I've heard of this kind of thing but never thought about it too much because of how contrived it seems. Reading this what strikes my is the underlying assumption about the decision maker's confidence in the outcome of their actions. This is where people really end up deluding themselves, and what ends up justifying all kinds of horrible stuff.
Shifting the conversation from morality to "we're sure this will work, eg we're sure they have wmds" ducks the moral question and just gets into a silly discussion about evidence.
Torture and capital punishment are both good examples. If your only "ethics" is whether it's ok to torture someone to save millions of lives, you basically end up being able to find justification for it in any situation.
Outcomes are intentionally made clear because the morality in this case can rule out options when uncertainty is involved.
In you example of capital punishment, If you say it is wrong, even with 100% certainty of guilt, then nobody has to discuss the situation where uncertainty is involved.
If you say it is moral with 100% confidence, then you can ask the separate question involving uncertainty.
The arrogant tone of the website is deeply frustrating. Especially since their examples are full of false premises. For example, in the nuke question they estimate torture’s efficacy at 70%. That’s way, way higher than the actual efficacy.
There are interesting questions here presented very poorly.
The real world effectiveness of torture actually doesn't matter for evaluating the given scenario. You should be able and willing to think about what you would do with the given constraints. Even if you think it's only 0.1% effective instead of 75%, you still have given as facts that the man confessed to planting the bomb and that a million people will definitely die if the bomb isn't found in less than 24 hours.
>You should be able and willing to think about what you would do with the given constraints.
Should I? That's basically claiming that I should have an answer for when I'm half way down any given slippery slope.
I would instead point out that (as a general matter and specifically in this case as the constraints bear no relation to reality) a slippery slope is a logical fallacy and I should no more be expected to have a pre-set contingency for every possible trolley problem than I should for waking up and finding out that 1+1=Banana.
If you care about the consistency of your framework, then yes, I believe so.
> I would instead point out that (as a general matter and specifically in this case as the constraints bear no relation to reality) a slippery slope is a logical fallacy
The presented scenarios are not any kind of slippery slope fallacy, so your statement is a non sequitur.
I think your point is right, but I admit I struggled here. It's trying to constrain the scenario, but it's both too unrealistic and not constrained enough for me to manage. Here's what I was thinking.
First off, torture as far as I understand has way worse outcomes than 75% success, and I don't think information like a full blown confession is very often had. But OK, 75% chance they find out where it is with a definite admission of guilt. No chance they get bad information during torture. Howw long do they do the torture to get that info? 1 minute? 23 hours and 59 minutes? Is that really enough time to dispatch the team? Let's say they have enough time though. The scenario says "if the bomb is found, there's a chance it can be disarmed." What kind of chance? 99%? .00001%? If they fail to disarm it, did you send a team of people off to die with the other 1 million? Or do they evacuate to survive? Is it worth potentially sacrificing more people on the chance you can help then?
It was just too much for me I guess. But truly I agree with you. Maybe over a longer period of time I can think more clearly about it with more constraints to help me.
> Howw long do they do the torture to get that info? 1 minute? 23 hours and 59 minutes? Is that really enough time to dispatch the team?
You won't know until you try.
> If they fail to disarm it, did you send a team of people off to die with the other 1 million?
This sounds like you're suggesting that nobody should ever try to disarm a bomb. I think it's safe to assume that the people who disarm bombs have chosen to do so and weren't forced into it.
I've always wondered about that. It's generally reported that torture is largely ineffective because people will say anything to make it stop. However, that also includes the people who are being tortured and genuinely do not know the answer.
Whereas the fat terrorist can be tortured and then his answers checked with the NEST team and can be tortured again until the right answer is found, or until the device explodes. He knows this, so once it's clear that they really will torture him and check his answers, he can indeed end it with the right answer.
The problem is that there's no way to tell the difference between someone saying anything to make the torture stop and someone who is simply lying and actually knows the answer. You need additional information to verify information, for example, A who confesses under torture that B, C, and D are also terrorists has very little or no way to verify without also torturing B, C, and D to confess to terrorism. But there's no way to verify that any of those confessions are true either unless they point to something, the easiest of which is pointing to other supposed terrorists...
Yes, exactly. But in this case (hidden nuke), you send the nuclear response team to the location and if they say it's there, you have your answer. If it's not, you start on the next finger and keep going until you find the bomb or the bomb explodes (or the terrorist dies).
Of course you're still vulnerable to fakeouts with a decoy device, booby traps or multiple devices, which a sufficiently paranoid fat terrorist would use if he thought he could be caught and tortured, but that's not the scenario given.
It's really easy to see the answer to the supposed contradiction here: the question about torture is asked first, with no additional information. The question about torturing the fat man presupposes that it increases the chance of finding the bomb by a large percentage. Torture is commonly cited as being far more effective than it is.
This ultimately comes down to how people deal with being presented hypotheticals with false information. Many people will ignore commonly repeated lies, even in hypotheticals. Some people follow the hypothetical. Some people ignore the lie. Some people come over here to HN to point out the lie, and point out that the "philosophers" who posit these kinds of questions should probably be the ones tortured if you're going to torture anybody.
Yeah I'm not a philosopher, but I am an artificial intelligence researcher, and I feel like these questions have some aspect of the old rules based or expert system thinking in them, where we try to codify human thinking and it doesn't work because there are always edge cases and new information.
These, and questions like them (typical of many surveys) assume that its possible to give an answer to a complex question based on limited (or closed) sets of facts. It's not in general, if it was we'd have human-like rules engines
I think this is the opposite of surveys. It gives you a set of unquestionable facts to eliminate edge cases. It is not claiming that it can be generalized to the edge cases or new information.
If anything, I think the failure that people make when taking it is an inability to distinguish between absolute questions and general question.
Without this skill, you end up with people saying "torture is unacceptable under any circumstance", then later saying "I didn't mean to include the circumstances where the stakes are high or it will work"
If you have a situation where you choose A and 1 dies, choose B and 4 die, do nothing and all 5 die, then any choice is fine, you are choosing to save people, there is no choice that would be worse for anyone than doing nothing. I’m not convinced you have an obligation to do something to save someone (a different conundrum), but there’s certainly nothing wrong with saving.
Imagine you could disarm a bomb, but only one bomb, and two were going to go off. You could try to run a maximal algorithm to decide which group to save (the 5 80 year olds or the 30 kids? The 3 kids or the 30 80 year olds, the 15 people or your child), but any choice you make is fine by me.
However if you deliberately choose to act and kill someone who wouldn’t die if you didn’t act, that’s a different situation.
Where it gets interesting is when it brings in probabilities, two identical groups, both will die in 10 minutes. You can make a choice before t-5 to save one group, but the other group will then die, however there is a chance (90%, 9%, 9 in a million) that if you do nothing someone else will be able to save both groups.
The other one, they will die after 10 minutes, you can save one group, but it will kill the other group 5 minutes earlier than they would die without your intervention. What if it’s not 5 minutes but 5 seconds. Or 5 milliseconds. How about if it’s not 10 minutes to doom, but 10 days, or 10 years, or 100 years.
Interesting test, I recommend people take it, but for purposes of discussion, I have put the first 4 questions here, along with my thoughts
>Question 1: Torture, as a matter of principle, is always morally wrong.
False. People have a right to be free of torture, but this right can be invalidated based on their actions
>Question 2: The morality of an action is determined by whether, compared to the other available options, it maximizes the sum total of happiness of all the people affected by it.
False. This is utilitarian thinking. I hold that there are certain rights an individual has that can not be violated unless they first have taken action to void them, see Q1.
Question 3: It is always, and everywhere, wrong to cause another person's death - assuming they wish to stay alive - if this outcome is avoidable.
False. See question 1. The word avoidable does a lot of heaving lifting here and is ambiguous. For example, how much cost and risk there is avoidance.
Question 4: If you can save the lives of innocent people without reducing the sum total of human happiness, and without putting your own life at risk, you are morally obliged to do so.
False. Again, this is utilitarianism. Sum human happiness is not the only or driving moral imperative.
This set of values produces a 100% compatibility with the following trolly questions. What was the experience of others? I'm guessing 100% utilitarian would do the same
T/F/F/F 100% consistent - I save no one and even in the last question refuse to give a pass torture, because I think it's wrong. In the proposed scenario, it will have dire consequences, but such are my beliefs.
"...most people agree with you that the fat man should not be thrown off the bridge. However, this view seems to be inconsistent with your earlier claim that there is a moral requirement to maximise the happiness of the greatest number of people."
I don't agree with that. If people could be "morally" randomly killed to save two and more people, the overall happiness of the population would decrease. They would be scared of going on bridges to stop trolleys or hospitals so that they don't become accidental organ donors.
I got 100% consistency. The site thinks it's weird that I would kill the fat man without feeling morally obligated to. What a weird conflation to make.
I think GP is reacting to the word "obligation" (I certainly did.)
I'd claim there are actions that may be "morally justifiable" yet not raise to a level that creates a "moral obligation." The word "obligation" implies (to me at least) some sort of guilt or culpability if you don't undertake the action.
In fact my instinct is that most morally justifiable actions aren't weighty enough to be obligations. Consider any "morally neutral" action (for example, eating an apple instead of an orange for a snack) - I would label such an action "morally justifiable" (since applying the rules of morality does not argue against it) but it's certainly not a "moral obligation" (even ignoring the fact that the alternative is also morally justifiable!)
You could perhaps argue there is no such thing as a "truly neutral" action to an omniscient observer, who can measure and predict the impact on total human happiness of any action with certainty. Such an individual would indeed have a much more restrictive set of "morally justifiable" actions available to them - fortunately we mortals aren't burdened with that level of insight or certainty!
Interesting, I can see something being morally neutral or unknowable, but still don't understand the distinction between moral and moral obligation. If you know an action is moral but you are not doing it, you're acting is implicitly not moral. The only remaining conclusion is admission that do not act in a moral way.
If I understand correctly, your distinction between moral and morally justifiable is just a matter of uncertainty of facts, and if more facts were known, it would become moral or not? Is that correct?
I _desire_ to save more lives; I don't feel obligated to.
Consider the difference between doing something out of love and doing something to avoid divine punishment (hell) which is the basis of morality for billions of often very shitty people.
Interesting, I consider morality a matter completely separate from religion and the basis of ethics. Is there a name for simply following one's desires with no basis in ethics? I would say hedonism, but I think even that ascribes a ethical imperative to seek pleasure
> I would say hedonism, but I think even that ascribes a ethical imperative to seek pleasure
Describing hedonism as an imperative is wrong IMO. Hedonism is a drive not a directive. Personally I think that it's the most accurate truth in the world that people do things only because they want to and not for any other reason[1]. I help others because doing things that I think helps others brings me pleasure, not actually because of anything else.
1 - Actions of course don't exist in a vacuum and often doing one thing precludes doing another. So certainly one might work a shitty demeaning job because they want to not starve to death on the street more than they want to not work that job. But a person doesn't give money to a homeless man if it doesn't make them feel good to do so.
This stood out to me too. You don't have to be morally obligated to do something to actually do it.
Though I don't think that in a crisis, I actually would push the fat man, because I'll also be aware that I'll probably be imprisoned for murder and I don't know if the gratitude of the 5 people and warm fuzzies makes up for that. I'm not sure if that makes me a psycho, or if pushing the fat man does.
I think we're meant to assume that there aren't any witnesses. Fear of repercussions obviously frequently stops people from doing things they feel would be reasonable.
Of course, what this line of question implies is that we are all morally obligated to get fat so that we can sacrifice ourselves with fully informed consent instead of committing murder.
Don't you understand? I have to eat these fries; a net +4 lives may depend on it.
Test: Would you torture someone to stop a nuke from exploding in a city?
Me: No.
Test: That's not necessarily inconsistent, but it is very strange. Are you sure you meant your first answer?
sigh This test was clearly written by a utilitarian who is not capable of seeing the world another way, who believes if you make the numbers big enough, deep down I must be a utilitarian too.
I am not a utilitarian.
I mean it!
I am not willing to do evil that good may result. I am not willing to do it if the good is small. I am not willing to do it if the good is big. I am not willing to do it!
I say "of course" torture is not always morally unacceptable because I can think of so many circumstances in which it's obviously acceptable. EMTs necessarily torture people all the time in the process of treating them. All the personal trainers I've had - at least the good ones - have tortured me, and frankly that's what I pay them for. Navy SEALs torture prospective members to find and create those who can handle the rigors of war. And on a very small scale, parents routinely and necessarily make their kids suffer in order to make them better people. While what parents do stretches the definition of the word, I am convinced God does things on a grand scale that easily qualify under a similar principle.
The crucial factor isn't net human happiness - I don't believe in sacrificing one person for another in a general context. It isn't even consent. It's love. I believe you can hurt someone, even profoundly, not always as a sufficient condition but as an often necessary one, if you believe it's best for them.
Utilitarians think morality lies in outcomes - the whats. But whatever I am is completely orthogonal - I believe it lies in the story surrounding the outcome, the hows and the whys.
> Utilitarians think morality lies in outcomes - the whats. But whatever I am is completely orthogonal - I believe it lies in the story surrounding the outcome, the hows and the whys.
I think your take is off the mark because I got the same style of "how strange" commentary from different inputs. I said that I'd torture the guy while feeling no moral obligation to do so. The site was equally perplexed about my choices despite me being 100% consistent in the end.
IMO the biggest issue is that the site appears to believe that people only act from obligation. As though nobody ever does anything without feeling obligated. It's an extremely narrow-minded view of action that probably couldn't be farther from the truth if it tried.
Actually... I might start using that...
"Hey, do you want to get pizza?" "Sorry, I feel morally obligated to eat Indian food tonight."
I think you just have a different definition of torture. Simple as that. I think they were asking about nonconsentual infliction of pain and suffering to achieve a goal orthogonal to the recipients.
A: Not enough information, even about easily observable characteristics which may make it a morally hazy conundrum: how old are they, are any of them friends or family? I'm forced to treat their lives as equivalent and so I divert the train
2. Should I push a fat man to save 5 people?
A: No, because I have no guarantee or even belief that pushing him in front of the train is enough to stop it.
3. Should I push a fat saboteur to save 5 people?
A: He sabotaged the train with the express intent to kill 5 people, so risking his life to attempt to mitigate the damage of his action seems like a fitting punishment for the crime. Over he goes.
4. Would I torture him to save a million people?
Oh look, pushing him in front of the train was indeed not a guaranteed success. I wouldn't torture him because I have no belief that doing so would yield timely or accurate information, and because I believe that living in a world where torture is tolerated for any reason truly is a net reduction to sum total happiness, even if preserving that comes with the risk of a million lives.
Yes exactly. These scenarios assume some perfect world where, in the spur of the moment, you have guarantees of your actions. In the real world we can't guarantee a fat man will stop this train. In the real world, does you causing torture to save 5 people cause downstream future effects that actually make the world worse?
Say I tortured someone to save 5 people. Sure, 5 people would be saved and the argument this website makes says that where we stop. But what about the future consequences of you causing torture? The person you tortured comes back later and kills others because of the psychological harm you caused. Or the media makes you a hero and people start to think "Hmm torture, not that bad of a thing.".
If we are maximizing happiness for the largest group of people, we need to think about it over the largest span of time. And that's when things get really uncertain and murky.
The runaway trolley experiment is an excellent example of how words have different interpretations, and how those then change behavior while maintaining a consistent moral view. Change the words and people change behavior, while morality tend to stay consistent.
Take the first example. Is it the button pressing train operator that is causing the deaths, or is it the fault of the malfunctioning train and the operator is simply mitigating the destruction. It would be similar to how a airplane pilot might decide to crash a malfunctioning plane in a low population area rather than high population area. I would not say that the pilot "caused" the crash damage, through the experiment might frame it as the pilot deciding to either kill 1 or 5.
44 comments
[ 2.6 ms ] story [ 102 ms ] threadShifting the conversation from morality to "we're sure this will work, eg we're sure they have wmds" ducks the moral question and just gets into a silly discussion about evidence.
Torture and capital punishment are both good examples. If your only "ethics" is whether it's ok to torture someone to save millions of lives, you basically end up being able to find justification for it in any situation.
In you example of capital punishment, If you say it is wrong, even with 100% certainty of guilt, then nobody has to discuss the situation where uncertainty is involved.
If you say it is moral with 100% confidence, then you can ask the separate question involving uncertainty.
There are interesting questions here presented very poorly.
Should I? That's basically claiming that I should have an answer for when I'm half way down any given slippery slope.
I would instead point out that (as a general matter and specifically in this case as the constraints bear no relation to reality) a slippery slope is a logical fallacy and I should no more be expected to have a pre-set contingency for every possible trolley problem than I should for waking up and finding out that 1+1=Banana.
If you care about the consistency of your framework, then yes, I believe so.
> I would instead point out that (as a general matter and specifically in this case as the constraints bear no relation to reality) a slippery slope is a logical fallacy
The presented scenarios are not any kind of slippery slope fallacy, so your statement is a non sequitur.
First off, torture as far as I understand has way worse outcomes than 75% success, and I don't think information like a full blown confession is very often had. But OK, 75% chance they find out where it is with a definite admission of guilt. No chance they get bad information during torture. Howw long do they do the torture to get that info? 1 minute? 23 hours and 59 minutes? Is that really enough time to dispatch the team? Let's say they have enough time though. The scenario says "if the bomb is found, there's a chance it can be disarmed." What kind of chance? 99%? .00001%? If they fail to disarm it, did you send a team of people off to die with the other 1 million? Or do they evacuate to survive? Is it worth potentially sacrificing more people on the chance you can help then?
It was just too much for me I guess. But truly I agree with you. Maybe over a longer period of time I can think more clearly about it with more constraints to help me.
You won't know until you try.
> If they fail to disarm it, did you send a team of people off to die with the other 1 million?
This sounds like you're suggesting that nobody should ever try to disarm a bomb. I think it's safe to assume that the people who disarm bombs have chosen to do so and weren't forced into it.
The site seems to think that morality is a set of unparameterized rules blindly followed. That is frustrating.
Whereas the fat terrorist can be tortured and then his answers checked with the NEST team and can be tortured again until the right answer is found, or until the device explodes. He knows this, so once it's clear that they really will torture him and check his answers, he can indeed end it with the right answer.
Of course you're still vulnerable to fakeouts with a decoy device, booby traps or multiple devices, which a sufficiently paranoid fat terrorist would use if he thought he could be caught and tortured, but that's not the scenario given.
This ultimately comes down to how people deal with being presented hypotheticals with false information. Many people will ignore commonly repeated lies, even in hypotheticals. Some people follow the hypothetical. Some people ignore the lie. Some people come over here to HN to point out the lie, and point out that the "philosophers" who posit these kinds of questions should probably be the ones tortured if you're going to torture anybody.
These, and questions like them (typical of many surveys) assume that its possible to give an answer to a complex question based on limited (or closed) sets of facts. It's not in general, if it was we'd have human-like rules engines
If anything, I think the failure that people make when taking it is an inability to distinguish between absolute questions and general question.
Without this skill, you end up with people saying "torture is unacceptable under any circumstance", then later saying "I didn't mean to include the circumstances where the stakes are high or it will work"
Imagine you could disarm a bomb, but only one bomb, and two were going to go off. You could try to run a maximal algorithm to decide which group to save (the 5 80 year olds or the 30 kids? The 3 kids or the 30 80 year olds, the 15 people or your child), but any choice you make is fine by me.
However if you deliberately choose to act and kill someone who wouldn’t die if you didn’t act, that’s a different situation.
Where it gets interesting is when it brings in probabilities, two identical groups, both will die in 10 minutes. You can make a choice before t-5 to save one group, but the other group will then die, however there is a chance (90%, 9%, 9 in a million) that if you do nothing someone else will be able to save both groups.
The other one, they will die after 10 minutes, you can save one group, but it will kill the other group 5 minutes earlier than they would die without your intervention. What if it’s not 5 minutes but 5 seconds. Or 5 milliseconds. How about if it’s not 10 minutes to doom, but 10 days, or 10 years, or 100 years.
>Question 1: Torture, as a matter of principle, is always morally wrong.
False. People have a right to be free of torture, but this right can be invalidated based on their actions
>Question 2: The morality of an action is determined by whether, compared to the other available options, it maximizes the sum total of happiness of all the people affected by it.
False. This is utilitarian thinking. I hold that there are certain rights an individual has that can not be violated unless they first have taken action to void them, see Q1.
Question 3: It is always, and everywhere, wrong to cause another person's death - assuming they wish to stay alive - if this outcome is avoidable.
False. See question 1. The word avoidable does a lot of heaving lifting here and is ambiguous. For example, how much cost and risk there is avoidance.
Question 4: If you can save the lives of innocent people without reducing the sum total of human happiness, and without putting your own life at risk, you are morally obliged to do so.
False. Again, this is utilitarianism. Sum human happiness is not the only or driving moral imperative.
This set of values produces a 100% compatibility with the following trolly questions. What was the experience of others? I'm guessing 100% utilitarian would do the same
I don't agree with that. If people could be "morally" randomly killed to save two and more people, the overall happiness of the population would decrease. They would be scared of going on bridges to stop trolleys or hospitals so that they don't become accidental organ donors.
What drives your decision if not morality?
I'd claim there are actions that may be "morally justifiable" yet not raise to a level that creates a "moral obligation." The word "obligation" implies (to me at least) some sort of guilt or culpability if you don't undertake the action.
In fact my instinct is that most morally justifiable actions aren't weighty enough to be obligations. Consider any "morally neutral" action (for example, eating an apple instead of an orange for a snack) - I would label such an action "morally justifiable" (since applying the rules of morality does not argue against it) but it's certainly not a "moral obligation" (even ignoring the fact that the alternative is also morally justifiable!)
You could perhaps argue there is no such thing as a "truly neutral" action to an omniscient observer, who can measure and predict the impact on total human happiness of any action with certainty. Such an individual would indeed have a much more restrictive set of "morally justifiable" actions available to them - fortunately we mortals aren't burdened with that level of insight or certainty!
If I understand correctly, your distinction between moral and morally justifiable is just a matter of uncertainty of facts, and if more facts were known, it would become moral or not? Is that correct?
Describing hedonism as an imperative is wrong IMO. Hedonism is a drive not a directive. Personally I think that it's the most accurate truth in the world that people do things only because they want to and not for any other reason[1]. I help others because doing things that I think helps others brings me pleasure, not actually because of anything else.
1 - Actions of course don't exist in a vacuum and often doing one thing precludes doing another. So certainly one might work a shitty demeaning job because they want to not starve to death on the street more than they want to not work that job. But a person doesn't give money to a homeless man if it doesn't make them feel good to do so.
Though I don't think that in a crisis, I actually would push the fat man, because I'll also be aware that I'll probably be imprisoned for murder and I don't know if the gratitude of the 5 people and warm fuzzies makes up for that. I'm not sure if that makes me a psycho, or if pushing the fat man does.
Don't you understand? I have to eat these fries; a net +4 lives may depend on it.
Me: Of course not.
Test: Would you torture someone to stop a nuke from exploding in a city?
Me: No.
Test: That's not necessarily inconsistent, but it is very strange. Are you sure you meant your first answer?
sigh This test was clearly written by a utilitarian who is not capable of seeing the world another way, who believes if you make the numbers big enough, deep down I must be a utilitarian too.
I am not a utilitarian.
I mean it!
I am not willing to do evil that good may result. I am not willing to do it if the good is small. I am not willing to do it if the good is big. I am not willing to do it!
I say "of course" torture is not always morally unacceptable because I can think of so many circumstances in which it's obviously acceptable. EMTs necessarily torture people all the time in the process of treating them. All the personal trainers I've had - at least the good ones - have tortured me, and frankly that's what I pay them for. Navy SEALs torture prospective members to find and create those who can handle the rigors of war. And on a very small scale, parents routinely and necessarily make their kids suffer in order to make them better people. While what parents do stretches the definition of the word, I am convinced God does things on a grand scale that easily qualify under a similar principle.
The crucial factor isn't net human happiness - I don't believe in sacrificing one person for another in a general context. It isn't even consent. It's love. I believe you can hurt someone, even profoundly, not always as a sufficient condition but as an often necessary one, if you believe it's best for them.
Utilitarians think morality lies in outcomes - the whats. But whatever I am is completely orthogonal - I believe it lies in the story surrounding the outcome, the hows and the whys.
Reminds me of virtue ethics, and understanding the complex story of the hows and the whys is called practical wisdom. https://plato.stanford.edu/entries/aristotle-ethics/
IMO the biggest issue is that the site appears to believe that people only act from obligation. As though nobody ever does anything without feeling obligated. It's an extremely narrow-minded view of action that probably couldn't be farther from the truth if it tried.
Actually... I might start using that...
"Hey, do you want to get pizza?" "Sorry, I feel morally obligated to eat Indian food tonight."
A: Not enough information, even about easily observable characteristics which may make it a morally hazy conundrum: how old are they, are any of them friends or family? I'm forced to treat their lives as equivalent and so I divert the train
2. Should I push a fat man to save 5 people?
A: No, because I have no guarantee or even belief that pushing him in front of the train is enough to stop it.
3. Should I push a fat saboteur to save 5 people?
A: He sabotaged the train with the express intent to kill 5 people, so risking his life to attempt to mitigate the damage of his action seems like a fitting punishment for the crime. Over he goes.
4. Would I torture him to save a million people?
Oh look, pushing him in front of the train was indeed not a guaranteed success. I wouldn't torture him because I have no belief that doing so would yield timely or accurate information, and because I believe that living in a world where torture is tolerated for any reason truly is a net reduction to sum total happiness, even if preserving that comes with the risk of a million lives.
Score: 100% consistent.
Say I tortured someone to save 5 people. Sure, 5 people would be saved and the argument this website makes says that where we stop. But what about the future consequences of you causing torture? The person you tortured comes back later and kills others because of the psychological harm you caused. Or the media makes you a hero and people start to think "Hmm torture, not that bad of a thing.".
If we are maximizing happiness for the largest group of people, we need to think about it over the largest span of time. And that's when things get really uncertain and murky.
Take the first example. Is it the button pressing train operator that is causing the deaths, or is it the fault of the malfunctioning train and the operator is simply mitigating the destruction. It would be similar to how a airplane pilot might decide to crash a malfunctioning plane in a low population area rather than high population area. I would not say that the pilot "caused" the crash damage, through the experiment might frame it as the pilot deciding to either kill 1 or 5.