The Trolley Problem Problem
aeon.co
aeon.co
When thinking about philosophy, a core part of philosophy is to reduce concepts, ideas, or beliefs into abstractions (or a "spirit" or "essence" of what their intention is), the thought experiment presents itself as a perfectly useful tool to help challenge those concepts.
Ethics is not about absolutes, but really about sussing out where gray areas exist in what some might believe are black and white situations. Additionally, most of the scenarios are presented as a thought experiment, because conducting a real experiment with those conditions would be wildly _un_ethical.
How are we supposed to determine the nuance of valuing human life if we were bound to doing actual experiments? Also, who in their right mind would conduct such as experiment?
Zimbardo faced enough flack testing the limits of obedience and authority in the Standford Prison Experiement.
I'm not a chicken so I don't really know.
Sure, you can do things that you consider evil just to see how you would really react to being presented with such an unpleasant binary choice, but if you chose to do it, was it in fact not evil to you, are you just pathological, or are you perhaps in toddlerhood, where you don't yet really have an intuition for your own relationship with causality, such as it is?
(I’m not asking this to refute your point, it’s well met, I just found the opportunity to poke at the imaginary ethical elephant too much to resist, and no I’m not proud of it-those were good hens)
Once you shake out enough thoughts from the thought tree, you can hopefully discover what they're all pointing to in order to find what might _actually_ be driving an ethical decision.
Essentially, people can’t make judgements with total information.
It seems like it was responsible for WWI, WWII, the Spanish flu, etc but would worse things happened instead? I doubt it, but maybe it indirectly prevented total nuclear war.
Which when you get down to it is true of everything, we need to act with imperfect information.
Your argument: we can’t possibly know everything, therefore assume nothing is correct. Except reality is a persistent illusion, so we work in that reality. Your fantasy world where there is a god that can see all futures is fantasy. Don’t confuse the imagination with reality.
Your philosophy is literally dependent on information that’s impossible to know which is a real limitation. In the real world we can estimate harms and benefits, but it’s the unknown factor which makes sacrificing someone ‘for the greater good’ so abhorrent.
If the rule is to minimize total probable harm, then you need to consider second order effects of choices. You further need to carefully consider/model repercussions rather than assuming each choice has obvious costs. Simply shooting infected people may reduce the risks of spreading a plague in the short term, but it’s got huge long term issues as people then try and hide their infection etc.
Reduces for who? Over what time frame? What's the scope, and why that scope and not a different scope?
I see no evidence of this claimed misunderstanding in the linked article. And there's nothing "subtle" about it saying thought experiments lack value because they're contrived, that is the central point its making.
Having read the same article, my takeaway from their characterization of thought experiments was that they misunderstood their real purpose. To pose thought experiments against what real professionals would do in particular scenarios really does undermine what the point of presenting thought experiments does. They’re used to challenge our notions about the choice we make and think about how we shape our moral code.
They ultimately are intended to make you think more about why you think a particular way about something.
So that is why it seems to be misunderstood.
This in itself can be see as problematic, as one can say that trying to reduce "concepts, ideas, or beliefs into abstractions" is futile, as those "things" are not black and white, they do not conform to Aristotelian logic of one thing/concept being either "true" of "false", there's a "continuum" (for lack of a better word) when trying to define each and everyone of those terms.
In other words one cannot be moral or not-moral (to go back to the trolley problem), there's a moral "continuum" which even varies with time (I know I'm less "moral" before I had the chance to drink my coffee early in the morning).
To use things, we usually reduce the single parts and use abstractions to deal with what's important (to us in that context). You don't deal with individual atoms, you look at clusters and clusters of clusters etc and understand and use them alltogether as a pen, just like you've similar pens before, even though they were 100% completely different if you look close enough.
After all, a thought experiment that only holds true given an impossible premise would have no rhetorical power if we weren't willing to extrapolate our conclusions beyond the premise, to the possible. Which we are - that's why the trolley problem is taught to people who aren't railway signalmen.
There's a risk there are arguments we think are convincing us because they have isolated a pure abstraction - but actually it's the false dilemma in the premise that convinced us, while the argument distracted us.
I view it like a black box where you can define inputs and observe the output and your goal is to understand what calculations are made inside. Thought experiments are essentially fuzzing it to understand what's happening. "What if it's a fat guy? What if it's a smoker? What if it's a politician? What if it's a homeless person? What if there are 10 people, what if there are a million?" and you're iterating over lots of options to get closer to understanding the formula.
On the other hand, we can imagine at some point, there may be some broad principles that engineers will have to consider when programming responses to an impending accident. Perhaps slamming on the brakes is essentially always the optimum solution from the point of driver safety if several people unexpectedly appear in front of the car even though braking will be insufficient to avoid hitting them albeit at a reduced speed. On the other hand, there's an unquantifiable but known risk (to the driver) to swerving in order to avoid the otherwise certain collision.
That a human would pick one or the other approach in a split second without really having time to think about it doesn't really inform what the programmed response of a vehicle that does have time to deliberately pick from several options should be.
Added: It is, of course, pretty much an academic exercise today in that cars can't reliably even be prevented from running into highway dividers. But, in a few decades or whenever, one can imagine being at a point where the technology is good enough that some framework is needed for making rules. Maximizing driver safety at all costs vs. trying to do the least harm overall may well not result in the same rules being programmed.
A simpler example: Killing a convicted murderer, is murder again! Oh, except the murderer has a whole different ethical context than say a child: the murderer is again a threat to more people; the murderer had agency and could have decided not to murder; the act has societal benefit instead of harm; and on and on.
And if you don't like my argument, then voila! Taking your ethical dilemma and recasting it as mine seems "not fair!" to you. Which ironically is my point and QED.
The answer to the trolley problem is that it doesn't matter if you pull the lever or not. You should figure out who or what created the situation and truly eliminate the problem at its source.
The Trolley Problem poses this question: if a trolley is about to run over and kill five people standing on the tracks, and you can shunt it to a different track where it would only kill one person, should you do it? Or what if you could save those five by throwing a person near you onto the tracks; should you do it?
Many people feel intuitively that it would be wrong to throw that person onto the tracks, and an argument attributes this to a supposed essential difference between killing someone and deciding to let that person die.
I disagree. I too believe it would be wrong to throw that person onto the tracks, in real life, but not in the hypothetical trolley problems. There is no ethically significant difference between killing a person and letting the person die, if (as supposed in the trolley problems) there is no doubt that the death will occur.
The reason, in real life, why killing someone is ethically different from letting someone die is that real life is full of surprises: the person might not really die. If you kill him, his death is pretty certain (though not totally; just recently a man was hanged in Iran and survived). If you merely don't take action to save him, he might survive anyway. He might jump off the track, for instance, or someone might pull him off. All sorts of things might happen. Likewise, throwing the one person onto the track might not succeed in saving the other five; how could you possibly be sure it would? You might find that you had done nothing but cause one additional death. Thus, in real life it is a good principle to avoid actively killing someone now, even if that might result in other deaths later.
The trolley problems invalidate the principle because of the unlikely certainty that they assume. Precisely for that reason, they are not a useful moral guide for most real situations. In general, difficult artificial moral conundrums are not very useful guides for real-life conduct. It's much more useful to think about real situations. In the free software movement I have often decided not to propose an answer to a general question until I had some real cases to think about.
For real driverless cars, the trolley problem never arises: the right thing to do, whenever there is a danger of collision, is to brake as fast as possible.
More generally, the goal is to make sure to avoid any trolley problem.
This actually isn't true in every scenario. What if there is a car following close behind? What if the car behind has more passengers than the car in front of you? What if swerving will kill pedestrians, but less than the passengers in the car you're about to hit?
There are a lot of variables, and it's really not as straightforward as you make it sound
> What if there is a car following close behind?
They may be able to stop fast enough. Whereas if you intentionally don't brake hard enough and strike the car in front of you, you (the computer) have traded a possibility for a certainty. You control your braking and following parameters, they control theirs. (EDIT: Yeah, if there's a lot of distance in front of you, then the car can make use of that. It doesn't need to screech to a halt in every case. But this isn't trolley-car territory, it's just the standard way adaptive cruise and emergency braking already works)
> What if the car behind has more passengers than the car in front of you?
Irrelevant. It is not the car's job to be counting the number of occupants in every other car on the highway.
> What if swerving will kill pedestrians, but less than the passengers in the car you're about to hit?
I want the self-driving car to not make that choice. The passengers in the car, however numerous, are much more likely to survive impact than unprotected pedestrians.
It really is straightforward. Don't burden self-driving cars with silly edge-cases. Apply the most effective mitigation known in the general sense, and see that everyone is happy because they can anticipate the car's choices a bit better.
It could be computed in the seconds, or minutes which lead up to the incident. A smart driver, human or AI, has a ready, already computed backup plan.
No, of course not. If I'm about to rear-end another car, even if it's a 1973 Ford Pinto with 4 occupants, I'm not going to swerve toward sidewalk pedestrians in lieu of slamming on the brakes. Even if I have all the information ahead of time on that car's faults and occupancy level. Nor would I want a computer to make that decision either.
You're not wrong. These are things a computer could do. But I'd say that they're not something we want them to be doing. We don't need programmers debating trolley-problem scenarios. They can focus on just reducing or eliminating impacts directly caused by the system they're building. In other words, just hit the brakes.
We're living the trolley problem problem in real life, right now, because human beings are bad at estimating risk. A lot of people are operating with a certainty that IF you are EXPOSED to the virus, you WILL contract it, and it WILL be terrible, and PROBABLY fatal. Therefore, no amount of safekeeping -- masks, distancing, wiping down delivered groceries, baking mail and newspapers -- can be too aggressive. For such people, diverting the "trolley" to make long-running, terrible financial hardships for a third of the population is an easy call, when they feel they are literally saving the life of everyone on forward track.
People who support aggressive measures don't view their actions as killing the bystander. 'Financial hardship' is far from letting people die (i.e. no equivalent good from their PoV) and they almost always support measures to alleviate economic damage.
People who oppose lockdown have differing arguments which don't map into a trolley either. An argument for lack of efficiency isn't relevant. Others' argument that governments' action is taking away liberty for security is closer, but no trolley either. For that, they'd have to weigh liberty of 'bystanders' again liberty of those that will get sick, and it's clear their concept of liberty does not accept the latter condition as a loss of liberty (no equivalent good either).
You might wish to drop my suggested equivalent good criteria. I'm not so sure. If diverting the trolley merely slightly inconveniences the bystander, are we still talking about a trolley problem?
But I think I still got the jist of your argument: That people supporting lockdown are applying trolley problem logic ("hurt the economical bystanders to save sick people").
What I wrote in response applies. That isn't trolley logic since the goods are noncomparable. I'll go further: financial harms in general aren't a good fit for a trolley problem, since one can always compensate later.
The point of discussions about scenarios whose parameters cannot be known is to find a useful way to elide the unknowns and come to conclusions that we can agree to and understand. Failing to see the forest for the trees is not a problem with the exercise; it's a problem with the participant.
But that's also literally what it's supposed to do. It's a device to poke around and figure out your positions, e.g. are you into consequentialism or do you prefer deontology, are there circumstances that can change your position etc.
I often find people (not you, people in general) rejecting thought experiments, because they are not comfortable with their intuitive decisions and they feel that it will be exposed when they're forced to apply it to a hypothetical situation that does not leave them an easy out ("this wouldn't happen, I don't walk near train tracks, ever").
It's also poor form to sling broad accusations about what "some" philosophers do without citing any examples.
It's hardly bizarre. Wittgenstein, one of the most famous philosophers of the 20th century, wrote that most philosophy was incorrect because philosophers failed to clearly define their terms.
"The right method of philosophy would be this. To say nothing except what can be said, i.e. the propositions of natural science, i.e. something that has nothing to do with philosophy: and then always, when someone else wished to say something metaphysical, to demonstrate to him that he had given no meaning to certain signs in his propositions. This method would be unsatisfying to the other — he would not have the feeling that we were teaching him philosophy — but it would be the only strictly correct method".
If the audience cannot enter into the problem, then its value as a tool is diminished.
Case studies are also fraught with peril, but the analysis of where the particulars end and the general principles begin seems the bulk of the exercise anyway.
In particular, we spend a lot of time thinking about what would have to be, in most formulations of the problem, a quickly-decided act (otherwise, why not simply untie the people?) and we have absolute certainty given us as to the consequences, vs. all the uncertainty in real life. The problem is used to get rid of those elements as distractions, but they're essential features of people's reasoning (and reasoning ability) so they're not so easily discarded.
I suspect that lots of people have actively made decisions while driving their own car that endangers their own life in order to potentially save the life of someone else.
Its often used as a sort of backdoor to squeeze in the argument that we should let X people die to autonomous cars because in the future they will definitely be safer, for sure. (some very shaky priors there)
These arguments are usually introduced with massive assumptions about the applicability and reliability of statistics gathered from cars that basically give up when even slightly challenged with anything remotely ambiguous or difficult (how many people live where there is perfect weather almost every day?), and then are applied to the entire gambit of circumstances human drivers deal with every day, by the millions.
When the machine/program seems perfect, then people get conditioned not to question it. There was a case where a UI issue caused the selection of an adjacent dosage to what was intended, and a nurse gave a patient something like 100 pills.
There were several checks in theory, but everybody was like "gee this seems weird, if the computer says so, it must be right".
Automation inherently multiplies actions, so the fear is that bugs don't just cause a single error, but many, until someone notices and has a chance to react.
The point isn't to change your mind about that. The point is to consider an example that has some similarities and ask why you reach an opposite conclusion. The insights gained from that help clarify the human moral instinct. The project is to understand how we think and feel, not to dictate what we should think or feel.
The violinist in the article is clearly trying to be a pure math- like counter example, but to me, misses the emotional aspect of pregnancy (at the least) - a fetus is not a random person, it is you (partly), and furthermore, many would say society (and the species) depend on having babies, whereas the society and the species do not depend on violin players.
Not saying I agree with these arguments only that the real world is substantially messier than pure math, and thus, pure math thinking may stumble when applied to real world problems.
Tl;dr: humans are not rational in our beliefs and the continued attempts on all disciplines to assume we are, are well, irrational.
The violinist experiment does have a lot of holes it it, but I think that one in particular almost turns it on its head, particularly because it circumvents the crux of the problem around consent (which I think is at the core of the problem).
It would seem you would have to change the thought experiment to have the violinist actually be related to you (say a sibling). Now, would a person feel _as_ upset that they had to allow their sibling to use their kidneys for nine months in order to stay alive? That really changes it.
But civilisation doesn’t depend on maximising births, and actually probably would benefit from more resource intensive to produce artists, musicians, dancers, etc - is the other side of that argument.
But the two situations are not equivalent. The hitman is almost certainly not in a position to save his intended victim so he is not in any meaningful sense letting him die.
But is that simply you justifying your actions?
The problem isn't absurd when posed to humans because, how about the scenario where you don't know either person and you need to make a decision in a matter of seconds. That's the point.
If my mother was on one of the tracks, I might justify my actions to pull the lever and have it veer into the other group, but that isn't necessarily a morally correct decision.
The author in the article says that the problem is absurd when posed to humans, then goes onto applying the same principles of the problem to AI and cars but does not say that the problem is also absurd with regards to AI, despite coming to the same conclusion.
This is where the problem is: life is way more complex and nuanced to distill down to such a simplistic scenario. Measuring morality is not an instantaneous action.
Ethics thought experiments are interesting and sometimes fun and you can't deny the value of getting people thinking about choices and behavior.
But it's always seemed to me that they, along with ethics in general, miss the point. I don't think absolute right and wrong, or even absolute "lesser of two evils" is a particularly useful goal.
It's the contextual framework that matters... Values, priorities, fears, desires, needs, the things that comprise a person's identity and worldview. Those things are going to win over ethics every time when real world decisions are being made.
IMO there's a lot more value in exploring those things, as opposed to ethics, if the ultimate goal is to impact the behavior of a individuals in a society.
And the point of the thought experiment is to uncover those things, and they make up that person's personal ethics. Maybe their ethics are "I'll always save my family, fuck everybody else", maybe their ethics are "I only save people who share my skin color", or maybe they are "I always maximize the damage, I hate people", using hypothetical situations allows you to figure that out without taking that person on daily walks and sacrificing lots of people on the train tracks to find out.
To add some formal language:
> I don't think absolute right and wrong, or even absolute "lesser of two evils" is a particularly useful goal.
This sounds like a metaethical argument - what underpins ethics is a very important question that a lot of people miss, but that argument is actually the base of ethical stances. The formal metaethical belief here is objectivism. Some other options are things like subjectivism, cultural relativism, error theory, and non-cognitivism.[1]
[1] A video alternative: https://www.youtube.com/watch?v=FOoffXFpAlU
> It's the contextual framework that matters... Values, priorities, fears, desires, needs, the things that comprise a person's identity and worldview. Those things are going to win over ethics every time when real world decisions are being made.
Ethics (once you get past metaethics) is almost always around a framework, and focuses exactly on everything you listed. Aristotle explicitly focused on values to built his ethical framework. Foucault talks a lot about fear and power. Most of consequentialism and utilitarianism focuses on needs and desires. Rawls and egalitarianism is an example of talking about priorities.
--------------------------------------------------
To me, it sounds like you have a gripe with the impracticality of philosophers talking about ethics but care quite a lot about ethics itself. If so, I'd be with you strongly on both accounts.
OK, what's blindly, blisteringly wrong here? First, the idea that a space ship small enough its fuel would be rationed out by the gram would have enough room in the crew compartment for someone to hide in is ludicrous. Second, launching without a checklist? Are you out of your tiny little mind? Third, allowing unknown people to bring unknown contaminants into a space ship? Having a "KEEP OUT" sign doesn't save the colonists from another plague, now, does it?
The moral of the story is that it's hard to keep your mind on the supposed lesson when the flaws jump up and down and yell at you.
Taking a different tack, by thinking too hard about the consequences of the thought experiment instead of the lead-up to it, there's the Jew in the attic. You know how it goes: A Jewish family is hidden in your attic or guest bedroom or someplace and the Nazis come knocking. Do you lie to save the Jews? "Of course", you say, and come up with a nice logical argument for why your ethical system demands you lie in this instance. All functional ethical systems can come up with such an argument with minimal fuss. However: The Nazis were not very nice, you know. If they thought a town was holding out on them, they'd initiate reprisals. They'd kill a whole town in a fit of fascist pique. Saving a half-dozen Jews could doom a few thousand innocents, likely including the original Jews. But that's out of scope for the thought experiment.
That's a new one to me. Any sources I can read up on that?
The Lidice massacre was the complete destruction of the village of Lidice, in the Protectorate of Bohemia and Moravia, now the Czech Republic, in June 1942 on orders from Adolf Hitler and Reichsführer-SS Heinrich Himmler.
In reprisal for the assassination of Reich Protector Reinhard Heydrich in the late spring of 1942, all 173 males from the village who were over 15 years of age were executed on 10 June 1942. A further 11 men from the village but who were not present at the time, were later arrested and executed soon afterwards, along with several others who were already under arrest.[2] The 184 women and 88 children were deported to concentration camps; a few children who were considered racially suitable and thus eligible for Germanisation were handed over to SS families and the rest were sent to the Chełmno extermination camp where they were gassed.
https://en.wikipedia.org/wiki/Lidice_massacre
Memorialised in Bohuslav Martinu's "Memorial to Lidice".
Anthropoid was the assassination of Reinhard Heydrich.
> More than 13,000 people were arrested, including Jan Kubiš' girlfriend Anna Malinová, who died in the Mauthausen-Gusen concentration camp. First Lieutenant Adolf Opálka's aunt Marie Opálková was executed in the Mauthausen camp on 24 October 1942; his father Viktor Jarolím was also killed.[52][53] According to one estimate, 5,000 people were murdered in the reprisals.[54]
https://en.wikipedia.org/wiki/Oradour-sur-Glane_massacre
> In February 1944, the SS Division Das Reich was stationed in the Southern French town of Valence-d'Agen,[1] north of Toulouse, waiting to be resupplied with new equipment and fresh troops. Following the Allied Normandy landings in June 1944, the division was ordered north to help stop the Allied advance. One of its units was the 4th SS Panzer Grenadier Regiment ("Der Führer"). Its staff included regimental commander SS-Standartenführer Sylvester Stadler, SS-Sturmbannführer Adolf Diekmann commanding the 1st Battalion and SS-Sturmbannführer Otto Weidinger, Stadler's designated successor who was with the regiment for familiarisation. Command passed to Weidinger on 14 June.[2]
> Early on the morning of 10 June 1944, Diekmann informed Weidinger that he had been approached by two members of the Milice, a collaborator paramilitary force of the Vichy Regime. They claimed that a Waffen-SS officer was being held prisoner by the Resistance in Oradour-sur-Vayres, a nearby village. The captured officer was claimed to be SS-Sturmbannführer Helmut Kämpfe, commander of the 2nd SS Panzer Reconnaissance Battalion (also part of the Das Reich division). He may have been captured by the Maquis du Limousin the day before.
> On 10 June, Diekmann's battalion sealed off Oradour-sur-Glane and ordered everyone within to assemble in the village square to have their identity papers examined. This included six non-residents who happened to be bicycling through the village when the SS unit arrived. The women and children were locked in the church, and the village was looted. The men were led to six barns and sheds, where machine guns were already in place.
[snip]
> Several days later, the survivors were allowed to bury the 642 dead inhabitants of Oradour-sur-Glane who had been killed in just a few hours. Adolf Diekmann said the atrocity was in retaliation for the partisan activity in nearby Tulle and the kidnapping of an SS commander, Helmut Kämpfe.
Edit: Haha, I clicked on a link to goodreads and the second comment compares it to "The Cold Equations".
The mass included in the initial acceleration has already been accelerated, so even if they space someone it's not making it to the destination.
Also; How would they eject someone into space without going off course or using more fuel?
It's not ridiculous at all to think that they'd fuel a supply ship with exactly the fuel it needed and not more, and that that supply ship might have some empty space in it, especially if it were very, very large.
Nitpick: it had excess fuel on board. Even beyond any institutional safety margins, the first stage landed on the drone ship. That portion of the mission required the stage carry fuel in excess of the requirements for its primary mission.
"By my weightings, freedom is more important than public health."
"My weightings say the opposite."
That's where we are today. Putting numbers on it, numbers that can't be determined except by our personal biases, doesn't help.
Here's a fun story about cost benefit analysis in the real world, a bureaucrat deciding which kinds of rape hurt enough to be worth preventing:
Criticism: https://gulcfac.typepad.com/georgetown_university_law/
Defense: https://www.theregreview.org/2013/09/13/13-sunstein-cost-ben...
Anyway, long live the Trolley Problem.