It looks like I was wrong

It is now widely agreed that people's moral judgments have some impact on their intuitions about whether a behavior was performed intentionally, but there has been considerable disagreement about precisely what sort of moral judgment is having an impact here.  Is it a judgment about whether the behavior itself was morally bad?  Or about whether the agent was blameworthy?  Or about whether it was wrong for the agent to perform the action? 

In earlier work, I suggested that the relevant moral judgment was a judgment as to whether the action itself was a bad one.  This hypothesis has been put to the test in a number of subsequent experiments, and I am sorry to say that the results overwhelmingly indicate that the hypothesis is false.  Indeed, my beloved hypothesis has been falsified independently in experimental work by Cushman, Machery, Nichols, Phelan & Sarkissian, Pizarro et al., Sinnott-Armstrong et al., and Wright & Bengson.   (Unfortunately, not all of the papers are available on the web.)

But it seems to me that these studies, taken collectively, also show something of deeper importance.  It's not just that one particular hypothesis turned out to be incorrect; it's that it hasn't been possible to identify any conscious moral judgment that can explain the full range of results.  That is, it doesn't seem possible to point to some specific type of conscious moral judgment and say, 'This particular type of moral judgment explains the full range of results we have amassed thus far.' 

I now think that the best way to understand these phenomena is to posit non-conscious moral judgments.  The basic process would then go something like this: As soon we encounter a behavior, we make an automatic and extremely swift moral judgment that remains entirely shielded off from consciousness.  It is this non-conscious moral judgment that influences intuitions about intentional action.  Of course, a person can then reflect and arrive at a conscious moral judgment after further thought, but this conscious moral judgment will not affect intuitions about intentional action; those intuitions will be determined entirely by the immediate, automatic non-conscious judgment.  (For a more detailed discussion, see the later sections of this new paper.)

In any case, whether or not you decide to read the full paper, I would love to hear your thoughts on these matters.  Does the new hypothesis look like a promising one?

Comments

[tnadelhoffer]

Joshua,

Thanks for the post. I will make a point of reading the paper soon. For now, I just wanted to briefly respond to the following remarks:

"I now think that the best way to understand these phenomena is to posit non-conscious moral judgments. The basic process would then go something like this: As soon we encounter a behavior, we make an automatic and extremely swift moral judgment that remains entirely shielded off from consciousness. It is this non-conscious moral judgment that influences intuitions about intentional action. Of course, a person can then reflect and arrive at a conscious moral judgment after further thought, but this conscious moral judgment will not affect intuitions about intentional action; those intuitions will be determined entirely by the immediate, automatic non-conscious judgment."

But this is loosely the view I defended in my dissertation (and to a lesser extent in my jury biasing paper)! Borrowing from the work of Alicke (and Haidt), I suggested that people's moral judgments--which often involve spontaneous, pre-potent, non-conscious, judgments of blame--sometimes influence their judgments about intentionality. So while I think that the idea you are putting forward is a promising one (not least of which because it is similar to my own!), I am unsure that it is properly called a "new" one. I will, however, read the paper soon so that I can give you more than this unhelpful knee-jerk reaction!

[dweisk]

Hi Josh. I haven't followed this literature as closely as I should, so this comment may be naive. My immediate question is what explanatory work is being done by the claim that the moral judgment here is non-conscious. Obviously this would explain why subjects find it hard to articulate the basis for their overall judgments about actions. But does its being non-conscious help to explain those observed patterns of judgment, from the theorist's point of view?

I don't see how. Here's why. Judgments, whether conscious or not, are distinguished by their content. To call a judgment non-conscious is to make a claim about its status in the hierarchy of mental processing. But it isn't to ascribe any particular content to it. Psychological explanations are content-sensitive, however. In particular, explanations about how people reason about cases are explained by the content of the judgments that subjects entertain in thinking about those cases. The claim that this thinking is non-conscious doesn't give us any explanation about how people are reasoning morally when confronted with those cases. So we don't have any detailed idea of what sort of reasoning process is taking place in them. We don't know what non-conscious contents they are using to reason morally in those cases. My impression was always that part of the goal was to explain what moral properties people attend to in reasoning about intentional action. I don't see how that goal is preserved in the new explanation.

Put it this way. If the judgments are non-conscious, why even think that they are moral judgments? Obviously because subjects evince sensitivity to moral facts. This indicates that we can have some grip on the content of their judgments: they have to do with one sort of property rather than another. But this means that we can get at part of the content of these non-conscious judgments, at least from the point of view of theorists. If we can get at part of it, why not all of it? Here, though, we run into the problem isolated by the studies you discuss. It seems too difficult to figure out any one moral property that people are sensitive to in these cases. (As a pluralist about concepts, I find it delightful and completely to be expected that this sort of thing happens, of course.) The problem of a diversity of principles is not solved by walling it off with the non-conscious properties. Neither does their being non-conscious seem to add anything to the phenomenon of people's sensitivity to a diversity of moral properties. If no single type of conscious moral judgment can explain the results, presumably no single type of non-conscious moral judgment can do it either, right?

So the bottom line is this. Probably there are a lot of non-conscious factors at work in making these judgments. There are everywhere else; and people are just awful at articulating their reasons (in some sense) for thinking and doing things. But adding that the judgments are non-conscious doesn't help explain the phenomenon of many moral principles. It just pushes it back into the realm of states that are harder to study because you can't simply read their content off of people's verbal reports or questionnaire replies. We still need to know what sorts of states are involved in this non-conscious moral reasoning, and whether anything systematic can be said about how they are activated and processed.

[Chris]

I think the theoretical capital you really buy when you posit unconscious over conscious processing is automaticity. That is, unconscious processes tend to be automatic and effortless. But more importantly, it opens several methodological doors.

The best way to test these, then, is to do something that might interrupt or alters those processes. For example, if emotion is a key to recruiting unconscious moral processing (as in Haidt's social intuitionist model), then manipulating mood should change behavior in tasks that ordinarily elicit those emotions and therefore recruit unconscious moral processing. If you just want to show that they're unconscious, you can show that, for example, working memory loads don't affect behavior on these tasks (because automatic processes don't require much in the way of working memory capacity).

[john a]

Joshua: Very interesting and refreshing.
Question: why not posit the 'shortcut' as a habitual way of thining, not an unconscious way of thinking? We can uncover habits of thinking or judgment making, but I wonder if we can ever uncover the unconscious. It seems to me that if the unconscious is uncovered it is now conscious and we are back were we began.

[tnadelhoffer]

Joshua,

Now that I have had the chance to read your paper, I wanted to offer you something more substantive than my earlier territorial pissing! The first thing I want to point out is that I think you have done a good job of opening up new vistas for experimental philosophers to investigate. More specifically, I found the pairing of the issues that arise with respect to reasons explanation with those that arise with respect to intentionality judgments both illuminating and fruitful. That being said, here are some disjointed observations:

First, it appears that our views are not very far off these days. Indeed, we both seem to be committed to the view that (a) spontaneous, affect-backed, non-conscious, pre-potent moral judgments sometimes drive intentionality judgments rather than the other way around, and (b) judgments about badness/blame have different effects on intentionality judgments than judgments about goodness/praise. However, I think we still disagree with respect to what we should say when these spontaneous moral judgments have this effect. Do you still want to defend a core competency model here, or are you coming over to the dark-side of the performance error theorists? After all, it seems to me that the moral heuristics you seem to rely on in developing your view will sometimes lead people to make mistakes with respect to their intentionality judgments. If nothing else, I am curious to know where you’re presently at on this issue ...

Second, you say the following:

“What we wish to show is that these intuitions can sometimes be influenced by people’s beliefs about the moral status of the behavior in question” (p. 7).

There is a general worry I have with the way you talk about beliefs and judgments in this paper. You make it sound as if the “transgression judgment” is non-conscious, but is that correct? Is the judgment itself non-conscious or is it simply that the process that leads to the judgment is non-conscious? How would you show that the judgment itself is non-conscious? Don’t participants conclude that the CEO is a transgressor in the HARM case? If so, are you suggesting that they make two transgressor judgments—one non-conscious (which drives the intentionality judgment) and one conscious (which follows from the intentionality judgment)? If so, then our views are even more similar than I originally thought. Keep in mind that I floated a very inchoate evolutionary explanation of how and why this might occur in my response to you and Gabriel. Here is what I said then:

"Imagine that long ago humans who were good at quickly detecting “harmers”—i.e. morally blameworthy individuals such as cheats, liars, thieves, rapists, murderers, and other scoundrels—were more apt to survive than those who did not. For humans living under these conditions, the best survival strategy would be to blame first and worry about exculpating or mitigating circumstances later. Given that this kind of survival strategy is at least possible—if not likely—it is clearly possible that our judgments about the blameworthiness of an action may come before our determination of whether the action was performed intentionally. More importantly, this would not undermine the usefulness of the concept of intentional action to the extent that this concept—along with a cluster of other closely related concepts such as purposely, knowingly, and accidentally—could still be used to amplify, verify, mitigate, or exculpate our antecedent attributions of moral responsibility. And even though in these situations our notion of intentional action would admittedly not play its usual role of fixing blame, it would nevertheless have an important role to play—viz., helping to ensure that our initial “harmer detection” reaction was not unjustified."

Isn't this partly what you have in mind when it comes to the view you're now defending?

Third, you say the following:

“The key thing to keep in mind here is that the relevant underlying capacities were not specifically engineered to do the best possible job at understanding reason explanations” (p.9).

But this is precisely what opens the door for performance error models! If, as you admit, participants' non-conscious moral judgments can conflict with their conscious moral judgments, then presumably these non-conscious judgments can lead to intentionality judgments that they themselves would reject upon closer consideration.

Fourth, I wish you would quit inserting an "agent doesn't care at all about x" clause into your vignettes. As I tried to show in one my earlier responses to your work, it creates an extra asymmetry with respect to praise and blame that I think you would do better to avoid. Minimally, I am curious to hear why you build no-caring clauses into the scenarios in the first place?

OK, that's it for now.

[Joshua Knobe]

Thanks for these comments; I think they really help to clarify some of the important issues. At this point, I thought it might be helpful just to say something on a more general level about how this new hypothesis is distinct from earlier ones.

Basically, it seems like we will end up adopting very different research programs depending on whether we focus on conscious judgments or on non-conscious ones. If we think that the effect is driven by conscious judgments, our aim will be to figure out precisely *which* judgments are involved. (That is, we will assume that we already know what judgments we are making and the task is simply to determine which of those judgments is driving the effect.) By contrast, if we look to non-conscious judgments, it seems like we will end up adopting a rather different strategy. We won't assume that we already know what judgments we are making and that the goal is just to figure out which of those judgments is driving the effect. Instead, we will look to the available data as a way of figuring out what sorts of judgments we were actually making in the first place.

That is the strategy I have been trying to pursue. I posit a non-conscious judgment (which I call a judgment of 'transgression'). The content of this judgment isn't identical to the content of any of our conscious judgments, but we can get a pretty good sense of how it works just by looking at people's intentional action intuitions. So the basic approach here is to look at the available data and then use it to unearth the criteria governing a type of judgment that we are not consciously aware of making.

Well, I imagine that this answer will not satisfy everyone, but I hope it at least helps to make it clear why I think the notion of non-conscious judgments can serve a useful role here.

[tnadelhoffer]

Joshua,

You say:

"Basically, it seems like we will end up adopting very different research programs depending on whether we focus on conscious judgments or on non-conscious ones. If we think that the effect is driven by conscious judgments, our aim will be to figure out precisely *which* judgments are involved. (That is, we will assume that we already know what judgments we are making and the task is simply to determine which of those judgments is driving the effect.) By contrast, if we look to non-conscious judgments, it seems like we will end up adopting a rather different strategy. We won't assume that we already know what judgments we are making and that the goal is just to figure out which of those judgments is driving the effect. Instead, we will look to the available data as a way of figuring out what sorts of judgments we were actually making in the first place.

That is the strategy I have been trying to pursue. I posit a non-conscious judgment (which I call a judgment of 'transgression'). The content of this judgment isn't identical to the content of any of our conscious judgments, but we can get a pretty good sense of how it works just by looking at people's intentional action intuitions. So the basic approach here is to look at the available data and then use it to unearth the criteria governing a type of judgment that we are not consciously aware of making."

I had a couple of quick questions.

First,what is the "rather different strategy" you have in mind?

Second, what would you count as disconfirming evidence of your new approach (e.g., what would need to be shown in order for you to give up this non-conscious model for some of its conscious model alternatives)?

Third, why do you say that the content of a non-conscious judgment is different than the content of its conscious counterparts. Why think this? For while their accessibility to consciousness is obviously different, it is unclear why the content need be any different. Imagine that in response to witnessing a moral transgression, I have a prepotent, affective, negative, non-concious judgment of the "blameworthy transgressor" type you mention. Now imganine that judgment leads me to an intentionality judgment, which in turn leads me to a conscious "blameworthy trangressor" moral judgment. Why think the second moral judgment has a different content than the first one?

Finally, you did not address my earlier question concerning the core comptetency vs. performance error issue. I wasn't trying to be unduly territorial--I was just unconvinced that the overall strategy you have adopted is an entirely new one (even if the particularities are quite novel). Overall, it seems that the two of us are in the same general camp now--even if we still disagree about a number of central issues. I suppose all I really wanted was a clarification concerning where are views come apart these days.

OK, that's it for now. I am keen to hear more.

[Joshua Knobe]

Thomas,

Thanks for these additional comments; I think they really help to bring out the important issues.

In my view, the non-conscious judgments at work here are not at all analogous to our conscious judgments of blame. For one example, take the case of the agent who violates a Nazi law that has been set up as a means of identifying victims to be sent to the concentration camps. The data indicate that people are non-consciously classifying this act as a 'transgression,' but it does not appear that there is any sense in which people are literally *blaming* the agent for disobeying the Nazi law, nor does it seem likely that people respond to such a case with feelings of anger or disapprobation that somehow interfere with the normal workings of their cognitive capacities.

For that reason, I find it improbable that the results are best explained by a model according to which our underlying concept of intentional action is entirely non-moral but our emotional responses end up clouding our judgment and leading to incorrect intuitions on these cases.

However, I should emphasize that I am still open to the possibility that a model of that type could eventually turn out to be correct, and I would love to see further data that might actually serve to vindicate it.

[tnadelhoffer]

Joshua,

Thanks for the reply. Unfortunately, I now have even more questions than I had before. While you conceded that an affect-driven performance error model could turn out to explain the data, you nevertheless suggest that you find such a model improbable. Yet you and Shaun seem to think an affect performance error model might best explain why people have compatibilist intuitions in response to morally laden scenarios but incompatibilist responses to scenarios where the emotional blow is softened. Why do you think it is improbable in the former case yet likely to be true in the latter? On my view, the very same tale roughly explains the asymmetry in both domains. In some sense, this is what one would expect given the recent research on proto-morality in primates. They have negative affective responses to transgressions, they adopt punitive strategies, they express what appear to be retributivistic emotions, and they even show sensitivity to issues of fairness, justice, and the like.

Of course, the judgments that make all of this kind of proto-moral behavior possible predate our concepts of free will, intentional action, moral responsibility, etc. As a result, the “old-brain” structures that are the seat of this aspect of our moral capacity can operate independently of the more sophisticated folk psychological concepts we use in understanding, predicting, and judging behavior. Indeed, this is the sort of story that Hauser, de Waal, and others tell when explaining how we came to be moral creatures in the first place. This is also the sort of story you are going to need in order to develop your own “non-conscious transgressor moral judgment model." After all, for thousands of years, presumably all our ancestors had were the non-conscious spontaneous judgments and the blaming behavior that went along with them—both of which were necessary for deterrence and conflict resolution.

But given that the underlying structures that are responsible for our non-conscious moral judgments and our conscious ones are entirely dissociable, why not think that sometimes negative affect—and the blame attribution that goes with it—can alter our intuitions about intentionality? After all, you have already acknowledged that it can have an effect on our judgments about agency, free will, and moral responsibility? Then why not intentionality? Wouldn’t it be surprising if an affect-driven performance error model best explained the free will/moral responsibility data but failed to play a part in the best explanation of the intentionality/moral responsibility data?

My current view—I hope to have a new paper (w/ studies!) soon—is that the affect driven performance error is one that can be shown to be erroneous by the participants’ own lights! Here is a way of putting it to the test. You run a reflective equilibrium study where people are first given a battery of all of our intentionality scenarios. We would then engage in some Socratic reasoning with them to see whether their judgments withstand their own analysis. What do you think? Some grad students at FSU tried something similar a few years ago. I always thought that with an improved experimental design, something like this could be feasible. In fact, Eddy and I talk about precisely this possibility in our forthcoming response to Antti. But that’s another story for another day.

[tnadelhoffer]

Joshua,

I said: "[W]hy not think that sometimes negative affect—and the blame attribution that goes with it—can alter our intuitions about intentionality? After all, you have already acknowledged that it can have an effect on our judgments about agency, free will, and moral responsibility? Then why not intentionality?"

Clearly you think these non-conscious moral judgments have an effect. The debate between the two of us is what to make of it. I meant to say:

"Why not think that negative affect—and the blame attribution that goes with it—sometimes distorts our intuitions about intentionality? After all, you have already acknowledged that it may have a distorting or biasing effect on our judgments about agency, free will, and moral responsibility? Then why not intentionality?"

[Joshua Knobe]

Thomas,

As you helpfully point out, there is lots of evidence indicating that people's affective responses can sometimes interfere with the normal workings of their underlying competencies. The key question now is just whether that sort of process can explain the particular effects we find for intentional action intuitions.

I am inclined to think that it can't. In particular, I think there is good reason to suspect that the process at work in intentional action intuitions is different from the process at work in cases where it seems that affective responses are distorting people's judgments.

Take the case of intuitions about moral responsibility and determinism. There, we find that people have strongly compatibilist intuitions when one does everything possible to tell a story that brings out people's emotional responses (e.g., telling a lurid tale about someone killing his own children), that the effect is diminished when one just gives a one-sentence description of such an event (e.g., 'George stalked and raped a stranger'), and that the effect goes away entirely when one discusses something relatively innocuous (e.g., cheating on one's taxes).

But the effect for intentional action intuitions looks very different. It continues to arise even in cases that are not at all emotionally loaded (e.g., decreasing sales in New Jersey) and can be found even in people how have suffered damage to the ventromedial prefrontal cortex.

So my tentative hypothesis is that the two effects are very different and that only one of them involves a performance error driven by affect.

[Antti Kauppinen]

Very interesting paper, as usual! The methodological issues it raises seem important indeed. But here's a smaller point first. Here is the conclusion of the terrorism example:

"Faced with this vignette, most subjects say that the terrorist did not save the Americans ‘intentionally’ but that he did save the Americans ‘in order to rescue his son’ (Knobe & Kelly 2006). In other words, it simply isn’t true that people only accept reason explanations for behaviors that they regard as intentional."

Now, it is true that the folk use the term 'intentional' and its derivatives, but it is also (and, as far as I know, originally) a technical term. This is important, since it means you can't simply infer from the folk not labeling something 'intentional' that they do not regard it as intentional, in the sense of 'regarding as intentional' that philosophers use. Indeed, it seems to me very likely that the folk do in fact regard the terrorist's sparing the Americans as intentional. Why? Because a) I think it is very likely that people would agree that the terrorist grudgingly or reluctantly spared the lives of the Americans, and b) if someone does something grudgingly or reluctantly, he does it intentionally, in the sense in which action theorists use the term. (Indeed, the folk might agree, if the question was put to them like that.) If that is the case, we don't have a counterexample to the claim that people accept reason explanations only for actions they regard as intentional.

If there indeed are adverbs that modify only intentional actions (or descriptions under which actions are intentional), this indirect strategy could be extended to chairman cases and others, enabling us to get rid of the problematic quasi-theoretical term 'intentional/ly'.

[Joshua Knobe]

Antti,

Good point. Do you think it would be ok if we used the phrase 'on purpose'? Or do you think there might be problems there too?

[Antti Kauppinen]

Joshua,

Let me elaborate a bit first on what I think philosophical talk about intentionality in action is doing. First, it no doubt marks a distinction between "Jack fell off the truck" and "Jack jumped off the truck", things that happen to you and things you do. Second, it no doubt marks a distinction between "Jack flipped on the light" and "Jack alerted the burglar". Seemingly, a simple way to characterize this distinction is as that between doing something on purpose and doing something by accident.

Questions arise when we get to side effects. Suppose that flipping on the light also turns on the radio, which Jack knows but doesn't aim to bring about. Does he turn on the radio intentionally? I think you get to decide, as long as you bear in mind what you're doing. If bringing something about 'intentionally' or 'on purpose' means in your theory that it is the aim of the action, Jack doesn't turn on the radio 'intentionally'. (This requires an independent grasp of the aim of the action; perhaps that is the event such that if you believed it wouldn't come about as a result of phi-ing, you would not phi. I do know at least one better view, but I'm not at the liberty to discuss it here.) If all it means is that the agents bring the side effect about non-accidentally, Jack does turn the radio on 'intentionally'. Ordinary language is probably ambiguous here. It doesn't seem implausible to me that moral considerations might make a difference to how the ambiguity is resolved in particular cases, but I'm not convinced that that is philosophically very interesting.

But to try to answer your question, I think Anscombe's thesis is about the second disambiguation of acting intentionally. It is a substantial claim, since acting for reasons is mentioned nowhere in explaining the theoretical role of acting intentionally. So, to have an argument against it, you need to make sure you're getting at just that sense of 'intentionally'. I suggested earlier that adverbial modifiers like 'reluctantly' seem to go together with it. (If Jack doesn't want to turn on the radio in the example, he may reluctantly turn it on by flipping the light switch. Perhaps he thinks it'll wake up the neighbors, but just can't make his way in the dark.) How about 'on purpose'? Well, at least that's not theoretical. But doesn't it suffer from the same ambiguity? It could mean either that the thing is aimed at or that it's not accidentally brought about. So somehow you'd need to ensure that the folk read it in the desired way. Instead, if you wanted to be really straightforward, you could ask: "Did the terrorist accidentally save the Americans?". The problem with this is that it vitiates the need to run a survey (and not just because everyone would be likely to disagree, but because those who didn't would thereby reveal a lack of competence). But, without checking up on Anscombe (so I might well be wrong), the only way to defeat Anscombe's thesis would be to show that the folk are willing to regard side effects that are brought about accidentally as done for a reason.

I realize this still doesn't go to the deeper issues in your paper, but I take it these questions are important enough in action theory to merit some discussion.

[Joshua Knobe]

Antti,

Interesting thoughts, but there's one thing I don't quite understand yet. If you think that the asymmetry for 'intentionally' is due to an ambiguity, how do you explain the fact that we get a parallel asymmetry for reason explanations? Do you think that reason explanations are ambiguous too?

[Antti Kauppinen]

Josh,

I must confess I wasn't thinking of the reasons explanations side at all (since it was sufficient to defend Anscombe to show that people could still regard the action as intentional under that description, and there was no asymmetry in that case), but now that you mention it, I think the whole business is very fishy. I would have gone for flat-out performance error (there is *no* reason explanation for an action under a known side-effect description), but a native speaker (Ben B.) thought "in order to" indeed is ambiguous, too, so perhaps it is.

Here's the story. If you ask me, "A phi-d in order to y" says that A phi-d as a *means* to *end* y. The end was her reason for taking (what she believed to be) the means. But... this just isn't the case in any of the vignettes. Let's start with my example. Jack knows that by flipping the switch he will not only turn on the light, which he needs to do to find his way through the room, but also turn on the radio. Does he turn on the radio *in order to* find his way through the room? Sounds pretty absurd to me, but if it is linguistically acceptable, we have an ambiguity, for here 'in order to' definitely doesn't refer to the same relation as it does in "Jack turned on the light in order to find his way through the room". The same goes for reason talk. It doesn't seem right to me to say that Jack's reason for turning on the radio was finding his way through the room - he had no reason to turn on the radio! But if he did, he certainly didn't have it in the same sense as he had a reason to turn on the light.

So what about the chairman? Why do people disambiguate differently in the two conditions? (In one sense he has no reason to either help or harm the environment - he says explicitly that he doesn't give a damn!) I'm really not sure, but here's a hypothesis that may be far-fetched. It seems very natural to say that given the story, the chairman was *willing* to harm the environment in order to increase profits - that is, had it been a necessary means for making money, he would have done it. So maybe this is what people are saying when they say what they do. Why not say the same in the help condition? Well, the simple explanation is that it's just odd to think, given our prototypical understanding of how the economy works, that helping the environment would increase profits, so there's no point in emphasizing that the chairman was willing to help the environment for that reason.