Paper on the Trade-off Hypothesis

Hagop and I have just finished a new paper on the concept of intentional action.  A large portion of the paper is devoted to challenging Edouard's Trade-off Hypothesis, though we also offer some more general considerations about the current debate concerning the concept of intentional action.  If you are interested in cutting to the chase and examining our new evidence against the trade-off view, you should look directly at section 2.3 (though 2.1 provides some relevant background to this section).  We would, of course, be excited to receive comments on any and all parts of the paper.  Hopefully, this will help to spur new interest in the topic, ahead of a panel on these issues at the Central APA in two weeks. 

The Paper: Download trading_for_tradeoff.doc

Comments

[tnadelhoffer]

Hagop and Mark,

I have not forgotten about you, I've just been swamped. I have printed and read the paper though--which I enjoyed. I will try to post something tomorrow.

[Edouard Machery]

Mark and Hagop,

This is a very enjoyable paper and it raises some interesting difficulties for the trade-off hypothesis (henceforth, TOH). I hope the paper will be submitted soon and that it will find a good home.

There are many things I agree with in the paper. First, I tend now to agree with the claim made in the last section: No simple explanation—including the TOH—will account for all the data about the intentional nature of side-effects. Second, I agree that the TOH seems poorly equipped to account for some of the evidence presented in the paper (but not all). So, the evidence does indeed suggest that the TOH might well be false.

Now, some comments/criticisms.

1. p. 5 and 6. You note that the TOH is ambiguous, as I myself did in the paper: It does not specify whether it is the values of the character described in the probe (the CEO, the lieutenant, etc.) or the values of the subject reading the probe that matter when the subject categorizes the side-effect as a cost. You argue that Josh’s findings with the harm case are more consistent with the latter than with the former. This is correct for the harm case.

Now, I don’t find it implausible that in some contexts, the values of the character described in the probe will matter, while in other contexts (maybe in most), the values of the subject will matter. Certainly, as a proponent of the TOH, I do not have to deny this possibility. What matters from my perspective is that the subject conceives of the side-effect as a cost which is incurred in order to get a benefit.

2. I find the discussion in the 2nd paragraph of p. 6 unconvincing. For the sake of the argument, suppose that most people say that the side-effect in the harm case and other cases is intentional because they conceptualize it as a cost. Suppose also that when they do so, it is their values—not the values of the character described in the probe—that matter. People might conceptualize something as a cost for various reasons, including because they judged it to be morally objectionable. If this story is correct, in every case—i.e. when the side-effect is morally objectionable and when it is not—the reason why people judge a side-effect to be intentional is because they conceptualize it as a cost and because they believe that costs are intentionally incurred in order to reap benefits. I find this story deflationary. A side-effect is not judged intentional because it is morally wrong qua morally wrong or because it is (in some respect) bad qua bad, but because it has been classified as a cost. That is, it is qua cost that it judged intentional.

3. The free cup case and the extra-dollar case target Josh’s first theory that a side-effect is judged intentional if it morally wrong. Although in some papers, Josh has proposed a different theory—a side-effect is judged intentional if it is in some respect bad—in other papers, he still seems to favor some version of his earlier theory. Thus, in his most recent paper, as far as I remember the draft I read, Josh proposes that side-effects are the objects of an unconscious moral judgment.

4. p. 8. I am not sure to understand why I am guilty of an “illicit” conflation when I say that negatively valued means and side-effects are “by-products” engendered in the pursuit of a goal. If the only reason you want to phi is because if you phi, then y happens and you want y (i.e., if your desire for phiing is merely instrumental), then certainly you phiing is a byproduct of your pursuit of y.

5. Your reply to the parsimony argument is interesting, but not entirely convincing (p. 8). It is unclear whether people’s judgments about the intentional status of negatively valued and positively valued means should be explained in the same way. After all, in the former case, but not in the latter case, the side-effect is not desired.

6. I am really not convinced by the objection to my conclusion from the dog and the worker cases. First, if some x is appropriate, it is good in some respect. Second, my intuition is that people would have given similar answers if I had replaced “appropriate” with “good”.

I move now to your new and fascinating data

7. The first thing to note is that these findings are problematic for almost all accounts of the intentional status of side-effects. First, they are problematic for Josh’s account because, presumably, causing the death of soldiers is bad in some respects. Second, they are problematic for your own account because the attitude of the character described in the probe does not make any difference in subjects’ judgments. Third, they are problematic for my account (more on this in point 8). Prima facie, only Thomas’ account would have predicted these findings. (But it has problems on its own.)

8. It is clear that I would not have predicted your data on behalf of the TOH. So, yes, prima facie, they constitute evidence against the TOH. (Ron Mallon has further problematic data for the TOH.) The first cell (important/caring) is, in my mind, particularly damning.
I am less convinced, however, by your claim that the TOH predicts that people should be more likely to say that the side-effect is intentional when the goal is, as you put it, “unimportant” than when it is important. In cases 2 and 4, taking the hill is explicitly a goal. The story says that much. So, subjects do conceptualize it as a benefit—either with respect to their own values or, more plausibly in cases 2 and 4, with respect to the values of the agent. Thus, the TOH does predict people’s judgments in these cases. What it does not predict is people’s judgments in cases 1 and 3.

9. Note that the percentage of people in these cases is substantially higher than the usual harm case. I do not know what to do with this.

Anyway, this reply is already pretty long and I’d better stop for now.

Cheers

Edouard

[hagop sarkissian]

Edouard,

Thanks for your comments! Indeed, we hope to submit this paper soon, and so we are grateful for the feedback. I won't be able to respond to all your comments right now, but let me chime in on the following points (numbers are keyed to your comments above).

3. On the exegetical point regarding Josh's view – I think Josh has been fairly consistent with regards to it (until his most recent post, as you point out). For example, in his first two papers in Analysis, the feature is 'badness'. Badness is also the feature he notes in his paper with Mendlow, and his forthcoming overview in Phil Studies. However, I can see why some would take it as 'moral badness' instead, since he talked generally of 'moral considerations' in those early papers. There's room for debate here.

4. Hmm… The reason we find the conflation illicit is that a by-product is, almost by definition, a side-effect. In most dictionaries, a standard definition of a 'by-product' will have something like "a secondary or incidental product, often unforeseen or unintended". A means, by contrast, is intended. So this is why we thought the move toward calling them both by-products to be illicit (perhaps 'misleading' would have been better).

6. I find it fairly clear that an action can have any of the following features: a) bad but otherwise appropriate, b) good and otherwise appropriate, c) bad and otherwise inappropriate, or d) good yet otherwise inappropriate. Now, I wouldn't deny what you say here – that appropriate actions are 'good in some respect'. But the fact that I think we can make the sorts of distinctions I just noted is what made us discuss your interpretation of the dog and worker results.

Our own thought was that if you asked subjects: 'Was the death of the lone worker good or bad?', subjects would answer 'bad'. Now, what if you asked: 'Was the death of the lone worker appropriate?' Here, I think subjects would answer 'yes'. But I think this shows that the same action can be both bad and appropriate under different descriptions. Perhaps a better experimental design will be able to pick out which is operative in subjects' judgments. In any event, this is what made us think that Knobe would have predicted the results of your cases. [Incidentally, this discussion brings up a general point about the interaction between features within a case, a point I return to in #7, below.]

7. To clarify our own position: we postulate attitude as being a feature that might have an effect in some cases - notably, the 1st case in the present paper and the city planner case in our previous paper. But attitude is one of many features that can lead to folk ascriptions of intentional action, and its effect on any particular case will depend on its interactions with (and relationship to) other features of the case. So we don't think it will always have an effect, but we believe it does in some cases.

8. I see your point (though I think you meant *less* instead of *more* in this statement: "I am less convinced, however, by your claim that the TOH predicts that people should be *more* likely to say that the side-effect is intentional when the goal is, as you put it, "unimportant" than when it is important.") Strictly speaking, on your view, it seems as though if an agent is pursuing a goal, then it is a benefit in some sense (whether to the agent or the subject); the importance of the goal is neither here nor there.

Our general idea in cases 2 and 4 was that the clearer the benefit in the vignette, the more likely subjects would take the deaths of the soldiers as a trade-off. Hence, in one vignette, taking the hill is described as 'imperative', and in the other it is described as fleeting and worthless. If the benefit is fleeting and worthless, then sacrificing the soldiers seems less like a 'trade-off' and more like a callous and uncaring act. So, we think that whatever explains the high numbers in that case, it's not that a trade-off has occurred. (Most probably, it's that subjects blame the lieutenant and find him negligent and deplorable.) Now, I don't think this explanation necessarily rules out your interpretation of the data, but it's something that we could explore in future work. (More generally, I think it fair to say that both Mark and I were unsure just how to present the data in that section of the paper, and your comments might lead us to make changes in that regard.)

Anyway, it seems as though this post is getting pretty long itself!, so I, too, will stop for now.

Best,

Hagop

[Edouard Machery]

Hagop,

A few remarks.

About 4. Let's not start a terminological dispute. I was looking for a generic term that could be applied to negatively valued means and side-effects. They have this in common that they are not desired. "By-product" might not be the best term for this, but I did not find a better term.

About 6. What about the question "Was it GOOD to cause the death of the workman on the side tracks in order to save the five workmen? Yes/No? What would subjects say?

I should also note that in your probes (and in some probes used by others), the intentionality question bears on an action--in your probes, causing the death of soldiers in an attempt to capture a hill--that is closely related to a means. Certainly, causing the death is not *in itself* a means for capturing the hill, but putting these soldiers at risk is a means that is closely related to causing their death.

I assume I could have replaced the means paying an extra-dollar with the event having one less dollar in one's bank account in the extra-dollar case, without changing the findings. The latter event is not in itself a means, but is closely related to a means.

More generally, I agree with the important point that a single event can be described in many ways and that people's evaluations of this event depend on how it is described.

About 8. Yes, you are correct, I meant *less*. Thanks for the correction.

Generally, I think these are great data worth thinking about seriously.

Edouard

[Mark Phelan]

Hey Edouard,

Thanks for your kind and insightful comments. Sorry it's taken me a while to respond to your comments. When you posted, I had to prepare for class, and last night my 15 month old daughter spilled a drink on my laptop (using my wife's computer now).

I have a few things to add to what Hagop said, along with some other points. I'll respond according to your number scheme.

2. You write, "People might conceptualize something as a cost for various reasons, including because they judged it to be morally objectionable. If this story is correct in every case—i.e. when the side-effect is morally objectionable and when it is not—the reason why people judge a side-effect to be intentional is because they conceptualize it as a cost and because they believe that costs are intentionally incurred in order to reap benefits. I find this story deflationary.”

Well, there's deflationary and then there's deflationary. Certainly, your view (unlike Joshua's and Thomas's maintains that the proximal causes of intentionality judgments are not moral in nature. But at certain points you seem to claim that if the TOH is correct we will not be tempted to conclude that our folk theory of mind has been shaped by moral considerations. But so long as moral considerations are (as you seem to suppose they are, in some cases) upstream of judgments of cost, which are upstream of judgments of intentionality, then it's not clear that the truth of the trade-off view would show that our folk theory of mind has not been shaped by moral considerations.

It seems that here, and at certain points in the paper (for example, figure 4), you suppose subjects have reasons to conceive some effect as a cost. They think it’s prudentially unadvisable or morally bad, and reckon that it’s endured for some benefit. This causes them to think it’s a cost. But, you could also simply suppose that people perceive costs directly—and simply define costs, technically, as bad things endured for benefits. The veracity of this second version of the trade-off view would definitely show that our theory of mind had not been shaped by morality, but I don’t believe the first one would.

5. You write that our reply to the parsimony argument is interesting, but not entirely convincing, because it's not clear that negatively-valued and positively-valued means should be explained in the same way. You claim that in the first case - but not the second - the means is not desired. But, similar to your discussion with Hagop on point 6, one effect can be described and perceived in a variety of ways, and assessments of that event will vary depending on how it is perceived. In-and-of-itself the negatively-valued means is not desired (while the positively-valued means is desired in-and-of-itself). But qua means-to-the-end, both the negatively and positively-valued means are desired. So far as I know, it is unsettled which level of analysis we should pay attention to when determining whether negatively and positively valued means should admit of the same explanation—or whether some other feature entirely is the relevant one.

6. You write, "I am really not convinced by the objection to my conclusion from the dog and the worker cases. First, if some x is appropriate, it is good in some respect. Second, my intuition is that people would have given similar answers if I had replaced "appropriate" with "good". "

In response to the first point, I would agree that if x is appropriate, it is good in some respect. An appropriate side effect might be considered good as a concomitant to some desired end—or, to put the same point another way, if something is appropriate, it would seem that it is considered part of a situation the totality of which is good. But I understood Joshua's view as suggesting that what matters is whether or not the side effect is good in-and-of-itself.

My intuition, along with Hagop’s, is that, given the option, subjects would judge killing the worker to be bad, and saving the dog to be good. But I also feel the force of what you say in your second point here. Had you simply asked, ‘Was it good to kill the worker,’ and, ‘Was it good to save the dog,’ (and not asked, for example, ‘Was it good or bad to kill the worker,’) you might well have gotten similar answer (though I’m not sure that you would have—this is something that would have to be tested). In so far as you would get the same answers, do you think that this might be due to some sort of pragmatic pull in these cases? For example, if asked the question, “Was it good to kill the worker,” do you think subjects might feel a pragmatic pull to answer the question, “Good given the totality of the situation?” I mean, the answer to the ‘good in-and-of-itself’ question would just be so obvious: Of course its not good in-and-of-itself to kill the worker, and of course it is good in-and-of-itself to save a dog.

9. I'm not quite sure what you mean here. Do you mean that the divergence between our negative (not important) and positive (important) cases is smaller than in cases such as those in Knobe '03? This is right, though there's also a lot of divergence on those numbers in later repetitions of the study (c.f. McCann, forthcoming) as Nichols and Ulatowski point out, so its not clear how substantial the difference is.

In any case, in our negative vignettes (i.e. where the main goal is unimportant), our numbers do line up fairly closely with the negative cases in Knobe '03. (They are in the low to mid 70's). So, it may be that whatever explains the judgment of intentionality in these cases is similar to what explains it in Knobe's cases (and perhaps has nothing to do with the importance of the goal.) In the other cases, however, it may be that the importance of the goal acts as a mitigating factor for some people, causing them to 'pull away' from the judgment of intentionality. This is all pretty speculative, of course, but this is the sort of complex interaction between concepts in assessments of intentionality that Hagop and I had in mind. It's beginning to seem that assessing intentionality is a very complicated matter. It is not simply an additive matter, as a prototype theorist might have imagined.

Best,

Mark

[hagop sarkissian]

Your comments are very much appreciated, Edouard.

4. Yes, I don't think much hangs on this issue.

6. That's a very interesting question. I find my intuitions conflicted. 'Good' doesn't seem quite right to me. But if it turned out the folk thought otherwise, it wouldn't be the first time I'd be wrong. The folk can be surprising :)

Finally, what you say in the second paragraph of 6 is something Mark and I have also discussed and have concerns about. Just when an effect is a *pure* side-effect and when it is a means is not entirely clear. I think we can clarify things somewhat by using carefully crafted cases, but many of the examples in the literature (including some of our own, as you point out) seem less than immune to 'means-end' interpretation.

Best, Hagop

[tnadelhoffer]

Mark and Hagop,

For now, I have just two responses:

First, you claim that because the chairman says “I don’t care at all about harming the environment” this suggests that the chairman does not view harming the environment as a cost—but I am unsure this is correct. After all, he could just mean “harming the environment is a cost that I am happy to incur in order to make money.” Indeed, this is along the lines of what I argued in my paper on praise—namely, that the “I don’t care at all” condition in Knobe’s original studies was influencing people’s evaluation of the agent’s blame (or praise). This, in turn, was (inappropriately) influencing their intuitions about intentionality. In both the harm and help cases, the CEO’s attitude toward the side effect seems morally problematic. After all, in the harm case he did not care that he would be doing something bad and in the help case he did not care that he would bring about something good. As a result, they are both likely to elicit negative character judgments in the eyes of participants. Or so I have suggested. In some respects, my view overlaps with your suggestion at the end of the paper that information about the agent’s attitude towards x-ing can influence their judgments about intentionality. Of course, there is nothing wrong with looking at P’s attitude towards x-ing when trying to figure out whether P x-ed intentionally. The problem arises when judgments about P’s attitudes lead to moral judgments about P’s character which in turn influence judgments about intentionality. P’s attitude toward x-ing is relevant to the question of whether P x-ed intentionally. Yet neither P’s character nor the moral badness (or goodness) of x should factor into our judgments concerning whether P x-ed intentionally.

Second, I believe that if the subject-centered hypothesis were correct, it would no longer make sense to think that subjects are expressing their competency with the concept of intentional action--instead, an error theory would be needed to explain folk ascriptions of intentionality in side effect cases. Keep in mind that on so-called subject-centered approaches the *subjects’* evaluative judgments about the side effect are influencing their ascriptions of intentionality rather than the subjects’ judgments about the *agent’s* evaluative judgments about the side effect. But how can the subjects’ judgments about the relative badness or goodness of a side effect be relevant to determining whether an agent intentionally brought about the side effect? Figuring out whether P x-ed intentionally involves looking at the mental states of P and trying to determine how they relate to some action x. On the standard view, in judging whether P x-ed intentionally, we will limit ourselves to P’s cognitive and conative mental states. The moral properties of x will be irrelevant except insofar as these properties are relevant to the kinds of attitudes P adopts toward x-ing. But the mere fact that a subject S judges that x is bad cannot be relevant to whether P x-ed intentionally. How could it be? S’s judgment concerning the badness (or goodness) of x, much like S’s judgment concerning the blameworthiness (or praiseworthiness) of P, may indeed influence S’s judgments concerning whether P x-ed intentionally—but why think that is it appropriate when they do? It is not enough to simply reply that: This is how people ought to make judgments about intentionality because this happens to be how people make judgments about intentionality. After all, people regularly draw conclusions based on the gambler’s fallacy, but that’s no reason to think these conclusions are appropriate.

OK, that’s it for now. I am still thinking through your new studies. Hopefully, we can hash out some of the details over drinks next weekend in Chicago!

[tnadelhoffer]

I should confess that my interest in the performance error vs. core competency issue waxes and wanes partly for reasons Edouard discusses in the first section of his helpful paper, and partly owing to the kinds of worries Antti expresses in his interesting criticism of experimental philosophy.

I am ultimately more interested in what Eddy and I call the project of experimental descriptivism than I am in what we call experimental analysis.

But I nevertheless think we must be careful not to slide too hastily from "property x factors into S's intentionality judgment about P's y-ing" to "property x ought to play such a role in S's deliberations."

For instance, even if it could be shown that the race of an agent influences people's judgments concerning whether the agent x-ed intentionally, it would not follow that race should have such an influence. I think the same thing can be said about the participants' own moral beliefs. Even if their moral beliefs concerning both the side effect and the agent in the vignettes do shape their ascriptions of intentional action, it is unclear why we should conclude that these beliefs are entering into the participants' deliberations appropriately.

[Edouard Machery]

Mark,

A few comments (more tomorrow, hopefully).

1. You write that I "seem to claim that if the TOH is correct we will not be tempted to conclude that our folk theory of mind has been shaped by moral considerations".

I have a few issues with the claim that "our folk theory of mind has been shaped by moral considerations". The most important one is that proponents of this claim have done nothing to explain what it means. Nobody doubts that in some cases, mentalistic concepts are applied as a result of normative judgments. If you tell me that John is guilty of poisoning Jim, I will conclude that John *knew* that the white substance he put in Jim's coffee cup was not sugar. Here the concept of knowledge is applied as a result of a normative judgment. Nobody wants to deny this trivial fact. So, what do proponents of the claim "our folk theory of mind has been shaped by moral considerations" really claim?

I assume that by arguing that "our folk theory of mind has been shaped by moral considerations", you want to say something more than the trivial fact that some mentalistic concepts are applied as a result of normative judgments. But if this is so, the TOH really is deflationary, for the TOH is merely committed to the trivial fact.

More on the other points tomorrow.

Edouard

[Mark Phelan]

Hey Edouard,

I don't like the metaphor, "our folk theory of mind has been shaped by moral considerations," anymore than you do. And I agree that the claim is underdescribed so much as to be unhelpful. (I believe it originates in Knobe '06, so maybe Joshua can be of some help here.)

I don't personally want to argue that our folk theory of mind is shaped by moral considerations. I was only trying to argue, in the previous comment, that the truth of (your version of) the TOH doesn't show that our folk theory of mind wasn't so shaped. Without knowing what exactly is entailed by the suspect metaphor, so long as we suppose that moral assessments are still playing a triggering role for our concept of intentional action, we might leave the way open for one who wants to claim that moral considerations shape the folk theory of mind.

I was suggesting that a different version of the TOH might close off this avenue, so far as intentionality judgments are concerned (though of course it says nothing about knowledge judgments, or other folk concepts of mind).

Mark

[Mark Phelan]

Hey Thomas,

Some brief comments on your comments:

1. I think we're in broad agreement on this point. As Hagop and I tried to make clear in the paper, that the chairman says, "I don't care at all about the environment," is consistent with her nonetheless caring something about the environment. There are issues to be worked out regarding the semantics of this statement, but those needn't deter us so far as the topic at hand is concerned.

2. I'm inclined to agree that if the subject-centered view were correct we should suppose that subjects are simply incompetent with the concept 'intentional action'.

On a related point, you write, "S’s judgment concerning the badness (or goodness) of x, much like S’s judgment concerning the blameworthiness (or praiseworthiness) of P, may indeed influence S’s judgments concerning whether P x-ed intentionally—but why think that is it appropriate when they do? It is not enough to simply reply that: This is how people ought to make judgments about intentionality because this happens to be how people make judgments about intentionality."

I agree with this, and I think Hagop and I are more engaged in the project of 'experimental descriptivism' in this paper. Nonetheless, I would still endorse the sentiment from our previous paper that decisions as to how concepts should be used starts with the descriptive project of determining how they are used. This doesn't end the debate, it just begins it.

Mark

[Edouard Machery]

Mark,

Sorry to be slow to reply. I had numerous problems with my travel plans.

On your reply to 6.
You write: "But I understood Joshua's view as suggesting that what matters is whether or not the side effect is good in-and-of-itself."

I do not understand what this means. Any action can be subsumed under an infinite number of types and it is *always* evaluated with respect to a type. In the present case, the type could be killing the worker on the side-tracks, pushing a lever, causing the death of the lone worker on the sidetracks, causing the death of the lone worker in order to save five other workers, etc. etc.

The question I asked was: "Was it appropriate to cause the death of the workman on the side tracks in order to save the five workmen?" Under that description of the character's action, subjects judge that it is appropriate, and, I submit, they would also judge it is good (in spite of Hagop's disagreeing intuitions). So, under that description, wouldn't Josh predict that the action should be judged not intentional?

Of course, under other descriptions, people might judge that the action was bad (causing the death of someone is bad). So, maybe, you might reply that as long as the action is judged bad under some respect or other (viz. under some other description p), then it will be judged to be intentional, even under description q. (variant: under some SALIENT respect of other).

Suppose this is the view. A worry is that most actions are bad in some respect, but not intentional. Consider the free cup case in my paper. The character gets a free mug. That's good in some respect, but it might also bad in some other respect: e.g., the character has to carry the mug with him... But these actions are judged to be non-intentional. (Of course, you might say that this respect is not salient in the story.)

A more general comment
You are right that we need to develop a more complex picture of how people decide whether an action is intentional. Chris (from Mixing Memory) has often noted that the psychological explanations provided by experimental philosophers are typically tenuous. He is right.