What's the difference between actually being in a situation and just thinking about the situation? In other words, what's the difference between a real-life ethical dilemma and a hypothetical ethical dilemma?

There are many differences.

  • Some differences are phenomenological. A real-life dilemma feels very different from a hypothetical dilemma.
  • Some differences are decision-theoretic. For example, the action-space of the real-life dilemma isn't isomorphic to the action-space of the hypothetical dilemma.
  • Some differences are cognitive. There are particular cognitive processes/algorithms that can be used to solve real-life dilemmas that can't be used to solve hypothetical dilemmas. And vice-versa.
  • Some differences are social. The social norms governing real-life dilemmas differ from those governing hypothetical dilemmas.
  • E.t.c, e.t.c.

In this post, I'll discuss two decision-theoretic differences, and I'll reach this conclusion:

For almost every real-life dilemma, there will not exist a hypothetical dilemma which has the same decision-theoretic properties as the real-life dilemma.

In subsequent posts, I might discuss other differences, and I expect to reach these conclusions:

  • Many people underestimate the difference between hypothetical dilemmas and real-life dilemmas.
  • Often people should replace thinking about hypothetical dilemmas with thinking about corresponding real-life dilemmas (especially for ethics, decision theory, and LessWrong).

What counts as a decision-theoretic difference?

A good hypothetical dilemma should capture all the decision-theoretic properties of the real-life dilemma. Here's what I mean:

Suppose Alice is faced with a real-life dilemma — she sees a child drowning in a pond, but saving the child would ruin her expensive shoes. So Alice has two options: save the child, or leave the child.

And suppose Bob is faced with the corresponding hypothetical dilemma: "you see a child drowning, but saving the child would ruin your expensive shoes." Bob has two options: hypothetically save the child, or hypothetically leave the child.

This is a faithful hypothetical dilemma if and only if Bob's decision is decision-theoretically equivalent to Alice's. In formal terms, there must exist an order-preserving isomorphism between Alice's options and Bob's options.

But this is rarely the case. I'll' explain two ways that Bob's decision differs from Alice's.

1. Bob's Secret Third Option

Firstly, there is a secret third option open to Bob which isn't open to Alice — namely, disengaging from the hypothetical. Here's the difference between Alice's and Bob's dilemmas:

  • Alice has two options: (A1) save the child, or (A2) leave the child.
  • Bob has three options: (B1) hypothetically save the child, (B2) hypothetically leave the child, or (B3) cease hypothesising anything.

Alice is faced with a real-life dilemma, so she must make a decision between A1 and A2. This "must" is a physical necessity — Alice must make a decision in the same way that she must have a mass.

Bob is faced with a hypothetical dilemma, but he doesn't actually need to make a hypothetical decision (i.e. pick B1 or B2). It's physically possible for Bob to pick B3, even though that would violate the rules of the game.

Of course, we might tell Bob "you must make a decision!". But this "must" isn't a physical necessity — instead, it's a ludic necessity (where "ludic" means "related to games"). In other words, Bob must make a hypothetical decision in the same way that a chess player must move the bishop along diagonal squares. It's physically possible, but not ludically possible, to move the bishop along orthogonal squares.

Is Bob's secret third option decision-theoretically relevant? I think so.

Both of Alice's options are harmful — in A1 her shoes are ruined and in A2 a child dies. She picks the least harmful option, which presumably will be A1. This is the option most people would pick without much hesitation.

But which option is best for Bob? B1 causes "psychic harm" by imagining one's shoes ruined. B2 causes "psychic harm" by imagining a child dying. And B3 causes social harm by violating the rules-of-the-game. Now, the psychic harm of B1 is pretty mild, so Bob will probably pick B1. But there are hypothetical ethical dilemmas where B3 is better than both B1 and B2. This is often the case when the hypothetical dilemma is between two really bad options, such as a trade-off between sacred values.

I think B3 is a very common move. What does B3 look like in practice?

  • Bob might question various presuppositions, even if he wouldn't question those presuppositions in more convenient contexts.
  • Bob might query the meaning of certain words, even if he wouldn't query the meaning of those words in more convenient contexts.
  • Bob might claim "we shouldn't engage in hypotheticals", even if he engages in hypotheticals in more convenient contexts.

If someone picks B3 but you want them to pick a proper hypothetical option, then you should try to make the dilemma real! This is especially useful if the "someone" is yourself. Does this mean actually drowning a child in a pond? Kind of. It means considering a situation in your own life which already has the appropriate form of the hypothetical dilemma. Instead of considering the question "should I hypothetically save a hypothetical child at the hypothetical cost of my hypothetical shoes?", Bob should instead consider the question "should I actually save an actual child at the actual cost of my actual shoes?"

(Of course, sometimes the hypothetical dilemma has no real-life equivalent, and the hypothetical can't be actualised without severe negative externalities. In these cases, a hypothetical might be the next best thing.)

2. Real life is particular, but the hypothetical is universal.

There's a second important decision-theoretic difference between real-life dilemmas and hypothetical dilemmas: the real-life dilemma is particular, but the hypothetical is (by design) universal.

Alice's dilemma is about a particular situation . She's watching a particular child drown (the -child) and she is wearing a particular pair of shoes (her -shoes). Whether she saves the child (A1) or leaves the child (A2), then that will be a particular event. There's only one configuration of atoms conforming to Alice's situation .

But Bob's dilemma isn't about a particular situation, but rather it's about a universal situation. This universal is a set of situations . Note that there are quintillions of configurations of atoms conforming to . This is because Bob's brain is finite, so his imagination can only pin down a small amount of information about the situation. This problem doesn't face Alice because her environment contains the remaining information required to pin down her exact situation.

How is that difference decision-theoretically relevant? Let me explain.

When Alice picks either A1 or A2, she is determining what transpires in her particular situation . But there are also other situations  which are very similar to  (maybe the atoms are configured slightly differently), and what transpires in  is heavily correlated with what transpires in . So Alice is also determining whether the children in these slightly different situations are saved or not.

Compare this with Bob. When Bob picks either B1 or B2, he is determining what transpires in the hypothetical situation . There are many situations  which are elements of the set . What transpires in the situations  is slightly correlated with Bob's hypothetical decision. And what transpires in situations  is also slightly correlated with Bob's hypothetical decision, where  is a situation which is very similar to 

So we can see a difference between Alice's and Bob's decisions: Alice will heavily influence a small number of situations. Bob will weakly influence a large number of situations.

Normally, when you add up all the influences, the real-life decision is more influential than the hypothetical decision. But in some cases, the hypothetical decision is more influential than the real-life decision.

Is there a trick to circumvent this problem?

Well, there are two kinds of problems with hypothetical dilemmas.

  1. The hypothetical is too influential.
  2. The hypothetical isn't influential enough.

Maybe the hypothetical is too influential — Bob's hypothetical decision will pathologically influence the decisions of his parallel selves. In this case, Bob should make the dilemma more particular. If there exists a similar dilemma in his actual life, he should focus on that. Or if there are no negative externalities to actualising the dilemma, then he should do that. Or if there's a historic example, then he should consider that one. This way the hypothetical situation  will cover fewer situations.

Maybe the hypothetical isn't influential enough — Bob's hypothetical decision will fail to influence the decisions of his parallel selves. In this case, Bob should make the dilemma more universal. This way the hypothetical situation  will cover more situations. Moreover, he should commit more strongly to his hypothetical decision, which increases the correlation between his hypothetical decision and the outcomes in those situations.

Conclusion and further work.

I think that these two decision-theoretical differences establish that hypothetical dilemmas and real-life dilemmas are quite different. You should be careful not to conflate them. I might explore some other differences in later posts.

New Comment
3 comments, sorted by Click to highlight new comments since:

There is another major difference. A hypothetical scenario consists of a combination of various elements and a simultaneous exclusion of other elements — especially, exclusion of all responses but the ones presented. The combination may not be possible in reality, and even if possible, the ludic constraint itself is unique to the hypothetical situation. Reality is not a game with rules someone dreamt up.

For example, Alice can slip off her shoes before entering the water. Indeed, if her shoes are impractical for wading through a muddy pond (as expensive women's shoes typically are) she will do that anyway.

The rules of the game that the questioner is proposing are a part of the questioner's agenda, and it is always prudent and legitimate to question them.

Thanks for this.  I wish you'd framed it more as "what can thought experiments show and what they can't".  There are two HUGE advantages that thought experiments have over real life:

  1. Thought experiments are cheap and scalable.  Nobody actually drowns or gets run over by a trolley, even if millions of people consider the thought experiment.  Thought experiments aren't even bound by physical feasibility.  The downside is that they don't prove anything, in the sense of demonstrating an actual outcome.  They remain in the domain of thought.
  2. Thought experiments are automatically abstract/general.  This is a strength AND a weakness - in truth, most interesting questions are a tension between competing values/desiderata, and the details matter a lot in determining exactly where the line is for this instance.  Thought experiments allow and require us to fill in unspecified details, and for the best ones, ask which details are important in evaluation.

Thought experiments are VERY GOOD at showing conflicts between intuitions.  They're just OK, but they're all we have, at extending intuitions into nonexistent situations so they can be examined and discussed.  

Thought experiments are NOT EVIDENCE of anything in the real world, but they ARE EVIDENCE of consistency and details of one's model of the world.

I broadly agree. Maybe my title is a bit clickbaity.

My claim is that, on the margin, hypothetical dilemmas are overrated (especially in ethics, decision theory, and lesswrong), and that most discussions about these hypothetical dilemmas should be replaced by discussions about real-life dilemmas.

e.g. rather than discussing the "child in a pond" problem, people should discuss the "should I donate £1000 to AMF?" problem. The benefit of focusing on the second problem is that you can actually execute your decision.

The second problem is also cheap and scalable!

Other examples: people waste time on "would I hypothetically hide a hypothetical Jew from hypothetical Nazis in my hypothetical attic?" rather than on "should I actually risk my life to actually save an actual person from actual persecution?"