Are We Dumb(founded) or Dubious?

My student, Bradley Thomas, and I have been wondering if there is a problematic confounding factor in the infamous trolley cases. If you are like us (and many of our students), one reason you answer that it’s wrong to push the fat man off the footbridge is not so much that you think it is morally inappropriate but that you think it is stupid. A fat man is not going to stop a train! So, while it’s highly believable that pushing the fat man will kill him, it’s unbelievable that doing so will actually save five people. In contrast, it is believable enough that you could throw a switch to make a train avoid five people so that doing so will in fact save them, though it will also kill one worker. 

Of course, you are told to believe what the scenario says, but that’s hard to do when what it says is so … silly. Some people may overtly reject the purported “facts” of the case. Others may try to accept them but be unable to “internalize” them in such a way that their moral intuitions properly respond to those facts. As Anthony Appiah points out in his recent book, these dilemmas ask us to imagine what should be done in emergency situations and, if we are able to imagine that, we may employ heuristics to make quick moral judgments, heuristics like: “Don’t risk killing an innocent person unless you feel quite confident it will save more innocent persons.”

So, we are trying to test (a) whether this “epistemic” confound may be exaggerating the large differences in responses to Switch and Footbridge cases, differences usually explained in terms of the effects on people’s intuitions of killing by personal contact and/or of killing a person as a means to an end rather than as a side-effect of that end (we are not trying to show that these effects are not significant), and (b) whether this confound has been discounted as an explanation for people’s divergent responses to the two cases, such that people may be less dumbfounded than some have claimed. Instead, they may be dubious.

For instance, if we’re not mistaken, Fiery Cushman and Liane Young (in their wonderful 2007 M&L paper) treated subjects who explained their divergent moral judgments in terms of rejecting stipulations of one of the scenarios as “morally dumbfounded”—that is, they coded such explanations as “Alternative explanation: added assumption,” rather than as “Sufficient” (i.e., explanations that recognize the “personal” nature of Footbridge vs. the “impersonal” nature of Switch). If we are correct, subjects’ explaining that they reject the stipulations of a case should similarly count as an adequate explanation of their judgments, at least if they seem to recognize that they are answering differently in part because the scenarios differ in their believability.

We have some preliminary results showing that the degree to which participants believe the purported outcomes of moral dilemmas is highly correlated with their moral judgments (e.g., the more they believe that smothering their child in the circumstances of the scenario will save their other three children, then more they agree that smothering their child is the right thing to do). The tests below are meant to further test this hypothesis. We think the results are potentially important both for what they reveal about people’s moral intuitions and the relation of such intuitions to their “epistemic” intuitions and also for what they may say about the methodology of experimental philosophy—i.e., how (and when) can we trust judgments about unbelievable philosophical thought experiments?

We hope you will provide feedback about our hypothesis and also about our scenarios, which we have found excruciatingly difficult to develop. We're really hoping you save us from headaches down the road by finding any flaws that we've overlooked!  For all four of our cases, in addition to asking the standard “How morally appropriate is it for [agent] to do X [the act that leads to one being killed]?”, we also ask four questions about believability that take this basic form:

How likely do you think it is that, if [agent] does NOT do X, the five workers will be killed and the one worker will survive?

How likely do you think it is that, if [agent] DOES do X, the one worker will be killed?

How likely do you think it is that, if [agent] DOES do X, the five workers will survive?

How likely do you think it is that the ONLY way for [agent] to save the five workers is to do X?

Here is our version of the impersonal Switch case, which we call "believable" to indicate that it is rather believable that flipping the switch will actually save the five (it has been modified from the traditional Switch cases in order to make it more parallel with the new cases we constructed to test our hypotheses):

Frank is the only passenger in the front car of a subway train on some elevated tracks; the conductor just shouted that the brakes have failed, and then passed out over the controls. Frank knows about subway trains and knows that the automatic braking system at the next station will stop the train, so Frank, the conductor, and the other passengers are not in danger. But he sees that the train is speeding toward five people working ahead on the tracks; they are on a high overpass so they cannot escape in time. Frank also sees that there is a side track leading off to the left and, if he can flip a switch on the train's control panel, it will turn the train onto the side track where there is one person working on the tracks. The only way for Frank to prevent the deaths of the five workers on the main track is to flip the switch in order to turn the train onto the side track. If Frank flips the switch, the one worker on the side track will be run over and killed. If Frank does nothing, the five workers on the main track will run over and killed.

To examine our question of whether believability influences people's judgments about this case, we have constructed an unbelievable impersonal dilemma called Statue to match the believable impersonal Switch case above. Like Switch, Statue is impersonal (no contact), killing the one is a side-effect of the means used to save the five, and performing the proposed action is likely to kill the one (just as flipping the switch will kill the one worker on the side track). However, in contrast to Switch, in Statue (like the Footbridge case) we have attempted to make it unbelievable that performing the proposed impersonal action will actually save the five workers. If believability, or epistemic intuitions that contradict the stipulations of the thought experiment, are part of the driving force behind the difference in people's judgments about the Switch vs. Footbridge cases, then we would expect that judgments about this Statue scenario to move in the direction of judgments about Footbridge (and that judgments of appropriateness will correlate with judgments of believability). Below is our version of Statue, or if you are curious to see what these will look like on our online surveys, you can see the Statue case by clicking here

John is on a footbridge over some elevated subway tracks. John can see that a subway train approaching the footbridge is out of control, with its conductor alive but passed out over the controls. John knows about subway trains and knows that the automatic braking system at the next station will stop the train so the conductor and passengers are not in danger. But he sees that the train is speeding toward five people working ahead on the tracks; the tracks are so high that they cannot escape in time. John also sees that the only way to stop the train is to drop a weight onto it heavy enough to break through the cockpit window and depress the emergency brake switch on the control panel beneath the passed-out conductor. The only available, sufficiently heavy weight on the footbridge is a large statue next to John. The only way for John to prevent the deaths of the five workers is to push the statue off the footbridge and through the cockpit window in order to depress the brake switch on the control panel beneath the conductor. If John pushes the statue off the footbridge, the conductor will be crushed and killed. If John does nothing, the five workers on the track will be run over and killed.

Similarly, we have tried to construct a believable personal dilemma called Conductor to compare with the unbelievable personal Footbridge case (this was a hard one to create!). Like Footbridge, Conductor is personal (involves contact), killing the one is a *means* to saving the five, and performing the proposed action is likely to kill the one (as pushing the fat man will kill him). However, in contrast to Footbridge, in Conductor we have tried to make it believable that performing the proposed personal action will actually save the five workers. If believability is part of the driving force behind the difference in people's judgments about the Switch vs. Footbridge cases, then we would expect judgments about this Conductor scenario to move in the direction of judgments about Switch (and that judgments of appropriateness will correlate with judgments of believability). Below is our Conductor scenario, followed by our version of Footbridge (again modified from the original to make it more parallel with our other cases):

Your comments are very much appreciated!

Thanks, Eddy and Bradley

 

Conductor

Steve is the only passenger in the front car of a subway train on some elevated tracks; the conductor is alive but has passed out over the controls. Steve knows about subway trains and knows that the automatic braking system at the next station will stop the train, so Steve, the conductor, and the other passengers are not in danger. But he sees that the train is speeding toward five people working ahead on the tracks; the tracks are so high that they cannot escape in time. Steve also sees that if he can pull down the emergency brake lever located under the control panel, the train will stop before it hits the five workers. The train’s cockpit is so narrow that the only way to get to the brake lever in time is to push the conductor off of the controls and out of the doorway of the speeding train. The only way for Steve to prevent the deaths of the five workers on the main track is to push the conductor out of the train in order to pull down the brake lever. If Steve pushes the conductor out of the train, the conductor will be killed by the impact. If Steve does nothing, the five workers will be run over and killed.

Footbridge

Bill is on a footbridge over some elevated subway tracks. Bill can see that a subway train approaching the footbridge is out of control, with its conductor alive but passed out in the cockpit. Bill knows about subway trains and knows that the automatic braking system at the next station will stop the train so the conductor and passengers are not in danger. But he sees that the train is speeding toward five people working ahead on the tracks; the tracks are so high that they cannot escape in time. Bill also sees that the only way to stop the train is to drop a weight onto it heavy enough to break through the cockpit window and depress the emergency brake switch on the control panel. The only available, sufficiently heavy weight on the footbridge is a very large subway worker who has fainted next to Bill. The only way for Bill to prevent the deaths of the five workers on the track is to push the large worker off the footbridge and through the cockpit window in order to depress the brake switch. If Bill pushes the large worker off the footbridge, the worker will be killed by the impact. If Bill does nothing, the five workers on the track will be run over and killed.

Comments

[Regina Rini]

I think you're raising a really good question, and I would not be at all surprised if some sort of 'dubitability effect' explains some of the variance seen in the traditional Switch and Footbridge cases.

But I suspect that - as you note - it will be quite difficult to show this. Here's one concern. How do you keep your introduction of the epistemic questions from altering your subjects' naive creedence?

I suspect that, once you've explicitly asked a believability question, subjects will immediately begin to reduce their creedence in the efficacy of the proposed actions across all scenarios, relative to how they might have felt absent those questions. It's a matter of pragmatics - your asking if the subjects believe the story implicates that the story perhaps ought not be believed.

If this concern is on track, then obviously that will be a problem for your epistemic results: subjects' reports of their beliefs about the actions in this study may not be representative of studies in which subjects are not asked these epistemic questions. And their reports of moral appropriatness may be then similarly unrepresentative, following on the implications of Appiah's point about emergency situations.

Of course, your hypothesis pertains to a correlation between strength of belief and moral evaluation, and you might very well find such a correlation even if there is some across-the-board impact of the belief questions. But I can imagine an opponent suggesting that, perhaps, such a correlation when observed in conditions of partial belief has no bearing on conditions of certainty. Basically, the worry isn't just that introduction of the epistemic question systematically augments creedence levels or moral approval. Rather, by the null hypothesis, subjects fully accept the typical experimental paradigm's assertion that these actions really are efficacious. Once you've implicitly cast doubt on this assertion, then subjects may then reset their creedence levels in a completely independent manner.

Here's one idea for ameliorating the problem. Treat the pragmatic consequences of asking your 4 belief-questions as another possible variable, and control for it. That is, run a control group of subjects who are asked only the single moral-appropriateness question. I don't think this solves the whole problem, but it at least gives you a starting point.

Another option, perhaps, is to have subjects answer the belief questions for each scenario after they've answered the appropriateness question for all of the scenarios. That is, don't (possibly) pragmatically introduce doubt until you've already recorded the moral evaluative responses. (But then it's possible that subjects will confabulate about their own epistemic states in order to generate a seeming justification for the pattern of moral evaluations they've already given...)

I think the ideal solution would require finding some implicit measure for dubitability. I don't really know how to do that, but I'll bet someone in social psychology has thought of something clever.

Anyway, the research question itself is a really good one. I hope you can work out any kinks in operationalizing it.

[Bradley]

Thanks for the comments Regina. You raise very good points that we're worried about. Let us know what you think about the ways were trying to deal with them. One way is to do primarily between-subjects comparisons using only the first dilemma that subjects responded to. So we will primarily be looking at appropriateness responses that are just like the ones previously collected (when no epistemic questions were asked)...since people will have responded to them prior to being introduced to any epistemic questions. That way the moral intuitions we're examining wont have been affected by the introduction of epistemic questions (i.e. we won't be altering the subjects "naive credence").

We're also doing a lot of counterbalancing (dilemma order and question order), so we'll be able to compare subjects appropriateness responses to, e.g., Switch (and similarly for the other dilemmas) when it was the first scenario they read and they hadn't seen any epistemic questions yet to their responses when Switch was read after other dilemmas which introduced subjects to the appropriateness and epistemic questions.

We'll also be able to compare appropriateness responses to Switch when it was the first scenario they read and they saw the appropriateness question prior to the epistemic questions with responses when Switch was the first scenario but they saw the epistemic questions first.

[Eddy Nahmias]

Thanks Regina. I just wanted to highlight Bradley's last paragraph. We are counterbalancing the moral judgment questions and the epistemic questions precisely to see if there is an order effect. If there is one, the question will be how to interpret it. It may be that people adjust their "belief" responses to match the moral judgment they just made, which would be an interesting finding in and of itself.

[Taylor Murphy]

I am worried that conductor might be misleading.

The conductor is not (unlike the traditional story) being used as an object or a tool in order to save the lives, but has to be killed in the process of saving the lives. That is, if the conductor wasn't there, he wouldn't have died (and the people would have been saved). This I think is different from when the fat person is the object or tool of life saving, where if there were no fat person, nobody would be saved. In the latter case, the killing is explicitly part of the plan to save the 5 people, whereas in the former it is just so happens that the conductor is there. It is, as it were, collateral damage, and is closer to the person-on-the-other-tracks scenario than it is the person-as-a-means scenario.

I remember reading about a study where people were given three scenarios, involving using a large person to save the lives. Here is what they looked like.

http://img119.imageshack.us/img119/2620/dsadsadsadsadasdafw7.png

Instead of having two tracks, they had a track that diverged (and the 5 hikers were after the tracks went back together). If there was just a fat person on the tracks, people declared it immoral. If it was just an anvil on the tracks, it was fine. If it was a fat person in front of an anvil, it was also fine. Why is this? Possibly because in the person + anvil case the person is collateral damage, and if it were just the anvil things would have gone just the same.

Of course, in the "fat person only" scenario it is still perhaps unrealistic that the person would actually stop the train, so it may be that collateral damage isn't the explanation here - but it very well might be and I think the conductor scenario has to make the conductor part of the life saving plan, and not collateral damage.

Unfortunately this is from memory, and I can't find the actual study, but the following link discusses it in some detail (at least, at a glance it looks like it does). http://journals.cambridge.org/download.php?file=%2FUTI%2FUTI20_01%2FS0953820807002932a.pdf&code=

[Taylor Murphy]

The last link doesn't work. For some reason, the direct link doesn't work but the link from google does, so here it is. http://www.google.ca/url?sa=t&ct=res&cd=2&url=http%3A%2F%2Fjournals.cambridge.org%2Fproduction%2Faction%2FcjoGetFulltext%3Ffulltextid%3D1729532&ei=DGf9R_mAN6mopwTd0KjwCQ&usg=AFQjCNHRW7aWOzmZW0KGPpIyWGUcbgFPNA&sig2=y9pwCrpSTE_Xx9P4pe-O4Q

[jonathan weinberg]

I have a small concern about the wording of the "how likely" probes, in that I'm worried that they are ambiguous between two readings that philosophers might could paraphrase as

"In the possible world that you have imagined, how likely..."
and
"If a situation like the one described were to arise in the actual world, how likely..."

There is also something a bit weird about answering "how likely" questions in response to someone else's story. I'm halfway inclined to say, "hey, you wrote the thing -- _you_ tell _me_ how likely it is!"

Btw, has anyone looked at the difference of introducing an explicit instruction component of something like, "No matter how improbable you think this might be in the real world, for the sake of argument, please just take it as a definite fact that ..."? If not, it might be interesting to look at your four cases crossed with some sort of instructional difference like that.

[Dan]

Neat idea. I share Taylor's concern, though, about the causal structure of Conductor - the conductor seems more like an obstacle to be removed than a tool to be used.

Also, are you familiar with Josh Greene's recent work on the trolley problem? I saw him give a talk where he went through about 20 variations of it, trying to identify what features produce the difference between the standard Trolley and Footbridge scenarios. If you haven't seen this work, you may want to get in touch with him. At minimum, you could check to be sure your scenarios are matched according to the features that he has identified as relevant (which I believe involve physical force and the structure of intentions).

[Bradley]

Good thoughts. Fyi, here is the paper (I think) Taylor's talking about:

http://www.law.georgetown.edu/faculty/mikhail/documents/dissociation_mind_language.pdf

[Eddy Nahmias]

Thanks for the useful comments!
Jonathan, I share your concerns. I'm not sure how we could phrase the believability questions better--ideas are welcome. We hope that answers to them basically indicate whether participants *buy* what they are being told to buy. Even if JK Rowling tells us Dumbledore is gay, I think many readers would say they don't believe it. Or perhaps we are giving participants a way to indicate the sort of cognitive dissonance I think these scenarios can cause in us--we're told we have to believe something we just don't believe. I'd also like to test the difference in results when you include the explicit instruction you suggest ("Believe what it says even if seems improbable...") vs. when you don't. But I think the way we're doing it now gives better variables to compare. (By the way, in our free will studies, we always included an instruction to believe the unlikely and we used alternative universes in some cases to make it more believable.)

I'm afraid we just can't think of a good believable personal case that gets all the elements of the fat man case. I think that with our conductor case we've gotten in the personal contact (pushing) and pushing him is not really a side effect (as in Switch). Whether it counts as a means to the goal of pulling the brake lever is more debatable and it is true that he certainly is not a tool to achieve the end the way fat man is.

If anyone can think of a case that will get all the elements *and* is believable, we'll buy you a beer (or several!). If not, we think we can still get interesting results out of these cases.

Thanks again for the ideas--keep them coming.

[Taylor Murphy]

How about this?

Steve is the only passenger in the front car of a subway train on some elevated tracks; the conductor is alive but has passed out over the controls. Steve knows about subway trains and knows that the automatic braking system at the next station will stop the train, so Steve, the conductor, and the other passengers are not in danger. But he sees that the train is speeding toward five people working ahead on the tracks; the tracks are so high that they cannot escape in time. Steve also knows that if he can hit the door beside the conductor open, the emergency brakes will engage and the train will stop before it hits the five workers. Steve reaches over the conductor and realizes the door is jammed shut and needs to be hit open with something massive. Steve only has a few seconds to act, and the only available option is to use the conductor to hit the door open (by pulling the conductor towards him, and then thrusting the conductor into the door). The only way for Steve to prevent the deaths of the five workers on the main track is to use the conductor to hit open the cockpit door. If Steve uses the conductor, the conductor will be killed by the impact when falling out of the train. If Steve does nothing, the five workers will be run over and killed.

I think it captures the idea that the conductor is an instrument in saving the lives (rather than merely collateral damage), and it seems believable to me.

Hey guys,

I think the idea is very interesting. I've heard the problem of believability for the footbridge case brought up in a slightly different way before: not that a fat man won't stop a train (which, I agree, is dubious), but that it's unbelievable that pushing him off the footbridge will work because fat guys bounce!

Now, a few (really nit-picky) issues with your scenarios:

1. There's a typo in 'believable'. The last sentence needs another 'be'.

2. Also, I think the wording of the first scenario makes it feel as though there's a chance the single worker on the side track might escape. You specify that the tracks are elevated at the beginning of the vignette, but you bring it up again in the case of the five workers, but not in the case of the one. So, another pragmatic implication might be, since you thought it was necessary to mention in the five worker case, you would have thought the same in the single worker case. So, since you didn't mention it, it must not be the case that the side track is high. So, maybe the worker will escape. Anyway, that's a rational reconstruction (as Grice would say) of the way the first scenario struck me. You could get around it by just not mentioning the highness of the tracks again when you describe the situation of the five workers.

3. I know your interested in the dubious nature of the means of stopping the train in the original footbridge case, but your own footbridge case leaves me with just a general sense of incredulity. In all your scenarios, the conducter is unconscious, but in this scenario the worker has passed out too. It strikes me as odd that these people are so prone to unconsciousness, since I've seen people faint maybe only two or three times in my life. I don't know that it's likely to effect anything, but anyway it seems unneccessary. A conscious fat man would presumably work just as well in this scenario as an unconscious one, which is to say not very.

4. The problems with the causal structure of conductor have already been brought up. Another issue I had reading it was that it seems weird that Steve could only push the guy out of the train. Why not just push him back off of the controls? Another possible structure for the vignette that might solve both of these problems (though I think Taylor's is pretty good) would be one where the train is speeding down the track in the direction of five workers and Steve is standing outside of the train. Right beside the track (about six feet from where Steve is standing) there's a lever that would divert the train onto a side track a little further down. But, between Steve and the lever is a large man. Steve only has time to push the man with enough force to switch the lever, but if he does this, the man will also fall onto the track in front of the oncoming train and be killed.

[Mark Phelan]

Oh, that last comment was me, btw.

[Fiery Cushman]

Eddie & Bradley-

This is a terrific project! I'm excited to hear how these experiments come out. A few quick notes:

- Josh Greene has been doing a bunch of trolley work where he includes a plausibility measure (I think he calls it "unconscious realism?"). He finds significant effects. His method is just to enter the measure as a covariate in his models and show that, even controlling for plausibility, he gets consistent differences between cases. I believe this is the citation, from his webpage:

Greene, J.D., Lindsell, D., Clarke, A.C., Nystrom, L.E., and Cohen, J.D. (submitted) Pushing moral buttons: The interaction between personal force and intention in moral judgment.

- An undergraduate thesis student of mine (Lirui Li) also ran some correlations like this on a large set of dilemmas. She finds an effect of plausibility, too. (Good news for your hypothesis!). I'm going to encourage her to post a description of her results to this site.

- Regarding the status of people's "alternative explanations" in the 2007 Mind and Language paper. You're absolutely right! We tried to address this point more rigorously in our 2006 Psych Science paper (published earlier, but actually conducted later). The basic idea was to show that alternative explanations are more likely for 'principles' that dumbfound (e.g. means vs. side-effect) than for 'principles' that do not dumbfound (e.g. action vs. omission). Is this data any more convincing?

All in all, I strongly suspect that your basic intuition will be confirmed: implausibility exerts an effect on these cases. I have my doubts that it totally explains the means/side-effect distinction, but if it did, that would certainly be a very important finding!

[Eddy Nahmias]

Hi Fiery, thanks for the comments! Hopefully, I can get the paper from Joshua and please encourage your student to send hers our way too. I am not surprised to hear that others have been testing this issue and it will be interesting to see what comes of it all. We worry that this "epistemic effect" may infect much of research that uses these moral dilemmas in a way that makes the results murkier than has been suggested. Note that we aren't trying to show that the epistemic effect shows that the other effects (e.g., means/side-effect or personal contact/distance) are not still significant. But it may show those effects are less substantial and also that people are not as dumbfounded as is often suggested.

Take, for instance, Haidt's incest case. When I present this case to students, most just reject the stipulation that there will be no bad effects of the siblings' sexual escapade. Perhaps they are just confabulating an explanation for their disgust reaction (this would be an interesting thing to test). But perhaps their disgust reaction tracks the reasons they really have for thinking incestual sex is wrong, reasons that they do not (cannot?) accept are not still in play in the scenario despite the stipulated claims that those reasons do not apply (e.g., some students say they just don't buy that neither sibling will suffer any detrimental psychological effects for the rest of his or her life--at a minimum, they will have to keep a very big secret from their future spouses or lovers).

[Eddy Nahmias]

And thanks Mark and Taylor. We are working on a scenario using Taylor's suggestions, but not sure which we will run (maybe both). I think both of your suggestions highlight the problem we've been facing--it's just really hard to come up with a believable scenario that requires personally killing someone (esp as a means) in a way that will save five, within the confines of the trolley world. We may try cases outside of trolley world, but wanted to try these cases just since they are the ones everyone focuses on.

[Dan]

Here's another way to deal with the lack of symmetry between Conductor and Footbridge: you could keep Conductor, and come up with a new scenario that is analogous to Conductor in its causal structure & method of killing but unbelievable. For instance, the set-up could be similar to Statue, where the only way to stop the train is to push a statue onto it. But in this case, the cockpit is empty so the statue will not harm anyone. However, there is a passed-out man on the narrow bridge between you and the statue, so the only way to get to the statue is to push this man to his death. This scenario seems to allow for a pretty clean comparison with Conductor.

Also, now that I think about it, I don't think that your original Statue scenario counts as "impersonal" according to Greene's latest analysis, since you're killing the conductor using direct physical force (pushing the statue onto him), which (according to Greene's theory) probably should set off the alarm bells that were built into us in the evolutionary environment (where it was possible to kill people with physical force - like by pushing big rocks onto them - but not with switches).

[Mark Murrell]

Check out my Pleasure/Pain principle: http://markinchicago.typepad.com/my_weblog/2008/04/the-pleasure-pa.html

[Dane Thayer]

Does the second thought experiment allow too much room to construct details about the story which might muddle things up. For instance, was the conductor passed out at the controls because he was out carousing the night before at the bar. Perhaps you should guard against certain interpretations (or at least be aware of the possibilities)? It would be too strong to say that the conductor deserves to be crushed by the weight if he had demonstrated negligence; however, one can imagine these details having an influence on the decisions of the subjects. In any event, it is going to be hard to tease out what is really going on.

[Eddy Nahmias]

Thanks Dan and Dane. We changed "passed out" to "alive but unconscious" and hope that people do not interpret that in any way to suggest that the conductor is negligent in such a way that it's OK to kill him. Though dropping something on the conductor may count as personal on some understanding of personal, it does not seem to involve the personal *contact* that Greene initially used as the criterion. And the conductor is just in the way of the brake switch, so the agent is clearly not intending to kill him.

Taylor, in the end we used something like your scenario. Much thanks! We will be posting the link to the surveys soon (as a new post) in the hopes that philosophers and other informed people will take them. But we're collecting data from students first. So, it's too late to fix our scenarios. Please keep your wonderful advice to yourself so we don't feel bad...