultimate biological purpose of preserving genes. However, our commonsense intuition is that truth is not equal to „whatever that serves the purpose
of preserving genes‟, for there are literally false beliefs with great survival value even when the humans and environments involved are completely normal biologically. One of
the major challenges for teleosemantics is to explain away this apparent counter-evidence against teleosemantics‟ fundamental claims.
Papineau‟s teleosemantics tries to characterize the satisfaction condition of a
desire by referring to the biological function of the desire, that is, as the condition that the desire is evolutionally designed to bring about. Papineau (1993, pp. 61-65) admits that a coward‟s desire to avoid any scratch in battle is harmful and evolution compensates it with the false belief that entering the battle will not get any scratch. That is a case where a false belief plus a harmful desire has survival value for the entire bio-system. This analysis of how evolution works is certainly correct, but it shows exactly that we cannot differentiate between truth and falsity by referring to the outcome of the entire bio-system alone. That is, we cannot differentiate between the cases where one has a harmful desire compensated by a false belief preventing the desire from being satisfied, and the cases where one has a healthy desire with a true belief bringing about the satisfaction condition of the desire, because they have exactly the same outcome for the entire bio-system.
More abstractly speaking, the situation is this. We have a bio-system S, and
intuitively, we recognize three related parameters b, d, and q for the system. b can take
values „true‟ or „false‟, indicating that the system S is having a belief that is true or false;
and d can take values „satisfied‟ or „unsatisfied‟, indicating that the system S is having a
related desire whose satisfaction condition obtains or not obtains; and q can take values
„good‟ or „bad‟, indicating that the desire is healthy or harmful. The problem for us is that the combination (b=„true‟ and d=„satisfied‟ and q=„good‟) may generate the same
outcome for the entire system S as (t=„false‟ and d=„unsatisfied‟ and q=„bad‟) will
generate. Therefore, by referring to the outcome of the entire system alone (i.e. whether or not it survives), we cannot differentiate between the two cases.
It does not help to call one of the two types of cases „special cases‟, because we
cannot even divide all the cases into two different types without an independent
characterization of „true‟, „satisfied‟, or „good‟. If we had an independent characterization
of „satisfaction condition‟, then we could define a healthy desire as a desire that helps the entire system to survive when the satisfaction condition obtains, and similarly define a harmful desire as a desire that helps the system to survive when the satisfaction condition does not obtain. Then, we could further define a true belief as a belief that helps the system to survive when combining with a healthy desire, or as a belief that damages the system when combining with a harmful desire. Similarly, if we had an independent characterization of „healthy‟, then we could define a desire‟s satisfaction condition as the
condition that helps the system to survive when the desire is known healthy, or as the condition that damages the system when the desire is known harmful. Moreover, if we had an independent characterization of „true‟, then we could define a healthy desire as a
desire that will help the system to survive when combining with true beliefs, and then we could further define the satisfaction condition of a desire. With any one of the three, we can characterize the others. However, we cannot get off the ground without any independent characterizations of any one of them.
Papineau (2003) suggests that a desire D represents the external condition X
whose failure is a malfunctioning of D. „Malfunctioning‟ here means a malfunctioning of
a particular desire as a desire for something, not a malfunctioning of a desire as a
component of the entire bio-system in the sense of not helping the entire bio-system to
survive. For instance, not dead is a malfunctioning of the desire to die as „a desire to die‟,
but it is not a malfunctioning of the desire as a component of the entire bio-system in the
sense of not helping the bio-system to survive. In the latter sense, the desire is actually not malfunctioning when committing suicide fails. Therefore, this „malfunctioning‟ refers to something beyond the outcome of the entire bio-system. However, what does „the
malfunctioning of a particular desire as a desire for the desired‟ mean? It seems that it
can only mean „the condition represented by that desire does not obtain‟. Therefore, this
characterization of desire is essentially circular. It cannot help us to get off the ground.
This seems to indicate a general problem for this kind of teleological approaches. That is, if we refer only to observable and obviously non-intentional characteristics of the entire bio-system, e.g. its survival, then we cannot distinguish between different intentional inner states, namely, true beliefs plus satisfied healthy desires vs. false beliefs plus unsatisfied harmful desires. However, if we try to refer to some property of a particular intentional inner state, for instance, a particular desire‟s malfunctioning as a
desire for something, then it actually presupposes the representation relation and it does not really offer anything new in characterizing that representation relation.
Millikan‟s teleosemantic account relies on the concept of Normal mechanism or
Normal explanation and the concept of semantic mapping function between inner
representations and external world affairs (Millikan 1993, 89-90; 2004, Chapter 5, 6). It
is alleged that the semantic mapping function is selected by evolution, and the evolutionally selected Normal mechanism guarantees that the consumers of the representations will exploit the semantic correspondences to generate responses with survival value. This appears to have reduced the semantic norm to the biological norm. That is, when Normal mechanism is working, there are no false beliefs with survival
value and truth is just whatever that serves the ultimate biological purpose. This appears to be able to define truth by referring only to non-intentional characteristics of the entire bio-system (i.e. its survival), at least under the assumption that Normal mechanism is
However, it seems that this is based on an ambiguity in the concept of „Normal mechanism‟. If a Normal mechanism is any mechanism that happens to be selected in the
evolutional history in the universe, then intuitively it is obvious that there are false
beliefs with survival value when Normal mechanism is working. These may include some
literally false religious beliefs that encourage ethical social behaviours, or some superstitions that happen to protect people from dangers. One may claim that the mechanism is not „Normal enough‟ in such cases. “Systems whose jobs were to distort certain beliefs would have to ride on more general systems whose jobs were to produce true beliefs” (Millikan 2004, pp. 86). However, if an accidental natural holocaust
occurred and humans were extinct before another „more normal‟ mechanism ever
evolved, then this would be the „most normal‟ mechanism in the universe. Now, if the „most normal‟ mechanism that ever existed in the universe is still not „normal enough‟,
there must be some implicit and un-stated extra normativity condition in „being a Normal mechanism‟. What could that be except for „being able to get truths, real truths, not just
to help preserving genes‟? This will be another interpretation of „Normal mechanism‟, that is, a Normal mechanism is a mechanism that helps us to get truths when everything
is Normal. If that is what „Normal mechanism‟ means, then it is circular for characterizing truth.
Resorting to some „ideal‟ normal mechanism does not help either. What is the
standard for being „ideal‟? If Homo sapiens‟ current biological status is still not „ideal‟,
what could make it biologically more „ideal‟? Is it wiping out all other species so that
human is the only species that survives, or is it instead living harmoniously with other species? Intuitively, the latter does not require humans to know many literal truths about nature. It may even require humans to know fewer truths about nature (considering the possibility of a human-made nuclear holocaust). Therefore, an unstated condition is still that an „ideal‟ mechanism is a mechanism that helps to get more literal truths, not just to
help humans to survive better. Resorting to „ideal‟ presupposes some direction to project onto some „ideal future‟, and any condition for characterizing that direction for our purpose here must guarantee that it gets more truths and is not just a religiously, ethically, or otherwise more desirable direction. That actually presupposes the concept of truth and it cannot be used to characterize truth.
In general, to guarantee truth, we need a normativity condition that can exclude the cases of false beliefs with survival value as abnormal. The intuitive conviction that there is such a condition is perhaps right, but it actually comes from our semantic intuition that there is a distinction between true beliefs and false beliefs, independent of their effects on the entire bio-system. The problem for us is that we cannot state that condition without circularity, by referring to non-intentional biological conditions of the
entire bio-system (e.g. its survival) alone. The situation is similar to Papineau‟s case. If
we had an independent characterization of „true‟, then we could define „Normal mechanism‟ as the mechanism that gets truths, not just the mechanism that helps the
entire bio-system to survive (for an Abnormal mechanism can also help the system to survive combined with false beliefs). Similarly, if we had an independent characterization of „Normal mechanism‟, then we could define truth as whatever we obtain under Normal mechanism, not just whatever that helps the entire bio-system to survive (for false beliefs can also do so in Abnormal situations). We cannot get off the ground without any independent characterizations of any of the two.
Another idea could be that true beliefs with survival value are in some sense more stable than false beliefs with survival value. However, it is again unclear if one can offer a characterization of that stability condition in non-intentional terms, so that it can comeback to define truth.
More generally, in sciences, we normally start with postulating internal structures and then refer to the structures to explain global behaviours of the system, for instance, a system‟s degree of adaptability to its environments. We normally do not refer to global behaviours to characterize internal processes if we can avoid it, and if we have to do so, we will consider that branch of science immature. Behaviourism in psychology used to be that, but in contemporary cognitive psychology, we certainly prefer postulating internal neural or higher-level cognitive mechanisms to explain observed cognitive behaviours. The same perhaps applies to characterizing the semantic representations. It might not be true that it is logically impossible to characterize the semantic representation relation using teleological terms or other terms referring to the behaviours of the entire bio-
system only. If referring to the survival of the entire bio-system alone cannot differentiate between the true and false beliefs, one might try to consider more characteristics of the entire bio-system. The real point is that it is perhaps a better scientific methodology to postulate the internal structures of inner representations and human cognitive architectures, and then use them to characterize the semantic representation relation and to explain how false beliefs with survival value can arise in biologically completely normal situations. That is, pursuing a structural theory of content is perhaps a better scientific methodology.
On the other side, the idea that some sort of biological normativity guarantees truths is not completely off the mark. At least for some simplest beliefs critical for human survival, such as beliefs involved in recognizing foods or immediate dangers, biological normativity should guarantee truth, for otherwise humans would not have flourished. The genuinely difficult question for a theory of content is to describe how exactly evolution selects truths for simple beliefs but truth in general can diverge from serving the biological purpose for complex beliefs. This also implies that we must characterize the contents and truth conditions of complex concepts and beliefs using terms other than teleological terms. Only with an independent characterization available can we identify when and where the two diverge.
2. A Summary of a Structural Theory of Content
Ye (2007) proposes a structural theory of content. It postulates structures for inner representations such as concepts, and postulates structural semantic mapping rules for characterizing the representation relation. Here I will give a brief summary of the theory.
Then, in the next section, I will come back to explain the role of evolution in selecting the semantic representation relation according to this theory, and explain how false beliefs can also serve the biological purpose while granting that the basic semantic
representation relation is selected by evolution.
According to the theory (Ye 2007), basic constituents of concepts are inner maps,
which are like perceptual mental images. An inner map is composed of map points and
map attributes on points. An inner map can represent an object (or more accurately, a 4-dimentional time-section of an object), with map points representing perceptually very small parts of the object. There are three types of map attributes on a map point. (1)
Sensory property attributes represent the sensory properties of a small part of an external object corresponding to the map point, such as its colour, brightness and so on. (2) Absolute spatiotemporal attribute represents the spatiotemporal location of the small part relative to the body hosting the inner map. (3) Relational spatiotemporal attributes
represent the spatiotemporal relations between two small parts. Map attributes are the simplest, atomic inner representations. The entire inner map represents (a time section of) an object in case the map points and the map attributes on the map points can be successfully mapped to the small parts of the object and their sensory properties, their locations relative to the body hosting the inner map, and their mutual spatiotemporal relations. Therefore, the content of an inner map is determined by some structural and compositional semantic mapping rules, together with the semantic mappings of map attributes as the most primitive inner representations.
Concepts belong to a higher-level in the hierarchy of inner representations. A basic concept is a composition of inner maps, and basic concepts are in turn components
of composite concepts. Most lexical concepts (e.g. PET, FISH) are basic concepts. The theory integrates the summary feature list theory and the exemplar theory of concepts (Murphy 2002; Laurence and Margolis 1999) and assumes that a basic concept consists
of a list of summary features and a collection of exemplars. A summary feature is again a
concept and can be a composite concept expressed by a linguistic expression. A summary feature represents a property of the things represented by the concept, including their perceptual or functional properties but not limited to these. An exemplar consists of a list of summary features (again!) and an inner map. An exemplar typically represents an
object instance that one encountered and applied the basic concept before, in learning or using the concept. Typically, the inner map in an exemplar encodes one‟s perceptual memory about the instance, and the summary features in the exemplar encode one‟s
conceptualized memories about other properties of the instance. For instance, my concept DOG may contain summary features such as „can bark‟, „has four legs‟ and so on, and it
may contain an exemplar representing my neighbour‟s dog, with an inner map
representing its appearance and with the summary feature „belonging to my neighbour‟
describing that instance.
The theory assumes that a concept can have other meta-attributes encoding information about the concept itself, not the things represented. In particular, a concept has a meta-attribute indicating its type, namely, whether it is an essentialist concept,
functional concept, perceptual appearance concept, or singular concept. The type of a
concept decides its compositional semantic mapping rules.
Summary features and exemplars in a basic concept have weights assigned. The semantic mapping rule for a basic concept says that an entity is represented by the basic