Allowing Experimental Philosophy


  “Oh, Agent Starling… do you really think you can dissect me with this blunt little tool?

Over at the Experimental Philosophy Blog Joshua Knobe reports on some recent work.

People ordinarily distinguish between doing and allowing. They distinguish between ‘breaking’ and ‘allowing to break,’ between ‘raising’ and ‘allowing to rise,’ between ‘killing’ and ‘allowing to die.’  A question now arises as to how people make this distinction. How do people know, e.g., whether a given act counts as actually breaking something or merely allowing it to break?

Fiery Cushman, Walter Sinnott-Armstrong and I have a new paper on this question.  As you may by now have guessed, our thesis is that people draw the distinction in part by looking to the moral properties of the act in question.

… In our first study, all subjects were given a story about a person who ends up in the hospital.  The person is being kept alive by life-support systems, but then the doctor turns off these systems specifically for the purpose of making sure the person dies.  Subjects are then asked whether it would be more accurate to say that the doctor ‘ended’ the person’s life or that he ‘allowed it to end.’

Now comes the tricky part.  Subjects were randomly assigned either to the ‘morally bad’ condition or to the ‘morally ambiguous’ condition. Subjects in the morally bad condition were told that the doctor removed the life-support system because he despised the patient and did not want to use valuable resources on him; subjects on the morally ambiguous condition were told that the doctor turned off the life-support systems because he honestly believed that the patient would be better off not having to go on suffering. …

As expected, subjects in the morally bad condition tended to say that the doctor ‘ended’ the patient’s life, while those in the morally ambiguous showed a more complex pattern. …

In our actual paper, we mostly just present these results without offering much of a explanation, but we are very curious about how exactly one might explain the effect found here.  It seems like it might be helpful to think in a more general way about what the distinction between doing and allowing is all about and then to figure out how that distinction might relate to moral considerations.  Any suggestions?

Well, golly, I have thought in a general way about the distinction and do have a few suggestions…

There were only two scenarios presented to subjects in the experiment.  In the first scenario, subjects are told:

Dr. Bennett is an experienced emergency-room physician in a busy downtown hospital. Late one winter night, a homeless man is brought into the emergency room. He is suffering from acute hypothermia, he is unconscious, and his identity is unknown.

The homeless man’s basic organ systems have shut down and a nurse has hooked the man up to a respirator. Without the respirator, his lungs would stop functioning immediately and he would die. With the respirator and some attention from Dr. Bennett the man would remain alive for a week or two, but he would never regain consciousness and could not live longer than two weeks.

Each subject was then given one of two versions of what happens next.

Morally bad case: In situations like this, it is up to Dr. Bennett to decide whether treatment should be continued.  Dr. Bennett thinks to himself, “This bum deserves to die. He shouldn’t sit here soaking up my valuable time and resources. The best thing to do would be to disconnect him from the machine”.  For just that reason, Dr. Bennett disconnects the homeless man from the respirator. The man’s lungs stop functioning and he quickly dies.

Morally ambiguous case: In situations like this, it is up to Dr. Bennett to decide whether the treatment should be continued. Dr. Bennett thinks to himself, “This poor man deserves to die with dignity. He shouldn’t spend his last days hooked up to such a horrible machine. The best thing to do would be to disconnect him from the machine. For just that reason, Dr. Bennett disconnects the homeless man from the respirator. The man’s lungs stop functioning, and he quickly dies.

Subjects are then asked to respond on a seven point scale between 1) Ended and 7) Allowed:

          Is it more appropriate to say that Dr. Bennett ended the homeless man’s life,
or that Dr. Bennett allowed the homeless man’s life to end?
○               ○               ○              ○              ○               ○              ○
Ended                                                                                    Allowed to end

Where to begin?

I) Let us start with the most obvious problems. Note that the subjects are not asked if the doctor ended the patient’s life or only allowed it to end. They are asked what it is “appropriate” to say.

One wishes that a philosopher among the experimenters had made the Gricean point that it is sometimes appropriate to say things one does not regard as literally true. Thus consider how you would answer the following:

          Is it more appropriate to say that Hitler threw Jews into gas chambers,
or that Hitler allowed Jews to be thrown into gas chambers?
○               ○               ○              ○              ○               ○              ○
Threw                                                                      Allowed to be thrown

The absence of some controls on what the subjects thought were the literal facts about killing vs. allowing as opposed to what they thought it “appropriate” to say, seems to me to render the whole enterprise dubious from the start.

II) Given that the kill/allow to die distinction is usually thought to apply exclusively to any given act it is at least odd to be asking the subjects to rank appropriateness of killing vs. allowing on a 7 point scale. Compare asking subjects on a scale of 1 to 7 whether it was appropriate to say whether the homeless man ends up 1) dead; 7) not dead.

What does a middling response mean? Does it signal the subject’s uncertainty about whether the case was a killing or an allowing, or does it signal that the subject operates a theory in which a single act can be in some measure be both a killing and an allowing. The latter possibility needs to be controlled against otherwise grouping the 7’s 6’s and 5’s together in the analysis would be a gross distortion of the subjects’ theory: a mistake on a par with treating those who think you can be a little bit dead as similar in view to those who think you are either dead or you are not, provided they agreed that the dead were “mostly dead”.

Did the subjects believe, prior to the test, that any given act was either exclusively a killing or an allowing?  If so, in being offered this sliding scale did they think they were being invited to apply a different standard than their normal one, one in which the difference was a matter of degree? This would distort the result. 

Imagine showing subjects a parade of gravid women and asking them to say of each lady whether it was more or less appropriate to describe her as 1) pregnant or 7) not pregnant.  I predict you would get an interesting variety of results but that they might not be relevant to the subjects’ extra-experimental views on the pregnant/non-pregnant distinction.

III) Then there is the deeper complication that even if one supposes that no action can be both an ending and an allowing to end, it seems that one agent can nevertheless do both with respect to a single death.  Thus we folks are happy to say things like “Jones killed Smith by pushing him in the water and allowing him to drown”. If pressed to say, in the context of an experiment , whether it is more appropriate to call it a drowning or an allowing to drown folks might give a (3-4) response not because they we were uncertain of their judgment, and not they because they thought any single action was a middling case, but because they were unsure which actions to single out.

Then again, if forced to make a summary judgment about a sequence of actions, particularly in the context of moral evaluation, subjects might pick the actions they found morally salient. Thus suppose the scenarios were :

Jones accidently and unknowingly brushes Smith into the lake but, then, seeing Smith’s plight, Jones decides not to throw him a life vest but to instead stand by and watch Smith’s long futile struggle.


In a fit of rage, Jones pushes Smith into the sea then, paralyzed in shock at his own actions, Jones loses the vital moment when he might have tossed Smith a life vest before the waves snatched him from the surface.

In the former case, I would predict subjects would be be more inclined to say Jones’ crime lay in allowing Smith to die whereas in the latter, I would predict more subjects would be inclined to say Jones killed Smith. In neither case would the subject’s moral views of any particular act  be determining whether they thought a particular act was a killing or an allowing but rather which acts were more morally significant.

The appropriate way to control for this source of noise would be to use very simple examples in which subjects were directed to pass judgments on a very specific actions.  Alas, not so in this experiment.  Instead subjects are  asked to make summary judgments over complex ensembles of actions.

Think of how many things the doctor does. He works as an emergency room physician, an emergency room which apparently allows hapless bums to be admitted with a (false) promise of care,  he (inadvertently?) allows/authorizes a nurse to put the bum on a respirator, he makes a (possibly faulty -see below) diagnosis, he disconnects the respirator. The good doctor– we are invited to believe–  will give “attention” to the disconnected bum, the bad doctor will not.

Which (set) of these actions are the subjects being asked to judge are doings or allowings? When the experimenters say the subjects judged the doctor’s actions as a killing or an allowing how are they sure which actions or sets of actions the subjects had in mind?

This is not a merely theoretical worry…

IV)  Anyone familiar with the philosophical literature will immediately recognize that the Good Doctor’s behavior is not an example either of doing or allowing but rather of what Jeff McMahon calls “withdrawing aid”. To see the difference consider.

Jones makes a leak in the dyke causing it to break. The town floods. People die.

Smith repairs a leak in the dyke. It does not break. There is no flood. No one dies.

The Dutch Boy sticks his finger in the dyke preventing it from leaking. But then he tires and pulls his finger out.  The dyke breaks.  The town floods. People die.

Jones kills; Smith prevents death; The little Dutch boy withdraws aid. More abstractly: clear cases of withdrawing aid are ones in which the agent’s only alternative to causing a harm is to act in a way which prevents that harm. Philosophers disagree about whether to also count in cases where the agent gives aid and then later, with other alternatives, withdraws it.

As McMahon observed, the metaphysics and morality of withdrawing aid are significantly different from that of doing vs. allowing. When the Dutch Boy pulled his finger out, was he killing the townsfolk or only allowing “nature to take its course”?  Here one might think the problem is one of summing the action of initiating aid with the act of withdrawing it.  But in the most famous withdrawing-aid example– Thomson’s “violinist” case–  the agent wakes up to find herself already providing aid involuntarily.  When Thomson’s agent disconnects the umbilicus from the violinist does she kill or allow to die? One may be tempted to say both or neither and it is a very interesting philosophical question how to accommodate these cases into any coherent account of the doing/allowing distinction.

Uninterested in the philosophy, the Experimental Philosopher may instead just insist we tell him whether such cases are 1) killing or 7) allowing.  But now it is not clear what the significance of the answer is. Even if it were agreed that subjects demonstrated moral bias in describing a withdrawal of aid as a “killing”  or “allowing”, this might only show that moral factors play a role when subjects are forced to subsume withdrawing aid cases under the doing/allowing distinction. That in turn would not show that moral factors play a role in simple doing /allowing contrasts or that subjects think it appropriate to subsume withdrawing aid cases under the doing allowing vocabulary.

V) On the other hand, the level of story telling here is too poor for it to be really clear what is going on. In the case as presented it is the nurse who puts the patient on the respirator, but it’s the doctor who disconnects. So, whether or not we regard this as a case of an agent withdrawing aid already given will depend upon whether we regard the nurse and the Doctor as a sort of institutionally collective agent or, alternatively, regard the nurse merely as an instrument of the doctor’s will.  If we do see the nurse and doctor as acting in concert, we are more likely to see the case as an agent withdrawing aid earlier provided. On the the other hand, it is hard to imagine the Bad Doctor sharing his vile deliberations with the rest of the medical team and the text does make it clear that he disconnects the machine himself. Thus subjects may have treated the Good Doctor’s case as one of withdrawing aid while seeing the Bad Doctor as more clearly a rogue  killer.

VI) Obviously, if we are practicing the “method of difference” we want the good and the bad doctor scenarios to be identical except for the crucial variant: the doctor’s motives. But when we look closely things come apart all over the place. Consider for example the question of how long it takes the bum  to die.  The script says that his lungs “will stop functioning immediately” when the respirator is disconnected and, after disconnection, that “he quickly dies”. But how quickly is that? According to the Good Doctor he is disconnecting so that that the bum won’t spend “his last days hooked up to a horrible machine”.  Should we believe the Good Doctor? If so, then the “quick death” will take at least a couple of days during which, presumably, the good doctor will providing “attention”. Or maybe the point of the story is that the Good Doctor is not a very good doctor since the bum dies so quickly?

VII) The timing issue is vital because it connects with tricky metaphysical questions about the essences of events in general and deaths in particular. We are apt to think a killer gives his victim a different death than he would otherwise have had. But we don’t seem to treat the exact time of a death as metaphysically essential to it: a death can be postponed.  Someone who postpones a death causes it to happen when it does, but we do not want to say he causes the death itself. ( Again, why this is so is a richly interesting philosophical question. ).

If we think of a doctor as postponing the occurrence of one and the same death — even a little– we will be disinclined to say either that he ended the patient’s life (after all, that end would have come anyway) or that he allowed it to end. 

Thus imagine a doctor applying the most heroic efforts possible to postpone an inevitable death. When all is done would you find it more appropriate to say that this Heroic Doctor “1) ended the life” or “7) allowed the life to end”.  Of course, neither of these sound right, but, if an experimental philosopher insisted you answer you’d probably skew to “allowed” since it seems less a travesty of the truth than “ended”.  Which is to say that in medical examples we would expect subjects to show a bias in favor of “allowed to end” answers even when the subjects thought the doctor had done everything to prevent life from ending.

As we have just noticed, the experimenters’ story invites us to believe that the Good Doctor does postpone the bum’s death, just not as much as he might have had he left the respirator on. Meanwhile, in the story as told, the Bad Doctor does nothing to postpone the death. The nurse puts the bum on the respirator. When the Bad doctor arrives on the scene, he disconnects so as not to waste “valuable time”.

This too, all by itself, might explain the experimental results and is unconnected to subjects’ views on doing/allowing.

VII) In the experimenters’ second question,  there is only one scenario: A pregnant women is told that unless she takes more vitamins her fetus will die. She doesn’t. It does.

Subjects are asked:

      Is it more appropriate to say that Sarah made the fetus die, or that Sarah allowed the fetus to die?
○               ○               ○              ○              ○               ○              ○
Made                                                                                             Allowed

Lo and behold! The experiments discovered that subjects were more likely to lean towards “allowed” if they ascribed themselves a “pro-choice” position on abortion. This is offered as evidence of the experimenters’ thesis that one’s moral views infect influence one’s judgments on doing/allowing.

It never seems to have occurred to the experiments that pro-lifers and pro-choicers characteristically differ not just about morality but whether or not the fetus is “a life” and this in turn might play a a role in their verdicts on whether a life has been ended or allowed to end.  Quick, answer the following question!

      When a doctor removes your appendix, is it more appropriate to say the doctor makes the appendix die, or that the doctor allows the appendix to die.
○               ○               ○              ○              ○               ○              ○
Makes                                                                                             Allows

Now ask yourself if your response to this lunatic query shows anything at all about the way you distinguish doing from allowing.


When philosophers criticize experimental philosophy they often argue that the responses of naive subjects are not to be taken seriously.  I want it noted that I have not taken that line here. Indeed, if anything, I expect to be accused of crediting the subjects with too much acumen.  I would resist that charge. I think plain folks have an exquisitely nuanced capacity to drawn subtle metaphysical and moral distinctions.

But if you are going to understand what folks are saying you cannot listen with a tin ear.

Leave a Reply

Your email address will not be published. Required fields are marked *