Dale Jacquette's third chapter in Philosophy of Mind is "Artificial Intelligence: Mechanism, Minds, and Machines." It is a fairly traditional discussion of the problem of whether or not a computer can be regarded as being a mind (the possibility of the computer having a mind, just a people have minds, is left undiscussed), for which Jacquette claims intentionality is a necessary and sufficient condition. To establish that computers cannot be intentional, Jacquette claims to show that they certainly cannot understand (as opposed to imitate) language, for which intentionality is a necessary but not sufficient condition. It is debatable whether Jacquette establishes if either of these claims is true, i.e. that intentionality is necessary for the understanding of language, and that computers cannot (in principle - no one denies that today's computer cannot) understand language. However, even should he have shown that computers cannot understand language, this does not entail that they are incapable of intentionality. In fact, by yet another criteria which Jacquette claims demonstrates intentionality, that of having a 'mental map', computers with the right kind of sensory apparatus are clearly capable of intentionality. In this paper we trace Jacquette's argument against computers being capable of intentionality, and how this jives with the claim that having a mental map entails intentionality. Finally, I hope to determine where Jacquette has made his mistake, and determine on the basis of Jacquette's arguments, whether or not computers can be said, given the right capabilities, to be a mind.

THE ARGUMENT

1. Minds (and other mental phenomena) are distinguished from non-mental (physical) phenomena by intentionality (i.e. intentionality is the necessary and sufficient condition).
2. Intentionality is a necessary condition for understanding language.
3. The Turing test establishes whether or not a thing understands language¹.
4. Computers must necessarily fail the Turing Test. Therefore:
---
5. Therefore, computers are not minds.

This observation points in the direction of Jacquette's other major claim concerning intentionality and language, that intentionality is theoretically prior to language. This position was debated by Roderick Chisholm and Wilfrid Sellars in a series of letters written in 1956². Jacquette sums up their debate concisely:

The idea that language should have meaning without thought seems absurd, as Chisholm maintains. But what is thought, if it is not, as Sellars suggests, a kind of inner speech or mental language? (Jacquette, pg 104)

Thought without language is most likely a nonverbal association of thing with thing, a nonlinguistic memory and categorization of objects or experiences by similarity of appearance. This can eventually support the use of one object to represent those that resemble it, and, at more advanced levels, evolve into the use of spoken or written signs more symbolically as objects to stand for other objects. (Jacquette, pg 105)

THE TURING TEST

Alan Turing invented what he called the "Imitation Game" (Turing), but which is now more commonly referred to as the "Turing Test." He described it in the following way: an "interrogator" sits at a computer terminal and types messages which are transmitted to two other terminals (separated from him/her by a wall). At one of the terminals sits a person, and the other is hooked up to a computer - this person and the computer are the "participants". The interrogator can direct his/her questions or statements to either participant - say by a convenient set of labels like 'P1' and 'P2'. The idea of the game³ is for the interrogator to determine which of the participants is the computer and which is the person.

There is a great deal of confusion on Jacquette's part concerning what the result of the Turing Test is, i.e. what is to be said if the interrogator cannot distinguish the computer from the person? Jacquette says it quite clearly himself by quoting Turing:

Turing does not say that any machine that passes the test or wins the game is therefore intelligent. On the contrary, he maintains that whether or not a machine can pass the test should replace the original question "Can machines think?" which he says, "I believe to be too meaningless to deserve discussion." (Jacquette, pg 67; and Turing, pg 13)

SEARLE'S CHINESE ROOM

John Searle poses an interesting objection to the Turing Test in the form of his "Chinese Room" (Searle). Searle asks the reader to imagine a Turing test situation in which the interrogator submits messages in Chinese to a single participant, a person, who does not speak, read or write a single word of Chinese. However, the participant does have a "crib book", which has Chinese sentences grouped into two columns. The column on the left is labeled 'submission', the column on the right is labeled 'response', and the sentences, all in Chinese, are paired off so that each submission has a corresponding response. Now, upon receiving a message from the interrogator, the participant finds this message in the 'submission' column and responds with its corresponding 'response'. We are asked to set aside practical concerns like the speed and accuracy with which the participant can lick up the correct response for any given submission, for after all, Searle's real target is a computer which can access data very quickly indeed and never make mistakes (at least within the limits of their programming).

Now clearly, nobody but the interrogator understands Chinese in this situation - the participant does not understand Chinese, and no one is willing to say that the crib book understands Chinese. Once we grant Searle this, his next move is to say, 'let us imagine now that it is not a human with a crib book, but just a computer with the crib book program into its memory'. Would we say that the computer understands Chinese? We find ourselves quite rightly compelled to say, 'No.' There is no relevant difference in the two cases, and for the very same reasons we will not grant the human understanding, we will be reluctant to grant the computer understanding.

But Searle has really just pulled a fast one - we are asked to ignore practical concerns like 'lick-up speed' and 'lick-up accuracy', because Searle intends to make his point concerning the computer, not the human. There is practical concern which Searle sneaks past us along with speed and accuracy, and that is the size of the crib book and the nature of language.

The average American has a spoken vocabulary of 3000-4000 words and a reading vocabulary of around 10,000 words. But the number of unique sentences that (s)he is capable of speaking or reading is infinite. (Why? Syntax.) So if the computer is to win the Imitation Game, it must be able to 'understand' an infinite number of sentences, which means the crib book would have to have an infinite number of entries - this is not a practical concern that can be swept under the rug. The only way a computer would not necessarily lose the Imitation game is to be programmed with a vocabulary comparable to that of the interrogator (not a challenge at all, The Oxford English Dictionary does not require even one-half the memory of a CD ROM disk) and a complete working grammar. Without both a vocabulary and a grammar, the computer will necessarily lose. But not for the right reasons.

GÖDEL'S THEOREM⁴

Searle's Chinese Room objection was an attempt to offer a counterexample which showed that winning Turing's Imitation Game did not warrant the conclusion that a computer understood language. It has been suggested that questioning a computer about Gödel's Theorem (Gödel) would demonstrate that a computer cannot understand language. Here is the idea: Gödel proved that any formal system capable of axiomatizing arithmetic is necessarily inconsistent - meaning that contradictions can be derived from true statements, or incomplete - meaning there are true sentences expressible in the arithmetic which cannot be formally proven within the arithmetic. This applies to computers as they employ exactly such a formal system of axiomatized arithmetic (both Turing and John von Neumann are credited with developing the formal system of axiomatized arithmetic that all modern computers employ).

Gödel's proof makes use of a sentence like "This sentence cannot be proven.⁵" Note:

1) If the sentence is true, then since it says of itself that it is unprovable, there is a true but unprovable sentence of the logic.
2) If the sentence is false, then it is provable, since it says of itself that [it] is unprovable. But if the sentence is provable, then there is a provable sentence of the logic that implies its own unprovability. (Jacquette, pg 82)

Jacquette employs this kind of Gödel sentence to show that computers cannot understand the meaning of the sentence, and thus that they cannot understand language. The argument goes like this (Jacquette, pg 83-84): Jacquette asks us to imagine a conversation in which he asks a computer to consider a Gödel sentence⁷ (e.g. "This sentence is unprovable.") expressed formally in a consistent logic as S. This sentence must be true, and Jacquette grants that the computer could recognize this. He then asks the computer to consider a another sentence S', this time that says, " is provable." (Note: this means S' is equivalent to not-S.) But then he asks the computer to treat "provable" as though it meant "unprovable" and vice versa. (Note: this means S' is equivalent to S. ) He then asks the computer if S' is true. He claims the conversation would have to go something like:

The general principle is that all sentences that say of themselves that they are unprovable are true if they belong to a consistent logic. But now "provable" means "unprovable." The principle is transformed to read that all sentences that say of themselves that they are provable are true if they belong to a consistent system of logic. This is what S' says; so S' is true.

What about sentence not-S?

Since S is true, not-S is false.

But not-S says that S is provable. What does the general principle imply about the truth of not-S ?

Not-S is true.

The computer responds correctly up to that last question, where Jacquette asks what Not-S's truth-value is in light of the general principle. But its last statement clearly contradicts its previous response where it repeats its original evaluation of S. Jacquette and the computer started off discussing a general principle, i.e. Gödel's Theorem. Half way through the conversation, the general principle was transformed into (dare I say it?) not-Gödel's Theorem. But Gödel's Theorem and not-Gödel's Theorem are equivalent! This amounts to nothing more than saying that the computer cannot tell the difference between proven and unproven, i.e. that one is the negation of the other. But the computer makes this kind of distinction every time Jacquette asks it a question concerning Not-S!

How does a person get the correct answers to the 'Gödel problem'? Actually, the answer is in the (Note:...)'s of the paragraph preceding Jacquette's conversation with the computer. The key is in recognizing that both of the transformations that lead to not-Gödel's Theorem are in fact, themselves negations, and thus, there is no net change in the general principle. So what has happened is a very old mistake - what one hopes to prove is assumed in the proof. Jacquette hopes to prove, by means of the Gödel objection⁸, that computers cannot understand language, but assumes in doing so that they cannot understand language.

JACQUETTE'S CASE

So far, we've addressed two of Jacquette's objections against the notion that computers can, or could ever be made to understand language. The first one, Searle's Chinese Room objection, was an objection to Turing's Imitation Game - that a computer winning the game did not necessarily warrant the claim that it understood language. The second objection, Gödel's objection, was supposed to be a specific example of language a computer cannot understand. As Jacquette says,

There is no program currently that comes anywhere near to passing the Turing Test even for ordinary conversation. If machines were somehow able ... [to pass the test] ... the Turing Test is an unsatisfactory measure of intelligence." (Jacquette, pg 86).

... there are few mathematical logicians or philosophers of mind who agree ... that Gödel's theorem proves that the mind is not a machine and that no machine can duplicate the mind's reasoning abilities." (Jacquette, pg 86).

The missing ingredient that prevents machine considered only as such from being or having minds is the intrinsic intentionality, aboutness, or directedness of thought toward intended objects. (Jacquette, pg 94)

I have been very careful not to use the phrase 'Turing Test' except when quoting Jacquette. Recall Turing's quote concerning intelligence and thinking; Turing's Imitation Game tests for neither of those qualities, much less for the quality of possessing or being a mind. It tests for the quality of understanding language, and as we have shown, both of Jacquette's objections fall short of discrediting the claim that the Game does just this.

And even if either objection were valid, it would merely demonstrate that computers cannot use language, not that they lacked intentionality, intelligence or minds. Intentionality may be a necessary condition for understanding language, but it cannot be both necessary and sufficient. Otherwise, thinking, which is intentional, would not enjoy the relation to language Chisholm attributes to it,

"... if there were no thinkers, there would be no language, but if there were no language, there could still be thinkers who had no linguistic way of expressing their thoughts." (Jacquette, pg 103).

That this is so is demonstrated in Jacquette's following chapter "Intentionality and the Nature of Thought." In this chapter, Jacquette argues, amongst other things, that having a mental map of one's surroundings is a demonstration of intentionality. A map is about something, something that I have sensed in the world about me. I construct a map on the basis of my sensations, and this map represents, or is about, the world around me. This aboutness or direction of thought toward a thing is the hallmark of intentionality, and thus, anything capable of having a map of its surroundings is capable of intentionality.

Many robots have been built that could map a room. They have employed many kinds of sensors, i.e. ways of sensing the objects in the room. For example, early models had tactile sensors - bumpers that registered pressure when the robot came into contact with an object. When such a robot was set down in a room with various objects placed on the floor, say a couch in one corner, and an ottoman in front of a recliner in another, and a coffee table in center, the robot would move along a wall until it bumped into the first obstacle it encountered. It would then proceed to circumnavigate the obstacle until it knew its extent, and then continue, eventually moving away from the wall, until it had encountered all the obstacles in the room - much in the same way a blind person might learn the layout of a room with which (s)he was unfamiliar. The resulting map would be stored in its memory, and used later to navigate the room⁹, i.e. to move from one point to another without running into any of the furniture. More advanced robots use ultrasound as a kind of sonar, visible light and a digitized image of the room from a camera mounted on their body, or a combination of these three (other sensor mechanisms are of course possible, but these three are the most common because they most closely mimic the way animals sense their environment).

If a map is intentional, then certainly these robots would qualify as intentional. And if intentionality a necessary and sufficient condition for having or being a mind, then they also are minds. Jacquette says as much of "lower" animals that are capable of acting on a 'mental map'. For example, he says of his tropical fish that they

... perceive the objects in and surrounding the [fish]tank, that they want and expect to be fed in the morning, that they do no just react by fleeing but fear capture in the net [as an even "lower" animal like a mollusk might not]. (Jacquette, pg 115)

While this may be anthropomorphic for what might better be described as purely nonintentional cause and effect mechanisms, it may also in some cases be the best explanation of the animal's behavior. If so, it is undoubtedly intentional, since mental maps, like mental languages, images, and other kinds of representation, are intentional entities par excellence. A map always means [his emphasis] something. (Jacquette, pg 115)

On one hand, Jacquette is forced to agree that computers, with the right kind of sensory apparatus and programming, are capable of intentionality. On the other hand, Jacquette wants to claim that computers are incapable of intentionality because they are incapable of understanding language. He has made a simple logical mistake concerning necessary conditions, and his objections against the Turing's Game have been shown to be flawed. If one accepts Jacquette's notion of intentionality, it is clear that computers, maybe not any old computer, but at the very least those with the right kind of sensory apparatus, are capable of being minds.

So why all the effort to show this is not the case? Since behaviorism arrived on the scene in the first decades of this century, and as neurophysiology makes greater and greater advances in understanding the brain, reductivist and eliminativist theories of behavior have increasingly threatened the plausibility of the thing we call the mind and the moral quality of personhood. The more advanced computers become, the closer they come to being to able to do things that have traditionally seemed to be the purview of persons, the more nervous we become that the behaviorists are right. The more we question if computers are becoming more like us, and thus should acquire the rights and responsibilities of personhood, the more we are tempted to see ourselves as nothing more than very complex mechanisms devoid of personhood.

The answer, I think, is fairly obvious, and Jacquette even points to it himself - just replace "fish" with 'computers':

If it seems wrong to say that neurophysiologically simple creatures like fish have intentions, it may only be because this appears to imply that fish are self-conscious beings, or (finny) persons capable of deliberate or intentional action. If fish act intentionally, they nevertheless are not persons, nor are they morally responsible for what they do. (Jacquette, pg 115)

¹...understands language...
Jacquette speaks as though the Turing test directly tests for intelligence, intentionality, and possession of or the quality of being a mind (the three of which he seems to treat interchangeably). For reasons that will become clear later, this simply is not and cannot be the case - it is certainly not what Turing claimed it tested. I choose to be as generous as possible toward Jacquette's position, and present his argument, when possible, as though he were not confused in this way.

²...letters written in 1956...
These letters were published in 1958, and while I have not quoted them directly here, Jacquette uses them extensively . The citation given is:

Chisholm, R M; Sellars, W. "Chisholm-Sellars Correspondence on Intentionality", Minnesota Studies in the Philosophy of Science 2, Concepts, Theories, and the Mind-Body Problem, ed: H Feigel, M Scriven, G Maxwell, Minneapolis: University of Minnesota Press (1958).

⁴GÖDEL'S THEOREM
I'm acutally quite reluctant to discuss this objection at all - Jacquette ends his discussion with the statement that the concensus today is that Gödel's Theorem does not prove anything about the status of computers. Many Intuitionists initially seized on Gödel's Theorem as a refutation of any kind of Formalism, but in the⁴ intervening years, a more critical eye has been turned turned toward the implications of Gödel's Theorem with respect for philosophy. Still, it is so often addressed in discussions of AI that I feel I must address it here.

⁵This sentence cannot be proven.
Actually, this a gross simplification of the kind of 'sentence' required to yield Gödel's results. In fact, this sentence a variation of the "Liar's Paradox", and itself fails to yield the paradox unless one unless one makes no distinction between "provable" and "true" (Van Heijenoort, pg 356), a dinstinction Jacquette ignores without pointing out to the reader. If the reader sees the difference, (s)he might be left wondering why Jacquette finds his 'Gödel's Theorem' so significant.

⁶It would be disastrous for logic to be logically inconsistent.
It will be very amusing when the first computer capable of passing the conversational Turing Test claims that not-S is true. And that when told it must therefore not be a mind, responds by saying, "No, it just means I'm not as committed to a consistent logic as you."

⁷...a Gödel sentence...
I say "a Gödel sentence" because it ought to be clear that there a re an infinite number of syntactically equivalent ways of expressing such a sentence while using very different semantics.

⁸...the Gödel objection...
By calling this the 'Gödel objection' I do not mean to attribute the use of the Gödel Theorem as an objection to the claim that computers can be made to understand language to Gödel himself - I do not know of Gödel ever addressed the subject himself. J R Lucas was the first to employ the Gödel Theorem in this fashion (Lucas, J R, "Minds, Machines, and Gödel", Philosophy 36, 1961, pg 112-127.)

⁹The resulting map would be stored in its memory, and used later to navigate the room...
In my previous life as a Computer Engineering major, I wrote a piece of software to control just such a robot.

Chisholm, R M. "Intentionality", Encyclopedia of Philosophy, Vol 3, NY: Macmillan Publishing Co (1972) 201-204.

Gödel, K. "Some Metamathematical Results on Completeness and Consistency, On Formally Undecidable Propositions of Principia Mathematica and Related Systems I, and On Completeness and Consistency", Frege and Gödel: Two Fundamental Texts in Mathematical Logic, ed: J van Heijenoort, Cambridge, MA: Harvard University Press (1970) 83-108.

van Heijenoort, J. "Gödel's Theorem", Encyclopedia of Philosophy, vol 3, NY: Macmillan Publishing Co (1972) 348-357.

Jacquette, D. Philosophy of Mind, Englewood Cliffs, NJ: Prentice Hall (1994).

Scheer, R K. "Philosophy of Mind" Lectures, KS: Kansas State University, January 12 - May 2, 1995.

Searle, J R. "Minds, Brains and Programs", The Behavioral and Brain Sciences, vol 3 (1980) 417-424; 450-456.

Turing, A M. "Computing Machinery and Intelligence", Mind, vol 59 (1950) 433-460.