Philip Kitcher

Columbia University


Every day, in laboratories in countries all around the globe, molecular biologists and their technical assistants manufacture new organisms. Some of these organisms are chimeras, expressing quite different properties in different clusters of their cells Ė flies or mice, for example, that contain both male and female tissues. Others are designed as factories for the manufacture of specific substances; thus itís routine to build bacteria with special genetic fragments inserted into them, and to use the organisms so engineered to churn out large quantities of proteins for medical, agricultural, or experimental purposes. This wide-ranging ability to create new forms of life depends on representations of biologically significant molecules, nucleic acids and proteins. The bio-engineers draw on maps that show the arrangement of genes, on reams of printout that identify the sequences of bases in particular regions of DNA, on pictures of molecular structures and on general claims about DNA replication, transcription, and translation.

These activities inspire an obvious thought: how could the practice of manufacturing organisms be so successful unless the maps, the sequence data, the pictures, and the main claims of molecular genetics were not at least approximately correct? When anti-realists of various stripes cavil at the "miracle argument", one is inclined to invite them (urge them? force them?) to look closely at the range and the precision of the accomplishments. "Go to the laboratory", one is tempted to say, "consider its ways, and be wise!". This, of course, is a blunt response, in the style of Dr. Johnsonís celebrated reply to Berkeley, and, although anti-realists would surely benefit from a greater awareness of the details of their opponentsí premise, much more needs to be said. In what follows, Iíll try to say some of it.

Letís begin with a statement of the realist position I hope to defend. Many of the statements made in everyday contexts and in the sciences are true in the sense that: (1) the names that occur in them refer to entities that exist independently of the person (or people) who make(s) such statements; (2) the variables that occur in them also range over such entities; (3) the predicates that occur in them denote sets of such entities (sets of ordered pairs of such entities, etc.); and (4) the truth values of the statements are generated in the fashion made familiar to us by Tarski. Iíll abbreviate these conditions by saying (perhaps tendentiously) that this version of realism adopts a modest Correspondence Theory of Truth. The correspondence part is, I trust, obvious; the modesty comes from the fact that thereís no commitment to the enterprise of giving a general account of reference or of offering a physicalist reduction of semantic notions.

Realists donít, of course, suppose that all the statements we make are true (in the modest correspondence sense), nor that all parts of science are to be treated in the same way. They recommend a particular treatment of those statements that have passed appropriate scrutiny of their epistemic credentials. According to realism, one way in which epistemic desert can be established is through the role that the pertinent statements play in successful practices. Specifically, realism endorses a Success-to-Truth rule, licensing inferences of the form:

S plays a crucial role in a systematic practice of fine-grained prediction and intervention.

S is approximately true.

Plainly, both this rule and the formulation of the modest correspondence theory contain terms that need clarification. Iíll let the clarifications emerge from a brief consideration of some important questions and objections.

There are at least six major concerns about the position Iíve sketched. First, what sense can be given to the notion of "independence" (or "mind-independence") that figures in the favored account of truth? Second, what do the qualifiers in the schematic premise of the rule mean? Third, whatís the notion of approximate truth that figures in the schematic conclusion? Fourth, isnít the success-to-truth rule belied by the history of inquiry, in which numerous successful theories have turned out to be false? Fifth, is there a justification for adopting the success-to-truth rule that avoids begging the question against the anti-realist? Sixth, even if a success-to-truth rule is accepted, why should it license the attribution of truth in the realistís preferred sense? In this essay, I hope to provide a detailed response to the last of these worries. But it will be helpful to start with a much briefer reply to the first five.


My version of realism is a homely doctrine, vested in commonsense appraisals rather than weighty metaphysics. Our notion of independence is generated from everyday situations in which we observe others interacting with entities, and recognize that the entities would persist with most of their properties intact, even if the others were absent. From such situations we form the conception not only of objects that exist independently of each of us, but also of objects that exist independently of all of us. Now it must be admitted that many, if not most, of the objects with which we deal daily are not entirely independent of human beings. Hardly any part of our environment is unaffected by the presence of Homo sapiens. Nevertheless, the facts that the chair on which I sit was made on an assembly line and that the street scene outside my window contains cars, buildings, lampposts, and even trees that are shaped by human hands hardly testifies to the kind of mind-dependence that anti-realists emphasize. None of these entities depends on me or on my thoughts, and their dependence on other people is thoroughly mundane and comprehensible. Even if the makers of the chair were in the room, their judgments about the chair would be judgments about an object that has its properties independently of their current representations and cogitations. Although it may be tricky to disentangle completely the notion of mind-independence that figures in realism from the quotidian causal dependence of objects on their makers and maintainers, itís not hard to appreciate that thereís an important distinction to be drawn here, even if the precise formulation of it is left to other occasions.

On to the second concern. I emphasize that the success-to-truth rule is concerned with statements that play a "crucial role" in the generation of "systematic" and "fine-grained" predictions and interventions for obvious reasons. Statements that belong to a scientific theory but that are never put to work in the derivation of forecasts or attempts to control the world gain no credit by association with the elements of the theory that are genuinely employed. The free-riding statements could be dropped from the theory without affecting the success of the practice. Similarly, unless we can proceed successfully across a wide range of instances and unless we can do things that are exact Ė and therefore exacting Ė there is room for worry that our achievements are too easy, that they could be the products of any number of rival systems, even quite misguided ones. The example with which I began, the practice of bioengineering, was deliberately chosen. I didnít offer a brief for the "whole of science" (whatever that might be) or for an isolated hypothesis, but for a body of knowledge in which some key statements and pictures are put to work, again and again, to yield a wide-ranging set of successes (across bacteria, fungi, plants, and animals) where the tasks undertaken are extremely demanding (localize properties to just these tissues, make a bacterial factory for just this hormone). I illustrate the "middle range" in which the success-to-truth rule applies by pointing to this instance as paradigmatic, and claiming only that cases that share enough of its features are plausible candidates for the use of the rule. Ultimately, realists should try to delimit the proper domain of the rule more precisely. For present purposes, I rely on a gesture towards an example and an invitation to assess similarity along the dimensions Iíve specified.

The third worry focuses on the notion of approximate truth. Here again my approach begins with our everyday notion, not with some technical construct born of efforts to articulate a philosophical thesis. There are two familiar contexts in which we make judgments about approximate truth, one in which we are concerned with a generalization and one in which weíre interested in ascribing a value to a quantity. "All trains for New Haven stop at 125th Street" says the sign Ė and we believe it until a pedant points out that, under some circumstances (fire at the station or some other emergency) the train passes through. Because we recognize that the circumstances are rare, we hail the original generalization as approximately true, and our reasons give clues to our meaning: "almost all" instances have the property ascribed, or, more precisely, the probability that something falling under the antecedent will fall under the consequent is close to 1. Assignments of magnitudes are handled in a similarly straightforward fashion. I believe that the statement "The gap for the new stove is 30 inches" is approximately true because the width of the gap is close to 30 inches. There is nothing mysterious about either of these cases, and, given an everyday understanding of Ďapproximately trueí in these two contexts, realists can rest content with their formulation of the rule they favor, unless or until itís shown that their uses of the rule commit them to conclusions that canít be assimilated to one or the other.

I end my brief (breezy?) response to the first five worries with some remarks about the two most prominent anti-realist arguments. Historically-minded anti-realists are fond of seeing the history of science as a series of ventures in which our predecessors confidently concluded the truth of "their theories" from the practical successes to which those theories were put Ė even though their successors (but our predecessors) were forced to revise those theories. One mistake in this line of argument is to move swiftly from noting that a past theory is not entirely true to declaring it false, and thus insinuating that it is entirely false. We do better to think from the beginning about the individual statements employed by past theorists and about the use of such statements in generating predictions and interventions. The next step is to focus on the items in the dismal catalogue of past embarrassments with which the anti-realist confronts us. Itís anti-realist hyperbole to suggest, for example, that such things as the "humoral theory of disease" or "theories of blending inheritance" were successful. The hard cases are those, like the wave theory of light, in which thereís a systematic practice of fine-grained prediction and intervention. In such examples, itís important to recognize that the parts of the theory that are involved in the successes are typically not abandoned in the march of inquiry: Fresnelís mathematics of wave propagation lives on in the contemporary physics textbook, and itís the mathematics that does the work in generating his experimental results (not any speculations about the ether). Once the success-to-truth rule is properly formulated, so that we see that the statements with which itís concerned are those that play a "crucial role" in the successful practice, the hard cases are no longer genuine counterexamples to the realist claim.

Even if this is admitted, the fifth worry is still pertinent. Itís one thing to show that the success-to-truth rule is consonant with serious reflection on the history of inquiry, quite another to convince those who are skeptical about inferences to the best explanation that the rule can be justified. After all, if someone has doubts about inferences to the best explanation in everyday scientific practice, why should that person think that the rule is sound? Once again, I see the rule as grounded in an everyday scenario. People sometimes find themselves trying to achieve something or to predict the behavior of a system when the entities on whose properties the outcome depends are inaccessible to them. In some such situations, outsiders (or the agentís later self) can recognize the entities and their pertinent properties and can watch to see how the accuracy of the agentís understanding of the causal structure is reflected in practical success. We all have experience of such occasions, perhaps most purely in the context of parlor games, and those experiences support the idea that, when the task is demanding, then the agent is highly unlikely to do well in a series of different versions of it unless that agent has an approximately correct view of the underlying entities. Indeed, Iíd suggest that the success-to-truth rule, as Iíve formulated it, is a natural generalization from our observations of others struggling with problems that involve entities they canít perceive and from our reflections on cases when weíve temporarily been in similar positions before gaining access to the pertinent entities. So I think we have inductive grounds for thinking that the success-to-truth rule is reliable.

I offer quick replies to five worries to motivate the idea that itís worth concentrating on the sixth. For the anti-realist could grant (wholeheartedly, or for the sake of argument) everything Iíve said in this section, and still deny the version of realism Iíve commended. Nothing that has been said shows that realists are entitled to use the success-to-truth rule to derive conclusions about the truth (or approximate truth) of the statements they endorse when truth is understood in the favored way. If realists want more than the right to attach Ďtrueí, conceived in a bland or deflationary fashion, to statements they like, then more work has to be done to show that correspondence truth is needed for explaining the successes that flow from inquiry.


There have been several penetrating critiques of the idea that realists can do the necessary work. Thus Michael Williams responds to the inference from success to truth as follows:

If I want a cold drink and believe that the refrigerator, rather than the oven, is the place to look, I will increase the likelihoood of getting what I want. This is because cold drinks are more likely to be found in the refrigerator than in the oven. To say that my having true beliefs makes it more likely that I will attain my goals is just a compact way of pointing to the indefinite number of mundane facts of this sort. It involves nothing so arcane as a physical correspondence theory of truth.

Michael Levin (a self-declared realist) offers a different line of objection. After proposing that explanations typically offer mechanisms for the phenomena explained, he continues:

And here is my problem: what kind of mechanism is truth? How does the truth of a theory bring about, cause or create, its issuance of successful predictions? Here, I think, we are stumped. Truth, like Mae Westís goodness, has nothing to do with it.

But the most sustained attack on the explanatory power of correspondence truth comes from Paul Horwich. In the course of a many-sided defense of a minimal theory of truth, Horwich responds directly to the realistís argument from success to correspondence truth:

Ö rather than saying

    1. The theory that nothing goes faster than light works well because it is true,
    2. we could equally have said

    3. The theory that nothing goes faster than light works well because nothing goes faster than light.

No further explanatory depth is achieved by putting the matter in terms of truth.

Horwichís rejoinder, which incorporates Levinís complaint about lack of explanatory power, rests on a detailed analysis of the kind of case that Williams cites, to wit an everyday situation in which an agent with a true instrumental belief stands a higher chance of attaining his goal. I think that weíll become clearer about the issues raised by all three critics if we begin by scrutinizing Horwichís attempt to show that explanations of everyday success in terms of true beliefs can always be replaced, without loss, by explanations that only deploy a minimal conception of truth.

Horwich considers the predicament of an agent, Bill; all that Bill wants is to have a beer; Bill thinks that merely by nodding he will get one. We suppose that, if Billís belief is true, then heíll get what he wants. Horwich offers the following derivation as an explanation.

    1. Bill wants <Bill has a beer>.
    2. Bill believes <Bill nods ® Bill has a beer>.
    3. [Bill wants <Bill has a beer> & Bill believes <Bill nods® Bill has a beer>] ® Bill nods.
    4. Bill nods.
    5. Billís belief is true.
    6. <Bill nods ® Bill has a beer> is true.
    7. <Bill nods ® Bill has a beer> is true iff Bill nods ® Bill has a beer.
    8. Bill nods ® Bill has a beer.
    9. Bill has a beer.
    10. <Bill has a beer> is true iff Bill has a beer.
    11. <Bill has a beer> is true.
    12. Bill gets what he wants.

Apart from the empirical assumptions specific to the case (1, 2, 5), the only premises of this account are the "practical syllogism" (3), and the Tarski biconditionals that flow from Horwichís minimal theory of truth (7, 10). Every other step is obtained by elementary logic from previous steps.

This appears to be a very powerful debunking of the realistís insistence that correspondence truth was required to explain success, for we seem to be able to give an explanation without invoking any notion of correspondence. All thatís needed is the minimal theory of truth. But we should ask exactly what has been explained. The answer, surely, is that if an agent has a true belief about means-ends relations, then that agent is likely to be successful. That isnít quite the explanandum that realists have taken to be crucial in their defenses of correspondence truth. Correspondence truth has been supposed to be necessary because of the way in which true beliefs about means-ends relations, or behavior that is as if the agent has true beliefs about means-ends relations, result from true beliefs about the objects that figure in the desired goal-state. The problem with Horwichís explanation is that it stops at a rather shallow level. Indeed, itís akin to the classic paradigms of "explanation" that subsumed facts about bird plumage under putative ornithological laws. The major instances of scientific explanation go much deeper, and I think we can emulate them in this case.

Billís performance is extremely local. We hear about one success on an occasion in which Bill had a commonplace belief about means-ends relations. But Iíve insisted, from the beginning, that the realistís argument concerned systematic successes. What exactly does that mean? I answer that the cases of concern are those in which our agent has a device, or set of devices, from which beliefs about means-ends relations (or behavior that is as if the agent had beliefs about means-ends relations) flow. For my original example, the devices consist of genetic maps, pictures of molecular structures, reams of sequence data and lore about nucleic acids and proteins. Weíll eventually return to that example. But itíll be easier to start with something much simpler.

Consider Ophelia. She wants to get to various places in the neighborhood of Elsinore. Luckily she has a map. Among other things, she uses that map to find the cliff that "beetles oíer its base into the sea", to bring "dead menís fingers" to her fatherís grave, and, eventually, to go to the brook. More specifically, she plans her comings and goings by looking at the map and adjusting her behavior with respect to the objects and spatial relations indicated by it. Itís an interesting question whether she can be attributed beliefs about the proper means of arriving at the destinations at which she aims: does she believe <Ophelia crosses the meadow and goes straight® Ophelia arrives at the graveyard> or does she just believe <the way to the graveyard crosses the meadow and goes straight on>? I shanít take a stand on this psychological issue at this stage, but simply concede to Horwich the idea that beliefs about means-ends relations are involved (via the "practical syllogism") in the genesis of Opheliaís behavior.

Ophelia wants to get to the brook. She believes that the path to the brook leads past the willows, as indeed it does. She follows the path and successfully reaches the brook. Hereís a simplified Horwich-style explanation of her success.

    1. Ophelia wants <Ophelia gets to the brook>.
    2. Ophelia believes <Ophelia follows the path past the willows® Ophelia gets to the brook>.
    3. [Ophelia wants <Ophelia gets to the brook> & Ophelia believes <Ophelia follows the path past the willows® Ophelia gets to the brook>] ® Ophelia follows the path past the willows.
    4. Ophelia follows the path past the willows.
    5. Ophelia follows the path past the willows® Ophelia gets to the brook.
    6. Ophelia gets to the brook.
    7. Here, Iíve omitted some of the steps in Horwichís original explanation, since theyíll be inessential to the discussion.

      The crucial question to ask is why 14 holds. Obvious answer: Ophelia has looked at the map. More exactly, Ophelia is a competent map-user. She knows which symbols on the map stand for which things in the environs of Elsinore and she knows which spatial relations on the map correspond to spatial relations among the places in her vicinity. So we could deepen the Horwich-style explanation by exhibiting 14 as the consequence of underlying facts about the situation.

      13a. Ophelia believes <the map is accurate>.

      13b. The map shows a line running from the castle past the willows to the brook.

      13c. For any map of this type, if the map shows a line running from A to B to C, then A, B, C lie on a path.

      13d. Ophelia is competent to read a map of this type.

      13e. Any competent reader of a map of this type who sees a line on the map joining A, B, C will believe <the map is accurate® thereís a path joining A, B, C>.

      13f. Ophelia believes <the map is accurate® thereís a path joining the castle, the willows and the brook>.

      13g. Ophelia believes <thereís a path joining the castle, the willows and the brook>.

      13h. Ophelia believes <thereís a path joining the castle, the willows and the brook> ® Ophelia believes <Ophelia follows the path past the willows® Ophelia gets to the brook>.

      14. Ophelia believes <Ophelia follows the path past the willows® Ophelia gets to the brook>.

      I apologize for excessive pedantry, but itís important to recognize just whatís involved in the generation of Opheliaís belief about means-ends relations.

      We can now begin to discern just where the appeal to correspondence comes in. The ultimate source of Opheliaís success isnít 14 but 13a, her belief that the map is accurate. That belief leads Ophelia to trace a spatial route through the world that corresponds to a set of spatial relations on the map. Her feet trace a line (speaking loosely) from castle to willows to brook that (roughly) accords with a line on the map, and itís because of the (rough) accord between line-in-the-world and line-on-the-map that she succeeds. To put the point another way, her recognition of a line on the map causes her to follow a trajectory of a particular type, and the lineís conformity to a path in the physical world, one that joins castle to willows to brook, explains why she gets where she wants to go.

      The point will emerge with greater clarity if we think more systematically about Opheliaís behavior, and if we liberate ourselves from the tendency to think that beliefs about means-ends relations must be proximal to the behavior. Think of the map as a device that, in the context of particular desires, indicates to those who trust it and those who are competent to use it actions that are suited to those desires. Then we can simplify our previous treatment by proceeding as follows.

    8. Ophelia believes <the map is accurate>.
    9. Ophelia is competent to read the map.
    10. Ophelia desires q [possible instances: <Ophelia gets to the brook>, <Ophelia gets to the cliff>, <Ophelia gets to the graveyard>].
    11. [Ophelia believes <the map is accurate> & Ophelia is competent to read the map & Ophelia desires q ] ® Ophelia does A(q).
    12. The map is accurate.
    13. [The map is accurate & Ophelia does A(q)] ® q.
    14. q.

This schema allows us to explain Opheliaís navigational successes in terms of the accuracy of the map. In each instance, her reading of the map generates an action A(q), which we can think of as the action pertinent to q. The crucial point for understanding the role of correspondence is to see that this action can be conceived in two different ways.

For we can think of the idea of "pertinence" as a psychological relation, holding between an agentís beliefs and desires, on the one hand, and actions that are viewed by the agent as appropriate to those desires, given those beliefs, on the other; or we can envisage "pertinence" as a relation between actions and outcomes on the other. On the first conception, the notion could be implicitly defined by 22; on the second, it could be implicitly defined by 24. But we canít offer both definitions at once. To do so would be to assume that thereís always an action that honors two quite different constraints, being singled out as the appropriate one for agents with particular beliefs and desires, on the one hand, and being causally effective in satisfying those desires, on the other. The explanatory mystery is precisely why there should be any action that meets these two constraints.

Realists think we can solve the explanatory mystery by invoking the notion of correspondence, and it should now be clear why. The parallel between the psychological life of the subject and the effectiveness of the action comes about because of a correspondence between elements in the subjectís representations and elements in the world. Opheliaís decisions to orient her body in particular ways stem from her reading of the map in the light of her preferred destinations. Those decisions turn out to be effective. Why do actions that are psychologically pertinent (i.e. that satisfy 22) turn out to be causally pertinent (i.e. satisfy 24)? Because of the relations between parts of the map and parts of the world Ė relations of correspondence between symbols and places, lines and paths.

Another perspective on Opheliaís route-finding may prove helpful. She gets where she wants to go again and again. How does she do it? Horwich-style explanations, even if they turned out to be psychologically plausible in their invocation of beliefs about means-ends relations, wouldnít yield any insight into the pattern of success. They would be like the brute force explanation of the prevalence of male to female births that grinds out the details of a million copulations and pregnancies. A good explanation would show us what all the instances have in common Ė and, indeed, by using Fisherís sex-ratio argument, we can provide such an explanation. In the case at hand, we also have a unifying explanation: Ophelia proceeds again and again, by reading the map. Reading the map generates, if you like, beliefs about means-ends-relations Ė it surely leads, with or without this mediation, to her following particular trajectories. Why do these beliefs about means-end relations turn out to be true, the trajectories lead her to her goals? The answer is that the correspondence of the symbols on the map to places and spatial relations in the vicinity of Elsinore aligns Opheliaís intentional states with the causally effective course of action.

We can now respond to Levinís reasonable question and to Williamsí complaint that no arcane philosophical theory is required. The obvious force of Levinís request for a mechanism is that a significant amount of scientific explanation does proceed by identifying mechanisms (think of contemporary biochemistry, for example), and itís far from obvious that articulating a correspondence theory of truth would display anything that might be called a mechanism. Yet not all scientific explanation is mechanistic. On some occasions, insight is generated by showing how, whatever the details of underlying mechanisms, the behavior of certain types of systems will have particular properties. Fisherís sex-ratio argument is a case in point. Even without delineating the grubby details of the adjustment of the sex-ratio so that itís 1-1 at sexual maturity, Fisher shows why natural selection will always favor organisms with a propensity to have offspring of the underrepresented sex. (Thus, any departure from a 1-1 ratio at sexual maturity will create a selection pressure to restore the population to 1-1. The preponderance of male births is the result of differential mortality before puberty.)

The role of correspondence in the explanation of practical success is similar, for the point is to show that, whatever the details of the mechanisms, two processes are coordinated with one another. There are processes that go on in Opheliaís head as she reads the map and figures out what to do. There are processes that go on in the world as she moves through it. We donít explain why she reaches her goals by elaborating the details of the mechanisms that lead her to move as she does and that relate her movements to the physical environment. That, of course, would be the analogue of the "brute force" method of explaining the preponderance of male births by recounting the history of copulation, pregnancy, and birth. What needs explaining is why the mechanisms that lead from map-reading to intentions to actions (whatever they are) coordinate so well with the mechanisms that enable Ophelia to start out as she does and reach her various destinations (whatever the causal mechanisms are in the particular cases). The realist proposes that this coordination occurs because of the correspondence between map and world.

Finally, in response to Williams, I reiterate that the modest correspondence theory Iíve defended is no piece of arcane philosophy. Rather, as I suggested earlier, the notion of correspondence is rooted in familiar situations, and, in the example under discussion, it would be recognized by anyone who peered over Opheliaís shoulder and followed her comings and goings. More importantly, Williams, like Horwich, thinks that an indefinite bundle of "explanations" that appeal to beliefs about means-ends relations is all that we need to understand success. Now while there may be occasions on which this style of account is the best we can do Ė perhaps thereís no deeper explanation for the agentís having the instrumental belief he does Ė there are many other occasions when the beliefs about means-ends relations (or behavior that is as if there were beliefs about means-ends relations) result from the use of the same resources again and again. As Iíve been arguing, itís precisely in these latter instances that the realistís notion of correspondence can play an explanatory role.


The case of Ophelia is a useful illustration, but it artificially simplifies the situations that are of most interest. I return, in conclusion, to the example with which I began.

How exactly do the ventures of engineering new organisms and manufacturing proteins work? Iíll focus on the latter case. To build a bacterium that will churn out large quantities of a desired protein Ė clotting factor or insulin, letís say Ė an agent (or, more likely, a group of agents) must be able to produce large quantities of DNA with appropriate sequences, must be able to incorporate that DNA in a form in which it can be inserted into and retained by a host bacterium, must be able to maintain the colony of descendant bacteria, and must be able to harvest the desired product.

Each of these four tasks involves many subtasks. If you aim to produce insulin, then you have to be able to identify DNA fragments that code for insulin, as well as the sequences that regulate insulin production. Typically, youíll use your knowledge of the sequence, or part of the sequence, to make a probe that attaches to the right region of the human genome. You begin by chopping a complete complement of human DNA into lots of fragments using restriction enzymes (proteins that cut the DNA at sites where particular sequences are found); you then manufacture a radioactively-labelled DNA with a sequence thatís complementary to some distinctive sequence in the insulin gene; this will pick out the fragments that you are seeking (theyíll fluoresce). The fragments are then incorporated into a small circular DNA molecule (a plasmid) that can be injected into a bacterium. Of course, youíll have to choose your bacterium carefully so that the fragment will be retained even under replication and so that the production of an extra protein wonít interfere with the cellular metabolism.

When this complex task is undertaken by a single person (itís usually distributed over a group, since laboratories are no more inclined to build certain kinds of molecules from scratch than they are to blow their own glass), that person draws on lots of different kinds of resources: most obvious thereís the basic account of how nucleic acids are made, how they yield proteins through transcription and translation, and so forth; then there are maps of human and bacterial chromosomes, accompanied with vast amounts of information about sequences at specific sites; there is a catalogue of restriction enzymes with their particular sequences for splicing; there are descriptions of the properties and requirements of different bacteria; and, of course, there are instruments that are supposed to carry out or to record causal transactions among molecules, to separate out molecules of particular kinds or to reveal molecules that have been treated in particular ways. The insulin-maker proceeds by drawing on many of these resources. The person manufacturing clotting-factor will draw on a slightly different selection. The bioengineer who wants a particular kind of chimeric mouse will use another combination. And so it goes. In the case of Ophelia, there was a one-many relation between the underlying resource and the practical successes: the map was always used and the plan was formed in very similar ways in each case. In the practice of building new organisms and manufacturing proteins, the relation is many-many: overlapping selections are made from a vast body of lore and from an arsenal of equipment and techniques.

Furthermore, whereas the formation of Opheliaís plan was a rather simple matter Ė she looked at the map and picked out landmarks along her route Ė bioengineering (when itís not yet routine) involves considerable creativity. There are so many decisions. Which restriction enzyme should we use at the beginning? (We donít want our fragments to be too large or too small). How should we produce "sticky ends" on the fragment we want to insert? What is a good bacterium for the job at hand? Instead of thinking of a unique action thatís pertinent to the goal, there are typically many candidates, whose reliability the bioengineer must assess, and their promotion of or interference with other goals has to be evaluated by using the background set of resources.

In effect, then, when an individual scientist is undertaking a new project of manufacturing large quantities of a molecule, that scientist must engage in what Iíll call a rehearsal. The rehearsal consists in considering candidate causal processes and their consequences. In other words, thereís a psychological process in which various representational tokens occur in propositions that the scientist entertains or accepts or in images that the scientist contemplates. Some of these psychological representations may take a public form in the jottings and diagrams in a notebook. Because the process is complex, itís entirely possible that the first (second, third, etc.) plan that the scientist tries has unexpected consequences, that things donít behave quite in the anticipated way. But the record of this kind of research is that the set of resources on which the scientist originally drew can be employed to adjust the second rehearsal to the troubles that occurred in implementing the results of the first rehearsal (the first performance), and that, after a sequence of adjustments, a final rehearsal yields a final successful performance. The scientist ultimately makes large quantities of the desired molecule, or, to put the success in less tendentious terms, the diabetics eventually thrive on the manufactured stuff and the hemophiliacs can suffer cuts without uncontrollable bleeding.

The background resources thus allow for a sequence of rehearsals in which scientific ingenuity is manifested in different selections from them. Further, the failures, the partial successes, and the complete successes are comprehensible in terms of the lack of correspondence, partial correspondence and full correspondence between rehearsal and performance. What runs through the scientistís head is a prefiguring of the causal process that will play out in the laboratory, and success comes because of and to the extent that the psychological tokens correspond to the objects and relations in the causal process. Our scientist envisages that a particular enzyme, EcoRI, will cut a particular molecule between two specific sequences, CCG and AAT. Using symbols that stand for the various entities, the scientist prefigures the causal process that will unfold, and the performance works because of the correspondence between the symbols in the rehearsal and the entities and their properties in the world.

When thereís no such correspondence, the experiments break down. Vivid though the rehearsals may be, the courses of action to which they lead fail to achieve their intended ends. Out of the resources of molecular biology, through creativity and ingenuity, comes a tracing in thought of a route to a desired goal. Then the route is followed. If the course of the mental route corresponds to the route in the world, then success ensues; if not, not. That is the realistís intuition, and the explanatory role it finds for correspondence truth.