There hasn’t been nearly enough work put into creating good philosophical thought experiments, and as a result the ones that we have are either flawed or flat-out terrible. It’s easiest to explain the issues by looking at one of the terrible ones. So, in the footbridge or fat man question a trolley is out of control, and will kill its five passengers. You’re standing on a bridge over the tracks, and if you push someone standing next to you off the bridge, your action will kill that person, but stop the trolley, saving the lives of its passengers. Would you do so?

The 2020 PhilPapers Survey recently sent out 100 questions to all anglophone academic philosophers. The footbridge question was on the list, and only 22% of responders gave an answer other than yes or no. It doesn’t seem unfair to describe it as one of the best known philosophical thought experiments, or to claim that it is treated seriously by professional philosophers. It’s also witlessly awful.  

All of the specifics of the problem are in conflict with the definite consequences of your actions which the problem insists that you assume. Pushing someone off a bridge is exceptionally unlikely to stop a runaway trolley safely. Nor will it definitely kill the person you push. Nor will a runaway trolley definitely kill its passengers. If you are imagining an action which will stop a runaway trolley safely, save all its passengers, and kill someone else, you are not imagining pushing someone off a bridge. If, while standing on a bridge and observing with horror an unfolding tragedy below, another onlooker sidled up to you and suggested that a solution could be found by giving a hearty shove to a fellow gawper of larger than average girth, you would not react by seriously considering the merits of the proposal. In short, answers to this question can’t answer anything other than which parts of the question the person answering it has decided they should ignore.

The trolley problem isn’t as bad, but it still isn’t good enough. Firstly, nothing anyone could do to the controls of this railway would definitely make something happen. It’s a Roadrunner cartoon, Wild West type railway (or it wouldn’t have people tied to it). It’s just not reliable. In this problem, that’s crucial; you have to be made to put yourself in a situation where taking the decision to kill someone will absolutely, without any doubt, save five lives. If you have any doubt that the lives will be saved, you’re answering a different question. Secondly, while the trolley is real, and the people on the tracks are real, using points to switch the trolley isn’t.  The speeding vehicle, the screaming victims, the severed limbs, are arresting in a way that the work of the signalman can’t be.  There’s an imbalance in the physicality and immediacy of different parts of the problem. Thirdly, to the extent I can imagine switching points on an old-fashioned railway, I’m pulling a lever, and I’d try to pull the lever halfway so that the trolley derails. If I can’t do that, then whatever I’m doing I’m not switching points on an old-fashioned railway. Fourthly, the trolley is out of control, hurtling towards you, and you have little time to think. Why? The question isn’t designed to identify differences in knee-jerk and considered responses. It’s a necessary part of the question if the deadly contingency is a trolley, or you’d just untie the people from the tracks, but the clarity of the question is undermined. All these points could come across as nit-picking; I don’t think they are. Specifics have been provided which don’t just fail to illuminate the problem, they change it, in ways that make it either unclear or self-contradictory, and in ways that mean different people will understand the problem differently, and therefore answer different questions. The trolley problem isn’t awful, as the footbridge problem is. But it’s not great either. I think you can say the same about all of our current crop.

Thought experiments matter. They allow difficult and contentious problems to be understood, remembered, and referred to. Thought experiments about real-world morality are even more important; if you want to pinpoint, say, a disconnect between what people think the answer to a moral question is (for example, are five lives more valuable than one) and what they believe they should do in a situation arising from that question (should you kill one person to save five people) it’s vital that people are able to imagine themselves in that situation. That means it has to seem real, which means a good thought experiment. 

And insufficient effort seems to have gone into trying to design better philosophical thought experiments; ones that are sufficiently concrete and immediate, which accurately express the underlying moral dilemma that they’re trying to get at, and which will become successful memes.  Perfection is unattainable, but it must be worth trying to do better than we have done so far. We’ve left it to philosophers to attempt to create passages of writing which must, if they are to achieve the desired effect, be clear, vivid, and easily comprehensible. With the best will in the world, very little in the output of the profession over the millenia suggests that it’s a task they are likely to succeed in. It’s like asking Wittgenstein to teach mathematics to children. 

If you think you can do better, please comment with suggestions. It may be that there’s a community dedicated to this work already, somewhere on the internet, but if there is, I haven’t found it.  I’ll put a couple of efforts in the comments myself. I don’t think mine are any good. Making good ones is really difficult. If we succeeded, though, we might find that thought experiments could do things that aren’t even imaginable now. It might be like building a hotter oven so you can make bronze more quickly, and then discovering that it can be used to create a completely different shiny material. 

I’m not sure that the vital importance of creative artists to the future of the world in the plot of Watchmen (the comic) really works. But I love the idea of it. And I think designing thought experiments is an area where artistic genius could be essential. Succeeding seems likely to need creativity on the level of that required for writing a great poem. We should all be trying to do it. In the same way that two hundred years ago a fair percentage of educated English-speaking people were writing poetry which they rather hoped might turn out to be an equal to Elegy in a Country Churchyard, or On Westminster Bridge, bored students and office workers now should be thinking up scenarios which they fondly imagine might actually allow people to understand where they disagree with others on moral questions, communicate their position with clarity, and build a way of doing philosophy and understanding morality that matters to the broader culture, and makes a difference. Most of these attempts will be miserable failures, but some of them should be significantly better than anything we currently have. There’s always the chance that this might cause “more and bloodier wars than anything else in the history of creation”, but I think it’s worth the risk. 

There are also other areas of language where we could do better. It’s not ideal, for example, that in motte and bailey castles the motte was generally outside and often at some distance from the bailey. But that’s another post. 


New Comment
6 comments, sorted by Click to highlight new comments since: Today at 10:18 AM

The philosophic thought experiments we have are made by a class of philosophers who have high intelligence and who are incentivised to create good thought experiments. Creating a new thought experiments that gets used by other philosophers means that those will likely cite your papers and is thus valuable for any philosopher.

To the extend that the thought experiments we have are bad by your standards it's likely that the philosophic community has different standards of what makes a thought experiment worth talking about then you.

If you think there's a problem worth fixing it would be worth to understand where your standards differ from that of the philosophic community and whether you have arguments why your standards would be better.

So a "good" thought experiment in the philosophic community is one that generates lots of citations and academic prestige. This seems to have heavy network/first-mover/lock-in effects. Also, an emphasis on drama and counter-intuitiveness and communicating with academics. There's surely some selection pressure for thought experiments that "allow difficult and contentious problems to be understood, remembered, and referred to", for example, but it seems to be pretty weak compared to the other factors.

I share this sentiment to an extent. I remember, for instance, noticing that I actually never really understood the Prisoner's Dilemma until reading this post. Afterwards the original formulation of the dilemma just sounded straight up lazy. I'm supposed to pretend that money is equivalent to utilons and disregard human relations? C'mon, do you see any movie directors telling the audience to just suspend their disbelief instead of, y'know, actually making an effort to tell the story in a more believable manner?

One thing I might be able to contribute: Estimating one's own IQ can be a difficult task as one is often involuntarily biased by social conditioning to virtue signal either confidence or humility. A thought experiment that I found helpful as an intuition pump is imagining someone with a gadget that can read out your true IQ holding a gun to your head, telling you to predict your own IQ and threatening to kill you if you get it wrong by 10 points or more.

I think introducing an agent that is deliberately setting up a certain scenario helps in a lot of cases. For the trolley problem, you could imagine some sort of evil genius orchestrating the dilemma like a SAW movie or something.

For many audiences, presuming an adversarial agent changes the situation greatly - it's no longer about the competition between different intuitions of what do do, it's now about how to thwart the adversary.

This does occur in some of the Newcomb discussions - "Fuck Omega, it's cheating!" is probably not that common, but it's not unheard-of either.

I'm not sure why you think insufficient results imply insufficient effort, nor why you think this group is likely to do much better, especially without examples or criteria for a good thought experiment.

From my perspective, a whole lot of human effort has gone into philosophical thought experiments, and that these are the best we've found indicates that they're probably close to the best that are findable by "trying harder".  this doesn't make them GOOD; it may just indicate that good ones are impossible, or unfindable by current mechanisms.  

Different search mechanisms might come up with something, or might just provide evidence that there aren't any great ones.

I also disagree that these are all that bad.  IMO, the purpose of such experiments are to simplify the scenario to provide contrast between competing desiderata, in order to make them explicit.  The trolley problem does show the contrast between the intuition that non-action is preferable to harmful-action and the intuition that fewer deaths is better than more.  The fact that it's not a real-world choice is not that harmful in this purpose, and introducing all the complexities of reality really obscures the main distinction.

If you think you can do better, please comment with suggestions.

Sure. My target will be the Smoking Lesion hypothetical. The hypothetical supposes:

  • There is some behavior B, which is pleasant.
  • There is some disease D, which is unpleasant.
  • There is some root cause C that tends to cause B & D.
  • B does not cause D and D does not cause B.
  • There are no observable intermediaries between C and B.
  • Question: should you engage in behavior B?

In the traditional Smoking Lesion, we pick B=smoking, D=cancer, and C=genetic lesion. This has four main downsides for me:

  • We must imagine that smoking does not cause cancer. This is quite a sizable change to the universe, and would have hard-to-predict consequences in terms of the prevalence and perception of smoking, for example. There's quite a lot of mental effort required to keep this in mind while navigating the hypothetical.
  • We must imagine that smoking is pleasant, overall. Generally people say that smoking is an acquired taste and not initially pleasant, though there is some variation. There is also the unpleasantness of being addicted, of spending money, of non-cancer health effects, etc. So this is another sizable change to the universe.
  • Smoking is a somewhat taboo topic. Eg, smoking being pleasant, for anyone, can be taboo.
  • We must imagine that the genetic lesion directly causes smoking, with no observable intermediaries. So the lesion does not cause people to like the smell, or to be vulnerable to addiction, or to form habits faster, or to be convinced by evidential decision theory, or to have problems with weight that smoking helps with, or to have higher openness to experience, or anything like that.

If we imagine a "smoking" that is cheap, pleasant, non-addictive, not habit-forming, healthy, and doesn't reduce weight, where everyone likes it the same amount, plus all the changes downstream of that, it's unclear whether we should still call it "smoking".

My proposed fix: replace the Smoking Lesion with the Dancing Lesion. Dancing is pleasant, is not taboo or controversial, and is not carcinogenic. It doesn't completely fix the problem with observable intermediaries, but I suspect that is unfixable and a flaw in the hypothetical. Still, dancing is at least no worse than smoking in this respect. There are taboos in various cultures about certain types of dancing, but I'm not aware of any universal taboos and I think anyone can imagine a form of dance that is accepted in their culture.

New to LessWrong?