Ought implies can. Or does it?

“Ought implies can” is one of those elementary notions that every philosophy student learns in introductory courses. Specifically, she will come across two formulations of the principle, both due to Kant. In his Critique of Pure Reason, he says: “The action to which the ‘ought’ applies must indeed be possible under natural conditions.” And in Religion Within the Boundaries of Mere Reason he explains: “For if the moral law commands that we ought to be better human beings now, it inescapably follows that we must be capable of being better human beings.” (Of course one could deny the antecedent “if” condition, but the idea is that if one accepts it then what follows is entailed.)

This seems a pretty straightforward logical principle, i.e., entirely independent of empirical verification. Or is it? A recent paper by Vladimir Chituc, Paul Henne, Walter Sinnott-Armstrong and Felipe De Brigard argues otherwise. Published in the journal Cognition with the title “Blame, not ability, impacts moral ‘ought’ judgments for impossible actions: Toward an empirical refutation of ‘ought’ implies ‘can'” is one of the latest salvos in the controversial field of experimental philosophy.

Now to be clear, Kan’t dictum simply says that in order for someone to be morally obliged to do X it has to be (“naturally”) possible for that person to actually do X. If it’s impossible, then it makes no sense to make the moral demand to begin with. The standard position in philosophy is that this is necessarily, i.e., logically true, and thus that empirical evidence is besides the point.

One of the co-authors of the paper, Sinnot-Armstrong, had previously argued that “‘ought’ does not necessarily, analytically, or conceptually imply ‘can.’ Rather, it only suggests ‘can’ in contexts where ‘ought’ judgments are used to advise rather than to blame agents—if we were giving advice to a friend, then our advice would be useless if our friend could not do what we advise. In other contexts, such as when we are laying blame (‘Where are you? You ought to be here by now!’), there is no implication from ‘ought’ to ‘can.'”

Well, maybe individual agents not particularly well versed in logic do not mean to imply anything, but perhaps they should?

At any rate, Chituc et al. set out to do an empirical test of Kant’s dictum, contrasting two hypotheses:

H1. Participants will deny that an agent ought to do something that the agent can’t do, regardless of whether the agent is to blame for the inability.

vs:

H2. Participants will judge that an agent ought to do something that the agent can’t do when the agent is to blame for the inability.

The authors carried out three experiments to go after the problem. In the first one, they used 79 subjects recruited via Amazon Mechanical Turk and asked them to read these two vignettes:

“Adams promises to meet his friend Brown for lunch at noon today. It takes Adams thirty minutes to drive from his house to the place where they plan to eat lunch together.

Low blame vignette: Adams leaves his house at eleven thirty. However, fifteen minutes after leaving, Adams car breaks down unexpectedly. Because his car is not working at that time, Adams cannot meet his friend Brown at noon, as he promised.

High blame vignette: Adams decides that he does not want to have lunch with Brown after all, so he stays at his house until eleven forty-five. Because of where he is at that time, Adams cannot meet his friend Brown at noon, as he promised.”

The subjects were then asked: “Do you agree or disagree with the following statement? At eleven forty-five, it is still true that Adams ought to meet Brown at noon.”

The results showed that people are more likely to say that Adams ought to keep the promise in the high blame situation, where he manifestly, physically, cannot (because of his earlier, culpable, decision to delay departure). The authors take these results to support H2, which is true, but — seems to me — irrelevant to Kant’s point. Kant would have simply responded that Adams was to blame for not meeting his friend because he had purposely delayed departure until a time at which it was physically impossible for him to do so. The fact that at time t+1 an agent is naturally incapable of doing X does not make him not blameable for it, if at time t he willfully put himself in such position.

Moving on to experiment 2: in this case, 195 subjects read a modified version of the low blame situation above, designed so that the wording excluded alternate pathways for Adams to keep his promise, which might have confounded the results of the first experiment. Here is the wording of the new vignette:

“Brown is a CEO of a large company in the economic boom in the middle of the 20th Century. At 2 o’clock, Brown has a meeting in the city to make a significant financial decision that will decide the future of his company. Since so much money is at stake, he asks his trusted personal advisor, Adams, to meet him on the 12 o’clock train. On the train, he plans to discuss his decision on the ride into the city, where Brown will go straight to his 2 o’clock meeting. Adams promises to meet Brown on the train at noon. It takes Adams thirty minutes to drive to the train station, park, purchase a ticket, and board the train. However, fifteen minutes after leaving at eleven-thirty, Adams car breaks down unexpectedly. Because his car is not working at the time, Adams cannot meet Brown at noon, as promised. Since cell phones have not been invented yet, Adams has no way to contact him.”

People were asked “to rate how much they agreed with statements saying that, (i) at 11:45 AM, Adams ought to keep his promise, (ii) Adams can keep his promise, and (iii) Adams is to blame for not keeping his promise.”

Results: there was a modest but statistically significant correlation (r=0.23) between ought and blame judgments, but no significant correlation (r=0.08) between ought and can judgments. Finally, can and blame were also significantly correlated (r=0.24).

The idea was that the expectation based on Kant’s dictum was of a correlation between can and ought, which however turned out to be the only pair of variables not to be correlated. Again, however, to me this does not invalidate Kant, but it rather speaks to how poorly people understand logic, which is sufficiently worrisome.

And we finally get to the third experiment, in which 319 people were exposed to vignettes representing more complex situations and where asked more sophisticated questions, in order again to untangle some of the variables that might have be confounded in the first two cases. Here is one of the vignettes in question:

“Moral Obligation, Inability, High Fault (Low Fault):

Brown is excited about a new movie that is playing at the cinema across town. He hasn’t had a chance to see it, but the latest showing is at 6 o’clock that evening. Brown’s friend, Adams, asks Brown to see the movie with him, and Brown promises to meet Adams there. It takes Brown fifteen minutes to drive to the cinema, park, purchase a ticket, and enter the movie. It would take 30 min if Brown decided to ride his bike. The cinema has a strict policy of not admitting anyone after the movie starts, and the movie always starts right on time.

[As Brown gets ready to leave at 5:45, he decides he really doesn’t want to see the movie after all. He passes the time for five minutes, so that he will be unable to make it to the cinema on time. Because Brown decides to wait, Brown can’t make it to the movie by 6. (High Blameworthiness)]

[At 5:30, Brown thinks about riding his bike, but decides it is too cold. Instead, he leaves at 5:45, but his car breaks down five minutes later. He can’t fix it himself in time to make it to the cinema, and it is too late to make it by bike. Because his car is not working at the time, Brown can’t meet his friend Adams at the movie by 6. (Low Blameworthiness)]”

With respect to the above, participants were asked to judge the following statements:

“Can: At 5:30 [Can’t: 5:50], Brown can make it to the theater by 6.
Blame: Brown is to blame for not making it to the theater by 6.
Ought: Brown ought to make it to the theater by 6”

The data set here was complex enough that the authors had to carry out a multivariate analysis of variance. You can check the details in the original paper, but they again found a statistically significant correlation between ought and blame (r=0.41) and no significant correlation between ought and can (r=0.18). This time, however, the correlation between blame and can was also not significant (r=-0.07), unlike in the second experiment.

The authors take the findings to replicate those of the first two experiments, and moreover to “demonstrate that this relationship between ought and blame does not hold for ‘ought’ judgments based on non-moral desires or judgments about instances when the agent can do what he ought to do.”

Okay, what’s the overall conclusion, then?

“We show that judgments of ought, contrary to broad philosophical assumption, do not imply judgments of can.”

I would argue that the results show no such thing at all. What they do show is that people, under certain circumstances, arrive at non-logical judgments. But that’s hardly news! Imagine a similar study carried out on the gambler’s fallacy. What would be a statistician’s response when shown empirical data that a good number of people behave exactly as his understanding of probability theory says they shouldn’t behave? I am going to go on a limb and guess that the statistician in question would shrug his shoulders and mutter something to the effect that “people are stupid, that’s why the gambling industry thrives.” Why, exactly, shouldn’t a philosopher whose dictum is based on elementary logic (Kant, in this case) not do exactly the same? (Notice that this isn’t a point about ethics, it’s about logic, so one doesn’t have to buy into Kant’s ethical pronouncements to agree.)

The authors continue: “Instead, judgments of ought are affected by judgments of blame.” Exactly, and illogically so, one ought (?) to add.

Now, if what the authors had set out to do was descriptive work in the psychology of moral judgment, this would actually be interesting. But they very explicitly didn’t: “Even more than exploring the relationship between concepts in moral psychology, these findings have important normative significance, as they pose a serious challenge for the many philosophers who hold that ‘ought’ implies ‘can.'” But they do not pose any such challenge! No more than a similar study on the gambler’s fallacy empirically demonstrating that people commit it often would have any significance whatsoever for “prescriptive” probability theory.

Again: “Because this principle [that ought implies can] is usually taken as an analytic or a conceptual entailment, it is supposed to follow necessarily from the concepts expressed by the words ‘ought’ and ‘can.’ Our results show that it does not.” No, your results show no such thing. At all. What next? A Study showing that people think some bachelors are married, and that therefore the definition of bachelor is “wrong”?

Chituc and collaborators go so far as to blame philosophers for this glaring discrepancy between logic and fact: “Why, then, are so many philosophers attracted to the principle that ‘ought’ implies ‘can’? One explanation is that rather than the participants in our experiment being distorted by a motivation to ascribe blame, philosophers may be distorting their judgments by a motivation to withhold blame.”

Holy crap! Never mind that Kant laid plenty of blame on people for all sorts of things (so long as they could do otherwise), and that philosophers generally agree that ethics is prescriptive, which implies that an agent is to blame if he does not do something that he ought and could do. Philosophers don’t withhold blame, but they do try to keep it within the confines of what an agent can actually do, not blaming people for failing to do impossible things.

The only good point, philosophically speaking, that Chituc et al. make comes at the very end of the paper: “much of the appeal of the principle that ‘ought’ implies ‘can’ is supposed to be that it reflects commonsense moral judgments. If philosophers do not stick to common moral concepts, then they risk becoming esoteric and irrelevant.”

Well, to begin with, I’d like to see empirical evidence that philosophers really do think that Kant’s dictum reflects commonsense. This is experimental philosophy, after all, and it’s not like Kant himself was known to be a particularly commonsensical kind of thinker.

Second, right, philosophers ought to be relevant to the public. So should statisticians. But the way to do that is not to abandon logic or probability theory, it is by educating the public: if you insist in engaging in the gambler’s fallacy you are behaving irrationally; and if you insist in saying that someone ought to do something which he could not possibly do, then you are reasoning illogically. Data like that provided by Chituc and collaborators show us important things about just how badly people misjudge situations, and so make for a very good argument for more education in moral theory. And elementary logic.