Fred Dretske, "If You Can't Make One, You Don't Know How It Works"

From Dretske, F. "If You Can't Make One, You Don't Know How It Works.", in Perception, Knowledge, and Belief (Cambridge: Cambridge University Press, 2000)

There are things I believe that I cannot say - at least not in such a way that they come out true. The title of this essay is a case in point. I really do believe that, in the relevant sense of all the relevant words, if you can't make one, you don't know how it works. The trouble is I do not know how to specify the relevant sense of all the relevant words.

I know, for instance, that you can understand how something works and, for a variety of reasons, still not be able to build one. The raw materials are not available. You cannot afford them. You are too clumsy or not strong enough. The police will not let you.

I also know that you may be able to make one and still not know how it works. You do not know how the parts work. I can solder a snaggle to a radzak, and this is all it takes to make a gizmo, but if I do not know what snaggles and radzaks are, or how they work, making one is not going to tell me much about what a gizmo is. My son once assembled a television set from a kit by carefully following the instruction manual. Understanding next to nothing about electricity, though, assembling one gave him no idea of how television worked.

I am not, however, suggesting that being able to build one is sufficient for knowing how it works. Only necessary. And I do not much care about whether you can actually put one together. It is enough if you know how one is put together. But, as I said, I do not know how to make all the right qualifications. So I will not try. All I mean to suggest by my provocative title is something about the spirit of philosophical naturalism. It is motivated by a constructivist's model of understanding. It embodies something like an engineer's ideal, a designer's vision, of what it takes to really know how something works. You need a blueprint, a recipe, an instruction manual, a program. This goes for the mind as well as any other contraption. If you want to know what intelligence is, or what it takes to have a thought, you need a recipe for creating intelligence or assembling a thought (or a thinker of thoughts) out of parts you already understand.

In speaking of parts one already understands, I mean, of course, parts that do not already possess the capacity or feature one follows the recipe to create. One cannot have a recipe for cake that lists a cake, not even a small cake, as an ingredient. One can, I suppose, make a big cake out of small cakes, but recipes of this sort will not help one understand what a cake is (although it might help one understand what a big cake is). As a boy, I once tried to make fudge by melting fudge in a frying pan. All I succeeded in doing was ruining the pan. Don't ask me what I was trying to do - change the shape of the candy, I suppose. There are perfectly respectable recipes for cookies that list candy (e.g., gumdrops) as an ingredient, but one cannot have a recipe for candy that lists candy as an ingredient. At least it will not be a recipe that tells you how to make candy or helps you understand what candy is. The same is true of minds. That is why a recipe for thought cannot have interpretive attitudes or explanatory stances among the eligible ingredients - not even the attitudes and stances of others. That is like making candy out of candy - in this case, one person's candy out of another person's candy. You can do it, but you still will not know how to make candy or what candy is.

In comparing a mind to candy and television sets I do not mean to suggest that minds are the sort of thing that can be assembled in your basement or in the kitchen. There are things, including things one fully understands, things one knows how to make, that cannot be assembled that way. Try making Rembrandts or $100 bills in your basement. What you produce may look genuine, it may pass as authentic, but it will not be the real thing. You have to be the right person, occupy the right office, or possess the appropriate legal authority in order to make certain things. There are recipes for making money and Rembrandts, and knowing these recipes is part of understanding what money and Rembrandts are, but these are not recipes you and I can use. Some recipes require a special cook.

This is one (but only one) of the reasons it is wrong to say, as I did in the title, that if you cannot make one, you do not know how it works. It would be better to say, as I did earlier, that if you do not know how to make one, or know how one is made, you do not really understand how it works.

Some objects are constituted, in part, by their relationships to other objects. Rembrandts and $100 bills are like that. So are cousins and mothers-in-law. That is why you could not have built my cousin in your basement, while my aunt and uncle could. There is a recipe in this case, just not one you can use. The mind, I think, is also like that, and I will return to this important point in a moment.

It is customary to think of naturalistic recipes for the mind as starting with extensional ingredients and, through some magical blending process, producing an intentional product: a thought, an experience, or a purpose. The idea behind this proscription of intentional ingredients seems to be that since what we are trying to build - a thought - is an intentional product, our recipe cannot use intentional ingredients.

This, it seems to me, is a mistake, a mistake that has led some philosophers to despair of ever finding a naturalistic recipe for the mind. It has given naturalism an undeserved bad name. The mistake is the same as if we proscribed using, say, copper wire in our instruction manual for building amplifiers because copper wire conducts electricity - exactly what the amplifiers we are trying to build do. This, though, is silly. It is perfectly acceptable to use copper wire in one's recipe for building amplifiers. Amplifier recipes are supposed to help you understand how something amplifies electricity, not how something conducts electricity. So you get to use conductors of electricity, and in particular copper wire, as a part in one's amplifier kit. Conductors are eligible components in recipes for building amplifiers even if one does not know how they manage to conduct. An eligible part, once again, is an ingredient, a part, a component, that does not already have the capacity or power one uses the recipe to create. That is why one can know what gumdrop cookies are, know how to make them, without knowing how to make gumdrops or what, exactly, gumdrops are.

The same is true for mental recipes. As long as there is no mystery - not, at least, the same mystery - about how the parts work as how the whole is supposed to work, it is perfectly acceptable to use intentional ingredients in a recipe for thought, purpose, and intelligence. What we are trying to understand, after all, is not intentionality, per se, but the mind. Thought may be intentional, but that is not the property we are seeking a recipe to understand. As long as the intentionality we use is not itself mental, then we are as free to use intentionality in our recipe for making a mind as we are in using electrical conductors in building amplifiers and gumdrops in making cookies.

Consider a simple artifact - a compass. If it was manufactured properly (do not buy a cheap one), and if it is used in the correct circumstances (the good ones come with directions), it will tell you the direction of the Arctic Pole (I here ignore differences between magnetic and geographic poles). That is what the pointer indicates. But although the pointer indicates the direction of the Arctic Pole, it does not indicate the whereabouts of polar bears even though polar bears live in the Arctic. If you happen to know this fact about polar bears, that they live in the Arctic (not the Antarctic), you could, of course, figure out where the polar bears are by using a compass. But this fact about what you could figure out if you knew does not mean that the compass pointer is sensitive to the location of polar bears - thus indicating their whereabouts - in the way it indicates the location of the Arctic. The pointer on this instrument does not track the bears; it tracks the pole. If there is any doubt about this, try using Mill's Methods: move the bears around while keeping the pole fixed. The pointer on your compass will not so much as quiver.

Talking about what a compass indicates is a way of talking about what it tracks, what information it carries, what its pointer movements are dependent on, and a compass, just like any other measuring instrument, can track one condition without tracking another even though these conditions co-occur. Talk about what instruments and gauges indicate or measure creates the same kind of intensional (with an "s") context as does talk about what a person knows or believes. Knowing or believing that that is the North Pole is not the same as knowing or believing that that is the habitat of polar bears even though the North Pole is the habitat of polar bears. If we regard intensional (with an "s") discourse, referentially opaque contexts, as our guide to intentional (with a "t") phenomena, then we have, in a cheap compass, something we can buy at the local hardware store, intentionality. Describing what such an instrument indicates is describing it in intensional terms. What one is describing is, therefore, in this sense, an intentional state of the instrument.

It is worth emphasizing that this is not derived or in any way second-class intentionality. This is the genuine article - original intentionality, as some philosophers (including this one) like to say. The intentional states a compass occupies do not depend on our explanatory purposes, attitudes, or stances. To say that the compass (in certain conditions C) indicates the direction of the Arctic Pole is to say that, in these conditions, the direction of the pointer depends in some lawlike way on the whereabouts of the pole. This dependency exists whether or not we know it exists, whether or not anyone ever exploits this fact to build and use compasses. The intentionality of the device is not, like the intentionality of words and maps, borrowed or derived from the intentionality (purposes, attitudes, knowledge) of its users. The power of this instrument to indicate north to or for us may depends on our taking it to be a reliable indicator (and, thus, on what we believe or know about it), but its being a reliable indicator does not itself depend on us.

"Intentionality" is a much abused word, and it means a variety of different things. But one thing it has been used to pick out are states, conditions, and activities having a propositional content the verbal expression of which does not allow the substitution, salva veritate, of coreferring expressions. This is Chisholm's third mark of intentionality.' Anything exhibiting this mark is about something else under an aspect. It has, in this sense, an aspectual shape.' Compass needles are about geographical regions or directions under one aspect (as, say, the direction of the pole) and not others (as, say, the habitat of polar bears). This is the same way our thoughts are about a place under one aspect (as where I was born) but not another (as where you were born). If having this kind of profile is, indeed, one thing that is meant by speaking of a state, condition, or activity as intentional, then it seems clear that there is no need to naturalize intentionality. It is already a familiar part of our physical world. It exists wherever you find clouds, smoke, tree rings, shadows, tracks, light, sound, pressure, and countless other natural phenomena that carry information about how other parts of the world are arranged and constituted.

Intentional systems, then, are not the problem. They can be picked up for a few dollars at your local hardware store. We can, therefore, include them on our list of ingredients in our recipe for building a mind without fear that we are merely changing the shape of the candy or the size of the cake. What we are trying to build when we speak of a recipe for building a mind is not merely a system that exhibits intentionality. We already have that in systems and their information-carrying states that are in no way mental. Rather, what we are trying to build is a system that exhibits that peculiar array of properties that characterizes thought. We are, among other things, trying to build something that exhibits what Chishohn describes as the first mark of intentionality, the power to say that something is so when it is not so, the power to misrepresent how things stand in the world. Unlike information-providing powers, the capacity to misrepresent is not to be found on the shelves of hardware stores. For that we need a recipe.

MISREPRESENTATION

Let us be clear about what we seek a recipe to create. If we are trying to build a thought, we are looking for something that cannot only say that x is F without saying that x is G despite the coextensionality of "F" and "G"3; thus being about x under an aspect, we are looking for something that can say this, as a thought can say it, even when x is not F. Unless we have a recipe for this, we have no naturalistic understanding of what it is that we think, no theory of meaning or content. Meaning or content, the what-it-is that we think, is, like intelligence and rationality, independent of truth. So a recipe for thought, where this is understood to include what one thinks, is, of necessity, a recipe for building systems that can misrepresent the world they are about.

Without the capacity to misrepresent, we have no capacity for the kind of representation that is the stuff of intelligence and reason.

Jerry Fodor focused attention on what he calls the disjunction problem for naturalistic theories of representation.' The problem is one of explaining how, in broadly causal terms, a structure in the head, call it R, could represent, say, or mean that something was F even though (if misrepresentation is to be possible) non-F-ish things are capable of causing it. How, in roughly causal terms, can R mean that something is F (the way a thought can be the thought that something is F) when something's being F is (at bests) only one of the things capable of causing R? For someone trying to formulate an information-based recipe for thought, this is, indeed, a vexing problem. But I mention the problem here only to point out that this problem is merely another way of describing the problem (for naturalistic theories) of misrepresentation. For if one could concoct a recipe for building systems capable of misrepresentation - capable, that is, of saying of something that was not F that it was F - then one would have a recipe for meaning, for constructing structures having a content that was independent of causes in the desired sense. This is so because if R can misrepresent something as being F, then R is, of necessity, something whose meaning is independent of its causes, something that can mean cow even when it is caused by a distant buffalo or a horse on a dark night. It is, therefore, something whose meaning is less than the disjunction of situations capable of causing it. In the words of Antony and Levine, it is something whose meaning has been "detached" from its causes.6 A naturalistic recipe for misrepresentation, therefore, is a recipe for solving the disjunction problem.' One way of solving problems is to show that two problems are really, at bottom, the same problem. We are making progress.

For this problem artifacts are of no help. Although clocks, compasses, thermometers, and fire alarms - all readily available at the corner hardware store - can misrepresent the conditions they are designed to deliver information about, they need our help to do it. Their representational successes and failures are underwritten by - and, therefore, depend on - our purposes and attitudes, the purposes and attitudes of their designers and users. As representational devices, as devices exhibiting a causally detached meaning, such instruments are not therefore eligible ingredients in a recipe for making thought.

The reason the representational powers of instruments are not, like their indicative (information-carrying) powers, an available ingredient in mental recipes is, I think, obvious enough. I will, however, take a moment to expand on the point in order to set the stage for what follows.

Consider the thermometer. Since the volume of a metal varies lawfully with the temperature, both the mercury in the glass tube and the paper clips in my desk drawer carry information about the local temperature. Both are intentional systems in that minimal, that first, sense already discussed. Their behavior depends on a certain aspect of their environment (on the temperature, not the color or size, of their neighbors) in the same way the orientation of a compass needle depends on one aspect of its environment, not another. The only difference between thermometers and paper clips is that we have given the one volume of metal, the mercury in the glass tube, the job, the function, of telling us about temperature. The paper clips have been given a different job. Since it is the thermometer's job to provide information about temperature, it (we say) misrepresents the temperature when it fails to do its assigned job, just as (we say) a book or a map might misrepresent the matters of which they (purport to) inform us about. What such artifacts say or mean is what we have given them the job of indicating or informing us about, and since they do not lose their job - at least not immediately - merely by failing to satisfactorily perform their job, these instruments continue to mean that a certain condition exists even when that condition fails to exist, even when some other condition (a condition other than the one they have the job of informing about) is responsible for their behavior. For such measuring instruments, meanings are causally detached from causes for the same reason that functions are causally detached from (actual) performance. This is why thermometers can, while paper clips cannot, misrepresent the temperature. When things go wrong, when nothing is really 98', a paper clip fails to say, while the broken thermometer goes right on saying, that it is 98'.

But, as I said, thermometers cannot do this by themselves. They need our help. We are the source of the job, the function, without which the thermometer could not say something that was false. Take us away and all you have is a tube full of mercury being caused to expand and contract by changes in the temperature - a column of metal doing exactly what paper clips, thumb tacks, and flag poles do. Once we change our attitude, once we (as it were) stop investing informational trust in the instrument, it loses its capacity to misrepresent. Its meaning ceases to be detached. It becomes, like every other piece of metal, a mere purveyor of information.

NATURAL FUNCTIONS

Although representational artifacts are thus not available as eligible ingredients in our recipe for the mind, their derived (from us) power to misrepresent is suggestive. If an information-carrying element in a system could somehow acquire the function of carrying information, and acquire this function in a way that did not depend on our intentions, purposes, and attitudes, then it would thereby acquire (just as a thermometer or a compass acquires) the power to misrepresent the conditions it had the function of informing about. Such functions would bring about a detachment of meaning from cause. Furthermore, since the functions would not be derived from us, the meanings (unlike the meaning of thermometers and compasses) would be original, underived meanings. Instead of just being able to build an instrument that could, because of the job we give it, fool us, the thing we build with these functions could, quite literally, itself be fooled.

If, then, we could find naturalistically acceptable functions, we could combine these with natural indicators (the sort used in the manufacture of compasses, thermometers, pressure gauges, and electric eyes) in a naturalistic recipe for thought. If the word "thought" sounds too exalted for the mechanical contraption I am assembling, we can describe the results in more modest terms. What we would have is a naturalistic recipe for representation, a way of building something that would have, quite apart from its creator's (or anyone else's) purposes or thoughts, a propositional content that could be either true or false. If that is not quite a recipe for mental bearnaise sauce, it is at least a recipe for a passable gravy. I will come back to the bearnaise sauce in a moment.

What we need in the way of another ingredient, then, is some natural process whereby elements can acquire, on their own, apart from us, an information-carrying function. Where are these natural processes, these candyless functions, that will let us make our mental confections?8

As I see it, there are two retail suppliers for the required natural functions: one phylogenetic, the other ontogenetic.

If the heart and kidneys have a natural function, something they are supposed to be doing independently of our knowledge or understanding of what it is, then it presumably comes from their evolutionary, their selectional, history. If the heart has the function of pumping blood, if (following Larry Wright9) that is why the heart is there, then, by parity of reasoning, and depending on actual selectional history, the senses would have an information-providing function, the job of "telling" the animal in which they occur what it needs to know about the world in which it lives. If this were so, the natural function of sensory systems would be to provide information about an organism's optical, acoustic, and chemical surroundings. There would thus exist, inside the animal, representations of its environment, elements capable of saying what is false. Although I have put it crudely, this, I take it, is the sort of thinking that inspires biologically oriented approaches to mental representation.10

There is, however, a second, an ontogenetic, source of natural functions. Think of a system with certain needs, certain things it must have in order to survive." In order to satisfy those needs it has to do A in conditions C. Nature has not equipped it with an automatic A-response to conditions C. There is, in other words, no hard-wired, heritable instinct to A in circumstances C. Think of C as a mushroom that has recently appeared in the animal's natural habitat. Although attractive (to this kind of animal), the mushroom is, in fact, poisonous. The animal can see the mushrooms. It has the perceptual resources for picking up information about (i.e., registering) the presence of C (it looks distinctive), but it does not yet have an appropriate A response (in this particular case, A = avoidance) to C.

We could wait for natural selection, and a little bit of luck, to solve this problem for the species, for the descendants of this animal, but if the problem - basically a coordination problem - is to be solved at the individual level, by this animal, learning must occur. If this animal is to survive, what must happen is that the internal sign or indicator of C - something inside this animal that constitutes its perception of C - must be made into a cause of A (avoidance). Control circuits must be reconfigured by inserting the internal indicators of C (the internal sensory effects of C) into the behavioral chain of command. Short of a miracle - the fortuitous occurrence of A whenever C is encountered - this is the only way the coordination problem essential for survival can be solved. Internal indicators must be harnessed to effector mechanisms so as to coordinate output A to the conditions, C, they carry information about. Learning of this kind achieves the same result as do longer-term evolutionary solutions: internal elements that supply needed information acquire the function of supplying it by being drafted (in this case, through a learning process) into the control loop because they supply it. A supplier of information acquires the function of supplying information by being recruited for control duties because it supplies it."

Obviously this ingredient, this source of natural functions, whether it be phylogenetic or ontogenetic, cannot be ordered from a Sears catalog. There is nothing that comes in a bottle that we can squirt on thermally sensitive tissue that will give this tissue the natural function of indicating temperature, nothing we can rub on a photosensitive pigment that will give it the job of detecting light. If something is going to get the function, the job, the purpose, of carrying information in this natural way, it has to get it on its own. We cannot "assign" these functions, although we can (by artificial selection or appropriate training) encourage their development. If the only natural functions are those provided by evolutionary history and individual learning, then, no one is going to build thinkers of thoughts, much less a mind, in the laboratory. This would be like building a heart, a real one, in your basement. If hearts are essentially organs of the body having the biological function of pumping blood, you cannot build them. You can wait for them to develop, maybe even hurry things along a bit by timely assists, but you cannot assemble them out of ready-made parts. These functions are the result of the right kind of history, and you cannot - not now - give a thing the right kind of history. It has to have it. Although there is a recipe for building internal representations, structures having natural indicator functions, it is not a recipe you or 1, or anyone else, can use to build one.

THE DISJUNCTION PROBLEM

There are, I know, doubts about whether a recipe consisting of information and natural teleology (derived from natural functions - either phylogenetic or ontogenetic) is capable of yielding a mental product - something with an original power to misrepresent. The doubts exist even with those who share the naturalistic impulse. Jerry Fodor, for instance, does not think Darwin (or Skinner, for that matter) can rescue Brentano's chestnuts from the fire.13 He does not think teleological theories of intentionality will solve the disjunction problem. Given the equivalence of the disjunction problem and the problem of misrepresentation, this is a denial, not just a doubt, that evolutionary or learning-theoretic accounts of functions are up to the task of detaching meaning from cause, of making something say cow when it can be caused by horses on a dark night.

I tend to agree with Fodor about the irrelevance of Darwin for understanding mental representation. I agree, however, not (like Fodor) out of the general skepticism about teleological accounts of meaning, but because I think Darwin is the wrong place to look for the teleology, for the functions, underlying mental representations (beliefs, thoughts, judgments, preferences, and their ilk). Mental representations have their place in explaining deliberate pieces of behavior, intentional acts for which the agent has reasons. This is exactly the sort of behavior that evolutionary histories are unequipped to explain. We might reasonably expect Darwin to tell us why people blink, reflexively when someone pokes a finger at their eye, but not why they deliberately wink at a friend, the kind of behavior we invoke beliefs and desires (mental representations) to explain. I do not doubt that the processes responsible for blink (and a great many other) reflexes are controlled by elements having an information-providing function (derived from natural selection). After all, if the reflex is to achieve its (presumed) purpose, that of protecting the eye, there must be something in there with the job of telling (informing) the muscles controlling the eyelids that there is an object approaching. But the representations derived from these phylogenetic functions are not mental representations. We do not blink because we believe a finger is being jabbed at our eye. And even if we do believe it, we blink, reflexively, before we believe it and independent of believing it. So even if there are representations whose underlying functions are phylogenetic, these are not the representations we would expect to identify with mental representations, the representations that serve to explain intentional behavior. For that, I submit, one needs to look to the representations whose underlying functions are ontogenetic.

Nonetheless, wherever we get the teleology, Fodor thinks it is powerless to solve the disjunction problem and, hence, hopeless as an account of thought content. I disagree. There are, to be sure, some problems for which teleology is of no help. But there are, or so I believe, some aspects of the naturalization project for which functions are indispensable. Whether teleology helps specifically with the disjunction problem depends on what one identifies as the disjunction problem. Since I have heard various things singled out as the disjunction problem, I offer the following two problems. Both have some claim to be called the disjunction problem. I will indicate, briefly, the kind of solution I favor to each. Teleology helps with only one.14

1. If a token of type R indicates (carries the information that) A, it also indicates that A or B (for any B). If it carries the information that x is a jersey cow, for instance, it carries the information that x is either a jersey cow or a holstein cow (or a can opener, for that matter). It also carries the information that x is, simply, a cow - a jersey cow, a holstem cow, and so on. This being so, how does an information-based approach to meaning get a token of type R to mean that A rather than A or B? How can an event have the content JERSEY COW rather than, say, cow when any event that carries the first piece of information also carries the second? To this problem functions provide an elegant answer. A token of type R can carry information that it does not have the function of carrying - that it does not, therefore, mean (in the sense of "mean" in which a thing can mean that P when P is false). Altimeters, for instance, carry information about air pressure (that is how they tell the altitude), but it is not their function to indicate air pressure. Their function is to indicate altitude. That is why they represent (and can misrepresent) altitude and not air pressure.

2. If tokens of type R can be caused by both A and B, how can tokens of this type mean that A (and not A or B)? If R is a type of structure tokens of which can be caused by both cows and, say, horses on a dark night, how can any particular token of R mean cow rather than cow OR HORSE ON A DARK NIGHT? For this problem I think Fodor is right: teleology is of no help. What we need, instead, is a better understanding of information, how tokens of a type R can carry information (that x is a cow, for instance) even though, in different circumstances and on other occasions, tokens of this same type fail to carry this information (because x is not a cow; it is a horse on a dark night). The solution to this problem requires understanding the way information is relativized to circumstances, the way tokens of type R that occur in broad daylight at 10 feet, say, can carry information that tokens of this same type, in other circumstances, in the dark or at 1200 feet, fail to carry. 15

The problem of detaching meaning from causes - and thus solving the problem of misrepresentation - occurs at two distinct levels, at the level of types and the level of tokens. At the token level the problem is: how can tokens of a type all have the same meaning or content, F, when they have different causes (hence, carry different information)? Answer: each token, whatever information it happens to carry, whatever its particular cause, has the same information-carrying function, a function it derives from the type of which it is a token. Since meaning is identified with information-carrying function, each token, whatever its cause, has the same meaning, the job of indicating F. Teleology plays a crucial role here - at the level of tokens. The problem at the type level is: how can a type of event have, or acquire, the function of carrying

THE RECIPE

We have, then, the following recipe for making a thought-like entity. It does not give us a very fancy thought - certainly nothing like the thoughts we have every day: that tomorrow is my birthday, for example, or that I left my umbrella in the car. But one thing at a time. The recipe will do its job if it yields something - call it a "proto-thought" - that has belief-like features. I, personally, would be happy with a crude de re belief about a perceived object that it was, say, moving.

Recipe: Take a system that has a need for the information that F, a system whose survival or well-being depends on its doing A in conditions F. Add an element, or a detector system that produces elements, that carries information about condition F. Now, stir in a natural process, one capable of conferring on the F-indicator the function of carrying this piece of information. One does not quite "stir" these processes in (the metaphor is getting a bit strained at this point). Once you have got the right system, adding functions is more like waiting for the dough to rise. There is nothing more one can do. You sit back and hope that natural processes will take a favorable turn. Just as one cannot expect everything in which one puts yeast to rise (it does not work in sand), one cannot expect to get representational "bread" from everything in which needed indicators are placed. You need a reasonably sophisticated system, one with a capacity to reorganize control circuits so as to exploit information in coordinating its behavior to the conditions it gets information about. You need a system, in other words, capable of the right kind of learning. These are special systems, yes, but they are not systems that must already possess representational powers. We are not, in requiring such systems in our recipe, smuggling in tainted ingredients.

If all goes well, when the process is complete, the result will be a system with internal resources for representing - and, equally important from the point of view of modeling the mind, misrepresenting - its surroundings. Furthermore, that this system represents, as well as what it represents, will be independent of what we know or believe about it. For we, the cooks, are not essential parts of this process. The entire process can happen "spontaneously" and, when it does, the system will have its own cache of original intentionality.

RATIONALITY: THE FUNCTIONAL ROLE OF THOUGHT

Whether this is really enough to have supplied a recipe for thought depends, of course, on just what one demands of thought. What does it take to be a thought? If all it takes is possession of content, then, perhaps, we have supplied a recipe of sorts for making a thought. But the product is pretty disappointing, a mere shadow of what we know (in ourselves and others) to be the fullest and richest expression of the mind. What I have described might be realized in a snail. What we want (I expect to hear) is something more, something exhibiting the complex dynamics, both inferential and explanatory, that our thoughts have. To have a cow thought it is not enough to have an internal, isolated cow representation. To be a cow thought, this representation must actually do what cow thoughts do. It must be involved in reasoning and inference about cows. It must, together with cow-directed desires, explain cow-directed behavior and rationalize cow-related attitudes and intentions.

There is validity to this complaint. If we are going to make a thought, we want the product to both look and behave like a thought. What we have so far devised may (to be generous) look a bit like a thought. At least it has representational content of the sort we associate with thought. Nonetheless, there is nothing to suggest that our product will behave like a thought. Why, then, advertise the recipe as a recipe for thought? I have, after all, already conceded that there may be representations of this sort, mechanisms in the body having an indicator function, that are not mental representations at all. When the underlying functions are phylogenetic (e.g., in the processes controlling various reflexes), the representations are not thoughts. They have a content, yes, but they do not behave like thoughts. They do not, for instance, interact with desires and other beliefs to produce intelligent and purposeful action. Why, then, suppose that when the functions are ontogenetic, when they develop in learning, the results are any better qualified to be classified as mental?

Since I have addressed this issue elsewhere,17 I will merely sketch an answer. A system that acquires, in accordance with the recipe, and in its own lifetime, the power to represent the objects in its immediate environment will also, automatically, be an intelligent system, one capable of behaving (at least insofar as these representations are concerned) in a rational way. To see why this is so, consider a process by means of which an indicator of F might acquire the function of carrying information about the F-ness of things - becoming, thereby, a representation (possibly, on occasion, a misrepresentation) that something is F. In order to acquire this status, the element must acquire the job of supplying information about the F-ness of things. The only way an element can acquire this job description, I submit, is by being recruited to perform control-related services because it supplies this needed information. If R is drafted to shape output because it supplies needed information about when and where that output is appropriate, then, no matter what further services may be required of R, part of R's job, its function, is to supply this needed information. That is why it is there, directing traffic, in the way that it is.

In achieving its representational status, then, R becomes a determinant of need-related behavior, behavior that satisfies needs when R carries the information it is its function to carry. Since R represents the conditions (F) in which the behavior it is called upon to cause is need-satisfying, R must, when it is doing its job, produce intelligent (i.e., need-satisfying) output. Even when it is not doing its job, even when it misrepresents, the behavior it helps produce will be behavior that is rationalized by the F-facts that R (mis)represents as existing. According to this recipe for thought, then, something becomes the thought that F by assisting in the production of an intelligent response to F.

Something not only becomes the thought that F by assisting in the production of an intelligent response to F, it assists in the intelligent response because it signifies what it does. When the capacity for thought emerges in accordance with the preceding recipe, not only do thoughts (together with needs and desires) conspire to produce intelligent behavior, they produce this behavior because they are the thoughts they are, because they have that particular content. It is their content, the fact that they are thoughts that F, not thoughts that G, that explains why they were recruited to help in the production of those particular responses to F. This, it seems to me, vindicates, in one fell swoop, both the explanatory and rationalizing role of content. We do not need "rationality constraints" in our theory of content. Rationality emerges as a by-product from the process in which representational states are created. Our recipe yields a product having the following properties:

1. The product has a propositional content that represents the world in an aspectual way (as, say, F rather than G even when Fs are always G).
2. This content can be either true or false.
3. The product is a "player" in the determination of system output (thus helping to explain system behavior).
4. The propositional content of this product is the property that explains the product's role in determining system output. The system not only does what it does because it has this product, but what it is about this product that explains why the system does what it does is its propositional content.
5. Although the system can behave stupidly, the normal role of this product (the role it will play when it is doing the job for which it was created) will be in the production of intelligent (need and desire satisfaction) behavior.

This, it seems to me, is about all one could ask of a naturalistic recipe for thought.

1. Roderick M. Chisholm, Perceiving: A Philosophical Study (Ithaca, N.Y., 1957), chap. 11.
2. This is John Searle's way of putting it; see his The Rediscovery of Mind (Cambridge, Mass., 1992), 131, 156. I think Searle is wrong when he says (p. 161) that there are no aspectual shapes at the level of neurons. Indicators in the brain, those in the sensory pathways, are as much about the perceived world under an aspect as is the compass about the Arctic under an aspect.
3. Despite even the necessary coextensionality of "F' and "G." A thought that x is F is different from a thought that x is G even if F-ness and Gness are related in such a way that nothing can be F without being G. This, too, is an aspect of intentionality. In Knowledge and the Flow of Information (Cambridge, Mass., 1981), 173, 1 called this the second (for nomic necessity) and third (for logical necessity) orders of intentionality. Although measuring instruments exhibit first-order intentionality (they can indicate that x is F without indicating that x is G even when "F' and "G" happen to be coextensional), they do not exhibit higher levels of intentionality. If (in virtue of a natural law between F-ness and G ness) Fs must be G, then anything carrying information that x is F will thereby carry the information that it is G. Unlike thoughts, compasses cannot distinguish between nomically equivalent properties. My discussion has so far passed over this important dimension of intentionality. Although I will return to it briefly, the point raises too many complications to be addressed here.
4. Jerry Fodor, A Theory of Content and Other Essays (Cambridge, Mass., 1990) and, earlier, Psychosemantics (Cambridge, Mass., 1987).
5. "At best" because, with certain Fs ("unicorn," "miracle," "angel," etc.) something's being F will not even be among the things that cause R.
6. Louise Antony and Joseph Levine, "The Nornic and the Robust," in Meaning in Mind: Fodor and His Critics (Oxford, 1991), 1-16.
7. Fodor puts it a bit differently, but the point, I think, is the same: "Solving the disjunction problem and making clear how a symbol's meaning could be so insensitive to variability in the causes of its tokenings are really two ways of describing the same undertaking" (A Theory of Content and Other Essays, 91).
8. For the purpose of this essay, I ignore skeptics about functions - those who think, for example, that the heart only has the function o£ pumping blood because this is an effect in which we have (for whatever reason) a special interest. See, for example, John Searle, The Rediscovery of Mind, p. 238, and Dan Dennett's "Evolution, Error and Intentionality" in The Intentional Stance (Cambridge, Mass., 1987).
9. Larry Wright, "Functions," Philosophical Review 82 (1973): 139-168, and Teleological Explanations (Berkeley, 1976).
10. E.g., Ruth Millikan, Language, Thought, and Other Biological Categories: New Foundationsfor Realism (Cambridge, Mass., 1984) and "Biosemantics," Journal of Philosophy 86, no. 6 (1989); David Papineau, Reality and Representation (New York, 1987) and "Representation and Explanation," Philosophy of Science 51, no. 4 (1984): 550-572; Mohan Matthen. "Biological Functions and Perceptual Content," Journal of Philosophy 85, no. 1 (1988): 527; and Peter Godfrey Smith, "Misinformation," Canadian Journal of Philosophy 19. no. 4 (December 1989): 533-550 and "Signal, Decision, Action," Journal of Philosophy 88, no. 12 (December 1991): 709-722.
11. This may sound as though we are smuggling in the back door what we are not allowing in the front: a tainted ingredient, the idea of a needful system, a system, that, given its needs, has a use for information. I think not. All that is here meant by a need (for system of type S) is some condition or result without which the system could (or would) not exist as a system of type S. Needs, in this minimal sense, are merely necessary conditions for existence. Even plants have needs in this sense. Plants cannot exist (as plants) without water and sunlight.
12. This is a short and fast version of the story I tell in Explaining Behavior (Cambridge, Mass., 1988).
13. Fodor, A Theory of Content and Other Essays, 70.
14. I was helped in my thinking about these problems by Peter Godfrey-Smith's "Misinformation."
15. In Knowledge and the Flow of Information I called these circumstances, the ones to which the informational content of a signal was relative, "channel conditions."
16. I tackled that in Knowledge and the Flow of Information, 215fl:
17. Explaining Behavior, Chapters 4 and 5.