[Draft version of papers read at the 26th Annual meeting of the Society for Philosophy and Psychology, Barnard College, New York, June 16th 2000, and at The European Society for Philosophy and Psychology, Salzburg, Austria, 4th Sept 2000. Please do not quote.]
Abstract
In evaluating conditionals concerning what a person would have done in counterfactual circumstances, we suppose the counterfactual antecedent to be true, just as in what I loosely term the standard "Ramsey" procedure; but then we follow a different path--a simulative path--in evaluating the consequent. The simulative path imposes an implicit restriction on possible worlds, a procedural guarantee that the individual simulated is aware of or knows about the counterfactual condition. This difference makes clear the way in which reason explanations are implicitly cognitive and psychological.This implicit cognitivity has important consequences for conceptual development. If young children, even children of 2 or 3 years, follow the simulative path in interpreting counterfactuals about human action under counterfactual conditions, then they already give implicitly cognitive explanations. Their subsequent developmental task is chiefly to make explicit what they already ascribe implicitly. This will be is a process of subtraction, of shaving away some of the commitments a reason explanation makes.
Our understanding of others, according to the simulation theory, primarily depends on a procedure rather than a body of discursive knowledge or theory. My focus today is a simulative procedure for answering questions about action under counterfactual conditions. This procedure, I believe, is crucial to interpreting our everyday explanations of action.
1a) Gordon ran to the kitchen because there was smoke in the kitchen.
1b) The smoke detector sounded the alarm because there was smoke in the kitchen.
The explanations are not only linguistically similar; in some respects, at least, they are also logically similar. As with any 'because' explanation, each is true only if the explanans, the sentential clause that follows the 'because', is true. If it isn't true that there was smoke in the kitchen, then it isn't true that the smoke detector sounded the alarm because there was smoke, and it isn't true that I ran to the kitchen because there was smoke.
Further, the counterfactual commitments appear to be the same. If the smoke detector went off because there was smoke in the kitchen, then it wouldn't have gone off if there had not been smoke in the kitchen - leaving aside complications such as overdetermination. Likewise, if I ran to the kitchen because there was smoke in the kitchen, then I would not have done so if there had not been smoke in the kitchen - again with the same complications.
What is different is that, unlike the alarm's going off, my running to the kitchen is explained in terms of a reason. By this I mean a favorable consideration, a fact that, at least to my eyes, favored doing what I did. More specifically, the explanans is a premise of the reasoning that led me to act as I did - although in citing it I may make indexical adjustments such as change of tense1. I will call such explanations reason explanations. A reason explanation is true only if the explanans was a premise of the reasoning that actually led the agent to act. But it is usually noncommittal as to what the reasoning was that connected the premise with the action. In the example, the connective reasoning might be the following mix of theoretical and practical reasoning:
If there is smoke in the kitchen, then something must be burning. The best candidate is the pizza. If it is the pizza, then the best way to quickly stop the burning is to run into the kitchen, and once there, to turn off the oven. So, my decision plan: Run into the kitchen, and once there, turn off the oven.
But the reasoning might have been quite different. Imagine an employee who is simply following a rule:
If there is smoke in the kitchen and you are in a different room, run into the kitchen and turn off the oven.
There are also reason explanations of beliefs, desires, and various other propositional attitudes. For example,
He believed that something was burning because there was smoke in the kitchen.
What is strikingly different about reason explanations is that they are implicitly cognitive: If I ran into the kitchen because there was smoke, that being my reason, then it isn't sufficient that there was smoke. I had to know about it, I had to be aware that there was smoke; therefore, I had to believe there was. We're explaining the action by citing a non-psychological fact, namely that there is smoke in the kitchen, but doing so in a way that presupposes that I knew about it. A reason for action doesn't mention the agent's mental states, except in the rare case in which the agent's reason is the fact that he is in a certain mental state. So a reason explanation is not explicitly an explanation in terms of mental states such as belief and desire.
Because they are only implicitly cognitive and psychological, reason explanations look like non-cognitive, non-psychological explanations; the differences are hidden. Here are some other examples. The first of each pair is a reason explanation, the second not:
2a) The man is running because it is raining.
2b) The sewer is running because it is raining.
3a) I didn't strike the match, because it was so windy.
3b) The match didn't light, because it was so windy.
4a) She handed her car keys to the designated driver because she had had too much to drink.
4b) She told that joke because she had had too much to drink.2
C1a) If there had not been smoke in the kitchen, I would not have run into the kitchen,
I understand the antecedent to exclude the possibility that there was no smoke in the kitchen but I mistakenly believed that there was. For if I had had that belief, then I would have run into the kitchen anyway. To put the point in terms of possible worlds: In interpreting the counterfactual that corresponds to a reason explanation, we consider only possible worlds in which the counterfactual condition, described by the antecedent of the conditional, lies within the agent's epistemic horizon. Possible worlds in which there isn't smoke in the kitchen but I falsely believe there is are excluded.
One way to account for the difference between the A series and the B series of explanations is this: Intentional human action, my running to the kitchen, for example, is mediated by belief. My belief that there was smoke in the kitchen was an intermediate link - a representational link - between the situation, namely there being smoke in the kitchen, and my action. So in interpreting the A series of explanations we just take for granted the intermediate linkage. We rule out possible worlds in which the situation is different - there isn't smoke in the kitchen - but the intermediate link remains the same. We don't allow the counterfactual "facts" to vary independently of our beliefs.
Of course, matters are more complex. Even if we grant that beliefs cause actions, the causal relation must be of the right sort, excluding deviant causal chains - and there is no consensus among philosophers as to how to exclude these. Also, true beliefs are not generally caused by their truth conditions. For example, the necessary truths I believe aren't causes of my beliefs, nor are the moral truths, nor truths about the future. In short, the relation between reasons and the actions they explain is theoretically messy. If people are able to understand reason explanations, it is probably not by discursive knowledge of the complexities of that relationship.
But let's put aside these complexities for a moment. If a reason explanation is true only if the agent had the relevant knowledge or belief, what shall we say about explainers as young as 2 or 3 years old? There is something close to consensus that at some stage, at least, children do not have or at least do not employ the concepts of belief and knowledge. Yet they are able to give and to interpret explanations such as,
2a) The man is running because it is raining.
Some prominent developmental psychologists theorize that children at that stage understand actions to be caused directly by the external situation. Bartsch and Wellman speak of a desire psychology, which invests agents with desires but does not posit an intermediate link, belief, between desires and the world. The objects of desires are objects and states of affairs out there in the world. In his 1991 book, Understanding the Representational Mind, Josef Perner theorized that 3-year olds have a "Situation Theory." For them some aspect of the actual situation or state of affairs performs the role of belief. Thus actions are caused by desires in conjunction with some aspects of the actual situation.
There are differences among these authors, but all leave us with the question: Without using the concepts of belief and knowledge, how can young children understand that the man's running is an intentional and reason-driven action, whereas the sewer's running is not? If it were the presence of a belief intermediary that distinguished one from the other, then it would appear that young children interpret
5a) The boy ran to his mommy because he got stung by a bee,
precisely as they interpret explanations of natural phenomena:
5b) The boy's arm swelled up because he got stung by a bee.
Without employing cognitive concepts, how could they conceive an external event as causing someone's behavior by way of the agent's knowledge or awareness of it, not as a "brute" cause that has efficacy independently of the agent's awareness of it?
In my own past writing, I simply assumed that they made an implicit distinction. I suggested that young children explain and predict another's behavior as if the facts - that is, whatever they themselves believe to be so - were known to others as well. They treat the protagonist in a false belief story as knowing where the marble or the chocolate has been relocated. Because they fail to consider the possibility of false belief, they implicitly attribute knowledge to others, by default.
I didn't justify this claim. I might just as well have said that young children treat all explanations as non-cognitive. Thus they don't ascribe knowledge, even implicitly; they make no epistemological restriction of possible worlds - for how could they? Consequently, they don't explain actions in terms of reasons. Why choose the knowledge characterization over this non-cognitive characterization? I just took it for granted.
The distinction is between two ways of evaluating counterfactuals or answering counterfactual questions. One is the Ramsey test, the procedure proposed by Frank Ramsey and refined by Robert Stalnaker and others. The Ramsey procedure is roughly as follows: We add the antecedent of the conditional hypothetically to our existing store of beliefs - which is to say, essentially, we adopt a pretend belief. Then we make compensatory adjustments to our existing beliefs, the minimum adjustments needed to avoid inconsistency with the added belief. Some philosophers hold these adjustments to be justified by a possible world account: the minimal adjustments to beliefs correspond to the closest possible world in which the antecedent is true. The final steps are to ascertain whether the consequent is true or false, given our adjusted beliefs; and to evaluate the conditional accordingly.3
So, if there hadn't been smoke in the kitchen, would the smoke detector have gone off? Following the Ramsey procedure, I pretend-believe the following:
There isn't any smoke in the kitchen.
I know very little about the way smoke detectors work, but it is part of my background beliefs that they are generally designed to go off if and only if smoke crosses their sensor. So I take it for granted that my detector is in working order, I conclude that the alarm will not go off.
Now, would I have run into the kitchen, if there hadn't been smoke in the kitchen? Beginning as before, I pretend-believe that there isn't smoke in the kitchen, and I make adjustments to avoid inconsistency. Continuing down the Ramsey path, I might use background beliefs about the way people, or perhaps rational beings, are designed, the laws or general principles of their design; and then fill these in with metarepresentations, that is, with my beliefs about the beliefs, desires, and emotions that were operating in me at the time.
But here, in the case of predicting my own action, there is an alternative to the Ramsey procedure. Following this alternative path, I use my pretend belief,
There isn't any smoke in the kitchen,
in a different way. I add it to my existing beliefs - as well as to my existing desires and emotions and what have you - and I decide whether to run to the kitchen or not. In other words, following this path, I use the counterfactual antecedent as a premise of simulated reasoning. In general, when I follow the simulative path, I take the counterfactual antecedent as an input to reasoning that leads to a simulation of what is predicted: a pretend decision if the counterfactual concerns what I would decide to do, a pretend belief if the counterfactual concerns what I would believe, and so forth.
The simulative path is clearly available to children of 2 or 3 years.4 Children's pretend play sets up counterfactual premises for simulated reasoning leading to action - unlike the pure mind-play of hypothetico-deductive reasoning. When young children pretend that the banana is a telephone, they don't just speculate about the properties of their banana-telephone: they go on to use it as a telephone. They speak into it and they pretend to hear someone speaking through it. The counterfactual premise, "The banana is a telephone," becomes a premise of practical reasoning as well as of theoretical reasoning. They might reason: "This banana is a telephone. If it's a telephone, then if I pick it up, I can speak to Mommy. So I'll pick it up." Wherever their fancy takes them, whatever they pretend or suppose to be so, their practical reasoning keeps up; it tracks the counterfactual world. In this common form of pretense, possible worlds are limited to those in which I, the agent, exist and have the counterfactual "fact" available to my reasoning. Consider again:
5a) The boy ran to his mommy because he got stung by a bee.
5b) The boy's arm swelled up because he got stung by a bee.
Let's pretend the following:
The boy was not stung by a bee (or anything else).
Let's use this counterfactual premise as an input to a Ramsey procedure. The boy was not stung. Generally, people's arms don't swell without an insult such as a bee sting, and there was no other insult to the boy's arm. So the boy's arm did not swell up. Thus, if the boy had not been stung by a bee, his arm would not have swelled up. The explanation of the swelling, therefore, is that the boy was stung by a bee.
Now let's use this counterfactual premise as an input to a simulative procedure. I am playing in the sand with my friend. I have not been stung by a bee. Nothing else that is scary or painful is happening. Shall I run to my mommy? Of course not - Whatever for? So the boy does not run to his mommy. So, if the boy had not been stung by a bee, he would not have run to his mommy. The explanation of the boy's running to his mommy, therefore, is that he was stung by a bee.
Yes, it would. The task that lies before young children, if I am right, is to make explicit what they already ascribe implicitly. For this they need the capacity to shave away some of the many commitments a reason explanation makes. For example, to remove the commitment to truth while keeping - and making explicit - the commitment to belief: "I ran to the kitchen because I believed there was smoke in the kitchen" - that is, whether or not there actually was smoke in the kitchen. Or to leave the commitments to truth as well as belief, but to remove the further commitment to knowledge. Or to introduce a deviant causal chain leading from the belief to the action, so that the explanation, although cognitive, is not a reason explanation.
Let's focus on the important move from reason explanation to explanation in terms of belief. How do you weaken a reason explanation so as to remove the commitment to truth while keeping - and making explicit - the commitment to belief? You complicate the pretense. Let me explain. G. E. Moore noted that certain sentences were logically consistent, but yet, in some pragmatic way, inconsistent: for example,
(M) this mushroom is poisonous but I do not believe it is.
Hintikka made the interesting observation that the inconsistency of sentences like M disappears in a suppositional context. I can indeed suppose - or pretend or imagine - that a certain mushroom is poisonous and that I am ignorant of this fact: I picture myself innocently sautéing my False Morel, say, then eating it and getting very sick. This is, however, a difficult thing to imagine, for it requires feeding different premises into two parallel lines of suppositional reasoning - one for the first conjunct,
This mushroom is poisonous,
and a different one for the second conjunct,
I do not believe that this mushroom is poisonous.
The first conjunct becomes the input to a Ramsey procedure. Mushrooms such as this False Morel have the general property of making people who eat them sick. In the closest possible world in which I eat this mushroom, I get sick.
Corresponding to the second conjunct, I feed the premise,
This mushroom is not poisonous (it's a genuine Morel, say)
into a simulative procedure. In a children's game of make-believe, one would then act on the premise and, most likely, pretend-eat the mushroom. (It wouldn't be much fun otherwise.) Given the poisonous properties of the mushroom, however, you (or the person you are playing) are nomologically bound, in this closest possible world in which the mushroom is a False Morel, to be poisoned. Playing the role, you display in your behavior the operations of unseen forces: You begin to roll on the ground and moan in feigned sickness. Only then do the two pretenses, the inner simulative pretense and the outer Ramsey pretense, merge, bringing the dramatic irony to an end. You realize you've been hoodwinked by Mother Nature.5 The person you are playing is unaware of a fact that significantly affects the outcome of his behavior - like the protagonist of many false belief stories, who seeks an object in one place unaware that that it has been moved to another. The duplicity of a compartmentalized double pretense, a Ramsey procedure and a simultaneous simulative procedure with an independent input, is just what you need to remove the commitment to truth while keeping and making explicit the commitment to belief. Now, when you can perform such a pretense, you are prepared to give a new kind of explanation: I ate the poisonous mushroom because I believed it was not poisonous. Maxi went to the cupboard because he believed his chocolate was still there.
2.
Then there are ambiguous sentences. For example:
"He phoned his neurologist because his brain was in state S."
"She complained to her neurosurgeon because her amygdala was acting up again."
3. Several philosophers have pointed out that to accommodate counterfactual conditionals, as distinguished from indicative conditionals, the Ramsey test requires further adjustments to our existing stock of beliefs. But this complication will not affect my argument.
4. Note that it is not essential that I simulate the reasoning that would lead to the act or belief. There are shortcuts I can take. This way of reasoning with the antecedent presupposes that it is both counterfactually true and known to me.
5.
Another example that calls for thinking on two different semantic levels is the Ramsey test for the following counterfactual:
If I now believed I was about to win the lottery, I would be a fool.
For although I am supposing or pretending that I am about to win the lottery, I am also holding on to the facts (including the probabilities) as I actually see them: and it is in the light of these facts that I deem my counterfactual self a fool. Suppositions across two semantic levels, such as those of the form, "If I believed falsely that p," require parallel, or concurrent, suppositional tracks.
A different sort of split-level conditional is the following:
If I had won the lottery, then I would believe I had (that is, I would know about it).
Evaluating this conditional, and in general, the form, "If it were the case that p, then I would believe that p," calls for serial Ramsey tests. The first test (the "outer" pretense) takes as input p ("I won the lottery"). Its output - which may include evidence that p - is then taken as input to a second "inner" pretense. For example, if I had won the lottery, presumably there would be an official announcement saying so, such as a list of winning ticket numbers that included one mine. This output from the outer pretense then becomes input to the inner. Imagining such an announcement to have been made, I conclude (within the inner pretense) that I have won the lottery.