A Puzzle about Belief
In order to tackle the problem of meaning in general, we need to isolate some key concepts that philosophers use to isolate exactly what human (at least) understanding consists of. Philosophers make a distinction between epistemology (study of human knowledge) and linguistics (how symbols function as representations of other things like ideas or objects in the world). One way to describe this distinction is to say knowledge is internal and subjective while language is external and objective. This means that humans (as least) have an internal, subjective experience of the world that they take to be true. This experience cannot be objectified per se. Language, however, is open to anyone who can understand the symbols. So we’ll need to start our investigation drilling into belief and we’ll start our exploration of belief with a puzzle.
In his paper, “A Puzzle About Belief”[1] Saul Kripke contrasts the Millian view of names[2] with the Frege-Russell view and in doing so discovers what seems to be a problem that is not easily solved on either theory. Kripke largely is concerned with how names function in belief contexts. While an analysis of proper names will play a role in a larger theory about belief, it will not be our only concern. Still, I think examining Kripke’s puzzle will raise some important facts about belief that will set up the broader discussion. We will return to proper names throughout our analysis of belief as names play a key role in some types of Gettier cases.
According to Kripke, Mill held that proper names are simple: they pick out or refer to the thing that bears it. Names on this theory should not be associated with any properties of its bearer; they do not describe the objects that bear them. On the Frege-Russell theory, names are linguistic containers for definite descriptions. They are associated with bundles of properties possessed by the things that bear them. These properties constitute the sense of the name.
On Mill’s view, Mohammad Ali and Cassius Clay refer to the same individual and can be used salva veritate interchangeably. The two proper names pick out a single person and so can be substituted in sentences without changing the truth value of those sentences. On the Frege-Russell view however, referent of names are determined by descriptive properties the speaker identifies with the name. Aristotle, when described as the teacher of Alexander the Great and when described as the philosopher who studied under Plato, has two different senses though they do refer to the same object (p. 240). The way of describing Aristotle–the sense–however, does fix the reference and so the same name cum unique description could be said to refer to different objects. If Plato turned out to be the teacher of Alexander the Great, the name ‘Aristotle’ would refer to Plato since he satisfies the description.
In the first section of the paper, Kripke explores the Millian view he presented in Naming and Necessity[3] that names, but not descriptions, could be substituted salva veritate in modal contexts. He acknowledges that his position could be seen as decidedly non-Millian in belief contexts because in these contexts, it seems that neither names nor descriptions can be substituted and be guaranteed to preserve truth value. While arguments against Mill to tend to (sometimes strongly) favor Frege-Russell, Kripke cautions that the reason one ought to be critical of Mill’s position may not be due to the success of Frege-Russell. The reason is that common terms usually have common senses and these common senses could be identical even though the believer doesn’t identify the referent of the term. A person could believe that Cicero is bald and Tully is not and, when asked who Cicero and Tully are, respond with, “a famous Roman orator.” The description–the sense–is the identical but the believer does not understand that Cicero and Tully are the same famous Roman orator. This argument is meant to show that the failure of interchangeability of codesignative terms in believe contexts is not due to a difference in the senses of the terms due to counter-examples that show otherwise.
In section two, he presents a difficulty. If Kripke is correct in following Mill–“if reference is all there is to naming, what semantic difference can there be between” two names that reference the same thing? Is it possible for a person to believe some proposition is true without believing another proposition with a codesignative name is true? Could one believe that Allen Konigsberg directed Crimes and Misdemeanors and disbelieve that Woody Allen did? In order to address these questions, he introduces three principles:
The disquotational principle: If a normal English speaker, on reflection, sincerely assents to ‘p,’[4] then he believes that p.[5] That is, if a person assents to a statement, then that person believes that the proposition expressed by that statement is true.
The strengthened disquotational principle: A normal English speaker who is not reticent will be disposed to sincere reflective assent to ‘p’ if and only if he believes that p. (249). This principle establishes that assent indicates belief and that a lack of assent indicates a lack of belief.
The principle of translation: If a sentence of one language expresses a truth in that language, then any translation of it into any other language also expresses a truth (in that other language).
Given the disquotational principle, he posits that the principle of substitutivity is incorrect. For an English speaker using the language normally can believe ‘Cicero was bald’ and ‘Tully was not bald’ without contradiction (given that he may not know that Cicero and Tully are the same person). This problem, when applying the disquotational principle, seems to tacitly support Frege-Russell. But there are problems. If the substitution of codesignative names lead to contradictions in propositional contexts, it’s not clear how Frege-Russell solves them (see above — a believer can apply the same indefinite description in order to fill out the sense of a name and still not know that the two names refer to the same individual). Further, Kripke’s ‘puzzle’ will show that this mystery can be invoked by not using substitution of names at all. It appears in normal context when the general principles of disquotation and translation alone are applied.
In this section Kripke presents his puzzle. A “normal” French speaker, Pierre, comes to assent to the sentence “Londres est jolie” through looking at lovely pictures of the city and what he has heard. Sometime later, he moves to a particularly unattractive part of London not realizing that London and Londres refer to the same city. None of the people in his neighborhood know French so Pierre learns English by the “direct method” of not by translating French terms into English terms. Over time he learns the name of the city inhabits and assents to the sentence “London is not pretty.” Of course, while assenting to this sentence, he still assents to the French sentence about London above. What, then, does Pierre believe about the city? It would appear that he believes both even though they are contradictory.
One way to solve the puzzle might be to acknowledge that Frege and Russell were correct. If Pierre learns to identify both London and Londres by certain definite descriptions that are uniquely identifying he will be forced to conclude that both terms refer to the same city. But this need not follow. For if Pierre comes to learn each of the terms used in the definite descriptions in each language by direct method, he may not realize that the describing terms themselves have the same referent. The description of London in English would be just as isolated doxastically as the description of Londres in French. Kripke argues that such problems can exist even for natural kind terms which would seem to have a prima facie resistance to this problem.
Finally, he argues that the puzzle arises even in situations where the same name is used in the same language. He considers Peter who learns of a man named Paderewski and that he was a famous pianist (and who also was a statesman but this is a fact Peter did not come to learn). He assents to the sentence, “Paderewski had musical talent.” Then later, in a different context, he comes to learn of a man named Paderewski who was a Polish Prime Minister. Assuming that this man has the same name as the famous pianist but skeptical of the musical abilities of politicians, Peter assents to the sentence, “Paderewski had no musical talent.” Here, Peter believes that the referent of the name “Paderewski” is satisfied by two individuals even though metaphysically they refer to the same man. What does Peter believe?
Kripke’s concern is not with the “conventional judgment” that belief contexts are referentially opaque but with whether codesignative proper names are interchangeable salva veritate in belief contexts (like they are in modal contexts). Even if they are “Shakespearian” in this sense, he doesn’t believe that this is enough to establish the Frege-Russell theory of reference over Mill’s. Whether Kripke’s puzzle succeeds in demonstrating this is somewhat tangential to my purpose. What Kripke’s paper demonstrates is that there is a difference in the way reference is established in belief contexts and modal or logical contexts and this is the relevant point.
Kripke’s puzzle essentially is this: a believer can appear to assent to two contradictory propositions and believe them without contradiction. This is because the truth conditions of a proposition are not always accessible from the first person point of view. Kripke demonstrates that the truth conditions and relations that may be analyzable in logical, “third-person” analyses may not hold in belief contexts because beliefs are irreducibly phenomenological. As Kripke argues in Naming and Necessity, this is partly because in logical contexts we establish the truth conditions by linguistic fiat; we can describe propositions and their relations to other propositions and the world in whatever way suits our cases as long as those relations don’t violate logical rules. This is only possible if belief as a phenomenological act is distinct from a third-person, linguistic description the propositions believed. What is contradictory logically and even metaphysically, may not be contradictory doxastically which gives weight to the idea that belief that a proposition is true is constrained in ways that a logical analysis is not.
This puzzle illustrates, we think, that there is a distinction between how terms relate to objects (sense) in logical and belief contexts. Since computers can only operate in logical contexts, there will, by necessity, need to be a distinction in what we mean by “meaning” in these two contexts. It may not be possible (and we know is in fact many times not possible) to determine truth conditions for a given human’s belief about a term (as evidenced by SFRs for example). The same problem will not exist at the purely algorithmic level for computers since logical relations in computer science either operate deterministically or probabilistically where the probabilities are at least in principle 100% determinable given an exhaustive understanding of boundary and background conditions. This allows us to scope the content problem and the problems the semantic web is designed to solve to a very particular set of problems. We’ll return to this below.
To further illustrate the distinction between “meaning” in belief contexts and in logical contexts, we want to consider the work of philosopher John Searle, namely his work on intentionality and his famous (or infamous depending on your view) “Chinese Room” argument. In his most famous book, Intentionality, Searle explores the question of what it means for a belief to be about something. His work here provides the computer scientist with a rich framework for thinking about problems in computer semantics as well as providing a foundation for thinking about human-to-computer interaction particularly when it comes to content.
In our view, a theory of intentionality provides the best foundation on which to explore these questions. Put simply, intentionality is a theory of how a mind can be directed at some object or state of affairs external to it. Intentionality as a theory covers much more than belief and a full explication would include intentional states like desires, emotions (like fear, joy, elation, love, hate, depression) and things like admiration and aspiration. Belief is a type of intentional state and the only one that will concern us here. John Searle has done significant work in articulating a full theory of intentionality and it is his work that I will examine here. His theory is grounded on his work in linguistics (notably his treatise on speech acts in his book with the same title) and so incorporates both intentionality and linguistic representations of mental phenomena.
To introduce Searle’s theory of Intentionality I will begin by looking at his analysis of naming and reference. I start here both to relate Intentionality, by way of transition, to what has been said by Kripke but also to apply the theory to a specific problem before I examine the theory per se. Both Searle and Kripke view themselves as interlocutors in the Mill vs. Frege-Russell debate over naming and reference and so their differences will both serve to illustrate the particulars of the problem they are trying to solve and in so doing bring into even clearer relief the specific properties of belief that are important for our study.
In the last chapter of his book Intentionality, Searle applies his theory of Intentionality specifically to the problem of naming. In his treatment, he contrasts the view of Mill (which he calls “causalist”) with Frege’s which he calls descriptivist. He believes a main question in a theory of reference is how a person succeeds in referring to an object.[6] He says Mill’s answer is just that a name refers and that’s it. A name may refer to an object as a theoretical truth but the actual practice of referring by a doxastic agent and the success associated with this act may involve description.
According to Searle, Kripke’s picture of the causal theory of names denies that identity is established by associating a name (“Allen”) with a unique description (“the person that directed Crimes and Misdemeanors“). If this were true, in possible worlds where Allen didn’t direct Crimes and Misdemeanors he would fail to be Allen. Rather, a name is established by a “baptism” in which a sign (name) is associated with an object through ostension or with a description. Reference to the object named depends on this baptismal event.
The causal theory says that reference is established by a direct casual chain that originates with the baptism and the name is passed through the chain to each person who references the original object in terms of that causal chain. The baptism establishes the object as the recipient of the name and the name is passed through the chain as the referent of that object. [7]
Searle counters that the introduction of the name in the baptism is “entirely descriptivist” and that the causal account is an Intentional account and not an externalist one. By this he means that the “referring” is done by an Intentional act of believing the name is causally connected to an object at the end of the chain. “What fixes the reference is an Intentional content which may or may not also have an external causal connection to the object.”[8] Further, Searle says that on Kripke’s view, “each speaker must intend to refer to the same object as the person from whom he learned the name.”
For Searle, the type of causation on which the causal theory depends must be externalist in nature. While he disagrees with externalist causation, he does allow for a “descriptivist” causal connection which is Intentional. An Intentional causal connection is one where the object satisfies some Intentional state and the name is established through that satisfaction. This is the type of causation involved when an object is baptized to have a specific name (particularly through ostension).[9]
Searle’s descriptivism is an extension of Frege’s account of reference. Frege’s theory of reference involves intentional content and merely external causation. In order to account for how a proper name refers to an object we need to show how the object satisfies or fits the “descriptive” Intentional content that is associated with the name in the minds of speakers; some of this Intentionality will “normally be expressed or at least be expressible in words.”[10] He is clear that there is little distinction between the descriptivist and the causal theorist in the way an object satisfied a referent when it comes to the act of baptism. Both the causal theorist and the descriptivist hold that the person baptizes an object with a name by an Intentional act: an object in the world satisfies some Intentional content in the mind of the believer which the believer baptizes with a name.
According to Frege, when one speaks of the reference of a sentence, one is interested in its truth value. That is, one is interested not merely in speaking the sentence for the sentence’s sake (as in, say, an actor repeating a line on a stage — Frege’s example) but in whether the reference of the object exists. Because of this, signs can be replaced in sentences (assuming he means here signs in sentences in whose reference we’re interested) without altering its truth value (salva veritate). “If our supposition that the reference of a sentence is its truth value is correct, then the latter must remain unchanged when a part of the sentence is replaced by an expression having the same reference….the truth value of a sentence containing another as a part must remain unchanged when the part is replaced by another sentence having the same truth value.” [11]
Frege argues that in the case of “that” clauses, the subordinate phrase that follows the word that does not always have a reference and thus cannot be replaced by a phrase that does have a reference and retain truth value. In some cases, the phrase following a that clause indicates a thought and so the entire sentence, not just the subordinate phrase is included in the truth value of the sentence. These may be true in cases of belief.
“Copernicus believed that planetary orbits are circles” is a sentence whose truth value is about the belief of Copernicus and not one about planetary orbits. Thus replacing the subordinate phrase with “Copernicus believed that the apparent motion of the Sun is produced by the real motion of the Earth” does not affect its truth value because both sentences are true given that these are two things Copernicus believed. The reference of the entire sentence is the belief of Copernicus and the truth value of the subordinate phrase does not relate directly to the truth value of the referent of that phrase.
When one uses words used in the ordinary way, they mean to speak of the word’s reference, the object. But one can be talking about the words themselves or their sense.[12] Frege then appears to make the same distinction that Chisholm and Kripke make and one that Searle argues is central to understanding reference: reference is established through Intentional content in the mind of the believer and its relation to objects in the world and not merely by relationships between words. All belief is belief de re.
Further, counterexamples to descriptivism fail because they generally only refer to what “the agent might say and not at the total Intentional content he has in his head….Each counterexample is designed to show that a speaker will refer to an object in the utterance of a name even though the associated definite description is not satisfied by that object or is satisfied by something else or by nothing.”[13]
Where Kripke holds that only names are rigid designators, Searle allows that at least some definite descriptions can be rigid designators when those descriptions are used in identity conditions.[14] If they are then they can function like names in reference cases (Searle briefly addresses modal cases but that is not the primary concern here). The definite description, “the man who has 10 coins in his pocket” functions as a rigid designator when the intentional content of the person who believes that the reference is satisfied by the actual person who is described by those properties.[15]
Whether Kripke’s theory of names or Searle’s is correct is not the main concern here. This analysis is meant to illustrate the recurring theme that belief contexts differ from logical or linguistic or modal contexts in fundamental ways. For Searle, Intentionality is essential for understanding reference and as we’ll see shortly, beliefs are irreducibly Intentional. Intentionality establishes the doxastic connection between propositional content and the world and should provide the context for analyzing truth conditions in belief contexts. Smith’s belief that the man who has ten coins in his pocket will get the job needs to be analyzed differently than the proposition, the man who has 10 coins in his pocket will get the job. In order to understand why, we must turn to Searle’s theory of Intentionality.
Intentionality
Intentionality is a primitive in Searle’s ontology. An intentional state is a property of mental states (not sentences) in which the mental state has as its content some object or state of affairs (“the specification of the mental state . . . requires the specification of some object or state of affairs”) which is not identical with the mental state.[16] As with Chisholm, Searle views mental states as representations of the objects and states of affairs outside the mind. The state is neither identical to those objects and states of affairs nor are they fully captured in a linguistic token which represents those objects and states of affairs.
Searle observes that intentional states can be stated as intensional[17] sentences. “John believes God loves him” as a mental state of John has God as his intentional object (it specifies God as the object of his belief) but the sentence specifying the belief is intensional because it may fail the first test of extensionality.[18] He states this as a puzzle: how can an intentional state give rise to an intensional sentence and what is the relation(s) between an intentional state and the objects they are about?
He responds to this puzzle by arguing that intentional states (not the intentional object of those states) just are representations of the objects (state of affairs) they are about just as speech acts represent the objects (state of affairs) they are about. When Tom says “It is raining”, that statement is meant to represent a state of affairs that it is raining even if it is not in fact raining. The relation, then, is between the mental state qua representation and the actual state of affairs. The identity relation is between the representation and the state of affairs and so the mental state does not need to meet the extensibility requirements.[19] The sentence that specifies the mental state (say a belief) can be true even if the referring terms in the sentence cannot be generalized existentially (e.g. from the sentence “Tom believes in Santa Claus” one can infer there exists an x and that x=Santa Claus). “John believes God loves him” (read as “John has as an object of his mental state that there is a God and the state of affairs that God loves him) is true even if God does not exist.
The objects of intentional states do not need to have some odd ontological status (what is a unicorn as an object of thought?). Rather, intentional objects of a mental state are the actual objects and states of affairs the mental state is of and are represented by an intentional state. If the actual object doesn’t exist (the unicorn) the intentional state qua representation of the object still exists and this Searle calls the representative content of a mental state. The Intentional object is the actual object or state of affairs toward which the mental state is directed.[20] Jones has belief that it is raining. The object of his belief is the state of affairs it is raining and this constitutes the intentional object of the intentional state. The belief itself is a representation of that state of affairs and is what directs the mental state towards the actual state of affairs. The object of his belief is not the representative content but the actual state of affairs. If it is not raining (the state of affairs does not obtain), the belief has no intentional object though it does have representative content.[21]
Searle holds that an Intentional state is a representation of objects and states of affairs. Intentional states are the means by which a mental state is directed towards its object. An Intentional state is directed at an object or state of affairs and not at its representative content. That is, the mental state is not about the representation. The representation is the content of the mental state and the mental state is about the actual thing in the world. Searle incorporates propositions into his description of representative content and seems to identify propositions with the representative content itself. A proposition is the content of belief, not the object of belief.[22] A propositions is what a belief is, not what a belief is of. This is the distinction between intensional and intentional readings of propositional content.
Conditions of Satisfaction
It’s critical to note that Searle is not attempting to give his readers an ontology of belief. When he speaks of representative content, he is not attempting to describe what that content is made of or in what it consists. He emphasizes that his use is logical not ontological. By this he means representation does not involve a picture of the belief or ‘meaning’ or the like. Rather, representative content means that the Intentional state has propositional content which determines a set of conditions of satisfaction, that its mode (Intentional states are in a psychological mode which I take to mean the type of thing it is) determines the direction of fit (mind-to-world) of the propositional content.[23] “The key to understanding representation is conditions of satisfaction.”[24]
A state of affairs is not a particular and does not relate to belief in any way like a particular. Beliefs represent states of affairs and other intentional states (he uses the example of love and hate) represent objects. When John loves Sally, the object of his intentional state is the actual thing Sally. A belief that it is raining does not have a particular thing as the object of that belief but rather a set of conditions which the belief represents and can be “correct” about or not. The belief stands in a representative relation to those conditions just as a true statement that it is raining stands to those conditions as truth conditions. His idea seems to be that the intentional object of beliefs is not a particular thing that exists but set of conditions that, as a whole, obtains. Put another way, states of affairs are not functionally interchangeable with names in intentional contexts.
Searle compares Intentional states to speech acts. When one performs a speech act, the statement is true if it corresponds to a state of affairs in the world. He calls this a “condition of satisfaction” or condition of success.[25] Similarly, Intentional states are satisfied when the direction of fit (mind-to-world) corresponds. It is this sense in which a description of the conditions of satisfaction is a logical one and not an ontological one. The Intentional state itself determines what must be the case—what actual state of affairs must obtain or what object must exist—in order for the belief to be true. A state of affairs obtaining is not the condition for truth in belief. Rather, the truth conditions must be understood in terms of the belief. The content of the belief fully determines the truth conditions and a state of affairs that happens to be true of a description of a belief is not identical to the belief itself being true.[26]
Searle writes, “Those conditions which, as determined by the Intentional content, must obtain if the state is to be satisfied….Thus, if I have a belief that it is raining, the content of my belief is: that it is raining. And the conditions of satisfaction are: that it is raining–and not, for example, that the ground is wet or that water is falling out of the sky.” The conditions of satisfaction for Intentional states occur in networks of other intentional states and are relative to their position in that network.[27] The idea here is that the conditions for satisfaction for a given intentional state is relative to the circumstances in which the state occurs. I can’t believe that Barack Obama is president of the United States prior to the existence of Barack Obama.
The Network and the Background
The conditions of satisfaction of an intentional state cannot be understood in terms of the state itself. A belief occurs in a noetic context and is the product of the operation of a number of faculties of the mind and other beliefs. As we saw with names, beliefs are formed in a network of other beliefs that establish their conditions of satisfaction. If Jones believes that the man who has ten coins in his pocket will get the job, he already knows what it would mean to get a job, he has to be able to count to ten, and he has some understanding of the inside of relation.[28]
The Network functions as a set of intentional states–both conscious and unconscious—to which a given intentional state is related. There are similarities here to the epistemic “tree” of classic foundationalism in that a given intentional state stands in specific basing relations to other intentional states and which establish the conditions of satisfaction of the intentional state in question.[29] Discovering the network is difficult because many intentional states in the network are “submerged in the unconscious” and difficult to bring to the surface epistemically. The Background, distinct from the Network, is defined as a “pre-intentional” foundation of intentional states. The Background itself is not intentional but is a precondition of intentionality. He calls this the “bedrock” of intentionality and implies that it is unanalyzable.[30]
The Network is all the beliefs and desires that make the conditions of satisfaction possible for the intentional state under consideration. The Network and the Background together make up the context for determining the condition of satisfaction. The Background is a set of capacities while the Network is a set of other intentional states (beliefs) but both are relevant for determining the condition of satisfaction for a given belief. The Background, as a set of nonrepresentational mental capacities, makes representation possible[31]. These capacities consist of know how rather than a form of knowing that. What Searle calls the Deep Background are capacities common to all normal humans such as the capacity for moving about and performing typical biological functions.[32] It also includes what he calls the “pre-intentional stance” which “takes account” of facts like things are solid, and objects exist independently. The Local Background[33] or “local cultural practices” are capacities common to humans in a given local context (opening doors, drinking beer from a bottle and the like). These two nuances allow Searle to distinguish conditions of satisfaction that are universal to all humans and specific to each individual. The conditions of satisfaction of a specific Intentional state involve other beliefs that provide a context of meaning for a given proposition. I can believe the door is open because my epistemic context includes the existence of doors and all the practices associated with them (like opening and closing them). They also provide us with the means of why what may be logically entailed by a proposition may not be a part of the satisfaction conditions of a belief.[34]
In Alfred Hitchcock’s Dial ‘M’ for Murder, murderer Tony Wendice is caught by the police because of his knowledge that the key was still under the stair carpet. As he walks in the door, he is surprised to find Chief Inspector Hubbard along with his wife Margot and her lover Mark Halliday standing on the other side. As viewers, we know that he will get caught when he walks in because we are informed about the plot to catch Wendice by the Chief Inspector prior to the trap being sprung. But in the movie, Wendice walks into the trap because he did not have the requisite beliefs that would allow him to draw the proper inferences to evade capture. As viewers, these inferences might seem obvious after the Chief Inspector lays it all out. But Wendice lacks the proper network of beliefs and so walks directly into the trap. Had Wendice had the same information as we the viewers, he would have escaped. But our third person access to the whole scene allows us to draw inferences (as soon as he walks through the door he will be caught) that were not accessible to the character.
In sum, the thesis of the Background includes the following four claims[35]:
- Intentional states do not function autonomously. They do not determine their conditions of satisfaction independently.
- Each intentional state requires for its functioning a set of Background capacities. Conditions of satisfaction are determined only relative to these capacities.
- Among these capacities will be some that are capable of generating other conscious states. To these others, conditions 1 and 2 apply.
- The same type of intentional content can determine different conditions of satisfaction when it is manifest in different conscious tokens, relative to different Background capacities, and relative to some Backgrounds it determines none at all.
Belief and Language
As with Chisholm and Kripke, Searle has shown that propositions in belief contexts bear special logical relations to other propositions and to the world. The truth conditions of a belief are relative to the Intentional state of the individual who has it and are not identical to the truth conditions of the propositional content of the proposition simpliciter. This distinction is essential when analyzing Gettier cases as these cases rely on descriptions of the beliefs, the world, and certain logical implications of the relationship between them. The intuitions that drive Gettier cases are based on what a purely logical analysis of the propositional content distinct from the Intentional aspect of the belief seems to entail. Because the cases rely on these descriptions, I want to round out my discussion of Searle’s view of Intentionality by looking at his analysis of the relationship of language to propositional content and how a logico-linguistic analysis differs from a doxastic analysis.
Intensionality
Searle, building on Austin’s philosophy of language, makes a distinction between Intensionality (with an s) and intentionality (with a t). Intensionality is a property of a certain class of sentences in which the referential terms in the sentence can be understood in such a way that they fail to refer to real objects in the world (fail the test for extensionality). He provides two tests for Intensionality:
a. Existential generalization over referential terms do not constitute a valid form of inference. An existential generalization of “John is looking for the lost city of Atlantis” means that from that sentence one could infer that there exists an x and x=the lost city of Atlantis.
- The sentence does not allow for substitution of referring terms salva veritate for terms which normally have the same referent. The term “Cicero” in the sentence “Cicero was bald” can be substituted with “Tully” salva veritate. However the same term in “Jones believes that Cicero was bald” cannot be substituted if Jones does not believe that Cicero and Tully are identical.[36]
Searle is saying that the truth conditions for a belief that something is true (e.g. the belief that God exists) are different than they are for the statement that someone believes something is true (“John believes God exists.”). In the former case, the belief is a representation whose truth conditions depend on the truth of the state of affairs believed. But the latter case, the truth conditions are dependent upon the statement as representation of the belief (which is a representation). The latter is intensional while the former is intentional. Getter cases seem to be instances of the latter in that they describe what a person is said to believe and then analyze the truth conditions in terms of the former.
In intensional cases, philosophers have puzzled why such statements “do not permit the standard logical operations if… the words contained in the sentences have the meanings they normally have and if the logical properties of a sentence are functions of its meaning, and its meaning in turn is a function of the meaning of the meaning of its component words.” [37]
Searle is clear that [the truth conditions of] beliefs themselves are exemplars of extensionality (existential generalizations can be validly inferred from them–though this is not true about beliefs about beliefs (see p.26)). This is why they cannot be intensional and is why they should be considered intentional.
Incompleteness of Representations
Searle explains that many authors have failed to see the intentional objects are identical with actual objects because they confuse the intentional object with the representative content. Representative content (my thought of my wife) is incomplete in ways that my actual wife is not. So how can the intentional object be identical with the actual object? Put another way, intensional readings of intentional states of affairs has given rise to a confusion about how beliefs (mental states) can be about actual objects and states of affairs when these things have many more properties than beliefs about them.
Searle’s answer is that the intentional object of mental states are identical with the objects and states of affairs they’re about but the specification of (statements about) the intentional object is done by way of the representative content (not the intentional object itself) and because the specification is done in terms of the representative content, it is an intensional aspect. That is, the description of a mental state is intensional but the mental state itself is Intentional. He draws a comparison with descriptions in general. We might easily say that it’s possible to describe a state of affairs and that the description is about the actual state of affairs. We might then talk about our description. The description of the description is intensional but the description itself is not.[38] “Sentences about propositions are intensional and sentences about mental states are intensional but propositions are not in general intensional and mental states are not in general intensional.[39]
Philosophy of Meaning and Computer Science
So how does all this relate to content development and the semantic web? The primary take away from this philosophical discussion is that there is an important distinction between the way humans form beliefs (intentional states) and establish reference. If Searle is correct, then much of the work for establishing meaning and reference is done in the mind of the human in ways that are not even remotely understood. It also isolates the important distinctions in the way the human mind operates and he way computers systems work. Any workable model that hopes to bridge the gap between human users and digital systems must take this distinction as a fundamental axiom.
For example, Kripke’s view is that a name is associated with an object through a baptismal event in which a sign (name) is associated with an object via ostension or with a description. Naming, for Searle, involves an intentional act: an object in the world satisfies some Intentional content in the mind of the believer which the believer baptizes with a name. Frege, Chisholm, and Kripke all make generally the same argument: reference is established through Intentional content in the mind of the believer and its relation to objects in the world and not merely by relationships between words. All belief is belief de re.
But notice here that we take it as a fundamental truth that computers don’t operate this way. Binary systems can only draw relationships between objects in the system; they are not intentional in the way described here. If this is correct, this means that the way a human establishes reference and the way a computer system establishes reference will be fundamentally different and there will need to be a “bridging” model that allows the two to interoperate. This, we believe, is the vision of the semantic web.
If we add to this Searle’s concept of the Background as critical for establishing proper reference, the demand on digital systems becomes enormous. The operative principle now becomes volume (this is part of the reason why talk of “indexed pages” for search engines like Bing and Google are important—size does matter). The human mind leverages the Background almost effortless but in computer systems, the Background needs to be incorporated as an essential aspect for each term in a given domain. The Network and the Background together make up the context for determining the condition of satisfaction according to Searle. Again, the semantic web focuses on this problem and has developed technologies to deal with it.
A Way Out: Artificial Intelligence
We’re now at the point where it will be valuable to question an assumption we’ve been making up to this point: computers don’t have intentional states and don’t mimic the conscious structure of the human mind. Actually this point is even stronger. It is not only that computer systems don’t possess these properties today but they can’t in principle possess them. The semantic web is not a bridging technology in the sense that it will get us to the next major development in AI. Rather it is a bridging technology in that it provides a fundamentally necessary means of bringing digital systems and human knowers together—a technology that will no doubt evolve over time but always be needed.
To put this claim in terms relevant for this paper, the claim is that computers operate on a syntactical model: they use rules to process a symbolic language to produce other symbols. Humans possess a “semantics engine” that includes the ability to have intentional states that are about those symbols and understand what those symbols mean. This is a very strong claim. It essentially is the claim that artificial intelligence is not possible if by AI we mean replicating, to some degree, human mental function in a digital system. The debate over whether AI in this sense is possible is a long and drawn out one and not a debate we can enter into here. The assumption up to this point is that a digital system cannot replicate intentionality and so talk of a “semantic web” just means that the semantic engine of human mentality is mimicked for the purpose of that mentality. In other words, the semantic web, like many AI projects, is not an end in and of itself. The semantics in “semantic web” really focuses on human semantics even though ultimately the technology that provide the foundation for the semantic web are wholly syntactical systems.
This claim is relevant to the present topic because it sets boundaries around the goals of something like semantic web technology. If computer scientists can agree that the goal isn’t to replicate[40] human understanding but to accommodate it or provide a digital model that collaborates with it, content and software developers may find much more success in the short term even if more “intelligent” systems are a long-term goal. To this end, we want briefly to look at an argument by John Searle in which he attempts to demonstrate that digital systems don’t replicate human consciousness and as long as digital systems continue to operate in the way they do, they can’t replicate human consciousness.
The Turing Test
Searle’s argument begins by considering the Turing Test as a basis for answering questions about whether computer systems can replicate human understanding. One of the best descriptions of the Turing Test we’ve read comes from Brian Christian’s entertaining and informative book, The Most Human Human. We repeat his description here in its entirety.
Each year, the artificial intelligence (AI) community convenience for the field’s most anticipated in controversial annual event – a competition called the Turing test. The test is named for British mathematician Alan Turing, one of the founders of computer science, who in 1950 attempted one of the field earliest questions: Can machines think? That is, would it ever be possible to construct a computer so sophisticated that it could actually be said to be thinking, to the intelligent, to have a mind? And if indeed there were, someday, such a machine: How would we know?
Instead of debating this question on purely theoretical grounds, Turing proposed an experiment. A panel of judges poses questions by computer terminal to a pair of unseen correspondents, one a human “confederate,” the other a computer program, and attempts to discern which is which. There are no restrictions on what can be said: the dialogue can range from small talk to the facts of the world (e.g., how many legs ants have, what country Paris is in) to celebrity gossip and heavy-duty philosophy—the whole gamut of human conversation. Turing predicted that by the year 2000, computers would be able to fool 30 percent of human judges after five minutes of conversation, and that as a result “one will be able to speak of machines thinking without expecting to be contradicted.” [41]
Christian points out that Turing’s vision did not come to pass and has not come to pass: no artificially intelligent machine has passed the Turing test. But this provides no proof that no machine could pass the test. Searle believes he has come up with just such an argument. His argument intends to show that even if an artificial (mainly digital) system could pass the Turing test, this system does not “think” or, better, “understand” in the way we typically take that word. His argument goes like this:
The Chinese Room Argument
Well, imagine that you are locked in a room, and in this room are several baskets full of Chinese symbols. Imagine that you (like me) do not understand a word of Chinese, but that you are given a rule book in English for manipulating these Chinese symbols. The rules specify the manipulations of the symbols purely formally, in terms of their syntax, not their semantics. So the rule might say: ‘Take a squiggle-squiggle sign out of basket number one and put it next to a squoggle-squoggle sign from basket number two.’ Now suppose that some other Chinese symbols are passed into the room, and that you are given further rules for passing back Chinese symbols out of the room. Suppose that unknown to you the symbols passed into the room are called ‘questions’ by the people outside the room, and the symbols you pass back out of the room are called ‘answers to the questions’. Suppose, furthermore, that the programmers are so good at designing the programs and that you are so good at manipulating the symbols, that very soon your answers are indistinguishable from those of a native Chinese speaker. There you are locked in your room shuffling your Chinese symbols and passing out Chinese symbols in response to incoming Chinese symbols. On the basis of the situation as I have described it, there is no way you could learn any Chinese simply by manipulating these formal symbols.[42]
Searle’s argument draws on the strong intuition that manipulating symbols according to rules is much different than understanding the meaning of the symbols. A machine (say a computer) that can produce the correct answer to questions by following an algorithm—a set of rules for processing symbols–designed to produce those answers is much different than having understanding about what those symbols mean. Searle argues that intentional states like belief are irreducible to symbol manipulation so even if a machine can mimic human consciousness, we cannot conclude that brain functions are computational.
A common response to Searle’s argument is that it is question begging.[43] Daniel Dennett argues that if a machine passes the Turing test, it just is intelligent and to claim it’s not is to beg questions against what intelligence is. For Dennett, having a mind just is having the ability to compute a function so when Searle includes “phenomenological” notions of things like consciousness and understanding to the mix and assumes by these terms that these are over and above “computing a function” he begs questions.[44]
The authors of this paper think Dennett misses the salient point of Searle’s argument. Searle’s position is based on a known quantity: what it is like to know Chinese (or any language). There is an “internal” phenomenology to understanding that appears, intuitively, to be over and above symbol manipulation based on rules and the Chinese Room argument provides a relevant counterexample. Dennett’s materialism cannot support an internalist ontology like this[45] and so he must reject it out of hand even though he has good evidence for it (as Searle stated in his public exchange with Dennett, “Should I pinch the author to remind him that he is conscious? Or should I pinch myself and report the results in more detail?” It’s hard to know what would convince Dennett that he’s conscious in the way relevant for Searle’s argument.)
In addition to Dennett, Searle has had other detractors and has attempted to address each counterargument over the years. Each of the criticisms come down to whether intentional states should be seen as computational or as something else. Searle argues strongly that it cannot be a function of computation similar to the way a digital computer works.
Regardless of whether Searle’s argument ultimately works, it does support the earlier statements that there is at least a prima facie different in the way humans process information and the way information processing systems can be implemented in digital machines. We think it further supports this idea: that even if Searle’s argument fails, the failure is not obvious or clear. A machine that exhibits “strong AI” may be in our future but is certainly not in our present or near future and so working on Searlean assumptions seems most auspicious.
Copyright© 2011 by Paul F. Pardi
[1] (Kripke S. A., 1979)
[2] In philosophy, the problem of naming has to do with the relationship between a linguistic marker (e.g. “Cicero”) and the object it is meant to pick out (a famous Roman orator).
[3] (Kripke S. , 1980)
[4] The variable ‘p’ in each of these principles stands for any appropriate English sentence.
[5] (Kripke S. , 1980, pp. 248-249)
[6] This is essential to content problems we need to solve in computer science. When a term is used by a human to request content from a computer, how do computer scientists create algorithms that “understand” the referent of that term? And how do content developers create content that can be most easily consumed and processed by those algorithms?
[7] See also (Searle J. R., Intentionality, 1983, p. 44)
[8] (Searle J. R., Intentionality, 1983, p. 235). Note the problem here for computer science. If the causal chain is intentional and computers operate on a binary system (externalist and non-intentional), how does baptism work? If the referent of a sign is a function of a belief in someone’s head then there is no specific property of the sign that includes the causal chain. This is why the reference is not externalist. But it also means that a sign like “Cicero” has no implicit referent and is “meaningless” as a singular term to a digital system.
[9] (Searle J. R., Intentionality, 1983, p. 241)
[10] (Searle J. R., Intentionality, 1983, p. 234)
[11] (Frege, 1892)
[12] Kripke understand Frege as arguing that a description fixes the meaning of a name. Aristotle is the Stagarite who taught Alexander the Great where “the Stagarite that taught Alexander the Great” is the meaning of the name Aristotle. Kripke thinks this is false for the descriptive phrase could not have been true of Aristotle but that would not change the thing (Aristotle). The name fixes an object in all possible worlds. He suggests also that perhaps Frege meant that a description fixes a reference: one can pick out Aristotle by describing him. Kripke seems more open to this idea but notes that a description shouldn’t be said to “fix” a reference (“rigid reference” p. 58 in Naming and Necessity) even though it can be used to pick out an object. By fix a referent, I think he means that a given description can’t be said to pick out an object in all possible worlds (see his description on p. 61). If this is what Frege meant by using a definite description as a referent for an object, Kripke rejects this too. Kripke argues that the description could be different but then that description would be used to pick out Aristotle.
[13] (Searle J. R., Intentionality, 1983, p. 250)
[14] (Searle J. R., Intentionality, 1983, p. 258)
[15] Searle argues that this may be a lead to solving Kripke’s puzzle about belief (Intentionality, 256)
[16] (Searle J. R., 1979, p. 182)
[17] I discuss Intensionality below under Belief and Language.
[18] See the section “Intensionality” below for a brief description of extensionality.
[19] (Searle J. R., 1979, pp. 184 bot.-185)
[20] (Searle J. R., 1979, p. 185)
[21] See (Searle J. R., 1979, p. 185). See also, See also (Searle J. R., 1983, pp. 16,17)
[22] (Searle J. R., Meaning and Use, 1979, p. 185); (Searle J. R., 1983, p. 118)
[23] (Searle J. R., 1983, p. 12)
[24] (Searle J. R., 1983, p. 13)
[25] (Searle J. R., 1983, p. 10)
[26] Searle writes, “The expression ‘conditions of satisfaction’ has the usual process-product ambiguity as between the requirement and the thing required. So, for example, if I believe that it is raining then the conditions of satisfaction of my belief are that it should be the case that it is raining (requirement). That is what my belief requires in order that it be a true belief. And if my belief actually is a true belief then there will be a certain condition in the world, namely the condition that it is raining (thing required), which is the condition of satisfaction of my belief, i.e., the condition in the world which actually satisfies my belief.” (Intentionality, 13)
[27] (Intentionality, pp. 20, 21)
[28] In computer systems, these relations must be explicit aspects of the content model and consumable by the algorithms that process the content. Semantic web technologies seek to account for these relations. See OWL and SKOS for example.
[29] This idea captures some of what I believe Chisholm is after with his idea of “epistemic intimacy” (see above). While Chisholm’s idea is narrower than Searle’s he argues that de re belief involves knowledge of other propositions that would allow the believer to attribute properties to the object the proposition is about. Chisholm and Searle are drawing off of the same intuition here.
[30] Again this has all the earmarks of the foundationalist’s notion of basicality though Searle does not draw this parallel and perhaps would balk at the suggestion.
[31] It’s worth mentioning here that for Searle, understanding is different from grasping meanings. One could grasp the meanings of terms in a sentence but not understand the sentence (Intentionality p. 146, 147). One could grasp all the terms in the sentence, “Bill opened the mountain” but not understand the sentence. The Background is relevant for understanding conditions of satisfaction for a given intentional state. Conditions of satisfaction are determined relative to a given Background and no intentional state determines conditions of satisfaction apart from a given Background (see Searle’s The Rediscovery of the Mind, p. 176)
[32] (Intentionality, p. 144)
[33] Plantinga introduces a similar idea in his expanded view of warrant. What Searle calls the deep background and local background, Plantinga calls the “maxi-environment” and the “mini-environment” respectively. See Kvanvig, J., ed. 1996. Warrant in Contemporary Epistemology; Essays in Honor of Plantinga’s Theory of Knowledge. New York: Roman and Littlefield, Chapter 16, “Respondeo”
[34] At the risk of being redundant, this again presents a unique set of challenges for computer science. If conditions of satisfaction are a function of background beliefs and the network, how does a computer system account for these dynamics without being able to leverage things like intentional states?
[35] (Searle J. , The Rediscovery of the Mind, 1994, p. 190). Compare with list on (Searle J. R., 1983, p. 177)
[36] (Searle J. R., 1983, p. 23)
[37] Ibid.
[38] (Searle J. R., 1983, p. 187)
[39] (Searle J. R., 1983, p. 188)
[40] We’re making a distinction here between “replication” and “simulation.” We leave it as an open question whether the human mind can be simulated by a computer system. Intuitively it seems that this is not only possible but highly probable. But simulation is not replication where this latter term implies two things: first that the human mind
[41] (Christian, 2011, p. 3)
[42] (Searle J. , 1984, p. 32)
[43] One of the most vocal critics of Searle’s argument has been Daniel Dennett. For a good summary of the main points of disagreement along with key references where the formal debate has taken place, see an exchange between Dennett and Searle here: http://www.nybooks.com/articles/archives/1995/dec/21/the-mystery-of-consciousness-an-exchange/
[44] See (Dennett, 1987, p. 332 ff.)
[45] Oddly, Searle’s own ontology doesn’t appear to support internal states either but we’ll leave that problem for another time.
