The official title of the computer model of the mind is "Machine Functionalism." It arises from the ashes of Dualism and Behaviorism, and may be seen as an attempt to steer a reasonable middle course between the extremes that they represent. Let's begin by incorporating the diagnosis of Behaviorism's failure that we ended with in the last section. Patterns of inputs and outputs are part of the story, but there's also what goes on in between. According to Machine Functionalism, to define any mental state or process, one must specify three elements:
(i) stimuli (inputs) that typically cause the mental state or process;
(ii) cause and effect relationships that the given mental state or process typically has to other mental states and processes ("functional role"); and
(iii) responses (outputs) that the mental state or process typically causes.
But that's not yet much of a theory. We need to find out why the view's name is apt.
"Functionalism" derives from the theory's use of functional definition. I'll introduce the basic idea with a simple, non-psychological example and then return to the theory's use of the idea.
It must be nice for you to be out of prison. After an almost entirely successful career as a safecracker, you've been 'rehabilitated' and decided to go straight. You want to make the best use of the skills you've got in making an honest living, and what you're best at is opening locked things. So you open a locksmithing business. You're very good at it, and soon have more work than you can handle, so you hire an apprentice, whom you teach so well that the two of you have more work than you can handle. After a third and fourth apprentice, you decide it would be more efficient to develop a training program for locksmiths (which could also be marketed to trade schools). The first step in developing the curriculum is, of course, to develop a good theory of locks and keys. We're all familiar with those oddly shaped pieces of metal - we carry them in our bags and pockets - that open various doors. But those aren't the only possible lock and key combinations. At the bank, there is a large vault room where the safe deposit boxes are secured. When the manager opens the vault door in the morning, you can bet she doesn't haul out a huge key to turn in the enormous lock; rather, she punches in a sequence of numbers on the electronic keypad built into the vault door, this activates a servomechanism, and motors cause the door to swing slowly open. The geometrical pattern at the back of your eye (the retina) is unique to you. At some high security installations, retinal scanning is used in place of metal keys. You step up to a laser-scanner, your eye is harmlessly scanned, and the pattern compared to the one stored in computer memory; if it's a match, you are admitted; if not, you're turned away (at best). We can conceive of chemical lock/key combinations: pour the right stuff into a little hole in the door, and a chemical reaction triggers opening. And there are other possibilities. Any theory of locks and keys must reflect this diversity of possible materials. The following definitions are a good start:
L is a lock =def L secures a portal for which there is a key.
K is a key =def K opens a lock that secures a portal.
Despite its simplicity, this pair of definitions illustrates two important features of even the most complex functional definitions. First, there is 'indifference to material': neither definition says, "one must make keys or locks out of just the following material;" instead, "lock" and "key" are functional notions - locks and keys are defined by what they do, not what they're made out of. Naturally, the definitions place some constraints on materials. A key made of butter or a lock made of ice won't do. Second, these functional definitions form a (here, tiny) system of interlocking concepts. Each concept's definition makes essential reference to the other concept. It's not possible to say what a key is unless one also says what a lock is; either you take the whole package, or you get nothing at all.
According to Machine Functionalism, a given mental state is defined by what is does as part of a system of mental states, and what it does is to stand in a complex web of cause and effect relations which it's the job of its functional definition to specify. When we talked about Cartesian Dualism's Postulate of Interactionism, we considered the example of the kind of pain you feel when you're stuck with a pin. How would Machine Functionalism define that same mental state? The three-part recipe would yield something like this:
p is pin-prick pain =def p is typically caused by pin-pricks; p typically causes desire for relief, belief that one is in pain, anxiety, ...; and p typically causes wincing, withdrawal, noise ("Ouch!") ....
The ellipses ("...") conceal a great deal here because pain bears many more relations to many other mental states and is expressed at the skin-level in many other ways. The full definition would take more words than in this whole book, or in most libraries since it really can't be given in isolation from the interlocking functional definitions of all the other implicated mental states. Given what we learned from the failure of Behaviorism, that's as it should be.
The word "typically" is doing its fair share of work, too. This kind of pain can occur in a pin-free environment, and can even result from nervous system malfunction, in the complete absence of sharp, pointy objects.
But why is it called "Machine Functionalism?" How does the computer model come into play in describing the complex web of relationships verbally specified in the functional definitions?
"The basic idea of cognitive science is that intelligent beings are semantic engines - in other words, automatic formal systems with interpretations under which they consistently make sense."
John Haugeland, "Semantic Engines: An Introduction to Mind Design" (31).
So sayeth Professor Haugeland. Our mission, should we choose to accept it, is to try to understand what he means. To see what he means, we have to understand what automatic formal systems are and what an interpretation of one is.
We begin by characterizing the notion of a formal system. In order to define one, we must specify three things:
(1) what the tokens are
(2) what the starting position is
(3) what moves are allowed in any given position.
From this characterization it follows that any formal system will have three features: (i) it will be self-contained; (ii) it is perfectly definite; and (iii) it is finitely checkable. Let us say that any system with these three features is digital.
Notice how abstract this characterization is. No limitation whatever is placed on what we might call the system's "material realization". Any system that meets the particulars of a specification will be the same formal system, despite any differences in the materials and despite many possible differences in their arrangements. Formal systems are thus to some extent medium independent. They are, in essence, a kind of functional definition and so share indifference to material.
Some of the most familiar examples of formal systems are board games, like chess. Chess, we know, is usually played on a square board marked off in 64 squares, with 32 pieces, 16 of one color and 16 of another. The rules of chess specify a system of abstract relationships which we find it convenient to keep track of with chess sets. It is, however, the relationships that matter. Chess sets can be made out of any material that suits us, and some museums have large collections of chess sets fashioned from a wide variety of materials. We could even substitute live human beings for the pieces and use a building with (at least) 64 rooms as the playing field. That's not something I'd recommend, but the abstractness of the defining rules allows it. Chess, however, is a very complicated game. To get a better feel for the concept of interpreted formal system, let's look at some simpler examples.
(a) Suppose that there are two bathrooms on an airplane, located opposite one another at the head of the long central isle. There is a nearby display panel, for indicating occupancy. If at least one of the bathroom doors is unlocked, then the light indicating vacancy is on. If both are locked, "No vacancy" is illuminated.
|
System and method for providing reservations for restroom use IBM (filed Aug. 2000) |
|
Abstract: The present invention is an apparatus, system, and method for providing reservations for restroom use. In one embodiment, a passenger on an airplane may submit a reservation request to the system for restroom use. The reservation system determines when the request can be accommodated and notifies the passenger when a restroom becomes available. The system improves airline safety by minimizing the time passengers spent standing while an airplane is in flight. What is claimed is: 1. A method of providing reservations for restroom use, comprising: receiving a reservation request from a user; and notifying the user when the restroom is available for his or her use. |
(b) Suppose that the display breaks down - some passenger, furious at getting an olive rather than an onion in his drink, has smashed it. So a flight attendant rigs up a board with four squares in it. If a bathroom door is unlocked, he puts a check mark in the "Vacancy" square; if it is locked, he puts a check mark in the "No vacancy" square. If both "No Vacancy" squares are checked, you've got to wait a bit longer. If one of the other squares is checked, relief is moments away. Because of the way the board is positioned, it is hard for the passengers to see it, so the first flight attendant enlists the aid of a second, who runs up and down the aisles, waving a "No Vacancy" sign, or a "Vacancy" sign, depending upon what he sees on the board. (There are other ways the flight attendants might have done things to convey the same information.)
(c) Consider now a computer designer, Ms. A, who wishes to design a circuit to represent the logic of the mathematical use of the word "or". According to such usage, a statement of the form "p or q" is true if and only if at least one of the component statements is true (and not otherwise). She designs a circuit in which the absence of current flow represents falsehood. If both branches of the circuit are open - that is, if both p and q are false - then no current flows, and the whole statement is false. If one or both of the branches is closed, then current flows, and the whole statement is true. The circuit is a parallel circuit.
(d) Suppose on the other hand that computer designer Mr. B decides that the absence of current flow represents truth instead. If both p and q are true, the branches are both open, no current flows, and the statement is true. If one of p and q is false, then one of the branches is closed, current flows, and the whole statement is false. In this way, Mr. B's circuit represents the logic of the word "and": a statement of the form "p and q" is true if and only if both constituent statements are true (and false otherwise).
| Bthrm 1 | Bthrm 2 | Bthrm Availability |
| NV | NV | NV |
| V | V | V |
| NV | V | V |
| V | NV | V |
| P | Q | P or Q |
| F | F | F |
| T | T | T |
| F | T | T |
| T | F | T |
| P | Q | P and Q |
| T | T | T |
| F | F | F |
| T | F | F |
| F | T | F |
Let me introduce you to one of the dullest 'games' or formal systems you'll ever know. It's the *-game, and it has some very simple rules. Inputs and outputs are of two sorts, X and Y. Two inputs are required, and one output is the result. If both inputs are of sort X, then the output is of sort X; if the inputs are both or sort Y, or are mixed, then the output is of sort Y.
| P | Q | P*Q |
| X | X | X |
| Y | Y | Y |
| X | Y | Y |
| Y | X | Y |
The marvelous thing about the *-game is that, with the right interpretation, it's very useful.
Let X mean NV and Y mean V, and we've got the much desired news about bathroom availability on board the airplane.
Let X mean false and Y mean true, and we compute the meaning of "or."
With X interpreted as true, and Y as false, we compute the meaning of "and."
So despite their obvious differences in these four examples, they are all examples of the same formal system. The game has two tokens and two 'squares'. Of course, we give the squares and the tokens different names and meanings in each example, but that is irrelevant to their identity as formal systems. What makes them formal is precisely the fact that meaning is irrelevant.
When Machine Functionalism analyzes mental states, the number and types of basic components is far greater, and the relationships far more complex - so simple tabular displays are no longer useful or feasible. (It sometimes helps to think of networks of nerve cells.) But the basic idea remains the same. Every mental state can be modeled as some interpreted formal system.
MF is far richer in its resources than is Behaviorism. While Behaviorism is confined to inputs and outputs at the boundary between organism and world, MF allows one to look inside and to keep track of the internal processing. Thus MF adds a third element to the inputs and outputs of Behaviorism: it adds relations among internal mental states of the subject. Mr. Waldo and the Unintelligent Waldo Simulator have the same inputs and outputs (over a rather wide range, anyway), but the computations that occur internally are of very different sorts. The computations that occur in Mr. Waldo produce understanding, and the computations that occur in the Simulator do not. Or so MF would claim. MF thus seems to avoid the problem of pattern poverty: richer patterns are allowed, and we are told, precisely, of what sort the patterns may be. MF seems to make good on the computer model for the mind.
Behaviorism was based on a 'black box' model of mental processes: we can treat everything inside the body as if it were in an unopenable black box, and concentrate on the patterns of inputs and outputs that the body might exhibit. MF replaces the one big black box with a series of smaller black boxes, wired together in complicated ways. When we seek to discern the nature of internal processing that MF says is essential to defining mental states and processes, we break the process down into simpler sub-processes, much as one analyses a computer program using a flowchart. And just as we define the boxes in a flowchart by specifying their inputs and outputs, MF tells us to define the sub-processes by their inputs and outputs. We can go all the way down to the level of nerve cells, if we wish. When we talks about nerve cells, though, we must treat them as little black boxes which give a particular kind of output for a specified input: they are functionally specified, as are the pieces in a chess game, as members of a complex formal system. Where ingenuity is required is in figuring out how to wire the little boxes together to produce intelligent mental processing. So, if we are careful not to confuse MF and Behaviorism, it might be helpful to think of MF as a kind of "Internal Behaviorism", which directs us to look not only at skin-level behavior, but also at the behavior of the stuff inside.
It is this similarity between the two views that worries many people. How much good does it really do to increase the number of 'black boxes'?
First Objection: MF versus the Block Heads
Ned Block has given a famous objection to MF. He focuses on a particular sort of mental state: sensations, like pain or pleasure. Unlike, say, belief or knowledge, sensations have a characteristic feel. There is no particular way it feels to believe that 2+2=4, or to know that there is a book in front of you. But having a toothache produces a very definite, all too familiar sort of sensation, toothache pain, and its essential feature is that it hurts. In general, what makes a sensation the mental state that it is, is precisely the way that it feels.
Suppose that at some time in the not too distant future, psychologists have discovered enough about people to draw functionally accurate flowcharts for toothache pain. It is plausible to assume that 1 billion nerve cells are involved in producing that unpleasant sensation. Each of them has a certain array of inputs and outputs that contribute to toothache pain. MF says that all there is to toothache pain is the pattern of inputs and outputs involving the nerve cells wired together. The flowchart for toothache pain is therefore a certain formal system. This system, like any other such system, is capable of many different sorts of physical realizations. There is nothing in the nature of the system that says it must be built out of small, squishy organic parts. Any network of components that obeys the rules will suffice.
Drawing on these features of MF, Block asks us to perform a thought experiment. Imagine that we convince the government of China that it would greatly enhance its international prestige to participate in a great experiment in Cognitive Science. We equip each of the 1 billion or so individual members of the Chinese population with two-way radios, linked via satellite to Cognitive Science Central in Cambridge, MA. Each of them receives instructions about how to play the toothache pain game, each individual playing the role of an individual nerve cell. (It's a game with 1 billion pieces. To avoid costly litigation, we give them all anesthetic, so that they won't feel anything if they trip and fall during the simulation.) And, for the period of an hour, the Chinese population obediently and flawlessly plays the toothache game, just as it is played by your nerve cells.
About such a situation, MF must say that the Chinese population, considered as a whole, suffers from toothache pain during this time. Just as the flight attendants could substitute for the electrical components in the airplane bathroom status display, the Chinese people can substitute for your nerve cells. But something can suffer toothache pain only if it hurts. So MF entails that during the hour of the simulation, China hurts. That, says Block is absurd. You cannot produce that sensation by such functional simulation. So MF is wrong. We can summarize as follows:
Block's Absent Quality Objection to MF
If MF is true, then China hurts if it runs your toothache pain program
China does not hurt if it runs your toothache pain program
Therefore, MF is not true.
There is a common misunderstanding of this objection that we must avoid. Block is not claiming that participation in the experiment would produce pain in the individual players (-remember that they're full of anesthetic, anyway!). MF does not say that. Rather, it is the whole system - the group that is the entire Chinese population - that has the pain and hurts. MF does say that the system hurts. And that is what Block finds absurd. Pain (and other sensations), Block says, are partly a matter of functional role, but they also have an essential element of felt quality (sensation = functional role + felt quality), and this latter element is what MF leaves out. If Block is right, MF is bedeviled by pattern poverty, too: both external and internal behaviorism share a failing, if he is right.
Philosophical Inventions - the Zombie
MFs have often responded to this argument by claiming that since there is no more to mental activity than functional role, the roles imagined in this odd experiment do produce the sensation of pain. It sounds odd to say, "China hurts" because the whole situation is odd and unfamiliar; sometimes, however, the truth is odd. Some of the oddity can be dispelled by indulging in another science-fictional thought experiment, which I heard described in a talk by one of the best and funniest philosophers of science, Clark Glymour. We already have artificial substitutes for hearts, lungs, kidneys, joints and other organs or body parts. Some of these replacements don't work as well as the originals, but it's a matter of time before better designs are produced. While no neuroscientist would claim to know exactly how brain cells work, a great deal has been learned, and it is similarly a matter of time before we are able to produce artificial neurons; they'll be something like the microchips of today, though much smaller and perhaps more complex ("silicon patches"). What they will do is to replicate the patterns of signal processing in wet, squishy organic neurons; the patches will be functionally identical to biological brain cells, though made of different (perhaps more robust) material. I'd like you to imaginatively project yourself into the future era when such patches are commonplace; I'll bet that younger readers will live in this era. Suppose that after years of hard work, you are ready for a retirement full of the sports you never had time for, say, ping-pong. But your years of alcohol consumption have pickled the brain cells that control parts of your hands, and they no longer work as well as they once did for either pinging or ponging. Your neighborhood neurosurgeon offers you hope: "We can remove the pickled cells and patch in some functionally good-as-old silicon substitutes. It's a quick, virtually no-risk, office procedure done under local anesthetic, and your health insurance will pay for it. It'd be no worse for you than having a cavity filled by your dental robot. I've done thousands of these procedures myself with no problems at all. You can call my patients to check up on me. Are we going ahead, or are you going to sit by the sidelines and watch enviously while others enjoy themselves at the table?" (When I've presented this thought experiment to people, only about fifty percent say they want to go ahead at first. Then, I reassure them about the safety of the procedure, and the percentage jumps over ninety percent.) Suppose that you have the work done and are once again able to win at ping-pong. Years pass and you pickle more cells, which you also have replaced. Finally, you're a silicon head: all of the old neurons have literally gone down the drain and you have functionally perfect artificial neurons instead. In some ways, you're better off: you used to have to worry about being hit in the head with your partner's paddle, but now, the paddle breaks. From the inside, so to speak, things have remained as always, since the input-output patterns and the pattern of intercellular connections have all been carefully preserved. If all that sounds right to you, then you are having strong machine functionalist intuitions because all that's been preserved are the relevant functional roles, and that seems to be enough.
So the debate may seem to end in a draw. There is another objection that may prove more serious.
An Objection to MF, On Purpose
Even if Block's objection were satisfactorily answered, there is another, even deeper objection that the MF faces. The objection centers on the question of interpretation.
Recall the example of the "and" and "or" circuits, and the Bathroom Availability Display. In a sense we made clear, all of these were physical realizations of the same formal system. But this is not to imply that there is no difference in meaning among "and", "or" and "bathroom available". It is just that these differences do not emerge until we say how the elements of the system are to be interpreted, or assigned meaning. If we let the absence of current mean false, then the circuit represents "or"; if the absence of current mean true, then the circuit represents "and". Without an interpretation, the formal system is, strictly speaking, a meaningless game. It would make no sense to ask about a circuit, "What does its output mean?" if the formal system it realizes had not been given an interpretation.
The same holds for MF in its attempt to give an account of mental processes. They are not just automatic formal systems. They are automatic formal systems with an interpretation. But now an important question arises: what is it about the internal and external behavior of a physical system that determines which interpretation is the correct one? Suppose that Mr. Waldo is doing a proof in a mathematics class and thinking a thought of the form "p or q". He is not thinking of "and" and he is not on his way to the bathroom. Pretend that there is a parallel circuit in his brain that executes the relevant operation. If it is to be that thought, and not "p and q" (or "bathroom status"), then there must be some way of nailing down the interpretation of the physical states of the circuit. But there pretty clearly is nothing about the circuit that decides the matter: just as it is, it could (and does, according to our definition of "physically realizes") equally well realize "p and q" or "bathroom status". Why isn't it doing all three simultaneously?
Such questions are familiar in computer science. And there, they have familiar and reasonable answers: it is relative to the purposes of the designers and the programmers, who decide what the circuits are to represent. One and the same sequence of circuit flip-flops might be used to make weather predictions on Monday, and on Tuesday to simulate battle logistics in WWII. It all depends on what purpose the programmer had.
What happens if we try the same tack with the thinking things that we are? The MF then says that "S believes that p or q" means that S's body (brain) executes the belief program relative to purpose P; or B(S,P). We introduce an extra parameter, as the mathematicians say, and relativize all mental state interpretations to the purpose(s) of the interpreter(s). Now serious trouble arises for MF. Among the mental processes that MF seeks to analyze is that of having a (particular) purpose. And the same problem we had with "or" thoughts arises with the mental activity of having a purpose. So the "P" in "B(S,P)" should be analyzed as P(S,P). But the interpretation of the second "P" also depends on a purpose parameter, so it should be written P(S,P); and so it goes, indefinitely. MF is caught in a trap: it needs the notion of purpose to do any interpreting, and so it needs a notion of purpose to interpret "having a purpose". But it can't interpret "having a purpose" unless it first has the notion of purpose. So it must fix an interpretation of "having a purpose" before it fixes an interpretation of "having a purpose". And the latter is impossible. So MF cannot fix any interpretations, if it needs the notion of purpose to do so. Although this objection may seem like a trick, it is not. It presents a very serious problem for MF.
An Argument against MF, on Purpose
If MF is true, then every mental activity is an interpreted formal system (IFS)
If every mental activity is an IFS, then having a purpose is an IFS (=P*)
Every IFS presupposes P*
No interpretation can presuppose itself
Therefore, MF is false
There are several responses to this argument that are worth considering, mainly because they help to show how deep the problem is.
As long as we confine our view to the inside of the head, so to speak, the objection spells doom for MF. So let's expand our view and look outside the head. Perhaps there's something about the relationship between the head and the wider world that will supply the missing purpose parameter.
Suppose that we say, on behalf of the MF, that it is God, our designer, who determines the interpretations of our brain function. We can call this Theological Functionalism. [As a matter of sociological fact, no proponent of MF is going to want to take this route, since MF is meant as a way of saving a materialistic view of the mind; bringing God in at this point will seem (at best) like a cheat to the proponent.] Anyway, it won't do any good. Presumably, one of God's mental activities is having purposes. If there is significant similarity between his purposiveness and ours, then MF will apply to his as well, and the problem arises all over again (only now we're dealing with an incorporeal being, and it's pretty obscure how the computer model is supposed to apply there). If, on the other hand, God's being in that mental state is entirely dissimilar from any mental state any human being could have, then we will have no idea how to understand such religious claims as "God purposefully designed the Universe". Such claims play an important role in many religions, and in attempting to persuade people to worship God. If making such claims strictly unintelligible is the price of Theological Functionalism, then no believer will want to endorse the view; and that leaves no one else to endorse it.
A second, more popular, response is Evolutionary Functionalism. In a metaphorical sense, human beings are 'designed by Nature': their behavior is shaped by the forces of biological evolution, and this in turn shapes our brain and somehow supplies the missing purpose parameter. Since these forces work on behavior, Evolutionary Functionalism is Behaviorism augmented by a distinction between adaptive and non-adaptive behavior. Given what a miserable failure Behaviorism was, we ought to be skeptical about Evolutionary Functionalism. Is it really plausible to insist that all distinctions in thought must be tied to this distinction? Much of what we value most in human life - advanced science, art, music, literature - seem to have no direct bearing on biological survival, and some of it arguably gets in the way.
One standard objection to this view is the "Swampman [better: Swamperson] Objection." Suppose that a molecule-for-molecule exact replica of you were to form accidentally out of "swamp gas." It seems that s/he would surely be in the same mental states as you were at the time of formation and would even believe, for example, that she was you. But Evolutionary Functionalism says that, lacking an evolutionary history, Swamperson has no mental states at all, a strongly counter-intuitive result, to say the least!
We need to see the details of Evolutionary Functionalism before finding it as much as plausible. For what it's worth, I've seen lots of details, and I don't think it works.
One response to the argument is to reject the first premise and MF with it. We might make an exception for purposiveness, and endorse a version of Purpose-Directed Functionalism instead. We would need to defend that move, but we have the materials for a defense at hand. In effect, our discussion of theories of thought serves as a kind of argument for Purpose-Directed Functionalism.
To be honest - the best policy - Purpose-Directed Functionalism is better described as a kind of Dualism. Dualism comes in two varieties: there's Substance Dualism, according to which mental activity requires a special, nonphysical substance; and there's what I call Conceptual Dualism, according to which there is at least one concept unique to psychology that's needed to state and explain the facts about certain complex physical systems, namely, the fact that they think.
Purpose-Directed Functionalism is a brand of Conceptual Dualism. It shares with Cartesian Dualism the insight that physics won't suffice, but it rejects the reliance on soul-stuff.
Although I'm rather attracted to Purpose-Directed Functionalism, I do worry that it might be subject to a version of the Problem of Other Minds: how can I know about a particular physical system (say, my wife) that a basic notion of purpose applies to it - given that this notion cannot be analyzed in terms of concepts from physics, chemistry or computer science, useful though their concepts remain? (My wife believes that I worry too much. At least, I think she does.)
These worries aside, our investigation of attempts to reduce biology and psychology to physics provide strong support for the claim that some sort of purpose-directed explanation is needed in each. If that's right, then our last, best separation principle, Separation by Explanation, fails.
The investigation in this book leaves us with no compelling reason to think that there is any fundamental difference between religion and science. We therefore find ourselves on a level playing field where it would make as much sense to criticize a proposal or practice as either pseudoreligious or pseudoscientific. So the next time someone offers you an explanation, no matter what the subject (e.g., "Here's why you should live your life in this way ...."), don't waste your time finding out whether it has a religious or scientific origin, because those labels carry virtually no valuable information. Instead, evaluate the explanation on its own merits, using the universal criteria that embody the highest standards of thought. Why settle for less?
Go to next section: School Board Problems
